xref: /openbmc/linux/fs/namespace.c (revision e23feb16)
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/acct.h>		/* acct_auto_close_mnt */
20 #include <linux/init.h>		/* init_rootfs */
21 #include <linux/fs_struct.h>	/* get_fs_root et.al. */
22 #include <linux/fsnotify.h>	/* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include "pnode.h"
27 #include "internal.h"
28 
29 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
30 #define HASH_SIZE (1UL << HASH_SHIFT)
31 
32 static int event;
33 static DEFINE_IDA(mnt_id_ida);
34 static DEFINE_IDA(mnt_group_ida);
35 static DEFINE_SPINLOCK(mnt_id_lock);
36 static int mnt_id_start = 0;
37 static int mnt_group_start = 1;
38 
39 static struct list_head *mount_hashtable __read_mostly;
40 static struct list_head *mountpoint_hashtable __read_mostly;
41 static struct kmem_cache *mnt_cache __read_mostly;
42 static struct rw_semaphore namespace_sem;
43 
44 /* /sys/fs */
45 struct kobject *fs_kobj;
46 EXPORT_SYMBOL_GPL(fs_kobj);
47 
48 /*
49  * vfsmount lock may be taken for read to prevent changes to the
50  * vfsmount hash, ie. during mountpoint lookups or walking back
51  * up the tree.
52  *
53  * It should be taken for write in all cases where the vfsmount
54  * tree or hash is modified or when a vfsmount structure is modified.
55  */
56 DEFINE_BRLOCK(vfsmount_lock);
57 
58 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
59 {
60 	unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
61 	tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
62 	tmp = tmp + (tmp >> HASH_SHIFT);
63 	return tmp & (HASH_SIZE - 1);
64 }
65 
66 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
67 
68 /*
69  * allocation is serialized by namespace_sem, but we need the spinlock to
70  * serialize with freeing.
71  */
72 static int mnt_alloc_id(struct mount *mnt)
73 {
74 	int res;
75 
76 retry:
77 	ida_pre_get(&mnt_id_ida, GFP_KERNEL);
78 	spin_lock(&mnt_id_lock);
79 	res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
80 	if (!res)
81 		mnt_id_start = mnt->mnt_id + 1;
82 	spin_unlock(&mnt_id_lock);
83 	if (res == -EAGAIN)
84 		goto retry;
85 
86 	return res;
87 }
88 
89 static void mnt_free_id(struct mount *mnt)
90 {
91 	int id = mnt->mnt_id;
92 	spin_lock(&mnt_id_lock);
93 	ida_remove(&mnt_id_ida, id);
94 	if (mnt_id_start > id)
95 		mnt_id_start = id;
96 	spin_unlock(&mnt_id_lock);
97 }
98 
99 /*
100  * Allocate a new peer group ID
101  *
102  * mnt_group_ida is protected by namespace_sem
103  */
104 static int mnt_alloc_group_id(struct mount *mnt)
105 {
106 	int res;
107 
108 	if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
109 		return -ENOMEM;
110 
111 	res = ida_get_new_above(&mnt_group_ida,
112 				mnt_group_start,
113 				&mnt->mnt_group_id);
114 	if (!res)
115 		mnt_group_start = mnt->mnt_group_id + 1;
116 
117 	return res;
118 }
119 
120 /*
121  * Release a peer group ID
122  */
123 void mnt_release_group_id(struct mount *mnt)
124 {
125 	int id = mnt->mnt_group_id;
126 	ida_remove(&mnt_group_ida, id);
127 	if (mnt_group_start > id)
128 		mnt_group_start = id;
129 	mnt->mnt_group_id = 0;
130 }
131 
132 /*
133  * vfsmount lock must be held for read
134  */
135 static inline void mnt_add_count(struct mount *mnt, int n)
136 {
137 #ifdef CONFIG_SMP
138 	this_cpu_add(mnt->mnt_pcp->mnt_count, n);
139 #else
140 	preempt_disable();
141 	mnt->mnt_count += n;
142 	preempt_enable();
143 #endif
144 }
145 
146 /*
147  * vfsmount lock must be held for write
148  */
149 unsigned int mnt_get_count(struct mount *mnt)
150 {
151 #ifdef CONFIG_SMP
152 	unsigned int count = 0;
153 	int cpu;
154 
155 	for_each_possible_cpu(cpu) {
156 		count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
157 	}
158 
159 	return count;
160 #else
161 	return mnt->mnt_count;
162 #endif
163 }
164 
165 static struct mount *alloc_vfsmnt(const char *name)
166 {
167 	struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
168 	if (mnt) {
169 		int err;
170 
171 		err = mnt_alloc_id(mnt);
172 		if (err)
173 			goto out_free_cache;
174 
175 		if (name) {
176 			mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
177 			if (!mnt->mnt_devname)
178 				goto out_free_id;
179 		}
180 
181 #ifdef CONFIG_SMP
182 		mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
183 		if (!mnt->mnt_pcp)
184 			goto out_free_devname;
185 
186 		this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
187 #else
188 		mnt->mnt_count = 1;
189 		mnt->mnt_writers = 0;
190 #endif
191 
192 		INIT_LIST_HEAD(&mnt->mnt_hash);
193 		INIT_LIST_HEAD(&mnt->mnt_child);
194 		INIT_LIST_HEAD(&mnt->mnt_mounts);
195 		INIT_LIST_HEAD(&mnt->mnt_list);
196 		INIT_LIST_HEAD(&mnt->mnt_expire);
197 		INIT_LIST_HEAD(&mnt->mnt_share);
198 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
199 		INIT_LIST_HEAD(&mnt->mnt_slave);
200 #ifdef CONFIG_FSNOTIFY
201 		INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
202 #endif
203 	}
204 	return mnt;
205 
206 #ifdef CONFIG_SMP
207 out_free_devname:
208 	kfree(mnt->mnt_devname);
209 #endif
210 out_free_id:
211 	mnt_free_id(mnt);
212 out_free_cache:
213 	kmem_cache_free(mnt_cache, mnt);
214 	return NULL;
215 }
216 
217 /*
218  * Most r/o checks on a fs are for operations that take
219  * discrete amounts of time, like a write() or unlink().
220  * We must keep track of when those operations start
221  * (for permission checks) and when they end, so that
222  * we can determine when writes are able to occur to
223  * a filesystem.
224  */
225 /*
226  * __mnt_is_readonly: check whether a mount is read-only
227  * @mnt: the mount to check for its write status
228  *
229  * This shouldn't be used directly ouside of the VFS.
230  * It does not guarantee that the filesystem will stay
231  * r/w, just that it is right *now*.  This can not and
232  * should not be used in place of IS_RDONLY(inode).
233  * mnt_want/drop_write() will _keep_ the filesystem
234  * r/w.
235  */
236 int __mnt_is_readonly(struct vfsmount *mnt)
237 {
238 	if (mnt->mnt_flags & MNT_READONLY)
239 		return 1;
240 	if (mnt->mnt_sb->s_flags & MS_RDONLY)
241 		return 1;
242 	return 0;
243 }
244 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
245 
246 static inline void mnt_inc_writers(struct mount *mnt)
247 {
248 #ifdef CONFIG_SMP
249 	this_cpu_inc(mnt->mnt_pcp->mnt_writers);
250 #else
251 	mnt->mnt_writers++;
252 #endif
253 }
254 
255 static inline void mnt_dec_writers(struct mount *mnt)
256 {
257 #ifdef CONFIG_SMP
258 	this_cpu_dec(mnt->mnt_pcp->mnt_writers);
259 #else
260 	mnt->mnt_writers--;
261 #endif
262 }
263 
264 static unsigned int mnt_get_writers(struct mount *mnt)
265 {
266 #ifdef CONFIG_SMP
267 	unsigned int count = 0;
268 	int cpu;
269 
270 	for_each_possible_cpu(cpu) {
271 		count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
272 	}
273 
274 	return count;
275 #else
276 	return mnt->mnt_writers;
277 #endif
278 }
279 
280 static int mnt_is_readonly(struct vfsmount *mnt)
281 {
282 	if (mnt->mnt_sb->s_readonly_remount)
283 		return 1;
284 	/* Order wrt setting s_flags/s_readonly_remount in do_remount() */
285 	smp_rmb();
286 	return __mnt_is_readonly(mnt);
287 }
288 
289 /*
290  * Most r/o & frozen checks on a fs are for operations that take discrete
291  * amounts of time, like a write() or unlink().  We must keep track of when
292  * those operations start (for permission checks) and when they end, so that we
293  * can determine when writes are able to occur to a filesystem.
294  */
295 /**
296  * __mnt_want_write - get write access to a mount without freeze protection
297  * @m: the mount on which to take a write
298  *
299  * This tells the low-level filesystem that a write is about to be performed to
300  * it, and makes sure that writes are allowed (mnt it read-write) before
301  * returning success. This operation does not protect against filesystem being
302  * frozen. When the write operation is finished, __mnt_drop_write() must be
303  * called. This is effectively a refcount.
