]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/namespace.c
nilfs2: fix potential memory overrun on inode
[mirror_ubuntu-bionic-kernel.git] / fs / namespace.c
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/idr.h>
19 #include <linux/init.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
27 #include "pnode.h"
28 #include "internal.h"
29
30 static unsigned int m_hash_mask __read_mostly;
31 static unsigned int m_hash_shift __read_mostly;
32 static unsigned int mp_hash_mask __read_mostly;
33 static unsigned int mp_hash_shift __read_mostly;
34
35 static __initdata unsigned long mhash_entries;
36 static int __init set_mhash_entries(char *str)
37 {
38 if (!str)
39 return 0;
40 mhash_entries = simple_strtoul(str, &str, 0);
41 return 1;
42 }
43 __setup("mhash_entries=", set_mhash_entries);
44
45 static __initdata unsigned long mphash_entries;
46 static int __init set_mphash_entries(char *str)
47 {
48 if (!str)
49 return 0;
50 mphash_entries = simple_strtoul(str, &str, 0);
51 return 1;
52 }
53 __setup("mphash_entries=", set_mphash_entries);
54
55 static u64 event;
56 static DEFINE_IDA(mnt_id_ida);
57 static DEFINE_IDA(mnt_group_ida);
58 static DEFINE_SPINLOCK(mnt_id_lock);
59 static int mnt_id_start = 0;
60 static int mnt_group_start = 1;
61
62 static struct hlist_head *mount_hashtable __read_mostly;
63 static struct hlist_head *mountpoint_hashtable __read_mostly;
64 static struct kmem_cache *mnt_cache __read_mostly;
65 static DECLARE_RWSEM(namespace_sem);
66
67 /* /sys/fs */
68 struct kobject *fs_kobj;
69 EXPORT_SYMBOL_GPL(fs_kobj);
70
71 /*
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
74 * up the tree.
75 *
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
78 */
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
80
81 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
82 {
83 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
85 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
87 }
88
89 static inline struct hlist_head *mp_hash(struct dentry *dentry)
90 {
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
94 }
95
96 /*
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
99 */
100 static int mnt_alloc_id(struct mount *mnt)
101 {
102 int res;
103
104 retry:
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
106 spin_lock(&mnt_id_lock);
107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
108 if (!res)
109 mnt_id_start = mnt->mnt_id + 1;
110 spin_unlock(&mnt_id_lock);
111 if (res == -EAGAIN)
112 goto retry;
113
114 return res;
115 }
116
117 static void mnt_free_id(struct mount *mnt)
118 {
119 int id = mnt->mnt_id;
120 spin_lock(&mnt_id_lock);
121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
123 mnt_id_start = id;
124 spin_unlock(&mnt_id_lock);
125 }
126
127 /*
128 * Allocate a new peer group ID
129 *
130 * mnt_group_ida is protected by namespace_sem
131 */
132 static int mnt_alloc_group_id(struct mount *mnt)
133 {
134 int res;
135
136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
137 return -ENOMEM;
138
139 res = ida_get_new_above(&mnt_group_ida,
140 mnt_group_start,
141 &mnt->mnt_group_id);
142 if (!res)
143 mnt_group_start = mnt->mnt_group_id + 1;
144
145 return res;
146 }
147
148 /*
149 * Release a peer group ID
150 */
151 void mnt_release_group_id(struct mount *mnt)
152 {
153 int id = mnt->mnt_group_id;
154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
157 mnt->mnt_group_id = 0;
158 }
159
160 /*
161 * vfsmount lock must be held for read
162 */
163 static inline void mnt_add_count(struct mount *mnt, int n)
164 {
165 #ifdef CONFIG_SMP
166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
167 #else
168 preempt_disable();
169 mnt->mnt_count += n;
170 preempt_enable();
171 #endif
172 }
173
174 /*
175 * vfsmount lock must be held for write
176 */
177 unsigned int mnt_get_count(struct mount *mnt)
178 {
179 #ifdef CONFIG_SMP
180 unsigned int count = 0;
181 int cpu;
182
183 for_each_possible_cpu(cpu) {
184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
185 }
186
187 return count;
188 #else
189 return mnt->mnt_count;
190 #endif
191 }
192
193 static void drop_mountpoint(struct fs_pin *p)
194 {
195 struct mount *m = container_of(p, struct mount, mnt_umount);
196 dput(m->mnt_ex_mountpoint);
197 pin_remove(p);
198 mntput(&m->mnt);
199 }
200
201 static struct mount *alloc_vfsmnt(const char *name)
202 {
203 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
204 if (mnt) {
205 int err;
206
207 err = mnt_alloc_id(mnt);
208 if (err)
209 goto out_free_cache;
210
211 if (name) {
212 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
213 if (!mnt->mnt_devname)
214 goto out_free_id;
215 }
216
217 #ifdef CONFIG_SMP
218 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
219 if (!mnt->mnt_pcp)
220 goto out_free_devname;
221
222 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
223 #else
224 mnt->mnt_count = 1;
225 mnt->mnt_writers = 0;
226 #endif
227
228 INIT_HLIST_NODE(&mnt->mnt_hash);
229 INIT_LIST_HEAD(&mnt->mnt_child);
230 INIT_LIST_HEAD(&mnt->mnt_mounts);
231 INIT_LIST_HEAD(&mnt->mnt_list);
232 INIT_LIST_HEAD(&mnt->mnt_expire);
233 INIT_LIST_HEAD(&mnt->mnt_share);
234 INIT_LIST_HEAD(&mnt->mnt_slave_list);
235 INIT_LIST_HEAD(&mnt->mnt_slave);
236 INIT_HLIST_NODE(&mnt->mnt_mp_list);
237 #ifdef CONFIG_FSNOTIFY
238 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
239 #endif
240 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
241 }
242 return mnt;
243
244 #ifdef CONFIG_SMP
245 out_free_devname:
246 kfree_const(mnt->mnt_devname);
247 #endif
248 out_free_id:
249 mnt_free_id(mnt);
250 out_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 */
274 int __mnt_is_readonly(struct vfsmount *mnt)
275 {
276 if (mnt->mnt_flags & MNT_READONLY)
277 return 1;
278 if (mnt->mnt_sb->s_flags & MS_RDONLY)
279 return 1;
280 return 0;
281 }
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283
284 static 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
293 static 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
302 static 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
318 static 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 */
343 int __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 (ACCESS_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 */
382 int 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 }
392 EXPORT_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 */
406 int 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 }
416 EXPORT_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 */
425 int __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 - 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 int mnt_want_write_file(struct file *file)
441 {
442 int ret;
443
444 sb_start_write(file->f_path.mnt->mnt_sb);
445 ret = __mnt_want_write_file(file);
446 if (ret)
447 sb_end_write(file->f_path.mnt->mnt_sb);
448 return ret;
449 }
450 EXPORT_SYMBOL_GPL(mnt_want_write_file);
451
452 /**
453 * __mnt_drop_write - give up write access to a mount
454 * @mnt: the mount on which to give up write access
455 *
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
458 * __mnt_want_write() call above.
459 */
460 void __mnt_drop_write(struct vfsmount *mnt)
461 {
462 preempt_disable();
463 mnt_dec_writers(real_mount(mnt));
464 preempt_enable();
465 }
466
467 /**
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
470 *
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
474 */
475 void mnt_drop_write(struct vfsmount *mnt)
476 {
477 __mnt_drop_write(mnt);
478 sb_end_write(mnt->mnt_sb);
479 }
480 EXPORT_SYMBOL_GPL(mnt_drop_write);
481
482 void __mnt_drop_write_file(struct file *file)
483 {
484 __mnt_drop_write(file->f_path.mnt);
485 }
486
487 void mnt_drop_write_file(struct file *file)
488 {
489 mnt_drop_write(file->f_path.mnt);
490 }
491 EXPORT_SYMBOL(mnt_drop_write_file);
492
493 static int mnt_make_readonly(struct mount *mnt)
494 {
495 int ret = 0;
496
497 lock_mount_hash();
498 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
499 /*
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
502 */
503 smp_mb();
504
505 /*
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
510 *
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
520 */
521 if (mnt_get_writers(mnt) > 0)
522 ret = -EBUSY;
523 else
524 mnt->mnt.mnt_flags |= MNT_READONLY;
525 /*
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
528 */
529 smp_wmb();
530 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
531 unlock_mount_hash();
532 return ret;
533 }
534
535 static void __mnt_unmake_readonly(struct mount *mnt)
536 {
537 lock_mount_hash();
538 mnt->mnt.mnt_flags &= ~MNT_READONLY;
539 unlock_mount_hash();
540 }
541
542 int sb_prepare_remount_readonly(struct super_block *sb)
543 {
544 struct mount *mnt;
545 int err = 0;
546
547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb->s_remove_count))
549 return -EBUSY;
550
551 lock_mount_hash();
552 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
555 smp_mb();
556 if (mnt_get_writers(mnt) > 0) {
557 err = -EBUSY;
558 break;
559 }
560 }
561 }
562 if (!err && atomic_long_read(&sb->s_remove_count))
563 err = -EBUSY;
564
565 if (!err) {
566 sb->s_readonly_remount = 1;
567 smp_wmb();
568 }
569 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
572 }
573 unlock_mount_hash();
574
575 return err;
576 }
577
578 static void free_vfsmnt(struct mount *mnt)
579 {
580 kfree_const(mnt->mnt_devname);
581 #ifdef CONFIG_SMP
582 free_percpu(mnt->mnt_pcp);
583 #endif
584 kmem_cache_free(mnt_cache, mnt);
585 }
586
587 static void delayed_free_vfsmnt(struct rcu_head *head)
588 {
589 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
590 }
591
592 /* call under rcu_read_lock */
593 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
594 {
595 struct mount *mnt;
596 if (read_seqretry(&mount_lock, seq))
597 return false;
598 if (bastard == NULL)
599 return true;
600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
603 return true;
604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
606 return false;
607 }
608 rcu_read_unlock();
609 mntput(bastard);
610 rcu_read_lock();
611 return false;
612 }
613
614 /*
615 * find the first mount at @dentry on vfsmount @mnt.
616 * call under rcu_read_lock()
617 */
618 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
619 {
620 struct hlist_head *head = m_hash(mnt, dentry);
621 struct mount *p;
622
623 hlist_for_each_entry_rcu(p, head, mnt_hash)
624 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
625 return p;
626 return NULL;
627 }
628
629 /*
630 * find the last mount at @dentry on vfsmount @mnt.
631 * mount_lock must be held.
632 */
633 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
634 {
635 struct mount *p, *res;
636 res = p = __lookup_mnt(mnt, dentry);
637 if (!p)
638 goto out;
639 hlist_for_each_entry_continue(p, mnt_hash) {
640 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
641 break;
642 res = p;
643 }
644 out:
645 return res;
646 }
647
648 /*
649 * lookup_mnt - Return the first child mount mounted at path
650 *
651 * "First" means first mounted chronologically. If you create the
652 * following mounts:
653 *
654 * mount /dev/sda1 /mnt
655 * mount /dev/sda2 /mnt
656 * mount /dev/sda3 /mnt
657 *
658 * Then lookup_mnt() on the base /mnt dentry in the root mount will
659 * return successively the root dentry and vfsmount of /dev/sda1, then
660 * /dev/sda2, then /dev/sda3, then NULL.
