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