304  */
305 int __mnt_want_write(struct vfsmount *m)
306 {
307 	struct mount *mnt = real_mount(m);
308 	int ret = 0;
309 
310 	preempt_disable();
311 	mnt_inc_writers(mnt);
312 	/*
313 	 * The store to mnt_inc_writers must be visible before we pass
314 	 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
315 	 * incremented count after it has set MNT_WRITE_HOLD.
316 	 */
317 	smp_mb();
318 	while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
319 		cpu_relax();
320 	/*
321 	 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
322 	 * be set to match its requirements. So we must not load that until
323 	 * MNT_WRITE_HOLD is cleared.
324 	 */
325 	smp_rmb();
326 	if (mnt_is_readonly(m)) {
327 		mnt_dec_writers(mnt);
328 		ret = -EROFS;
329 	}
330 	preempt_enable();
331 
332 	return ret;
333 }
334 
335 /**
336  * mnt_want_write - get write access to a mount
337  * @m: the mount on which to take a write
338  *
339  * This tells the low-level filesystem that a write is about to be performed to
340  * it, and makes sure that writes are allowed (mount is read-write, filesystem
341  * is not frozen) before returning success.  When the write operation is
342  * finished, mnt_drop_write() must be called.  This is effectively a refcount.
343  */
344 int mnt_want_write(struct vfsmount *m)
345 {
346 	int ret;
347 
348 	sb_start_write(m->mnt_sb);
349 	ret = __mnt_want_write(m);
350 	if (ret)
351 		sb_end_write(m->mnt_sb);
352 	return ret;
353 }
354 EXPORT_SYMBOL_GPL(mnt_want_write);
355 
356 /**
357  * mnt_clone_write - get write access to a mount
358  * @mnt: the mount on which to take a write
359  *
360  * This is effectively like mnt_want_write, except
361  * it must only be used to take an extra write reference
362  * on a mountpoint that we already know has a write reference
363  * on it. This allows some optimisation.
364  *
365  * After finished, mnt_drop_write must be called as usual to
366  * drop the reference.
367  */
368 int mnt_clone_write(struct vfsmount *mnt)
369 {
370 	/* superblock may be r/o */
371 	if (__mnt_is_readonly(mnt))
372 		return -EROFS;
373 	preempt_disable();
374 	mnt_inc_writers(real_mount(mnt));
375 	preempt_enable();
376 	return 0;
377 }
378 EXPORT_SYMBOL_GPL(mnt_clone_write);
379 
380 /**
381  * __mnt_want_write_file - get write access to a file's mount
382  * @file: the file who's mount on which to take a write
383  *
384  * This is like __mnt_want_write, but it takes a file and can
385  * do some optimisations if the file is open for write already
386  */
387 int __mnt_want_write_file(struct file *file)
388 {
389 	struct inode *inode = file_inode(file);
390 
391 	if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
392 		return __mnt_want_write(file->f_path.mnt);
393 	else
394 		return mnt_clone_write(file->f_path.mnt);
395 }
396 
397 /**
398  * mnt_want_write_file - get write access to a file's mount
399  * @file: the file who's mount on which to take a write
400  *
401  * This is like mnt_want_write, but it takes a file and can
402  * do some optimisations if the file is open for write already
403  */
404 int mnt_want_write_file(struct file *file)
405 {
406 	int ret;
407 
408 	sb_start_write(file->f_path.mnt->mnt_sb);
409 	ret = __mnt_want_write_file(file);
410 	if (ret)
411 		sb_end_write(file->f_path.mnt->mnt_sb);
412 	return ret;
413 }
414 EXPORT_SYMBOL_GPL(mnt_want_write_file);
415 
416 /**
417  * __mnt_drop_write - give up write access to a mount
418  * @mnt: the mount on which to give up write access
419  *
420  * Tells the low-level filesystem that we are done
421  * performing writes to it.  Must be matched with
422  * __mnt_want_write() call above.
423  */
424 void __mnt_drop_write(struct vfsmount *mnt)
425 {
426 	preempt_disable();
427 	mnt_dec_writers(real_mount(mnt));
428 	preempt_enable();
429 }
430 
431 /**
432  * mnt_drop_write - give up write access to a mount
433  * @mnt: the mount on which to give up write access
434  *
435  * Tells the low-level filesystem that we are done performing writes to it and
436  * also allows filesystem to be frozen again.  Must be matched with
437  * mnt_want_write() call above.
438  */
439 void mnt_drop_write(struct vfsmount *mnt)
440 {
441 	__mnt_drop_write(mnt);
442 	sb_end_write(mnt->mnt_sb);
443 }
444 EXPORT_SYMBOL_GPL(mnt_drop_write);
445 
446 void __mnt_drop_write_file(struct file *file)
447 {
448 	__mnt_drop_write(file->f_path.mnt);
449 }
450 
451 void mnt_drop_write_file(struct file *file)
452 {
453 	mnt_drop_write(file->f_path.mnt);
454 }
455 EXPORT_SYMBOL(mnt_drop_write_file);
456 
457 static int mnt_make_readonly(struct mount *mnt)
458 {
459 	int ret = 0;
460 
461 	br_write_lock(&vfsmount_lock);
462 	mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
463 	/*
464 	 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
465 	 * should be visible before we do.
466 	 */
467 	smp_mb();
468 
469 	/*
470 	 * With writers on hold, if this value is zero, then there are
471 	 * definitely no active writers (although held writers may subsequently
472 	 * increment the count, they'll have to wait, and decrement it after
473 	 * seeing MNT_READONLY).
474 	 *
475 	 * It is OK to have counter incremented on one CPU and decremented on
476 	 * another: the sum will add up correctly. The danger would be when we
477 	 * sum up each counter, if we read a counter before it is incremented,
478 	 * but then read another CPU's count which it has been subsequently
479 	 * decremented from -- we would see more decrements than we should.
480 	 * MNT_WRITE_HOLD protects against this scenario, because
481 	 * mnt_want_write first increments count, then smp_mb, then spins on
482 	 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
483 	 * we're counting up here.
484 	 */
485 	if (mnt_get_writers(mnt) > 0)
486 		ret = -EBUSY;
487 	else
488 		mnt->mnt.mnt_flags |= MNT_READONLY;
489 	/*
490 	 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
491 	 * that become unheld will see MNT_READONLY.
492 	 */
493 	smp_wmb();
494 	mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
495 	br_write_unlock(&vfsmount_lock);
496 	return ret;
497 }
498 
499 static void __mnt_unmake_readonly(struct mount *mnt)
500 {
501 	br_write_lock(&vfsmount_lock);
502 	mnt->mnt.mnt_flags &= ~MNT_READONLY;
503 	br_write_unlock(&vfsmount_lock);
504 }
505 
506 int sb_prepare_remount_readonly(struct super_block *sb)
507 {
508 	struct mount *mnt;
509 	int err = 0;
510 
511 	/* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
512 	if (atomic_long_read(&sb->s_remove_count))
513 		return -EBUSY;
514 
515 	br_write_lock(&vfsmount_lock);
516 	list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
517 		if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
518 			mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
519 			smp_mb();
520 			if (mnt_get_writers(mnt) > 0) {
521 				err = -EBUSY;
522 				break;
523 			}
524 		}
525 	}
526 	if (!err && atomic_long_read(&sb->s_remove_count))
527 		err = -EBUSY;
528 
529 	if (!err) {
530 		sb->s_readonly_remount = 1;
531 		smp_wmb();
532 	}
533 	list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
534 		if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
535 			mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
536 	}
537 	br_write_unlock(&vfsmount_lock);
538 
539 	return err;
540 }
541 
542 static void free_vfsmnt(struct mount *mnt)
543 {
544 	kfree(mnt->mnt_devname);
545 	mnt_free_id(mnt);
546 #ifdef CONFIG_SMP
547 	free_percpu(mnt->mnt_pcp);
548 #endif
549 	kmem_cache_free(mnt_cache, mnt);
550 }
551 
552 /*
553  * find the first or last mount at @dentry on vfsmount @mnt depending on
554  * @dir. If @dir is set return the first mount else return the last mount.
555  * vfsmount_lock must be held for read or write.
556  */
557 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
558 			      int dir)
559 {
560 	struct list_head *head = mount_hashtable + hash(mnt, dentry);
561 	struct list_head *tmp = head;
562 	struct mount *p, *found = NULL;
563 
564 	for (;;) {
565 		tmp = dir ? tmp->next : tmp->prev;
566 		p = NULL;
567 		if (tmp == head)
568 			break;
569 		p = list_entry(tmp, struct mount, mnt_hash);
570 		if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
571 			found = p;
572 			break;
573 		}
574 	}
575 	return found;
576 }
577 
578 /*
579  * lookup_mnt - Return the first child mount mounted at path
580  *
581  * "First" means first mounted chronologically.  If you create the
582  * following mounts:
583  *
584  * mount /dev/sda1 /mnt
585  * mount /dev/sda2 /mnt
586  * mount /dev/sda3 /mnt
587  *
588  * Then lookup_mnt() on the base /mnt dentry in the root mount will
589  * return successively the root dentry and vfsmount of /dev/sda1, then
590  * /dev/sda2, then /dev/sda3, then NULL.