661 *
662 * lookup_mnt takes a reference to the found vfsmount.
663 */
664 struct vfsmount *lookup_mnt(struct path *path)
665 {
666 struct mount *child_mnt;
667 struct vfsmount *m;
668 unsigned seq;
669
670 rcu_read_lock();
671 do {
672 seq = read_seqbegin(&mount_lock);
673 child_mnt = __lookup_mnt(path->mnt, path->dentry);
674 m = child_mnt ? &child_mnt->mnt : NULL;
675 } while (!legitimize_mnt(m, seq));
676 rcu_read_unlock();
677 return m;
678 }
679
680 /*
681 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
682 * current mount namespace.
683 *
684 * The common case is dentries are not mountpoints at all and that
685 * test is handled inline. For the slow case when we are actually
686 * dealing with a mountpoint of some kind, walk through all of the
687 * mounts in the current mount namespace and test to see if the dentry
688 * is a mountpoint.
689 *
690 * The mount_hashtable is not usable in the context because we
691 * need to identify all mounts that may be in the current mount
692 * namespace not just a mount that happens to have some specified
693 * parent mount.
694 */
695 bool __is_local_mountpoint(struct dentry *dentry)
696 {
697 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
698 struct mount *mnt;
699 bool is_covered = false;
700
701 if (!d_mountpoint(dentry))
702 goto out;
703
704 down_read(&namespace_sem);
705 list_for_each_entry(mnt, &ns->list, mnt_list) {
706 is_covered = (mnt->mnt_mountpoint == dentry);
707 if (is_covered)
708 break;
709 }
710 up_read(&namespace_sem);
711 out:
712 return is_covered;
713 }
714
715 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
716 {
717 struct hlist_head *chain = mp_hash(dentry);
718 struct mountpoint *mp;
719
720 hlist_for_each_entry(mp, chain, m_hash) {
721 if (mp->m_dentry == dentry) {
722 /* might be worth a WARN_ON() */
723 if (d_unlinked(dentry))
724 return ERR_PTR(-ENOENT);
725 mp->m_count++;
726 return mp;
727 }
728 }
729 return NULL;
730 }
731
732 static struct mountpoint *new_mountpoint(struct dentry *dentry)
733 {
734 struct hlist_head *chain = mp_hash(dentry);
735 struct mountpoint *mp;
736 int ret;
737
738 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
739 if (!mp)
740 return ERR_PTR(-ENOMEM);
741
742 ret = d_set_mounted(dentry);
743 if (ret) {
744 kfree(mp);
745 return ERR_PTR(ret);
746 }
747
748 mp->m_dentry = dentry;
749 mp->m_count = 1;
750 hlist_add_head(&mp->m_hash, chain);
751 INIT_HLIST_HEAD(&mp->m_list);
752 return mp;
753 }
754
755 static void put_mountpoint(struct mountpoint *mp)
756 {
757 if (!--mp->m_count) {
758 struct dentry *dentry = mp->m_dentry;
759 BUG_ON(!hlist_empty(&mp->m_list));
760 spin_lock(&dentry->d_lock);
761 dentry->d_flags &= ~DCACHE_MOUNTED;
762 spin_unlock(&dentry->d_lock);
763 hlist_del(&mp->m_hash);
764 kfree(mp);
765 }
766 }
767
768 static inline int check_mnt(struct mount *mnt)
769 {
770 return mnt->mnt_ns == current->nsproxy->mnt_ns;
771 }
772
773 /*
774 * vfsmount lock must be held for write
775 */
776 static void touch_mnt_namespace(struct mnt_namespace *ns)
777 {
778 if (ns) {
779 ns->event = ++event;
780 wake_up_interruptible(&ns->poll);
781 }
782 }
783
784 /*
785 * vfsmount lock must be held for write
786 */
787 static void __touch_mnt_namespace(struct mnt_namespace *ns)
788 {
789 if (ns && ns->event != event) {
790 ns->event = event;
791 wake_up_interruptible(&ns->poll);
792 }
793 }
794
795 /*
796 * vfsmount lock must be held for write
797 */
798 static void detach_mnt(struct mount *mnt, struct path *old_path)
799 {
800 old_path->dentry = mnt->mnt_mountpoint;
801 old_path->mnt = &mnt->mnt_parent->mnt;
802 mnt->mnt_parent = mnt;
803 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
804 list_del_init(&mnt->mnt_child);
805 hlist_del_init_rcu(&mnt->mnt_hash);
806 hlist_del_init(&mnt->mnt_mp_list);
807 put_mountpoint(mnt->mnt_mp);
808 mnt->mnt_mp = NULL;
809 }
810
811 /*
812 * vfsmount lock must be held for write
813 */
814 void mnt_set_mountpoint(struct mount *mnt,
815 struct mountpoint *mp,
816 struct mount *child_mnt)
817 {
818 mp->m_count++;
819 mnt_add_count(mnt, 1); /* essentially, that's mntget */
820 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
821 child_mnt->mnt_parent = mnt;
822 child_mnt->mnt_mp = mp;
823 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
824 }
825
826 /*
827 * vfsmount lock must be held for write
828 */
829 static void attach_mnt(struct mount *mnt,
830 struct mount *parent,
831 struct mountpoint *mp)
832 {
833 mnt_set_mountpoint(parent, mp, mnt);
834 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
835 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
836 }
837
838 static void attach_shadowed(struct mount *mnt,
839 struct mount *parent,
840 struct mount *shadows)
841 {
842 if (shadows) {
843 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
844 list_add(&mnt->mnt_child, &shadows->mnt_child);
845 } else {
846 hlist_add_head_rcu(&mnt->mnt_hash,
847 m_hash(&parent->mnt, mnt->mnt_mountpoint));
848 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
849 }
850 }
851
852 /*
853 * vfsmount lock must be held for write
854 */
855 static void commit_tree(struct mount *mnt, struct mount *shadows)
856 {
857 struct mount *parent = mnt->mnt_parent;
858 struct mount *m;
859 LIST_HEAD(head);
860 struct mnt_namespace *n = parent->mnt_ns;
861
862 BUG_ON(parent == mnt);
863
864 list_add_tail(&head, &mnt->mnt_list);
865 list_for_each_entry(m, &head, mnt_list)
866 m->mnt_ns = n;
867
868 list_splice(&head, n->list.prev);
869
870 attach_shadowed(mnt, parent, shadows);
871 touch_mnt_namespace(n);
872 }
873
874 static struct mount *next_mnt(struct mount *p, struct mount *root)
875 {
876 struct list_head *next = p->mnt_mounts.next;
877 if (next == &p->mnt_mounts) {
878 while (1) {
879 if (p == root)
880 return NULL;
881 next = p->mnt_child.next;
882 if (next != &p->mnt_parent->mnt_mounts)
883 break;
884 p = p->mnt_parent;
885 }
886 }
887 return list_entry(next, struct mount, mnt_child);
888 }
889
890 static struct mount *skip_mnt_tree(struct mount *p)
891 {
892 struct list_head *prev = p->mnt_mounts.prev;
893 while (prev != &p->mnt_mounts) {
894 p = list_entry(prev, struct mount, mnt_child);
895 prev = p->mnt_mounts.prev;
896 }
897 return p;
898 }
899
900 struct vfsmount *
901 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
902 {
903 struct mount *mnt;
904 struct dentry *root;
905
906 if (!type)
907 return ERR_PTR(-ENODEV);
908
909 mnt = alloc_vfsmnt(name);
910 if (!mnt)
911 return ERR_PTR(-ENOMEM);
912
913 if (flags & MS_KERNMOUNT)
914 mnt->mnt.mnt_flags = MNT_INTERNAL;
915
916 root = mount_fs(type, flags, name, data);
917 if (IS_ERR(root)) {
918 mnt_free_id(mnt);
919 free_vfsmnt(mnt);
920 return ERR_CAST(root);
921 }
922
923 mnt->mnt.mnt_root = root;
924 mnt->mnt.mnt_sb = root->d_sb;
925 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
926 mnt->mnt_parent = mnt;
927 lock_mount_hash();
928 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
929 unlock_mount_hash();
930 return &mnt->mnt;
931 }
932 EXPORT_SYMBOL_GPL(vfs_kern_mount);
933
934 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
935 int flag)
936 {
937 struct super_block *sb = old->mnt.mnt_sb;
938 struct mount *mnt;
939 int err;
940
941 mnt = alloc_vfsmnt(old->mnt_devname);
942 if (!mnt)
943 return ERR_PTR(-ENOMEM);
944
945 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
946 mnt->mnt_group_id = 0; /* not a peer of original */
947 else
948 mnt->mnt_group_id = old->mnt_group_id;
949
950 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
951 err = mnt_alloc_group_id(mnt);
952 if (err)
953 goto out_free;
954 }
955
956 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
957 /* Don't allow unprivileged users to change mount flags */
958 if (flag & CL_UNPRIVILEGED) {
959 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
960
961 if (mnt->mnt.mnt_flags & MNT_READONLY)
962 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
963
964 if (mnt->mnt.mnt_flags & MNT_NODEV)
965 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
966
967 if (mnt->mnt.mnt_flags & MNT_NOSUID)
968 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
969
970 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
971 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
972 }
973
974 /* Don't allow unprivileged users to reveal what is under a mount */
975 if ((flag & CL_UNPRIVILEGED) &&
976 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
977 mnt->mnt.mnt_flags |= MNT_LOCKED;
978
979 atomic_inc(&sb->s_active);
980 mnt->mnt.mnt_sb = sb;
981 mnt->mnt.mnt_root = dget(root);
982 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
983 mnt->mnt_parent = mnt;
984 lock_mount_hash();
985 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
986 unlock_mount_hash();
987
988 if ((flag & CL_SLAVE) ||
989 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
990 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
991 mnt->mnt_master = old;
992 CLEAR_MNT_SHARED(mnt);
993 } else if (!(flag & CL_PRIVATE)) {
994 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
995 list_add(&mnt->mnt_share, &old->mnt_share);
996 if (IS_MNT_SLAVE(old))
997 list_add(&mnt->mnt_slave, &old->mnt_slave);
998 mnt->mnt_master = old->mnt_master;
999 }
1000 if (flag & CL_MAKE_SHARED)
1001 set_mnt_shared(mnt);
1002
1003 /* stick the duplicate mount on the same expiry list
1004 * as the original if that was on one */
1005 if (flag & CL_EXPIRE) {
1006 if (!list_empty(&old->mnt_expire))
1007 list_add(&mnt->mnt_expire, &old->mnt_expire);
1008 }
1009
1010 return mnt;
1011
1012 out_free:
1013 mnt_free_id(mnt);
1014 free_vfsmnt(mnt);
1015 return ERR_PTR(err);
1016 }
1017
1018 static void cleanup_mnt(struct mount *mnt)
1019 {
1020 /*
1021 * This probably indicates that somebody messed
1022 * up a mnt_want/drop_write() pair. If this
1023 * happens, the filesystem was probably unable
1024 * to make r/w->r/o transitions.
