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