591  *
592  * lookup_mnt takes a reference to the found vfsmount.
593  */
594 struct vfsmount *lookup_mnt(struct path *path)
595 {
596 	struct mount *child_mnt;
597 
598 	br_read_lock(&vfsmount_lock);
599 	child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
600 	if (child_mnt) {
601 		mnt_add_count(child_mnt, 1);
602 		br_read_unlock(&vfsmount_lock);
603 		return &child_mnt->mnt;
604 	} else {
605 		br_read_unlock(&vfsmount_lock);
606 		return NULL;
607 	}
608 }
609 
610 static struct mountpoint *new_mountpoint(struct dentry *dentry)
611 {
612 	struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
613 	struct mountpoint *mp;
614 	int ret;
615 
616 	list_for_each_entry(mp, chain, m_hash) {
617 		if (mp->m_dentry == dentry) {
618 			/* might be worth a WARN_ON() */
619 			if (d_unlinked(dentry))
620 				return ERR_PTR(-ENOENT);
621 			mp->m_count++;
622 			return mp;
623 		}
624 	}
625 
626 	mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
627 	if (!mp)
628 		return ERR_PTR(-ENOMEM);
629 
630 	ret = d_set_mounted(dentry);
631 	if (ret) {
632 		kfree(mp);
633 		return ERR_PTR(ret);
634 	}
635 
636 	mp->m_dentry = dentry;
637 	mp->m_count = 1;
638 	list_add(&mp->m_hash, chain);
639 	return mp;
640 }
641 
642 static void put_mountpoint(struct mountpoint *mp)
643 {
644 	if (!--mp->m_count) {
645 		struct dentry *dentry = mp->m_dentry;
646 		spin_lock(&dentry->d_lock);
647 		dentry->d_flags &= ~DCACHE_MOUNTED;
648 		spin_unlock(&dentry->d_lock);
649 		list_del(&mp->m_hash);
650 		kfree(mp);
651 	}
652 }
653 
654 static inline int check_mnt(struct mount *mnt)
655 {
656 	return mnt->mnt_ns == current->nsproxy->mnt_ns;
657 }
658 
659 /*
660  * vfsmount lock must be held for write
661  */
662 static void touch_mnt_namespace(struct mnt_namespace *ns)
663 {
664 	if (ns) {
665 		ns->event = ++event;
666 		wake_up_interruptible(&ns->poll);
667 	}
668 }
669 
670 /*
671  * vfsmount lock must be held for write
672  */
673 static void __touch_mnt_namespace(struct mnt_namespace *ns)
674 {
675 	if (ns && ns->event != event) {
676 		ns->event = event;
677 		wake_up_interruptible(&ns->poll);
678 	}
679 }
680 
681 /*
682  * vfsmount lock must be held for write
683  */
684 static void detach_mnt(struct mount *mnt, struct path *old_path)
685 {
686 	old_path->dentry = mnt->mnt_mountpoint;
687 	old_path->mnt = &mnt->mnt_parent->mnt;
688 	mnt->mnt_parent = mnt;
689 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
690 	list_del_init(&mnt->mnt_child);
691 	list_del_init(&mnt->mnt_hash);
692 	put_mountpoint(mnt->mnt_mp);
693 	mnt->mnt_mp = NULL;
694 }
695 
696 /*
697  * vfsmount lock must be held for write
698  */
699 void mnt_set_mountpoint(struct mount *mnt,
700 			struct mountpoint *mp,
701 			struct mount *child_mnt)
702 {
703 	mp->m_count++;
704 	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
705 	child_mnt->mnt_mountpoint = dget(mp->m_dentry);
706 	child_mnt->mnt_parent = mnt;
707 	child_mnt->mnt_mp = mp;
708 }
709 
710 /*
711  * vfsmount lock must be held for write
712  */
713 static void attach_mnt(struct mount *mnt,
714 			struct mount *parent,
715 			struct mountpoint *mp)
716 {
717 	mnt_set_mountpoint(parent, mp, mnt);
718 	list_add_tail(&mnt->mnt_hash, mount_hashtable +
719 			hash(&parent->mnt, mp->m_dentry));
720 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
721 }
722 
723 /*
724  * vfsmount lock must be held for write
725  */
726 static void commit_tree(struct mount *mnt)
727 {
728 	struct mount *parent = mnt->mnt_parent;
729 	struct mount *m;
730 	LIST_HEAD(head);
731 	struct mnt_namespace *n = parent->mnt_ns;
732 
733 	BUG_ON(parent == mnt);
734 
735 	list_add_tail(&head, &mnt->mnt_list);
736 	list_for_each_entry(m, &head, mnt_list)
737 		m->mnt_ns = n;
738 
739 	list_splice(&head, n->list.prev);
740 
741 	list_add_tail(&mnt->mnt_hash, mount_hashtable +
742 				hash(&parent->mnt, mnt->mnt_mountpoint));
743 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
744 	touch_mnt_namespace(n);
745 }
746 
747 static struct mount *next_mnt(struct mount *p, struct mount *root)
748 {
749 	struct list_head *next = p->mnt_mounts.next;
750 	if (next == &p->mnt_mounts) {
751 		while (1) {
752 			if (p == root)
753 				return NULL;
754 			next = p->mnt_child.next;
755 			if (next != &p->mnt_parent->mnt_mounts)
756 				break;
757 			p = p->mnt_parent;
758 		}
759 	}
760 	return list_entry(next, struct mount, mnt_child);
761 }
762 
763 static struct mount *skip_mnt_tree(struct mount *p)
764 {
765 	struct list_head *prev = p->mnt_mounts.prev;
766 	while (prev != &p->mnt_mounts) {
767 		p = list_entry(prev, struct mount, mnt_child);
768 		prev = p->mnt_mounts.prev;
769 	}
770 	return p;
771 }
772 
773 struct vfsmount *
774 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
775 {
776 	struct mount *mnt;
777 	struct dentry *root;
778 
779 	if (!type)
780 		return ERR_PTR(-ENODEV);
781 
782 	mnt = alloc_vfsmnt(name);
783 	if (!mnt)
784 		return ERR_PTR(-ENOMEM);
785 
786 	if (flags & MS_KERNMOUNT)
787 		mnt->mnt.mnt_flags = MNT_INTERNAL;
788 
789 	root = mount_fs(type, flags, name, data);
790 	if (IS_ERR(root)) {
791 		free_vfsmnt(mnt);
792 		return ERR_CAST(root);
793 	}
794 
795 	mnt->mnt.mnt_root = root;
796 	mnt->mnt.mnt_sb = root->d_sb;
797 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
798 	mnt->mnt_parent = mnt;
799 	br_write_lock(&vfsmount_lock);
800 	list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
801 	br_write_unlock(&vfsmount_lock);
802 	return &mnt->mnt;
803 }
804 EXPORT_SYMBOL_GPL(vfs_kern_mount);
805 
806 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
807 					int flag)
808 {
809 	struct super_block *sb = old->mnt.mnt_sb;
810 	struct mount *mnt;
811 	int err;
812 
813 	mnt = alloc_vfsmnt(old->mnt_devname);
814 	if (!mnt)
815 		return ERR_PTR(-ENOMEM);
816 
817 	if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
818 		mnt->mnt_group_id = 0; /* not a peer of original */
819 	else
820 		mnt->mnt_group_id = old->mnt_group_id;
821 
822 	if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
823 		err = mnt_alloc_group_id(mnt);
824 		if (err)
825 			goto out_free;
826 	}
827 
828 	mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
829 	/* Don't allow unprivileged users to change mount flags */
830 	if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
831 		mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
832 
833 	/* Don't allow unprivileged users to reveal what is under a mount */
834 	if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
835 		mnt->mnt.mnt_flags |= MNT_LOCKED;
836 
837 	atomic_inc(&sb->s_active);
838 	mnt->mnt.mnt_sb = sb;
839 	mnt->mnt.mnt_root = dget(root);
840 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
841 	mnt->mnt_parent = mnt;
842 	br_write_lock(&vfsmount_lock);
843 	list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
844 	br_write_unlock(&vfsmount_lock);
845 
846 	if ((flag & CL_SLAVE) ||
847 	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
848 		list_add(&mnt->mnt_slave, &old->mnt_slave_list);
849 		mnt->mnt_master = old;
850 		CLEAR_MNT_SHARED(mnt);
851 	} else if (!(flag & CL_PRIVATE)) {
852 		if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
853 			list_add(&mnt->mnt_share, &old->mnt_share);
854 		if (IS_MNT_SLAVE(old))
855 			list_add(&mnt->mnt_slave, &old->mnt_slave);
856 		mnt->mnt_master = old->mnt_master;
857 	}
858 	if (flag & CL_MAKE_SHARED)
859 		set_mnt_shared(mnt);
860 
861 	/* stick the duplicate mount on the same expiry list
862 	 * as the original if that was on one */
863 	if (flag & CL_EXPIRE) {
864 		if (!list_empty(&old->mnt_expire))
865 			list_add(&mnt->mnt_expire, &old->mnt_expire);
866 	}
867 
868 	return mnt;
869 
870  out_free:
871 	free_vfsmnt(mnt);
872 	return ERR_PTR(err);
873 }
874 
875 static inline void mntfree(struct mount *mnt)
876 {
877 	struct vfsmount *m = &mnt->mnt;
878 	struct super_block *sb = m->mnt_sb;
879 
880 	/*
881 	 * This probably indicates that somebody messed
882 	 * up a mnt_want/drop_write() pair.  If this
883 	 * happens, the filesystem was probably unable
884 	 * to make r/w->r/o transitions.