1025 */
1026 /*
1027 * The locking used to deal with mnt_count decrement provides barriers,
1028 * so mnt_get_writers() below is safe.
1029 */
1030 WARN_ON(mnt_get_writers(mnt));
1031 if (unlikely(mnt->mnt_pins.first))
1032 mnt_pin_kill(mnt);
1033 fsnotify_vfsmount_delete(&mnt->mnt);
1034 dput(mnt->mnt.mnt_root);
1035 deactivate_super(mnt->mnt.mnt_sb);
1036 mnt_free_id(mnt);
1037 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1038 }
1039
1040 static void __cleanup_mnt(struct rcu_head *head)
1041 {
1042 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1043 }
1044
1045 static LLIST_HEAD(delayed_mntput_list);
1046 static void delayed_mntput(struct work_struct *unused)
1047 {
1048 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1049 struct llist_node *next;
1050
1051 for (; node; node = next) {
1052 next = llist_next(node);
1053 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1054 }
1055 }
1056 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1057
1058 static void mntput_no_expire(struct mount *mnt)
1059 {
1060 rcu_read_lock();
1061 mnt_add_count(mnt, -1);
1062 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1063 rcu_read_unlock();
1064 return;
1065 }
1066 lock_mount_hash();
1067 if (mnt_get_count(mnt)) {
1068 rcu_read_unlock();
1069 unlock_mount_hash();
1070 return;
1071 }
1072 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1073 rcu_read_unlock();
1074 unlock_mount_hash();
1075 return;
1076 }
1077 mnt->mnt.mnt_flags |= MNT_DOOMED;
1078 rcu_read_unlock();
1079
1080 list_del(&mnt->mnt_instance);
1081 unlock_mount_hash();
1082
1083 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1084 struct task_struct *task = current;
1085 if (likely(!(task->flags & PF_KTHREAD))) {
1086 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1087 if (!task_work_add(task, &mnt->mnt_rcu, true))
1088 return;
1089 }
1090 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1091 schedule_delayed_work(&delayed_mntput_work, 1);
1092 return;
1093 }
1094 cleanup_mnt(mnt);
1095 }
1096
1097 void mntput(struct vfsmount *mnt)
1098 {
1099 if (mnt) {
1100 struct mount *m = real_mount(mnt);
1101 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1102 if (unlikely(m->mnt_expiry_mark))
1103 m->mnt_expiry_mark = 0;
1104 mntput_no_expire(m);
1105 }
1106 }
1107 EXPORT_SYMBOL(mntput);
1108
1109 struct vfsmount *mntget(struct vfsmount *mnt)
1110 {
1111 if (mnt)
1112 mnt_add_count(real_mount(mnt), 1);
1113 return mnt;
1114 }
1115 EXPORT_SYMBOL(mntget);
1116
1117 struct vfsmount *mnt_clone_internal(struct path *path)
1118 {
1119 struct mount *p;
1120 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1121 if (IS_ERR(p))
1122 return ERR_CAST(p);
1123 p->mnt.mnt_flags |= MNT_INTERNAL;
1124 return &p->mnt;
1125 }
1126
1127 static inline void mangle(struct seq_file *m, const char *s)
1128 {
1129 seq_escape(m, s, " \t\n\\");
1130 }
1131
1132 /*
1133 * Simple .show_options callback for filesystems which don't want to
1134 * implement more complex mount option showing.
1135 *
1136 * See also save_mount_options().
1137 */
1138 int generic_show_options(struct seq_file *m, struct dentry *root)
1139 {
1140 const char *options;
1141
1142 rcu_read_lock();
1143 options = rcu_dereference(root->d_sb->s_options);
1144
1145 if (options != NULL && options[0]) {
1146 seq_putc(m, ',');
1147 mangle(m, options);
1148 }
1149 rcu_read_unlock();
1150
1151 return 0;
1152 }
1153 EXPORT_SYMBOL(generic_show_options);
1154
1155 /*
1156 * If filesystem uses generic_show_options(), this function should be
1157 * called from the fill_super() callback.
1158 *
1159 * The .remount_fs callback usually needs to be handled in a special
1160 * way, to make sure, that previous options are not overwritten if the
1161 * remount fails.
1162 *
1163 * Also note, that if the filesystem's .remount_fs function doesn't
1164 * reset all options to their default value, but changes only newly
1165 * given options, then the displayed options will not reflect reality
1166 * any more.
1167 */
1168 void save_mount_options(struct super_block *sb, char *options)
1169 {
1170 BUG_ON(sb->s_options);
1171 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1172 }
1173 EXPORT_SYMBOL(save_mount_options);
1174
1175 void replace_mount_options(struct super_block *sb, char *options)
1176 {
1177 char *old = sb->s_options;
1178 rcu_assign_pointer(sb->s_options, options);
1179 if (old) {
1180 synchronize_rcu();
1181 kfree(old);
1182 }
1183 }
1184 EXPORT_SYMBOL(replace_mount_options);
1185
1186 #ifdef CONFIG_PROC_FS
1187 /* iterator; we want it to have access to namespace_sem, thus here... */
1188 static void *m_start(struct seq_file *m, loff_t *pos)
1189 {
1190 struct proc_mounts *p = proc_mounts(m);
1191
1192 down_read(&namespace_sem);
1193 if (p->cached_event == p->ns->event) {
1194 void *v = p->cached_mount;
1195 if (*pos == p->cached_index)
1196 return v;
1197 if (*pos == p->cached_index + 1) {
1198 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1199 return p->cached_mount = v;
1200 }
1201 }
1202
1203 p->cached_event = p->ns->event;
1204 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1205 p->cached_index = *pos;
1206 return p->cached_mount;
1207 }
1208
1209 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1210 {
1211 struct proc_mounts *p = proc_mounts(m);
1212
1213 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1214 p->cached_index = *pos;
1215 return p->cached_mount;
1216 }
1217
1218 static void m_stop(struct seq_file *m, void *v)
1219 {
1220 up_read(&namespace_sem);
1221 }
1222
1223 static int m_show(struct seq_file *m, void *v)
1224 {
1225 struct proc_mounts *p = proc_mounts(m);
1226 struct mount *r = list_entry(v, struct mount, mnt_list);
1227 return p->show(m, &r->mnt);
1228 }
1229
1230 const struct seq_operations mounts_op = {
1231 .start = m_start,
1232 .next = m_next,
1233 .stop = m_stop,
1234 .show = m_show,
1235 };
1236 #endif /* CONFIG_PROC_FS */
1237
1238 /**
1239 * may_umount_tree - check if a mount tree is busy
1240 * @mnt: root of mount tree
1241 *
1242 * This is called to check if a tree of mounts has any
1243 * open files, pwds, chroots or sub mounts that are
1244 * busy.
1245 */
1246 int may_umount_tree(struct vfsmount *m)
1247 {
1248 struct mount *mnt = real_mount(m);
1249 int actual_refs = 0;
1250 int minimum_refs = 0;
1251 struct mount *p;
1252 BUG_ON(!m);
1253
1254 /* write lock needed for mnt_get_count */
1255 lock_mount_hash();
1256 for (p = mnt; p; p = next_mnt(p, mnt)) {
1257 actual_refs += mnt_get_count(p);
1258 minimum_refs += 2;
1259 }
1260 unlock_mount_hash();
1261
1262 if (actual_refs > minimum_refs)
1263 return 0;
1264
1265 return 1;
1266 }
1267
1268 EXPORT_SYMBOL(may_umount_tree);
1269
1270 /**
1271 * may_umount - check if a mount point is busy
1272 * @mnt: root of mount
1273 *
1274 * This is called to check if a mount point has any
1275 * open files, pwds, chroots or sub mounts. If the
1276 * mount has sub mounts this will return busy
1277 * regardless of whether the sub mounts are busy.
1278 *
1279 * Doesn't take quota and stuff into account. IOW, in some cases it will
1280 * give false negatives. The main reason why it's here is that we need
1281 * a non-destructive way to look for easily umountable filesystems.
1282 */
1283 int may_umount(struct vfsmount *mnt)
1284 {
1285 int ret = 1;
1286 down_read(&namespace_sem);
1287 lock_mount_hash();
1288 if (propagate_mount_busy(real_mount(mnt), 2))
1289 ret = 0;
1290 unlock_mount_hash();
1291 up_read(&namespace_sem);
1292 return ret;
1293 }
1294
1295 EXPORT_SYMBOL(may_umount);
1296
1297 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1298
1299 static void namespace_unlock(void)
1300 {
1301 struct hlist_head head = unmounted;
1302
1303 if (likely(hlist_empty(&head))) {
1304 up_write(&namespace_sem);
1305 return;
1306 }
1307
1308 head.first->pprev = &head.first;
1309 INIT_HLIST_HEAD(&unmounted);
1310 up_write(&namespace_sem);
1311
1312 synchronize_rcu();
1313
1314 group_pin_kill(&head);
1315 }
1316
1317 static inline void namespace_lock(void)
1318 {
1319 down_write(&namespace_sem);
1320 }
1321
1322 /*
1323 * mount_lock must be held
1324 * namespace_sem must be held for write
1325 * how = 0 => just this tree, don't propagate
1326 * how = 1 => propagate; we know that nobody else has reference to any victims
1327 * how = 2 => lazy umount
1328 */
1329 void umount_tree(struct mount *mnt, int how)
1330 {
1331 HLIST_HEAD(tmp_list);
1332 struct mount *p;
1333
1334 for (p = mnt; p; p = next_mnt(p, mnt)) {
1335 hlist_del_init_rcu(&p->mnt_hash);
1336 hlist_add_head(&p->mnt_hash, &tmp_list);
1337 }
1338
1339 hlist_for_each_entry(p, &tmp_list, mnt_hash)
1340 list_del_init(&p->mnt_child);
1341
1342 if (how)
1343 propagate_umount(&tmp_list);
1344
1345 while (!hlist_empty(&tmp_list)) {
1346 p = hlist_entry(tmp_list.first, struct mount, mnt_hash);
1347 hlist_del_init_rcu(&p->mnt_hash);
1348 list_del_init(&p->mnt_expire);
1349 list_del_init(&p->mnt_list);
1350 __touch_mnt_namespace(p->mnt_ns);
1351 p->mnt_ns = NULL;
1352 if (how < 2)
1353 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1354
1355 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, &unmounted);
1356 if (mnt_has_parent(p)) {
1357 hlist_del_init(&p->mnt_mp_list);
1358 put_mountpoint(p->mnt_mp);
1359 mnt_add_count(p->mnt_parent, -1);
1360 /* old mountpoint will be dropped when we can do that */
1361 p->mnt_ex_mountpoint = p->mnt_mountpoint;
1362 p->mnt_mountpoint = p->mnt.mnt_root;
1363 p->mnt_parent = p;
1364 p->mnt_mp = NULL;
1365 }
1366 change_mnt_propagation(p, MS_PRIVATE);
1367 }
1368 }
1369
1370 static void shrink_submounts(struct mount *mnt);
1371
1372 static int do_umount(struct mount *mnt, int flags)
1373 {
1374 struct super_block *sb = mnt->mnt.mnt_sb;
1375 int retval;
1376
1377 retval = security_sb_umount(&mnt->mnt, flags);
1378 if (retval)
1379 return retval;
1380
1381 /*
1382 * Allow userspace to request a mountpoint be expired rather than
1383 * unmounting unconditionally. Unmount only happens if:
1384 * (1) the mark is already set (the mark is cleared by mntput())
1385 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1386 */
1387 if (flags & MNT_EXPIRE) {
1388 if (&mnt->mnt == current->fs->root.mnt ||
1389 flags & (MNT_FORCE | MNT_DETACH))
1390 return -EINVAL;
1391
1392 /*
1393 * probably don't strictly need the lock here if we examined
1394 * all race cases, but it's a slowpath.