885 	 */
886 	/*
887 	 * The locking used to deal with mnt_count decrement provides barriers,
888 	 * so mnt_get_writers() below is safe.
889 	 */
890 	WARN_ON(mnt_get_writers(mnt));
891 	fsnotify_vfsmount_delete(m);
892 	dput(m->mnt_root);
893 	free_vfsmnt(mnt);
894 	deactivate_super(sb);
895 }
896 
897 static void mntput_no_expire(struct mount *mnt)
898 {
899 put_again:
900 #ifdef CONFIG_SMP
901 	br_read_lock(&vfsmount_lock);
902 	if (likely(mnt->mnt_ns)) {
903 		/* shouldn't be the last one */
904 		mnt_add_count(mnt, -1);
905 		br_read_unlock(&vfsmount_lock);
906 		return;
907 	}
908 	br_read_unlock(&vfsmount_lock);
909 
910 	br_write_lock(&vfsmount_lock);
911 	mnt_add_count(mnt, -1);
912 	if (mnt_get_count(mnt)) {
913 		br_write_unlock(&vfsmount_lock);
914 		return;
915 	}
916 #else
917 	mnt_add_count(mnt, -1);
918 	if (likely(mnt_get_count(mnt)))
919 		return;
920 	br_write_lock(&vfsmount_lock);
921 #endif
922 	if (unlikely(mnt->mnt_pinned)) {
923 		mnt_add_count(mnt, mnt->mnt_pinned + 1);
924 		mnt->mnt_pinned = 0;
925 		br_write_unlock(&vfsmount_lock);
926 		acct_auto_close_mnt(&mnt->mnt);
927 		goto put_again;
928 	}
929 
930 	list_del(&mnt->mnt_instance);
931 	br_write_unlock(&vfsmount_lock);
932 	mntfree(mnt);
933 }
934 
935 void mntput(struct vfsmount *mnt)
936 {
937 	if (mnt) {
938 		struct mount *m = real_mount(mnt);
939 		/* avoid cacheline pingpong, hope gcc doesn't get "smart" */
940 		if (unlikely(m->mnt_expiry_mark))
941 			m->mnt_expiry_mark = 0;
942 		mntput_no_expire(m);
943 	}
944 }
945 EXPORT_SYMBOL(mntput);
946 
947 struct vfsmount *mntget(struct vfsmount *mnt)
948 {
949 	if (mnt)
950 		mnt_add_count(real_mount(mnt), 1);
951 	return mnt;
952 }
953 EXPORT_SYMBOL(mntget);
954 
955 void mnt_pin(struct vfsmount *mnt)
956 {
957 	br_write_lock(&vfsmount_lock);
958 	real_mount(mnt)->mnt_pinned++;
959 	br_write_unlock(&vfsmount_lock);
960 }
961 EXPORT_SYMBOL(mnt_pin);
962 
963 void mnt_unpin(struct vfsmount *m)
964 {
965 	struct mount *mnt = real_mount(m);
966 	br_write_lock(&vfsmount_lock);
967 	if (mnt->mnt_pinned) {
968 		mnt_add_count(mnt, 1);
969 		mnt->mnt_pinned--;
970 	}
971 	br_write_unlock(&vfsmount_lock);
972 }
973 EXPORT_SYMBOL(mnt_unpin);
974 
975 static inline void mangle(struct seq_file *m, const char *s)
976 {
977 	seq_escape(m, s, " \t\n\\");
978 }
979 
980 /*
981  * Simple .show_options callback for filesystems which don't want to
982  * implement more complex mount option showing.
983  *
984  * See also save_mount_options().
985  */
986 int generic_show_options(struct seq_file *m, struct dentry *root)
987 {
988 	const char *options;
989 
990 	rcu_read_lock();
991 	options = rcu_dereference(root->d_sb->s_options);
992 
993 	if (options != NULL && options[0]) {
994 		seq_putc(m, ',');
995 		mangle(m, options);
996 	}
997 	rcu_read_unlock();
998 
999 	return 0;
1000 }
1001 EXPORT_SYMBOL(generic_show_options);
1002 
1003 /*
1004  * If filesystem uses generic_show_options(), this function should be
1005  * called from the fill_super() callback.
1006  *
1007  * The .remount_fs callback usually needs to be handled in a special
1008  * way, to make sure, that previous options are not overwritten if the
1009  * remount fails.
1010  *
1011  * Also note, that if the filesystem's .remount_fs function doesn't
1012  * reset all options to their default value, but changes only newly
1013  * given options, then the displayed options will not reflect reality
1014  * any more.
1015  */
1016 void save_mount_options(struct super_block *sb, char *options)
1017 {
1018 	BUG_ON(sb->s_options);
1019 	rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1020 }
1021 EXPORT_SYMBOL(save_mount_options);
1022 
1023 void replace_mount_options(struct super_block *sb, char *options)
1024 {
1025 	char *old = sb->s_options;
1026 	rcu_assign_pointer(sb->s_options, options);
1027 	if (old) {
1028 		synchronize_rcu();
1029 		kfree(old);
1030 	}
1031 }
1032 EXPORT_SYMBOL(replace_mount_options);
1033 
1034 #ifdef CONFIG_PROC_FS
1035 /* iterator; we want it to have access to namespace_sem, thus here... */
1036 static void *m_start(struct seq_file *m, loff_t *pos)
1037 {
1038 	struct proc_mounts *p = proc_mounts(m);
1039 
1040 	down_read(&namespace_sem);
1041 	return seq_list_start(&p->ns->list, *pos);
1042 }
1043 
1044 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1045 {
1046 	struct proc_mounts *p = proc_mounts(m);
1047 
1048 	return seq_list_next(v, &p->ns->list, pos);
1049 }
1050 
1051 static void m_stop(struct seq_file *m, void *v)
1052 {
1053 	up_read(&namespace_sem);
1054 }
1055 
1056 static int m_show(struct seq_file *m, void *v)
1057 {
1058 	struct proc_mounts *p = proc_mounts(m);
1059 	struct mount *r = list_entry(v, struct mount, mnt_list);
1060 	return p->show(m, &r->mnt);
1061 }
1062 
1063 const struct seq_operations mounts_op = {
1064 	.start	= m_start,
1065 	.next	= m_next,
1066 	.stop	= m_stop,
1067 	.show	= m_show,
1068 };
1069 #endif  /* CONFIG_PROC_FS */
1070 
1071 /**
1072  * may_umount_tree - check if a mount tree is busy
1073  * @mnt: root of mount tree
1074  *
1075  * This is called to check if a tree of mounts has any
1076  * open files, pwds, chroots or sub mounts that are
1077  * busy.
1078  */
1079 int may_umount_tree(struct vfsmount *m)
1080 {
1081 	struct mount *mnt = real_mount(m);
1082 	int actual_refs = 0;
1083 	int minimum_refs = 0;
1084 	struct mount *p;
1085 	BUG_ON(!m);
1086 
1087 	/* write lock needed for mnt_get_count */
1088 	br_write_lock(&vfsmount_lock);
1089 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1090 		actual_refs += mnt_get_count(p);
1091 		minimum_refs += 2;
1092 	}
1093 	br_write_unlock(&vfsmount_lock);
1094 
1095 	if (actual_refs > minimum_refs)
1096 		return 0;
1097 
1098 	return 1;
1099 }
1100 
1101 EXPORT_SYMBOL(may_umount_tree);
1102 
1103 /**
1104  * may_umount - check if a mount point is busy
1105  * @mnt: root of mount
1106  *
1107  * This is called to check if a mount point has any
1108  * open files, pwds, chroots or sub mounts. If the
1109  * mount has sub mounts this will return busy
1110  * regardless of whether the sub mounts are busy.