1395 */
1396 lock_mount_hash();
1397 if (mnt_get_count(mnt) != 2) {
1398 unlock_mount_hash();
1399 return -EBUSY;
1400 }
1401 unlock_mount_hash();
1402
1403 if (!xchg(&mnt->mnt_expiry_mark, 1))
1404 return -EAGAIN;
1405 }
1406
1407 /*
1408 * If we may have to abort operations to get out of this
1409 * mount, and they will themselves hold resources we must
1410 * allow the fs to do things. In the Unix tradition of
1411 * 'Gee thats tricky lets do it in userspace' the umount_begin
1412 * might fail to complete on the first run through as other tasks
1413 * must return, and the like. Thats for the mount program to worry
1414 * about for the moment.
1415 */
1416
1417 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1418 sb->s_op->umount_begin(sb);
1419 }
1420
1421 /*
1422 * No sense to grab the lock for this test, but test itself looks
1423 * somewhat bogus. Suggestions for better replacement?
1424 * Ho-hum... In principle, we might treat that as umount + switch
1425 * to rootfs. GC would eventually take care of the old vfsmount.
1426 * Actually it makes sense, especially if rootfs would contain a
1427 * /reboot - static binary that would close all descriptors and
1428 * call reboot(9). Then init(8) could umount root and exec /reboot.
1429 */
1430 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1431 /*
1432 * Special case for "unmounting" root ...
1433 * we just try to remount it readonly.
1434 */
1435 if (!capable(CAP_SYS_ADMIN))
1436 return -EPERM;
1437 down_write(&sb->s_umount);
1438 if (!(sb->s_flags & MS_RDONLY))
1439 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1440 up_write(&sb->s_umount);
1441 return retval;
1442 }
1443
1444 namespace_lock();
1445 lock_mount_hash();
1446 event++;
1447
1448 if (flags & MNT_DETACH) {
1449 if (!list_empty(&mnt->mnt_list))
1450 umount_tree(mnt, 2);
1451 retval = 0;
1452 } else {
1453 shrink_submounts(mnt);
1454 retval = -EBUSY;
1455 if (!propagate_mount_busy(mnt, 2)) {
1456 if (!list_empty(&mnt->mnt_list))
1457 umount_tree(mnt, 1);
1458 retval = 0;
1459 }
1460 }
1461 unlock_mount_hash();
1462 namespace_unlock();
1463 return retval;
1464 }
1465
1466 /*
1467 * __detach_mounts - lazily unmount all mounts on the specified dentry
1468 *
1469 * During unlink, rmdir, and d_drop it is possible to loose the path
1470 * to an existing mountpoint, and wind up leaking the mount.
1471 * detach_mounts allows lazily unmounting those mounts instead of
1472 * leaking them.
1473 *
1474 * The caller may hold dentry->d_inode->i_mutex.
1475 */
1476 void __detach_mounts(struct dentry *dentry)
1477 {
1478 struct mountpoint *mp;
1479 struct mount *mnt;
1480
1481 namespace_lock();
1482 mp = lookup_mountpoint(dentry);
1483 if (!mp)
1484 goto out_unlock;
1485
1486 lock_mount_hash();
1487 while (!hlist_empty(&mp->m_list)) {
1488 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1489 umount_tree(mnt, 2);
1490 }
1491 unlock_mount_hash();
1492 put_mountpoint(mp);
1493 out_unlock:
1494 namespace_unlock();
1495 }
1496
1497 /*
1498 * Is the caller allowed to modify his namespace?
1499 */
1500 static inline bool may_mount(void)
1501 {
1502 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1503 }
1504
1505 /*
1506 * Now umount can handle mount points as well as block devices.
1507 * This is important for filesystems which use unnamed block devices.
1508 *
1509 * We now support a flag for forced unmount like the other 'big iron'
1510 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1511 */
1512
1513 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1514 {
1515 struct path path;
1516 struct mount *mnt;
1517 int retval;
1518 int lookup_flags = 0;
1519
1520 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1521 return -EINVAL;
1522
1523 if (!may_mount())
1524 return -EPERM;
1525
1526 if (!(flags & UMOUNT_NOFOLLOW))
1527 lookup_flags |= LOOKUP_FOLLOW;
1528
1529 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1530 if (retval)
1531 goto out;
1532 mnt = real_mount(path.mnt);
1533 retval = -EINVAL;
1534 if (path.dentry != path.mnt->mnt_root)
1535 goto dput_and_out;
1536 if (!check_mnt(mnt))
1537 goto dput_and_out;
1538 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1539 goto dput_and_out;
1540 retval = -EPERM;
1541 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1542 goto dput_and_out;
1543
1544 retval = do_umount(mnt, flags);
1545 dput_and_out:
1546 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1547 dput(path.dentry);
1548 mntput_no_expire(mnt);
1549 out:
1550 return retval;
1551 }
1552
1553 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1554
1555 /*
1556 * The 2.0 compatible umount. No flags.
1557 */
1558 SYSCALL_DEFINE1(oldumount, char __user *, name)
1559 {
1560 return sys_umount(name, 0);
1561 }
1562
1563 #endif
1564
1565 static bool is_mnt_ns_file(struct dentry *dentry)
1566 {
1567 /* Is this a proxy for a mount namespace? */
1568 return dentry->d_op == &ns_dentry_operations &&
1569 dentry->d_fsdata == &mntns_operations;
1570 }
1571
1572 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1573 {
1574 return container_of(ns, struct mnt_namespace, ns);
1575 }
1576
1577 static bool mnt_ns_loop(struct dentry *dentry)
1578 {
1579 /* Could bind mounting the mount namespace inode cause a
1580 * mount namespace loop?
1581 */
1582 struct mnt_namespace *mnt_ns;
1583 if (!is_mnt_ns_file(dentry))
1584 return false;
1585
1586 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1587 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1588 }
1589
1590 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1591 int flag)
1592 {
1593 struct mount *res, *p, *q, *r, *parent;
1594
1595 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1596 return ERR_PTR(-EINVAL);
1597
1598 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1599 return ERR_PTR(-EINVAL);
1600
1601 res = q = clone_mnt(mnt, dentry, flag);
1602 if (IS_ERR(q))
1603 return q;
1604
1605 q->mnt_mountpoint = mnt->mnt_mountpoint;
1606
1607 p = mnt;
1608 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1609 struct mount *s;
1610 if (!is_subdir(r->mnt_mountpoint, dentry))
1611 continue;
1612
1613 for (s = r; s; s = next_mnt(s, r)) {
1614 struct mount *t = NULL;
1615 if (!(flag & CL_COPY_UNBINDABLE) &&
1616 IS_MNT_UNBINDABLE(s)) {
1617 s = skip_mnt_tree(s);
1618 continue;
1619 }
1620 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1621 is_mnt_ns_file(s->mnt.mnt_root)) {
1622 s = skip_mnt_tree(s);
1623 continue;
1624 }
1625 while (p != s->mnt_parent) {
1626 p = p->mnt_parent;
1627 q = q->mnt_parent;
1628 }
1629 p = s;
1630 parent = q;
1631 q = clone_mnt(p, p->mnt.mnt_root, flag);
1632 if (IS_ERR(q))
1633 goto out;
1634 lock_mount_hash();
1635 list_add_tail(&q->mnt_list, &res->mnt_list);
1636 mnt_set_mountpoint(parent, p->mnt_mp, q);
1637 if (!list_empty(&parent->mnt_mounts)) {
1638 t = list_last_entry(&parent->mnt_mounts,
1639 struct mount, mnt_child);
1640 if (t->mnt_mp != p->mnt_mp)
1641 t = NULL;
1642 }
1643 attach_shadowed(q, parent, t);
1644 unlock_mount_hash();
1645 }
1646 }
1647 return res;
1648 out:
1649 if (res) {
1650 lock_mount_hash();
1651 umount_tree(res, 0);
1652 unlock_mount_hash();
1653 }
1654 return q;
1655 }
1656
1657 /* Caller should check returned pointer for errors */
1658
1659 struct vfsmount *collect_mounts(struct path *path)
1660 {
1661 struct mount *tree;
1662 namespace_lock();
1663 tree = copy_tree(real_mount(path->mnt), path->dentry,
1664 CL_COPY_ALL | CL_PRIVATE);
1665 namespace_unlock();
1666 if (IS_ERR(tree))
1667 return ERR_CAST(tree);
1668 return &tree->mnt;
1669 }
1670
1671 void drop_collected_mounts(struct vfsmount *mnt)
1672 {
1673 namespace_lock();
1674 lock_mount_hash();
1675 umount_tree(real_mount(mnt), 0);
1676 unlock_mount_hash();
1677 namespace_unlock();
1678 }
1679
1680 /**
1681 * clone_private_mount - create a private clone of a path
1682 *
1683 * This creates a new vfsmount, which will be the clone of @path. The new will
1684 * not be attached anywhere in the namespace and will be private (i.e. changes
1685 * to the originating mount won't be propagated into this).
1686 *
1687 * Release with mntput().
1688 */
1689 struct vfsmount *clone_private_mount(struct path *path)
1690 {
1691 struct mount *old_mnt = real_mount(path->mnt);
1692 struct mount *new_mnt;
1693
1694 if (IS_MNT_UNBINDABLE(old_mnt))
1695 return ERR_PTR(-EINVAL);
1696
1697 down_read(&namespace_sem);
1698 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1699 up_read(&namespace_sem);
1700 if (IS_ERR(new_mnt))
1701 return ERR_CAST(new_mnt);
1702
1703 return &new_mnt->mnt;
1704 }
1705 EXPORT_SYMBOL_GPL(clone_private_mount);
1706
1707 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1708 struct vfsmount *root)
1709 {
1710 struct mount *mnt;
1711 int res = f(root, arg);
1712 if (res)
1713 return res;
1714 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1715 res = f(&mnt->mnt, arg);
1716 if (res)
1717 return res;
1718 }
1719 return 0;
1720 }
1721
1722 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1723 {
1724 struct mount *p;
1725
1726 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1727 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1728 mnt_release_group_id(p);
1729 }
1730 }
1731
1732 static int invent_group_ids(struct mount *mnt, bool recurse)
1733 {
1734 struct mount *p;
1735
1736 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1737 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1738 int err = mnt_alloc_group_id(p);
1739 if (err) {
1740 cleanup_group_ids(mnt, p);
1741 return err;
1742 }
1743 }
1744 }
1745
1746 return 0;
1747 }
1748
1749 /*
1750 * @source_mnt : mount tree to be attached
1751 * @nd : place the mount tree @source_mnt is attached
1752 * @parent_nd : if non-null, detach the source_mnt from its parent and
1753 * store the parent mount and mountpoint dentry.