1111  *
1112  * Doesn't take quota and stuff into account. IOW, in some cases it will
1113  * give false negatives. The main reason why it's here is that we need
1114  * a non-destructive way to look for easily umountable filesystems.
1115  */
1116 int may_umount(struct vfsmount *mnt)
1117 {
1118 	int ret = 1;
1119 	down_read(&namespace_sem);
1120 	br_write_lock(&vfsmount_lock);
1121 	if (propagate_mount_busy(real_mount(mnt), 2))
1122 		ret = 0;
1123 	br_write_unlock(&vfsmount_lock);
1124 	up_read(&namespace_sem);
1125 	return ret;
1126 }
1127 
1128 EXPORT_SYMBOL(may_umount);
1129 
1130 static LIST_HEAD(unmounted);	/* protected by namespace_sem */
1131 
1132 static void namespace_unlock(void)
1133 {
1134 	struct mount *mnt;
1135 	LIST_HEAD(head);
1136 
1137 	if (likely(list_empty(&unmounted))) {
1138 		up_write(&namespace_sem);
1139 		return;
1140 	}
1141 
1142 	list_splice_init(&unmounted, &head);
1143 	up_write(&namespace_sem);
1144 
1145 	while (!list_empty(&head)) {
1146 		mnt = list_first_entry(&head, struct mount, mnt_hash);
1147 		list_del_init(&mnt->mnt_hash);
1148 		if (mnt_has_parent(mnt)) {
1149 			struct dentry *dentry;
1150 			struct mount *m;
1151 
1152 			br_write_lock(&vfsmount_lock);
1153 			dentry = mnt->mnt_mountpoint;
1154 			m = mnt->mnt_parent;
1155 			mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1156 			mnt->mnt_parent = mnt;
1157 			m->mnt_ghosts--;
1158 			br_write_unlock(&vfsmount_lock);
1159 			dput(dentry);
1160 			mntput(&m->mnt);
1161 		}
1162 		mntput(&mnt->mnt);
1163 	}
1164 }
1165 
1166 static inline void namespace_lock(void)
1167 {
1168 	down_write(&namespace_sem);
1169 }
1170 
1171 /*
1172  * vfsmount lock must be held for write
1173  * namespace_sem must be held for write
1174  */
1175 void umount_tree(struct mount *mnt, int propagate)
1176 {
1177 	LIST_HEAD(tmp_list);
1178 	struct mount *p;
1179 
1180 	for (p = mnt; p; p = next_mnt(p, mnt))
1181 		list_move(&p->mnt_hash, &tmp_list);
1182 
1183 	if (propagate)
1184 		propagate_umount(&tmp_list);
1185 
1186 	list_for_each_entry(p, &tmp_list, mnt_hash) {
1187 		list_del_init(&p->mnt_expire);
1188 		list_del_init(&p->mnt_list);
1189 		__touch_mnt_namespace(p->mnt_ns);
1190 		p->mnt_ns = NULL;
1191 		list_del_init(&p->mnt_child);
1192 		if (mnt_has_parent(p)) {
1193 			p->mnt_parent->mnt_ghosts++;
1194 			put_mountpoint(p->mnt_mp);
1195 			p->mnt_mp = NULL;
1196 		}
1197 		change_mnt_propagation(p, MS_PRIVATE);
1198 	}
1199 	list_splice(&tmp_list, &unmounted);
1200 }
1201 
1202 static void shrink_submounts(struct mount *mnt);
1203 
1204 static int do_umount(struct mount *mnt, int flags)
1205 {
1206 	struct super_block *sb = mnt->mnt.mnt_sb;
1207 	int retval;
1208 
1209 	retval = security_sb_umount(&mnt->mnt, flags);
1210 	if (retval)
1211 		return retval;
1212 
1213 	/*
1214 	 * Allow userspace to request a mountpoint be expired rather than
1215 	 * unmounting unconditionally. Unmount only happens if:
1216 	 *  (1) the mark is already set (the mark is cleared by mntput())
1217 	 *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1218 	 */
1219 	if (flags & MNT_EXPIRE) {
1220 		if (&mnt->mnt == current->fs->root.mnt ||
1221 		    flags & (MNT_FORCE | MNT_DETACH))
1222 			return -EINVAL;
1223 
1224 		/*
1225 		 * probably don't strictly need the lock here if we examined
1226 		 * all race cases, but it's a slowpath.
1227 		 */
1228 		br_write_lock(&vfsmount_lock);
1229 		if (mnt_get_count(mnt) != 2) {
1230 			br_write_unlock(&vfsmount_lock);
1231 			return -EBUSY;
1232 		}
1233 		br_write_unlock(&vfsmount_lock);
1234 
1235 		if (!xchg(&mnt->mnt_expiry_mark, 1))
1236 			return -EAGAIN;
1237 	}
1238 
1239 	/*
1240 	 * If we may have to abort operations to get out of this
1241 	 * mount, and they will themselves hold resources we must
1242 	 * allow the fs to do things. In the Unix tradition of
1243 	 * 'Gee thats tricky lets do it in userspace' the umount_begin
1244 	 * might fail to complete on the first run through as other tasks
1245 	 * must return, and the like. Thats for the mount program to worry
1246 	 * about for the moment.
1247 	 */
1248 
1249 	if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1250 		sb->s_op->umount_begin(sb);
1251 	}
1252 
1253 	/*
1254 	 * No sense to grab the lock for this test, but test itself looks
1255 	 * somewhat bogus. Suggestions for better replacement?
1256 	 * Ho-hum... In principle, we might treat that as umount + switch
1257 	 * to rootfs. GC would eventually take care of the old vfsmount.
1258 	 * Actually it makes sense, especially if rootfs would contain a
1259 	 * /reboot - static binary that would close all descriptors and
1260 	 * call reboot(9). Then init(8) could umount root and exec /reboot.
1261 	 */
1262 	if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1263 		/*
1264 		 * Special case for "unmounting" root ...
1265 		 * we just try to remount it readonly.
1266 		 */
1267 		down_write(&sb->s_umount);
1268 		if (!(sb->s_flags & MS_RDONLY))
1269 			retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1270 		up_write(&sb->s_umount);
1271 		return retval;
1272 	}
1273 
1274 	namespace_lock();
1275 	br_write_lock(&vfsmount_lock);
1276 	event++;
1277 
1278 	if (!(flags & MNT_DETACH))
1279 		shrink_submounts(mnt);
1280 
1281 	retval = -EBUSY;
1282 	if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1283 		if (!list_empty(&mnt->mnt_list))
1284 			umount_tree(mnt, 1);
1285 		retval = 0;
1286 	}
1287 	br_write_unlock(&vfsmount_lock);
1288 	namespace_unlock();
1289 	return retval;
1290 }
1291 
1292 /*
1293  * Is the caller allowed to modify his namespace?
1294  */
1295 static inline bool may_mount(void)
1296 {
1297 	return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1298 }
1299 
1300 /*
1301  * Now umount can handle mount points as well as block devices.
1302  * This is important for filesystems which use unnamed block devices.
1303  *
1304  * We now support a flag for forced unmount like the other 'big iron'
1305  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1306  */
1307 
1308 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1309 {
1310 	struct path path;
1311 	struct mount *mnt;
1312 	int retval;
1313 	int lookup_flags = 0;
1314 
1315 	if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1316 		return -EINVAL;
1317 
1318 	if (!may_mount())
1319 		return -EPERM;
1320 
1321 	if (!(flags & UMOUNT_NOFOLLOW))
1322 		lookup_flags |= LOOKUP_FOLLOW;
1323 
1324 	retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1325 	if (retval)
1326 		goto out;
1327 	mnt = real_mount(path.mnt);
1328 	retval = -EINVAL;
1329 	if (path.dentry != path.mnt->mnt_root)
1330 		goto dput_and_out;
1331 	if (!check_mnt(mnt))
1332 		goto dput_and_out;
1333 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
1334 		goto dput_and_out;
1335 
1336 	retval = do_umount(mnt, flags);
1337 dput_and_out:
1338 	/* we mustn't call path_put() as that would clear mnt_expiry_mark */
1339 	dput(path.dentry);
1340 	mntput_no_expire(mnt);
1341 out:
1342 	return retval;
1343 }
1344 
1345 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1346 
1347 /*
1348  *	The 2.0 compatible umount. No flags.
1349  */
1350 SYSCALL_DEFINE1(oldumount, char __user *, name)
1351 {
1352 	return sys_umount(name, 0);
1353 }
1354 
1355 #endif
1356 
1357 static bool is_mnt_ns_file(struct dentry *dentry)
1358 {
1359 	/* Is this a proxy for a mount namespace? */
1360 	struct inode *inode = dentry->d_inode;
1361 	struct proc_ns *ei;
1362 
1363 	if (!proc_ns_inode(inode))
1364 		return false;
1365 
1366 	ei = get_proc_ns(inode);
1367 	if (ei->ns_ops != &mntns_operations)
1368 		return false;
1369 
1370 	return true;
1371 }
1372 
1373 static bool mnt_ns_loop(struct dentry *dentry)
1374 {
1375 	/* Could bind mounting the mount namespace inode cause a
1376 	 * mount namespace loop?
1377 	 */
1378 	struct mnt_namespace *mnt_ns;
1379 	if (!is_mnt_ns_file(dentry))
1380 		return false;
1381 
1382 	mnt_ns = get_proc_ns(dentry->d_inode)->ns;
1383 	return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1384 }
1385 
1386 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1387 					int flag)
1388 {
1389 	struct mount *res, *p, *q, *r, *parent;
1390 
1391 	if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1392 		return ERR_PTR(-EINVAL);
1393 
1394 	if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1395 		return ERR_PTR(-EINVAL);
1396 
1397 	res = q = clone_mnt(mnt, dentry, flag);
1398 	if (IS_ERR(q))
1399 		return q;
1400 
1401 	q->mnt.mnt_flags &= ~MNT_LOCKED;
1402 	q->mnt_mountpoint = mnt->mnt_mountpoint;
1403 
1404 	p = mnt;
1405 	list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1406 		struct mount *s;
1407 		if (!is_subdir(r->mnt_mountpoint, dentry))
1408 			continue;
1409 
1410 		for (s = r; s; s = next_mnt(s, r)) {
1411 			if (!(flag & CL_COPY_UNBINDABLE) &&
1412 			    IS_MNT_UNBINDABLE(s)) {
1413 				s = skip_mnt_tree(s);
1414 				continue;
1415 			}
1416 			if (!(flag & CL_COPY_MNT_NS_FILE) &&
1417 			    is_mnt_ns_file(s->mnt.mnt_root)) {
1418 				s = skip_mnt_tree(s);
1419 				continue;
1420 			}
1421 			while (p != s->mnt_parent) {
1422 				p = p->mnt_parent;
1423 				q = q->mnt_parent;
1424 			}
1425 			p = s;
1426 			parent = q;
1427 			q = clone_mnt(p, p->mnt.mnt_root, flag);
1428 			if (IS_ERR(q))
1429 				goto out;
1430 			br_write_lock(&vfsmount_lock);
1431 			list_add_tail(&q->mnt_list, &res->mnt_list);
1432 			attach_mnt(q, parent, p->mnt_mp);
1433 			br_write_unlock(&vfsmount_lock);
1434 		}
1435 	}
1436 	return res;
1437 out:
1438 	if (res) {
1439 		br_write_lock(&vfsmount_lock);
1440 		umount_tree(res, 0);
1441 		br_write_unlock(&vfsmount_lock);
1442 	}
1443 	return q;
1444 }
1445 
1446 /* Caller should check returned pointer for errors */
1447 
1448 struct vfsmount *collect_mounts(struct path *path)
1449 {
1450 	struct mount *tree;
1451 	namespace_lock();
1452 	tree = copy_tree(real_mount(path->mnt), path->dentry,
1453 			 CL_COPY_ALL | CL_PRIVATE);
1454 	namespace_unlock();
1455 	if (IS_ERR(tree))
1456 		return ERR_CAST(tree);
1457 	return &tree->mnt;
1458 }
1459 
1460 void drop_collected_mounts(struct vfsmount *mnt)
1461 {
1462 	namespace_lock();
1463 	br_write_lock(&vfsmount_lock);
1464 	umount_tree(real_mount(mnt), 0);
1465 	br_write_unlock(&vfsmount_lock);
1466 	namespace_unlock();
1467 }
1468 
1469 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1470 		   struct vfsmount *root)
1471 {
1472 	struct mount *mnt;
1473 	int res = f(root, arg);
1474 	if (res)
1475 		return res;
1476 	list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1477 		res = f(&mnt->mnt, arg);
1478 		if (res)
1479 			return res;
1480 	}
1481 	return 0;
1482 }
1483 
1484 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1485 {
1486 	struct mount *p;
1487 
1488 	for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1489 		if (p->mnt_group_id && !IS_MNT_SHARED(p))
1490 			mnt_release_group_id(p);
1491 	}
1492 }
1493 
1494 static int invent_group_ids(struct mount *mnt, bool recurse)
1495 {
1496 	struct mount *p;
1497 
1498 	for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1499 		if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1500 			int err = mnt_alloc_group_id(p);
1501 			if (err) {
1502 				cleanup_group_ids(mnt, p);
1503 				return err;
1504 			}
1505 		}
1506 	}
1507 
1508 	return 0;
1509 }
1510 
1511 /*
1512  *  @source_mnt : mount tree to be attached
1513  *  @nd         : place the mount tree @source_mnt is attached
1514  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1515  *  		   store the parent mount and mountpoint dentry.
1516  *  		   (done when source_mnt is moved)
1517  *
1518  *  NOTE: in the table below explains the semantics when a source mount
1519  *  of a given type is attached to a destination mount of a given type.
1520  * ---------------------------------------------------------------------------
1521  * |         BIND MOUNT OPERATION                                            |
1522  * |**************************************************************************
1523  * | source-->| shared        |       private  |       slave    | unbindable |
1524  * | dest     |               |                |                |            |
1525  * |   |      |               |                |                |            |
1526  * |   v      |               |                |                |            |
1527  * |**************************************************************************
1528  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1529  * |          |               |                |                |            |
1530  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1531  * ***************************************************************************
1532  * A bind operation clones the source mount and mounts the clone on the
1533  * destination mount.
1534  *
1535  * (++)  the cloned mount is propagated to all the mounts in the propagation
1536  * 	 tree of the destination mount and the cloned mount is added to
1537  * 	 the peer group of the source mount.
1538  * (+)   the cloned mount is created under the destination mount and is marked
1539  *       as shared. The cloned mount is added to the peer group of the source
1540  *       mount.
1541  * (+++) the mount is propagated to all the mounts in the propagation tree
1542  *       of the destination mount and the cloned mount is made slave
1543  *       of the same master as that of the source mount. The cloned mount
1544  *       is marked as 'shared and slave'.
1545  * (*)   the cloned mount is made a slave of the same master as that of the
1546  * 	 source mount.
1547  *
1548  * ---------------------------------------------------------------------------
1549  * |         		MOVE MOUNT OPERATION                                 |
1550  * |**************************************************************************
1551  * | source-->| shared        |       private  |       slave    | unbindable |
1552  * | dest     |               |                |                |            |
1553  * |   |      |               |                |                |            |
1554  * |   v      |               |                |                |            |
1555  * |**************************************************************************
1556  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1557  * |          |               |                |                |            |
1558  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1559  * ***************************************************************************
1560  *
1561  * (+)  the mount is moved to the destination. And is then propagated to
1562  * 	all the mounts in the propagation tree of the destination mount.
1563  * (+*)  the mount is moved to the destination.
1564  * (+++)  the mount is moved to the destination and is then propagated to
1565  * 	all the mounts belonging to the destination mount's propagation tree.
1566  * 	the mount is marked as 'shared and slave'.
1567  * (*)	the mount continues to be a slave at the new location.
1568  *
1569  * if the source mount is a tree, the operations explained above is
1570  * applied to each mount in the tree.
1571  * Must be called without spinlocks held, since this function can sleep
1572  * in allocations.