1754 * (done when source_mnt is moved)
1755 *
1756 * NOTE: in the table below explains the semantics when a source mount
1757 * of a given type is attached to a destination mount of a given type.
1758 * ---------------------------------------------------------------------------
1759 * | BIND MOUNT OPERATION |
1760 * |**************************************************************************
1761 * | source-->| shared | private | slave | unbindable |
1762 * | dest | | | | |
1763 * | | | | | | |
1764 * | v | | | | |
1765 * |**************************************************************************
1766 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1767 * | | | | | |
1768 * |non-shared| shared (+) | private | slave (*) | invalid |
1769 * ***************************************************************************
1770 * A bind operation clones the source mount and mounts the clone on the
1771 * destination mount.
1772 *
1773 * (++) the cloned mount is propagated to all the mounts in the propagation
1774 * tree of the destination mount and the cloned mount is added to
1775 * the peer group of the source mount.
1776 * (+) the cloned mount is created under the destination mount and is marked
1777 * as shared. The cloned mount is added to the peer group of the source
1778 * mount.
1779 * (+++) the mount is propagated to all the mounts in the propagation tree
1780 * of the destination mount and the cloned mount is made slave
1781 * of the same master as that of the source mount. The cloned mount
1782 * is marked as 'shared and slave'.
1783 * (*) the cloned mount is made a slave of the same master as that of the
1784 * source mount.
1785 *
1786 * ---------------------------------------------------------------------------
1787 * | MOVE MOUNT OPERATION |
1788 * |**************************************************************************
1789 * | source-->| shared | private | slave | unbindable |
1790 * | dest | | | | |
1791 * | | | | | | |
1792 * | v | | | | |
1793 * |**************************************************************************
1794 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1795 * | | | | | |
1796 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1797 * ***************************************************************************
1798 *
1799 * (+) the mount is moved to the destination. And is then propagated to
1800 * all the mounts in the propagation tree of the destination mount.
1801 * (+*) the mount is moved to the destination.
1802 * (+++) the mount is moved to the destination and is then propagated to
1803 * all the mounts belonging to the destination mount's propagation tree.
1804 * the mount is marked as 'shared and slave'.
1805 * (*) the mount continues to be a slave at the new location.
1806 *
1807 * if the source mount is a tree, the operations explained above is
1808 * applied to each mount in the tree.
1809 * Must be called without spinlocks held, since this function can sleep
1810 * in allocations.
1811 */
1812 static int attach_recursive_mnt(struct mount *source_mnt,
1813 struct mount *dest_mnt,
1814 struct mountpoint *dest_mp,
1815 struct path *parent_path)
1816 {
1817 HLIST_HEAD(tree_list);
1818 struct mount *child, *p;
1819 struct hlist_node *n;
1820 int err;
1821
1822 if (IS_MNT_SHARED(dest_mnt)) {
1823 err = invent_group_ids(source_mnt, true);
1824 if (err)
1825 goto out;
1826 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1827 lock_mount_hash();
1828 if (err)
1829 goto out_cleanup_ids;
1830 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1831 set_mnt_shared(p);
1832 } else {
1833 lock_mount_hash();
1834 }
1835 if (parent_path) {
1836 detach_mnt(source_mnt, parent_path);
1837 attach_mnt(source_mnt, dest_mnt, dest_mp);
1838 touch_mnt_namespace(source_mnt->mnt_ns);
1839 } else {
1840 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1841 commit_tree(source_mnt, NULL);
1842 }
1843
1844 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1845 struct mount *q;
1846 hlist_del_init(&child->mnt_hash);
1847 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1848 child->mnt_mountpoint);
1849 commit_tree(child, q);
1850 }
1851 unlock_mount_hash();
1852
1853 return 0;
1854
1855 out_cleanup_ids:
1856 while (!hlist_empty(&tree_list)) {
1857 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1858 umount_tree(child, 0);
1859 }
1860 unlock_mount_hash();
1861 cleanup_group_ids(source_mnt, NULL);
1862 out:
1863 return err;
1864 }
1865
1866 static struct mountpoint *lock_mount(struct path *path)
1867 {
1868 struct vfsmount *mnt;
1869 struct dentry *dentry = path->dentry;
1870 retry:
1871 mutex_lock(&dentry->d_inode->i_mutex);
1872 if (unlikely(cant_mount(dentry))) {
1873 mutex_unlock(&dentry->d_inode->i_mutex);
1874 return ERR_PTR(-ENOENT);
1875 }
1876 namespace_lock();
1877 mnt = lookup_mnt(path);
1878 if (likely(!mnt)) {
1879 struct mountpoint *mp = lookup_mountpoint(dentry);
1880 if (!mp)
1881 mp = new_mountpoint(dentry);
1882 if (IS_ERR(mp)) {
1883 namespace_unlock();
1884 mutex_unlock(&dentry->d_inode->i_mutex);
1885 return mp;
1886 }
1887 return mp;
1888 }
1889 namespace_unlock();
1890 mutex_unlock(&path->dentry->d_inode->i_mutex);
1891 path_put(path);
1892 path->mnt = mnt;
1893 dentry = path->dentry = dget(mnt->mnt_root);
1894 goto retry;
1895 }
1896
1897 static void unlock_mount(struct mountpoint *where)
1898 {
1899 struct dentry *dentry = where->m_dentry;
1900 put_mountpoint(where);
1901 namespace_unlock();
1902 mutex_unlock(&dentry->d_inode->i_mutex);
1903 }
1904
1905 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1906 {
1907 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1908 return -EINVAL;
1909
1910 if (d_is_dir(mp->m_dentry) !=
1911 d_is_dir(mnt->mnt.mnt_root))
1912 return -ENOTDIR;
1913
1914 return attach_recursive_mnt(mnt, p, mp, NULL);
1915 }
1916
1917 /*
1918 * Sanity check the flags to change_mnt_propagation.
1919 */
1920
1921 static int flags_to_propagation_type(int flags)
1922 {
1923 int type = flags & ~(MS_REC | MS_SILENT);
1924
1925 /* Fail if any non-propagation flags are set */
1926 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1927 return 0;
1928 /* Only one propagation flag should be set */
1929 if (!is_power_of_2(type))
1930 return 0;
1931 return type;
1932 }
1933
1934 /*
1935 * recursively change the type of the mountpoint.
1936 */
1937 static int do_change_type(struct path *path, int flag)
1938 {
1939 struct mount *m;
1940 struct mount *mnt = real_mount(path->mnt);
1941 int recurse = flag & MS_REC;
1942 int type;
1943 int err = 0;
1944
1945 if (path->dentry != path->mnt->mnt_root)
1946 return -EINVAL;
1947
1948 type = flags_to_propagation_type(flag);
1949 if (!type)
1950 return -EINVAL;
1951
1952 namespace_lock();
1953 if (type == MS_SHARED) {
1954 err = invent_group_ids(mnt, recurse);
1955 if (err)
1956 goto out_unlock;
1957 }
1958
1959 lock_mount_hash();
1960 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1961 change_mnt_propagation(m, type);
1962 unlock_mount_hash();
1963
1964 out_unlock:
1965 namespace_unlock();
1966 return err;
1967 }
1968
1969 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1970 {
1971 struct mount *child;
1972 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1973 if (!is_subdir(child->mnt_mountpoint, dentry))
1974 continue;
1975
1976 if (child->mnt.mnt_flags & MNT_LOCKED)
1977 return true;
1978 }
1979 return false;
1980 }
1981
1982 /*
1983 * do loopback mount.
1984 */
1985 static int do_loopback(struct path *path, const char *old_name,
1986 int recurse)
1987 {
1988 struct path old_path;
1989 struct mount *mnt = NULL, *old, *parent;
1990 struct mountpoint *mp;
1991 int err;
1992 if (!old_name || !*old_name)
1993 return -EINVAL;
1994 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1995 if (err)
1996 return err;
1997
1998 err = -EINVAL;
1999 if (mnt_ns_loop(old_path.dentry))
2000 goto out;
2001
2002 mp = lock_mount(path);
2003 err = PTR_ERR(mp);
2004 if (IS_ERR(mp))
2005 goto out;
2006
2007 old = real_mount(old_path.mnt);
2008 parent = real_mount(path->mnt);
2009
2010 err = -EINVAL;
2011 if (IS_MNT_UNBINDABLE(old))
2012 goto out2;
2013
2014 if (!check_mnt(parent))
2015 goto out2;
2016
2017 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2018 goto out2;
2019
2020 if (!recurse && has_locked_children(old, old_path.dentry))
2021 goto out2;
2022
2023 if (recurse)
2024 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2025 else
2026 mnt = clone_mnt(old, old_path.dentry, 0);
2027
2028 if (IS_ERR(mnt)) {
2029 err = PTR_ERR(mnt);
2030 goto out2;
2031 }
2032
2033 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2034
2035 err = graft_tree(mnt, parent, mp);
2036 if (err) {
2037 lock_mount_hash();
2038 umount_tree(mnt, 0);
2039 unlock_mount_hash();
2040 }
2041 out2:
2042 unlock_mount(mp);
2043 out:
2044 path_put(&old_path);
2045 return err;
2046 }
2047
2048 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2049 {
2050 int error = 0;
2051 int readonly_request = 0;
2052
2053 if (ms_flags & MS_RDONLY)
2054 readonly_request = 1;
2055 if (readonly_request == __mnt_is_readonly(mnt))
2056 return 0;
2057
2058 if (readonly_request)
2059 error = mnt_make_readonly(real_mount(mnt));
2060 else
2061 __mnt_unmake_readonly(real_mount(mnt));
2062 return error;
2063 }
2064
2065 /*
2066 * change filesystem flags. dir should be a physical root of filesystem.
2067 * If you've mounted a non-root directory somewhere and want to do remount
2068 * on it - tough luck.
2069 */
2070 static int do_remount(struct path *path, int flags, int mnt_flags,
2071 void *data)
2072 {
2073 int err;
2074 struct super_block *sb = path->mnt->mnt_sb;
2075 struct mount *mnt = real_mount(path->mnt);
2076
2077 if (!check_mnt(mnt))
2078 return -EINVAL;
2079
2080 if (path->dentry != path->mnt->mnt_root)
2081 return -EINVAL;
2082
2083 /* Don't allow changing of locked mnt flags.
2084 *
2085 * No locks need to be held here while testing the various
2086 * MNT_LOCK flags because those flags can never be cleared
2087 * once they are set.
2088 */
2089 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2090 !(mnt_flags & MNT_READONLY)) {
2091 return -EPERM;
2092 }
2093 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2094 !(mnt_flags & MNT_NODEV)) {
2095 /* Was the nodev implicitly added in mount? */
2096 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2097 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2098 mnt_flags |= MNT_NODEV;
2099 } else {
2100 return -EPERM;
2101 }
2102 }
2103 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2104 !(mnt_flags & MNT_NOSUID)) {
2105 return -EPERM;
2106 }
2107 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2108 !(mnt_flags & MNT_NOEXEC)) {
2109 return -EPERM;
2110 }
2111 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2112 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2113 return -EPERM;
2114 }
2115
2116 err = security_sb_remount(sb, data);
2117 if (err)
2118 return err;
2119
2120 down_write(&sb->s_umount);
2121 if (flags & MS_BIND)
2122 err = change_mount_flags(path->mnt, flags);
2123 else if (!capable(CAP_SYS_ADMIN))
2124 err = -EPERM;
2125 else
2126 err = do_remount_sb(sb, flags, data, 0);
2127 if (!err) {
2128 lock_mount_hash();
2129 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2130 mnt->mnt.mnt_flags = mnt_flags;
2131 touch_mnt_namespace(mnt->mnt_ns);
2132 unlock_mount_hash();
2133 }
2134 up_write(&sb->s_umount);
2135 return err;
2136 }
2137
2138 static inline int tree_contains_unbindable(struct mount *mnt)
2139 {
2140 struct mount *p;
2141 for (p = mnt; p; p = next_mnt(p, mnt)) {
2142 if (IS_MNT_UNBINDABLE(p))
2143 return 1;
2144 }
2145 return 0;
2146 }
2147
2148 static int do_move_mount(struct path *path, const char *old_name)
2149 {
2150 struct path old_path, parent_path;
2151 struct mount *p;
2152 struct mount *old;
2153 struct mountpoint *mp;
2154 int err;
2155 if (!old_name || !*old_name)
2156 return -EINVAL;
2157 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2158 if (err)
2159 return err;
2160
2161 mp = lock_mount(path);
2162 err = PTR_ERR(mp);
2163 if (IS_ERR(mp))
2164 goto out;
2165
2166 old = real_mount(old_path.mnt);
2167 p = real_mount(path->mnt);
2168
2169 err = -EINVAL;
2170 if (!check_mnt(p) || !check_mnt(old))
2171 goto out1;
2172
2173 if (old->mnt.mnt_flags & MNT_LOCKED)
2174 goto out1;
2175
2176 err = -EINVAL;
2177 if (old_path.dentry != old_path.mnt->mnt_root)
2178 goto out1;
2179
2180 if (!mnt_has_parent(old))
2181 goto out1;
2182
2183 if (d_is_dir(path->dentry) !=
2184 d_is_dir(old_path.dentry))
2185 goto out1;
2186 /*
2187 * Don't move a mount residing in a shared parent.
2188 */
2189 if (IS_MNT_SHARED(old->mnt_parent))
2190 goto out1;
2191 /*
2192 * Don't move a mount tree containing unbindable mounts to a destination
2193 * mount which is shared.
2194 */
2195 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2196 goto out1;
2197 err = -ELOOP;
2198 for (; mnt_has_parent(p); p = p->mnt_parent)
2199 if (p == old)
2200 goto out1;
2201
2202 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2203 if (err)
2204 goto out1;
2205
2206 /* if the mount is moved, it should no longer be expire
2207 * automatically */
2208 list_del_init(&old->mnt_expire);
2209 out1:
2210 unlock_mount(mp);
2211 out:
2212 if (!err)
2213 path_put(&parent_path);
2214 path_put(&old_path);
2215 return err;
2216 }
2217
2218 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2219 {
2220 int err;
2221 const char *subtype = strchr(fstype, '.');
2222 if (subtype) {
2223 subtype++;
2224 err = -EINVAL;
2225 if (!subtype[0])
2226 goto err;
2227 } else
2228 subtype = "";
2229
2230 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2231 err = -ENOMEM;
2232 if (!mnt->mnt_sb->s_subtype)
2233 goto err;
2234 return mnt;
2235
2236 err:
2237 mntput(mnt);
2238 return ERR_PTR(err);
2239 }
2240
2241 /*
2242 * add a mount into a namespace's mount tree
2243 */
2244 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2245 {
2246 struct mountpoint *mp;
2247 struct mount *parent;
2248 int err;
2249
2250 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2251
2252 mp = lock_mount(path);
2253 if (IS_ERR(mp))
2254 return PTR_ERR(mp);
2255
2256 parent = real_mount(path->mnt);
2257 err = -EINVAL;
2258 if (unlikely(!check_mnt(parent))) {
2259 /* that's acceptable only for automounts done in private ns */
2260 if (!(mnt_flags & MNT_SHRINKABLE))
2261 goto unlock;
2262 /* ... and for those we'd better have mountpoint still alive */
2263 if (!parent->mnt_ns)
2264 goto unlock;
2265 }
2266
2267 /* Refuse the same filesystem on the same mount point */
2268 err = -EBUSY;
2269 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2270 path->mnt->mnt_root == path->dentry)
2271 goto unlock;
2272
2273 err = -EINVAL;
2274 if (d_is_symlink(newmnt->mnt.mnt_root))
2275 goto unlock;
2276
2277 newmnt->mnt.mnt_flags = mnt_flags;
2278 err = graft_tree(newmnt, parent, mp);
2279
2280 unlock:
2281 unlock_mount(mp);
2282 return err;
2283 }
2284
2285 /*
2286 * create a new mount for userspace and request it to be added into the
2287 * namespace's tree
2288 */
2289 static int do_new_mount(struct path *path, const char *fstype, int flags,
2290 int mnt_flags, const char *name, void *data)
2291 {
2292 struct file_system_type *type;
2293 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2294 struct vfsmount *mnt;
2295 int err;
2296
2297 if (!fstype)
2298 return -EINVAL;
2299
2300 type = get_fs_type(fstype);
2301 if (!type)
2302 return -ENODEV;
2303
2304 if (user_ns != &init_user_ns) {
2305 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2306 put_filesystem(type);
2307 return -EPERM;
2308 }
2309 /* Only in special cases allow devices from mounts
2310 * created outside the initial user namespace.
2311 */
2312 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2313 flags |= MS_NODEV;
2314 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2315 }
2316 }
2317
2318 mnt = vfs_kern_mount(type, flags, name, data);
2319 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2320 !mnt->mnt_sb->s_subtype)
2321 mnt = fs_set_subtype(mnt, fstype);
2322
2323 put_filesystem(type);
2324 if (IS_ERR(mnt))
2325 return PTR_ERR(mnt);
2326
2327 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2328 if (err)
2329 mntput(mnt);
2330 return err;
2331 }
2332
2333 int finish_automount(struct vfsmount *m, struct path *path)
2334 {
2335 struct mount *mnt = real_mount(m);
2336 int err;
2337 /* The new mount record should have at least 2 refs to prevent it being
2338 * expired before we get a chance to add it
2339 */
2340 BUG_ON(mnt_get_count(mnt) < 2);
2341
2342 if (m->mnt_sb == path->mnt->mnt_sb &&
2343 m->mnt_root == path->dentry) {
2344 err = -ELOOP;
2345 goto fail;
2346 }
2347
2348 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2349 if (!err)
2350 return 0;
2351 fail:
2352 /* remove m from any expiration list it may be on */
2353 if (!list_empty(&mnt->mnt_expire)) {
2354 namespace_lock();
2355 list_del_init(&mnt->mnt_expire);
2356 namespace_unlock();
2357 }
2358 mntput(m);
2359 mntput(m);
2360 return err;
2361 }
2362
2363 /**
2364 * mnt_set_expiry - Put a mount on an expiration list
2365 * @mnt: The mount to list.
2366 * @expiry_list: The list to add the mount to.
2367 */
2368 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2369 {
2370 namespace_lock();
2371
2372 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2373
2374 namespace_unlock();
2375 }
2376 EXPORT_SYMBOL(mnt_set_expiry);
2377
2378 /*
2379 * process a list of expirable mountpoints with the intent of discarding any
2380 * mountpoints that aren't in use and haven't been touched since last we came
2381 * here
2382 */
2383 void mark_mounts_for_expiry(struct list_head *mounts)
2384 {
2385 struct mount *mnt, *next;
2386 LIST_HEAD(graveyard);
2387
2388 if (list_empty(mounts))
2389 return;
2390
2391 namespace_lock();
2392 lock_mount_hash();
2393
2394 /* extract from the expiration list every vfsmount that matches the
2395 * following criteria:
2396 * - only referenced by its parent vfsmount
2397 * - still marked for expiry (marked on the last call here; marks are
2398 * cleared by mntput())
2399 */
2400 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2401 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2402 propagate_mount_busy(mnt, 1))
2403 continue;
2404 list_move(&mnt->mnt_expire, &graveyard);
2405 }
2406 while (!list_empty(&graveyard)) {
2407 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2408 touch_mnt_namespace(mnt->mnt_ns);
2409 umount_tree(mnt, 1);
2410 }
2411 unlock_mount_hash();
2412 namespace_unlock();
2413 }
2414
2415 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2416
2417 /*
2418 * Ripoff of 'select_parent()'
2419 *
2420 * search the list of submounts for a given mountpoint, and move any
2421 * shrinkable submounts to the 'graveyard' list.
2422 */
2423 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2424 {
2425 struct mount *this_parent = parent;
2426 struct list_head *next;
2427 int found = 0;
2428
2429 repeat:
2430 next = this_parent->mnt_mounts.next;
2431 resume:
2432 while (next != &this_parent->mnt_mounts) {
2433 struct list_head *tmp = next;
2434 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2435
2436 next = tmp->next;
2437 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2438 continue;
2439 /*
2440 * Descend a level if the d_mounts list is non-empty.
2441 */
2442 if (!list_empty(&mnt->mnt_mounts)) {
2443 this_parent = mnt;
2444 goto repeat;
2445 }
2446
2447 if (!propagate_mount_busy(mnt, 1)) {
2448 list_move_tail(&mnt->mnt_expire, graveyard);
2449 found++;
2450 }
2451 }
2452 /*
2453 * All done at this level ... ascend and resume the search
2454 */
2455 if (this_parent != parent) {
2456 next = this_parent->mnt_child.next;
2457 this_parent = this_parent->mnt_parent;
2458 goto resume;
2459 }
2460 return found;
2461 }
2462
2463 /*
2464 * process a list of expirable mountpoints with the intent of discarding any
2465 * submounts of a specific parent mountpoint
2466 *
2467 * mount_lock must be held for write
2468 */
2469 static void shrink_submounts(struct mount *mnt)
2470 {
2471 LIST_HEAD(graveyard);
2472 struct mount *m;
2473
2474 /* extract submounts of 'mountpoint' from the expiration list */
2475 while (select_submounts(mnt, &graveyard)) {
2476 while (!list_empty(&graveyard)) {
2477 m = list_first_entry(&graveyard, struct mount,
2478 mnt_expire);
2479 touch_mnt_namespace(m->mnt_ns);
2480 umount_tree(m, 1);
2481 }
2482 }
2483 }
2484
2485 /*
2486 * Some copy_from_user() implementations do not return the exact number of
2487 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2488 * Note that this function differs from copy_from_user() in that it will oops
2489 * on bad values of `to', rather than returning a short copy.