1573  */
1574 static int attach_recursive_mnt(struct mount *source_mnt,
1575 			struct mount *dest_mnt,
1576 			struct mountpoint *dest_mp,
1577 			struct path *parent_path)
1578 {
1579 	LIST_HEAD(tree_list);
1580 	struct mount *child, *p;
1581 	int err;
1582 
1583 	if (IS_MNT_SHARED(dest_mnt)) {
1584 		err = invent_group_ids(source_mnt, true);
1585 		if (err)
1586 			goto out;
1587 	}
1588 	err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1589 	if (err)
1590 		goto out_cleanup_ids;
1591 
1592 	br_write_lock(&vfsmount_lock);
1593 
1594 	if (IS_MNT_SHARED(dest_mnt)) {
1595 		for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1596 			set_mnt_shared(p);
1597 	}
1598 	if (parent_path) {
1599 		detach_mnt(source_mnt, parent_path);
1600 		attach_mnt(source_mnt, dest_mnt, dest_mp);
1601 		touch_mnt_namespace(source_mnt->mnt_ns);
1602 	} else {
1603 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1604 		commit_tree(source_mnt);
1605 	}
1606 
1607 	list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1608 		list_del_init(&child->mnt_hash);
1609 		commit_tree(child);
1610 	}
1611 	br_write_unlock(&vfsmount_lock);
1612 
1613 	return 0;
1614 
1615  out_cleanup_ids:
1616 	if (IS_MNT_SHARED(dest_mnt))
1617 		cleanup_group_ids(source_mnt, NULL);
1618  out:
1619 	return err;
1620 }
1621 
1622 static struct mountpoint *lock_mount(struct path *path)
1623 {
1624 	struct vfsmount *mnt;
1625 	struct dentry *dentry = path->dentry;
1626 retry:
1627 	mutex_lock(&dentry->d_inode->i_mutex);
1628 	if (unlikely(cant_mount(dentry))) {
1629 		mutex_unlock(&dentry->d_inode->i_mutex);
1630 		return ERR_PTR(-ENOENT);
1631 	}
1632 	namespace_lock();
1633 	mnt = lookup_mnt(path);
1634 	if (likely(!mnt)) {
1635 		struct mountpoint *mp = new_mountpoint(dentry);
1636 		if (IS_ERR(mp)) {
1637 			namespace_unlock();
1638 			mutex_unlock(&dentry->d_inode->i_mutex);
1639 			return mp;
1640 		}
1641 		return mp;
1642 	}
1643 	namespace_unlock();
1644 	mutex_unlock(&path->dentry->d_inode->i_mutex);
1645 	path_put(path);
1646 	path->mnt = mnt;
1647 	dentry = path->dentry = dget(mnt->mnt_root);
1648 	goto retry;
1649 }
1650 
1651 static void unlock_mount(struct mountpoint *where)
1652 {
1653 	struct dentry *dentry = where->m_dentry;
1654 	put_mountpoint(where);
1655 	namespace_unlock();
1656 	mutex_unlock(&dentry->d_inode->i_mutex);
1657 }
1658 
1659 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1660 {
1661 	if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1662 		return -EINVAL;
1663 
1664 	if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1665 	      S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1666 		return -ENOTDIR;
1667 
1668 	return attach_recursive_mnt(mnt, p, mp, NULL);
1669 }
1670 
1671 /*
1672  * Sanity check the flags to change_mnt_propagation.
1673  */
1674 
1675 static int flags_to_propagation_type(int flags)
1676 {
1677 	int type = flags & ~(MS_REC | MS_SILENT);
1678 
1679 	/* Fail if any non-propagation flags are set */
1680 	if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1681 		return 0;
1682 	/* Only one propagation flag should be set */
1683 	if (!is_power_of_2(type))
1684 		return 0;
1685 	return type;
1686 }
1687 
1688 /*
1689  * recursively change the type of the mountpoint.
1690  */
1691 static int do_change_type(struct path *path, int flag)
1692 {
1693 	struct mount *m;
1694 	struct mount *mnt = real_mount(path->mnt);
1695 	int recurse = flag & MS_REC;
1696 	int type;
1697 	int err = 0;
1698 
1699 	if (path->dentry != path->mnt->mnt_root)
1700 		return -EINVAL;
1701 
1702 	type = flags_to_propagation_type(flag);
1703 	if (!type)
1704 		return -EINVAL;
1705 
1706 	namespace_lock();
1707 	if (type == MS_SHARED) {
1708 		err = invent_group_ids(mnt, recurse);
1709 		if (err)
1710 			goto out_unlock;
1711 	}
1712 
1713 	br_write_lock(&vfsmount_lock);
1714 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1715 		change_mnt_propagation(m, type);
1716 	br_write_unlock(&vfsmount_lock);
1717 
1718  out_unlock:
1719 	namespace_unlock();
1720 	return err;
1721 }
1722 
1723 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1724 {
1725 	struct mount *child;
1726 	list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1727 		if (!is_subdir(child->mnt_mountpoint, dentry))
1728 			continue;
1729 
1730 		if (child->mnt.mnt_flags & MNT_LOCKED)
1731 			return true;
1732 	}
1733 	return false;
1734 }
1735 
1736 /*
1737  * do loopback mount.
1738  */
1739 static int do_loopback(struct path *path, const char *old_name,
1740 				int recurse)
1741 {
1742 	struct path old_path;
1743 	struct mount *mnt = NULL, *old, *parent;
1744 	struct mountpoint *mp;
1745 	int err;
1746 	if (!old_name || !*old_name)
1747 		return -EINVAL;
1748 	err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1749 	if (err)
1750 		return err;
1751 
1752 	err = -EINVAL;
1753 	if (mnt_ns_loop(old_path.dentry))
1754 		goto out;
1755 
1756 	mp = lock_mount(path);
1757 	err = PTR_ERR(mp);
1758 	if (IS_ERR(mp))
1759 		goto out;
1760 
1761 	old = real_mount(old_path.mnt);
1762 	parent = real_mount(path->mnt);
1763 
1764 	err = -EINVAL;
1765 	if (IS_MNT_UNBINDABLE(old))
1766 		goto out2;
1767 
1768 	if (!check_mnt(parent) || !check_mnt(old))
1769 		goto out2;
1770 
1771 	if (!recurse && has_locked_children(old, old_path.dentry))
1772 		goto out2;
1773 
1774 	if (recurse)
1775 		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
1776 	else
1777 		mnt = clone_mnt(old, old_path.dentry, 0);
1778 
1779 	if (IS_ERR(mnt)) {
1780 		err = PTR_ERR(mnt);
1781 		goto out2;
1782 	}
1783 
1784 	mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1785 
1786 	err = graft_tree(mnt, parent, mp);
1787 	if (err) {
1788 		br_write_lock(&vfsmount_lock);
1789 		umount_tree(mnt, 0);
1790 		br_write_unlock(&vfsmount_lock);
1791 	}
1792 out2:
1793 	unlock_mount(mp);
1794 out:
1795 	path_put(&old_path);
1796 	return err;
1797 }
1798 
1799 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1800 {
1801 	int error = 0;
1802 	int readonly_request = 0;
1803 
1804 	if (ms_flags & MS_RDONLY)
1805 		readonly_request = 1;
1806 	if (readonly_request == __mnt_is_readonly(mnt))
1807 		return 0;
1808 
1809 	if (mnt->mnt_flags & MNT_LOCK_READONLY)
1810 		return -EPERM;
1811 
1812 	if (readonly_request)
1813 		error = mnt_make_readonly(real_mount(mnt));
1814 	else
1815 		__mnt_unmake_readonly(real_mount(mnt));
1816 	return error;
1817 }
1818 
1819 /*
1820  * change filesystem flags. dir should be a physical root of filesystem.
1821  * If you've mounted a non-root directory somewhere and want to do remount
1822  * on it - tough luck.
1823  */
1824 static int do_remount(struct path *path, int flags, int mnt_flags,
1825 		      void *data)
1826 {
1827 	int err;
1828 	struct super_block *sb = path->mnt->mnt_sb;
1829 	struct mount *mnt = real_mount(path->mnt);
1830 
1831 	if (!check_mnt(mnt))
1832 		return -EINVAL;
1833 
1834 	if (path->dentry != path->mnt->mnt_root)
1835 		return -EINVAL;
1836 
1837 	err = security_sb_remount(sb, data);
1838 	if (err)
1839 		return err;
1840 
1841 	down_write(&sb->s_umount);
1842 	if (flags & MS_BIND)
1843 		err = change_mount_flags(path->mnt, flags);
1844 	else if (!capable(CAP_SYS_ADMIN))
1845 		err = -EPERM;
1846 	else
1847 		err = do_remount_sb(sb, flags, data, 0);
1848 	if (!err) {
1849 		br_write_lock(&vfsmount_lock);
1850 		mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1851 		mnt->mnt.mnt_flags = mnt_flags;
1852 		br_write_unlock(&vfsmount_lock);
1853 	}
1854 	up_write(&sb->s_umount);
1855 	if (!err) {
1856 		br_write_lock(&vfsmount_lock);
1857 		touch_mnt_namespace(mnt->mnt_ns);
1858 		br_write_unlock(&vfsmount_lock);
1859 	}
1860 	return err;
1861 }
1862 
1863 static inline int tree_contains_unbindable(struct mount *mnt)
1864 {
1865 	struct mount *p;
1866 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1867 		if (IS_MNT_UNBINDABLE(p))
1868 			return 1;
1869 	}
1870 	return 0;
1871 }
1872 
1873 static int do_move_mount(struct path *path, const char *old_name)
1874 {
1875 	struct path old_path, parent_path;
1876 	struct mount *p;
1877 	struct mount *old;
1878 	struct mountpoint *mp;
1879 	int err;
1880 	if (!old_name || !*old_name)
1881 		return -EINVAL;
1882 	err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1883 	if (err)
1884 		return err;
1885 
1886 	mp = lock_mount(path);
1887 	err = PTR_ERR(mp);
1888 	if (IS_ERR(mp))
1889 		goto out;
1890 
1891 	old = real_mount(old_path.mnt);
1892 	p = real_mount(path->mnt);
1893 
1894 	err = -EINVAL;
1895 	if (!check_mnt(p) || !check_mnt(old))
1896 		goto out1;
1897 
1898 	if (old->mnt.mnt_flags & MNT_LOCKED)
1899 		goto out1;
1900 
1901 	err = -EINVAL;
1902 	if (old_path.dentry != old_path.mnt->mnt_root)
1903 		goto out1;
1904 
1905 	if (!mnt_has_parent(old))
1906 		goto out1;
1907 
1908 	if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1909 	      S_ISDIR(old_path.dentry->d_inode->i_mode))
1910 		goto out1;
1911 	/*
1912 	 * Don't move a mount residing in a shared parent.