2490 */
2491 static long exact_copy_from_user(void *to, const void __user * from,
2492 unsigned long n)
2493 {
2494 char *t = to;
2495 const char __user *f = from;
2496 char c;
2497
2498 if (!access_ok(VERIFY_READ, from, n))
2499 return n;
2500
2501 while (n) {
2502 if (__get_user(c, f)) {
2503 memset(t, 0, n);
2504 break;
2505 }
2506 *t++ = c;
2507 f++;
2508 n--;
2509 }
2510 return n;
2511 }
2512
2513 int copy_mount_options(const void __user * data, unsigned long *where)
2514 {
2515 int i;
2516 unsigned long page;
2517 unsigned long size;
2518
2519 *where = 0;
2520 if (!data)
2521 return 0;
2522
2523 if (!(page = __get_free_page(GFP_KERNEL)))
2524 return -ENOMEM;
2525
2526 /* We only care that *some* data at the address the user
2527 * gave us is valid. Just in case, we'll zero
2528 * the remainder of the page.
2529 */
2530 /* copy_from_user cannot cross TASK_SIZE ! */
2531 size = TASK_SIZE - (unsigned long)data;
2532 if (size > PAGE_SIZE)
2533 size = PAGE_SIZE;
2534
2535 i = size - exact_copy_from_user((void *)page, data, size);
2536 if (!i) {
2537 free_page(page);
2538 return -EFAULT;
2539 }
2540 if (i != PAGE_SIZE)
2541 memset((char *)page + i, 0, PAGE_SIZE - i);
2542 *where = page;
2543 return 0;
2544 }
2545
2546 char *copy_mount_string(const void __user *data)
2547 {
2548 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2549 }
2550
2551 /*
2552 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2553 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2554 *
2555 * data is a (void *) that can point to any structure up to
2556 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2557 * information (or be NULL).
2558 *
2559 * Pre-0.97 versions of mount() didn't have a flags word.
2560 * When the flags word was introduced its top half was required
2561 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2562 * Therefore, if this magic number is present, it carries no information
2563 * and must be discarded.
2564 */
2565 long do_mount(const char *dev_name, const char __user *dir_name,
2566 const char *type_page, unsigned long flags, void *data_page)
2567 {
2568 struct path path;
2569 int retval = 0;
2570 int mnt_flags = 0;
2571
2572 /* Discard magic */
2573 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2574 flags &= ~MS_MGC_MSK;
2575
2576 /* Basic sanity checks */
2577 if (data_page)
2578 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2579
2580 /* ... and get the mountpoint */
2581 retval = user_path(dir_name, &path);
2582 if (retval)
2583 return retval;
2584
2585 retval = security_sb_mount(dev_name, &path,
2586 type_page, flags, data_page);
2587 if (!retval && !may_mount())
2588 retval = -EPERM;
2589 if (retval)
2590 goto dput_out;
2591
2592 /* Default to relatime unless overriden */
2593 if (!(flags & MS_NOATIME))
2594 mnt_flags |= MNT_RELATIME;
2595
2596 /* Separate the per-mountpoint flags */
2597 if (flags & MS_NOSUID)
2598 mnt_flags |= MNT_NOSUID;
2599 if (flags & MS_NODEV)
2600 mnt_flags |= MNT_NODEV;
2601 if (flags & MS_NOEXEC)
2602 mnt_flags |= MNT_NOEXEC;
2603 if (flags & MS_NOATIME)
2604 mnt_flags |= MNT_NOATIME;
2605 if (flags & MS_NODIRATIME)
2606 mnt_flags |= MNT_NODIRATIME;
2607 if (flags & MS_STRICTATIME)
2608 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2609 if (flags & MS_RDONLY)
2610 mnt_flags |= MNT_READONLY;
2611
2612 /* The default atime for remount is preservation */
2613 if ((flags & MS_REMOUNT) &&
2614 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2615 MS_STRICTATIME)) == 0)) {
2616 mnt_flags &= ~MNT_ATIME_MASK;
2617 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2618 }
2619
2620 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2621 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2622 MS_STRICTATIME);
2623
2624 if (flags & MS_REMOUNT)
2625 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2626 data_page);
2627 else if (flags & MS_BIND)
2628 retval = do_loopback(&path, dev_name, flags & MS_REC);
2629 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2630 retval = do_change_type(&path, flags);
2631 else if (flags & MS_MOVE)
2632 retval = do_move_mount(&path, dev_name);
2633 else
2634 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2635 dev_name, data_page);
2636 dput_out:
2637 path_put(&path);
2638 return retval;
2639 }
2640
2641 static void free_mnt_ns(struct mnt_namespace *ns)
2642 {
2643 ns_free_inum(&ns->ns);
2644 put_user_ns(ns->user_ns);
2645 kfree(ns);
2646 }
2647
2648 /*
2649 * Assign a sequence number so we can detect when we attempt to bind
2650 * mount a reference to an older mount namespace into the current
2651 * mount namespace, preventing reference counting loops. A 64bit
2652 * number incrementing at 10Ghz will take 12,427 years to wrap which
2653 * is effectively never, so we can ignore the possibility.
2654 */
2655 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2656
2657 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2658 {
2659 struct mnt_namespace *new_ns;
2660 int ret;
2661
2662 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2663 if (!new_ns)
2664 return ERR_PTR(-ENOMEM);
2665 ret = ns_alloc_inum(&new_ns->ns);
2666 if (ret) {
2667 kfree(new_ns);
2668 return ERR_PTR(ret);
2669 }
2670 new_ns->ns.ops = &mntns_operations;
2671 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2672 atomic_set(&new_ns->count, 1);
2673 new_ns->root = NULL;
2674 INIT_LIST_HEAD(&new_ns->list);
2675 init_waitqueue_head(&new_ns->poll);
2676 new_ns->event = 0;
2677 new_ns->user_ns = get_user_ns(user_ns);
2678 return new_ns;
2679 }
2680
2681 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2682 struct user_namespace *user_ns, struct fs_struct *new_fs)
2683 {
2684 struct mnt_namespace *new_ns;
2685 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2686 struct mount *p, *q;
2687 struct mount *old;
2688 struct mount *new;
2689 int copy_flags;
2690
2691 BUG_ON(!ns);
2692
2693 if (likely(!(flags & CLONE_NEWNS))) {
2694 get_mnt_ns(ns);
2695 return ns;
2696 }
2697
2698 old = ns->root;
2699
2700 new_ns = alloc_mnt_ns(user_ns);
2701 if (IS_ERR(new_ns))
2702 return new_ns;
2703
2704 namespace_lock();
2705 /* First pass: copy the tree topology */
2706 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2707 if (user_ns != ns->user_ns)
2708 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2709 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2710 if (IS_ERR(new)) {
2711 namespace_unlock();
2712 free_mnt_ns(new_ns);
2713 return ERR_CAST(new);
2714 }
2715 new_ns->root = new;
2716 list_add_tail(&new_ns->list, &new->mnt_list);
2717
2718 /*
2719 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2720 * as belonging to new namespace. We have already acquired a private
2721 * fs_struct, so tsk->fs->lock is not needed.
2722 */
2723 p = old;
2724 q = new;
2725 while (p) {
2726 q->mnt_ns = new_ns;
2727 if (new_fs) {
2728 if (&p->mnt == new_fs->root.mnt) {
2729 new_fs->root.mnt = mntget(&q->mnt);
2730 rootmnt = &p->mnt;
2731 }
2732 if (&p->mnt == new_fs->pwd.mnt) {
2733 new_fs->pwd.mnt = mntget(&q->mnt);
2734 pwdmnt = &p->mnt;
2735 }
2736 }
2737 p = next_mnt(p, old);
2738 q = next_mnt(q, new);
2739 if (!q)
2740 break;
2741 while (p->mnt.mnt_root != q->mnt.mnt_root)
2742 p = next_mnt(p, old);
2743 }
2744 namespace_unlock();
2745
2746 if (rootmnt)
2747 mntput(rootmnt);
2748 if (pwdmnt)
2749 mntput(pwdmnt);
2750
2751 return new_ns;
2752 }
2753
2754 /**
2755 * create_mnt_ns - creates a private namespace and adds a root filesystem
2756 * @mnt: pointer to the new root filesystem mountpoint
2757 */
2758 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2759 {
2760 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2761 if (!IS_ERR(new_ns)) {
2762 struct mount *mnt = real_mount(m);
2763 mnt->mnt_ns = new_ns;
2764 new_ns->root = mnt;
2765 list_add(&mnt->mnt_list, &new_ns->list);
2766 } else {
2767 mntput(m);
2768 }
2769 return new_ns;
2770 }
2771
2772 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2773 {
2774 struct mnt_namespace *ns;
2775 struct super_block *s;
2776 struct path path;
2777 int err;
2778
2779 ns = create_mnt_ns(mnt);
2780 if (IS_ERR(ns))
2781 return ERR_CAST(ns);
2782
2783 err = vfs_path_lookup(mnt->mnt_root, mnt,
2784 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2785
2786 put_mnt_ns(ns);
2787
2788 if (err)
2789 return ERR_PTR(err);
2790
2791 /* trade a vfsmount reference for active sb one */
2792 s = path.mnt->mnt_sb;
2793 atomic_inc(&s->s_active);
2794 mntput(path.mnt);
2795 /* lock the sucker */
2796 down_write(&s->s_umount);
2797 /* ... and return the root of (sub)tree on it */
2798 return path.dentry;
2799 }
2800 EXPORT_SYMBOL(mount_subtree);
2801
2802 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2803 char __user *, type, unsigned long, flags, void __user *, data)
2804 {
2805 int ret;
2806 char *kernel_type;
2807 char *kernel_dev;
2808 unsigned long data_page;
2809
2810 kernel_type = copy_mount_string(type);
2811 ret = PTR_ERR(kernel_type);
2812 if (IS_ERR(kernel_type))
2813 goto out_type;
2814
2815 kernel_dev = copy_mount_string(dev_name);
2816 ret = PTR_ERR(kernel_dev);
2817 if (IS_ERR(kernel_dev))
2818 goto out_dev;
2819
2820 ret = copy_mount_options(data, &data_page);
2821 if (ret < 0)
2822 goto out_data;
2823
2824 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2825 (void *) data_page);
2826
2827 free_page(data_page);
2828 out_data:
2829 kfree(kernel_dev);
2830 out_dev:
2831 kfree(kernel_type);
2832 out_type:
2833 return ret;
2834 }
2835
2836 /*
2837 * Return true if path is reachable from root
2838 *
2839 * namespace_sem or mount_lock is held
2840 */
2841 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2842 const struct path *root)
2843 {
2844 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2845 dentry = mnt->mnt_mountpoint;
2846 mnt = mnt->mnt_parent;
2847 }
2848 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2849 }
2850
2851 int path_is_under(struct path *path1, struct path *path2)
2852 {
2853 int res;
2854 read_seqlock_excl(&mount_lock);
2855 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2856 read_sequnlock_excl(&mount_lock);
2857 return res;
2858 }
2859 EXPORT_SYMBOL(path_is_under);
2860
2861 /*
2862 * pivot_root Semantics:
2863 * Moves the root file system of the current process to the directory put_old,
2864 * makes new_root as the new root file system of the current process, and sets
2865 * root/cwd of all processes which had them on the current root to new_root.
2866 *
2867 * Restrictions:
2868 * The new_root and put_old must be directories, and must not be on the
2869 * same file system as the current process root. The put_old must be
2870 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2871 * pointed to by put_old must yield the same directory as new_root. No other
2872 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2873 *
2874 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2875 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2876 * in this situation.
2877 *
2878 * Notes:
2879 * - we don't move root/cwd if they are not at the root (reason: if something
2880 * cared enough to change them, it's probably wrong to force them elsewhere)
2881 * - it's okay to pick a root that isn't the root of a file system, e.g.
2882 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2883 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2884 * first.
2885 */
2886 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2887 const char __user *, put_old)
2888 {
2889 struct path new, old, parent_path, root_parent, root;
2890 struct mount *new_mnt, *root_mnt, *old_mnt;
2891 struct mountpoint *old_mp, *root_mp;
2892 int error;
2893
2894 if (!may_mount())
2895 return -EPERM;
2896
2897 error = user_path_dir(new_root, &new);
2898 if (error)
2899 goto out0;
2900
2901 error = user_path_dir(put_old, &old);
2902 if (error)
2903 goto out1;
2904
2905 error = security_sb_pivotroot(&old, &new);
2906 if (error)
2907 goto out2;
2908
2909 get_fs_root(current->fs, &root);
2910 old_mp = lock_mount(&old);
2911 error = PTR_ERR(old_mp);
2912 if (IS_ERR(old_mp))
2913 goto out3;
2914
2915 error = -EINVAL;
2916 new_mnt = real_mount(new.mnt);
2917 root_mnt = real_mount(root.mnt);
2918 old_mnt = real_mount(old.mnt);
2919 if (IS_MNT_SHARED(old_mnt) ||
2920 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2921 IS_MNT_SHARED(root_mnt->mnt_parent))
2922 goto out4;
2923 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2924 goto out4;
2925 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2926 goto out4;
2927 error = -ENOENT;
2928 if (d_unlinked(new.dentry))
2929 goto out4;
2930 error = -EBUSY;
2931 if (new_mnt == root_mnt || old_mnt == root_mnt)
2932 goto out4; /* loop, on the same file system */
2933 error = -EINVAL;
2934 if (root.mnt->mnt_root != root.dentry)
2935 goto out4; /* not a mountpoint */
2936 if (!mnt_has_parent(root_mnt))
2937 goto out4; /* not attached */
2938 root_mp = root_mnt->mnt_mp;
2939 if (new.mnt->mnt_root != new.dentry)
2940 goto out4; /* not a mountpoint */
2941 if (!mnt_has_parent(new_mnt))
2942 goto out4; /* not attached */
2943 /* make sure we can reach put_old from new_root */
2944 if (!is_path_reachable(old_mnt, old.dentry, &new))
2945 goto out4;
2946 /* make certain new is below the root */
2947 if (!is_path_reachable(new_mnt, new.dentry, &root))
2948 goto out4;
2949 root_mp->m_count++; /* pin it so it won't go away */
2950 lock_mount_hash();
2951 detach_mnt(new_mnt, &parent_path);
2952 detach_mnt(root_mnt, &root_parent);
2953 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2954 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2955 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2956 }
2957 /* mount old root on put_old */
2958 attach_mnt(root_mnt, old_mnt, old_mp);
2959 /* mount new_root on / */
2960 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2961 touch_mnt_namespace(current->nsproxy->mnt_ns);
2962 /* A moved mount should not expire automatically */
2963 list_del_init(&new_mnt->mnt_expire);
2964 unlock_mount_hash();
2965 chroot_fs_refs(&root, &new);
2966 put_mountpoint(root_mp);
2967 error = 0;
2968 out4:
2969 unlock_mount(old_mp);
2970 if (!error) {
2971 path_put(&root_parent);
2972 path_put(&parent_path);
2973 }
2974 out3:
2975 path_put(&root);
2976 out2:
2977 path_put(&old);
2978 out1:
2979 path_put(&new);
2980 out0:
2981 return error;
2982 }
2983
2984 static void __init init_mount_tree(void)
2985 {
2986 struct vfsmount *mnt;
2987 struct mnt_namespace *ns;
2988 struct path root;
2989 struct file_system_type *type;
2990
2991 type = get_fs_type("rootfs");
2992 if (!type)
2993 panic("Can't find rootfs type");
2994 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2995 put_filesystem(type);
2996 if (IS_ERR(mnt))
2997 panic("Can't create rootfs");
2998
2999 ns = create_mnt_ns(mnt);
3000 if (IS_ERR(ns))
3001 panic("Can't allocate initial namespace");
3002
3003 init_task.nsproxy->mnt_ns = ns;
3004 get_mnt_ns(ns);
3005
3006 root.mnt = mnt;
3007 root.dentry = mnt->mnt_root;
3008 mnt->mnt_flags |= MNT_LOCKED;
3009
3010 set_fs_pwd(current->fs, &root);
3011 set_fs_root(current->fs, &root);
3012 }
3013
3014 void __init mnt_init(void)
3015 {
3016 unsigned u;
3017 int err;
3018
3019 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3020 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3021
3022 mount_hashtable = alloc_large_system_hash("Mount-cache",
3023 sizeof(struct hlist_head),
3024 mhash_entries, 19,
3025 0,
3026 &m_hash_shift, &m_hash_mask, 0, 0);
3027 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3028 sizeof(struct hlist_head),
3029 mphash_entries, 19,
3030 0,
3031 &mp_hash_shift, &mp_hash_mask, 0, 0);
3032
3033 if (!mount_hashtable || !mountpoint_hashtable)
3034 panic("Failed to allocate mount hash table\n");
3035
3036 for (u = 0; u <= m_hash_mask; u++)
3037 INIT_HLIST_HEAD(&mount_hashtable[u]);
3038 for (u = 0; u <= mp_hash_mask; u++)
3039 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3040
3041 kernfs_init();
3042
3043 err = sysfs_init();
3044 if (err)
3045 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3046 __func__, err);
3047 fs_kobj = kobject_create_and_add("fs", NULL);
3048 if (!fs_kobj)
3049 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3050 init_rootfs();
3051 init_mount_tree();
3052 }
3053
3054 void put_mnt_ns(struct mnt_namespace *ns)
3055 {
3056 if (!atomic_dec_and_test(&ns->count))
3057 return;
3058 drop_collected_mounts(&ns->root->mnt);
3059 free_mnt_ns(ns);
3060 }
3061
3062 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3063 {
3064 struct vfsmount *mnt;
3065 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3066 if (!IS_ERR(mnt)) {
3067 /*
3068 * it is a longterm mount, don't release mnt until
3069 * we unmount before file sys is unregistered
3070 */
3071 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3072 }
3073 return mnt;
3074 }
3075 EXPORT_SYMBOL_GPL(kern_mount_data);
3076
3077 void kern_unmount(struct vfsmount *mnt)
3078 {
3079 /* release long term mount so mount point can be released */
3080 if (!IS_ERR_OR_NULL(mnt)) {
3081 real_mount(mnt)->mnt_ns = NULL;
3082 synchronize_rcu(); /* yecchhh... */
3083 mntput(mnt);
3084 }
3085 }
3086 EXPORT_SYMBOL(kern_unmount);
3087
3088 bool our_mnt(struct vfsmount *mnt)
3089 {
3090 return check_mnt(real_mount(mnt));
3091 }
3092
3093 bool current_chrooted(void)
3094 {
3095 /* Does the current process have a non-standard root */
3096 struct path ns_root;
3097 struct path fs_root;
3098 bool chrooted;
3099
3100 /* Find the namespace root */
3101 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3102 ns_root.dentry = ns_root.mnt->mnt_root;
3103 path_get(&ns_root);
3104 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3105 ;
3106
3107 get_fs_root(current->fs, &fs_root);
3108
3109 chrooted = !path_equal(&fs_root, &ns_root);
3110
3111 path_put(&fs_root);
3112 path_put(&ns_root);
3113
3114 return chrooted;
3115 }
3116
3117 bool fs_fully_visible(struct file_system_type *type)
3118 {
3119 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3120 struct mount *mnt;
3121 bool visible = false;
3122
3123 if (unlikely(!ns))
3124 return false;
3125
3126 down_read(&namespace_sem);
3127 list_for_each_entry(mnt, &ns->list, mnt_list) {
3128 struct mount *child;
3129 if (mnt->mnt.mnt_sb->s_type != type)
3130 continue;
3131
3132 /* This mount is not fully visible if there are any child mounts
3133 * that cover anything except for empty directories.
3134 */
3135 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3136 struct inode *inode = child->mnt_mountpoint->d_inode;
3137 if (!S_ISDIR(inode->i_mode))
3138 goto next;
3139 if (inode->i_nlink > 2)
3140 goto next;
3141 }
3142 visible = true;
3143 goto found;
3144 next: ;
3145 }
3146 found:
3147 up_read(&namespace_sem);
3148 return visible;
3149 }
3150
3151 static struct ns_common *mntns_get(struct task_struct *task)
3152 {
3153 struct ns_common *ns = NULL;
3154 struct nsproxy *nsproxy;
3155
3156 task_lock(task);
3157 nsproxy = task->nsproxy;
3158 if (nsproxy) {
3159 ns = &nsproxy->mnt_ns->ns;
3160 get_mnt_ns(to_mnt_ns(ns));
3161 }
3162 task_unlock(task);
3163
3164 return ns;
3165 }
3166
3167 static void mntns_put(struct ns_common *ns)
3168 {
3169 put_mnt_ns(to_mnt_ns(ns));
3170 }
3171
3172 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3173 {
3174 struct fs_struct *fs = current->fs;
3175 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3176 struct path root;
3177
3178 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3179 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3180 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3181 return -EPERM;
3182
3183 if (fs->users != 1)
3184 return -EINVAL;
3185
3186 get_mnt_ns(mnt_ns);
3187 put_mnt_ns(nsproxy->mnt_ns);
3188 nsproxy->mnt_ns = mnt_ns;
3189
3190 /* Find the root */
3191 root.mnt = &mnt_ns->root->mnt;
3192 root.dentry = mnt_ns->root->mnt.mnt_root;
3193 path_get(&root);
3194 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3195 ;
3196
3197 /* Update the pwd and root */
3198 set_fs_pwd(fs, &root);
3199 set_fs_root(fs, &root);
3200
3201 path_put(&root);
3202 return 0;
3203 }
3204
3205 const struct proc_ns_operations mntns_operations = {
3206 .name = "mnt",
3207 .type = CLONE_NEWNS,
3208 .get = mntns_get,
3209 .put = mntns_put,
3210 .install = mntns_install,
3211 };