1913 	 */
1914 	if (IS_MNT_SHARED(old->mnt_parent))
1915 		goto out1;
1916 	/*
1917 	 * Don't move a mount tree containing unbindable mounts to a destination
1918 	 * mount which is shared.
1919 	 */
1920 	if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1921 		goto out1;
1922 	err = -ELOOP;
1923 	for (; mnt_has_parent(p); p = p->mnt_parent)
1924 		if (p == old)
1925 			goto out1;
1926 
1927 	err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
1928 	if (err)
1929 		goto out1;
1930 
1931 	/* if the mount is moved, it should no longer be expire
1932 	 * automatically */
1933 	list_del_init(&old->mnt_expire);
1934 out1:
1935 	unlock_mount(mp);
1936 out:
1937 	if (!err)
1938 		path_put(&parent_path);
1939 	path_put(&old_path);
1940 	return err;
1941 }
1942 
1943 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1944 {
1945 	int err;
1946 	const char *subtype = strchr(fstype, '.');
1947 	if (subtype) {
1948 		subtype++;
1949 		err = -EINVAL;
1950 		if (!subtype[0])
1951 			goto err;
1952 	} else
1953 		subtype = "";
1954 
1955 	mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1956 	err = -ENOMEM;
1957 	if (!mnt->mnt_sb->s_subtype)
1958 		goto err;
1959 	return mnt;
1960 
1961  err:
1962 	mntput(mnt);
1963 	return ERR_PTR(err);
1964 }
1965 
1966 /*
1967  * add a mount into a namespace's mount tree
1968  */
1969 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1970 {
1971 	struct mountpoint *mp;
1972 	struct mount *parent;
1973 	int err;
1974 
1975 	mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1976 
1977 	mp = lock_mount(path);
1978 	if (IS_ERR(mp))
1979 		return PTR_ERR(mp);
1980 
1981 	parent = real_mount(path->mnt);
1982 	err = -EINVAL;
1983 	if (unlikely(!check_mnt(parent))) {
1984 		/* that's acceptable only for automounts done in private ns */
1985 		if (!(mnt_flags & MNT_SHRINKABLE))
1986 			goto unlock;
1987 		/* ... and for those we'd better have mountpoint still alive */
1988 		if (!parent->mnt_ns)
1989 			goto unlock;
1990 	}
1991 
1992 	/* Refuse the same filesystem on the same mount point */
1993 	err = -EBUSY;
1994 	if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1995 	    path->mnt->mnt_root == path->dentry)
1996 		goto unlock;
1997 
1998 	err = -EINVAL;
1999 	if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2000 		goto unlock;
2001 
2002 	newmnt->mnt.mnt_flags = mnt_flags;
2003 	err = graft_tree(newmnt, parent, mp);
2004 
2005 unlock:
2006 	unlock_mount(mp);
2007 	return err;
2008 }
2009 
2010 /*
2011  * create a new mount for userspace and request it to be added into the
2012  * namespace's tree
2013  */
2014 static int do_new_mount(struct path *path, const char *fstype, int flags,
2015 			int mnt_flags, const char *name, void *data)
2016 {
2017 	struct file_system_type *type;
2018 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2019 	struct vfsmount *mnt;
2020 	int err;
2021 
2022 	if (!fstype)
2023 		return -EINVAL;
2024 
2025 	type = get_fs_type(fstype);
2026 	if (!type)
2027 		return -ENODEV;
2028 
2029 	if (user_ns != &init_user_ns) {
2030 		if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2031 			put_filesystem(type);
2032 			return -EPERM;
2033 		}
2034 		/* Only in special cases allow devices from mounts
2035 		 * created outside the initial user namespace.
2036 		 */
2037 		if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2038 			flags |= MS_NODEV;
2039 			mnt_flags |= MNT_NODEV;
2040 		}
2041 	}
2042 
2043 	mnt = vfs_kern_mount(type, flags, name, data);
2044 	if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2045 	    !mnt->mnt_sb->s_subtype)
2046 		mnt = fs_set_subtype(mnt, fstype);
2047 
2048 	put_filesystem(type);
2049 	if (IS_ERR(mnt))
2050 		return PTR_ERR(mnt);
2051 
2052 	err = do_add_mount(real_mount(mnt), path, mnt_flags);
2053 	if (err)
2054 		mntput(mnt);
2055 	return err;
2056 }
2057 
2058 int finish_automount(struct vfsmount *m, struct path *path)
2059 {
2060 	struct mount *mnt = real_mount(m);
2061 	int err;
2062 	/* The new mount record should have at least 2 refs to prevent it being
2063 	 * expired before we get a chance to add it
2064 	 */
2065 	BUG_ON(mnt_get_count(mnt) < 2);
2066 
2067 	if (m->mnt_sb == path->mnt->mnt_sb &&
2068 	    m->mnt_root == path->dentry) {
2069 		err = -ELOOP;
2070 		goto fail;
2071 	}
2072 
2073 	err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2074 	if (!err)
2075 		return 0;
2076 fail:
2077 	/* remove m from any expiration list it may be on */
2078 	if (!list_empty(&mnt->mnt_expire)) {
2079 		namespace_lock();
2080 		br_write_lock(&vfsmount_lock);
2081 		list_del_init(&mnt->mnt_expire);
2082 		br_write_unlock(&vfsmount_lock);
2083 		namespace_unlock();
2084 	}
2085 	mntput(m);
2086 	mntput(m);
2087 	return err;
2088 }
2089 
2090 /**
2091  * mnt_set_expiry - Put a mount on an expiration list
2092  * @mnt: The mount to list.
2093  * @expiry_list: The list to add the mount to.
2094  */
2095 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2096 {
2097 	namespace_lock();
2098 	br_write_lock(&vfsmount_lock);
2099 
2100 	list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2101 
2102 	br_write_unlock(&vfsmount_lock);
2103 	namespace_unlock();
2104 }
2105 EXPORT_SYMBOL(mnt_set_expiry);
2106 
2107 /*
2108  * process a list of expirable mountpoints with the intent of discarding any
2109  * mountpoints that aren't in use and haven't been touched since last we came
2110  * here
2111  */
2112 void mark_mounts_for_expiry(struct list_head *mounts)
2113 {
2114 	struct mount *mnt, *next;
2115 	LIST_HEAD(graveyard);
2116 
2117 	if (list_empty(mounts))
2118 		return;
2119 
2120 	namespace_lock();
2121 	br_write_lock(&vfsmount_lock);
2122 
2123 	/* extract from the expiration list every vfsmount that matches the
2124 	 * following criteria:
2125 	 * - only referenced by its parent vfsmount
2126 	 * - still marked for expiry (marked on the last call here; marks are
2127 	 *   cleared by mntput())
2128 	 */
2129 	list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2130 		if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2131 			propagate_mount_busy(mnt, 1))
2132 			continue;
2133 		list_move(&mnt->mnt_expire, &graveyard);
2134 	}
2135 	while (!list_empty(&graveyard)) {
2136 		mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2137 		touch_mnt_namespace(mnt->mnt_ns);
2138 		umount_tree(mnt, 1);
2139 	}
2140 	br_write_unlock(&vfsmount_lock);
2141 	namespace_unlock();
2142 }
2143 
2144 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2145 
2146 /*
2147  * Ripoff of 'select_parent()'
2148  *
2149  * search the list of submounts for a given mountpoint, and move any
2150  * shrinkable submounts to the 'graveyard' list.
2151  */
2152 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2153 {
2154 	struct mount *this_parent = parent;
2155 	struct list_head *next;
2156 	int found = 0;
2157 
2158 repeat:
2159 	next = this_parent->mnt_mounts.next;
2160 resume:
2161 	while (next != &this_parent->mnt_mounts) {
2162 		struct list_head *tmp = next;
2163 		struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2164 
2165 		next = tmp->next;
2166 		if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2167 			continue;
2168 		/*
2169 		 * Descend a level if the d_mounts list is non-empty.
2170 		 */
2171 		if (!list_empty(&mnt->mnt_mounts)) {
2172 			this_parent = mnt;
2173 			goto repeat;
2174 		}
2175 
2176 		if (!propagate_mount_busy(mnt, 1)) {
2177 			list_move_tail(&mnt->mnt_expire, graveyard);
2178 			found++;
2179 		}
2180 	}
2181 	/*
2182 	 * All done at this level ... ascend and resume the search
2183 	 */
2184 	if (this_parent != parent) {
2185 		next = this_parent->mnt_child.next;
2186 		this_parent = this_parent->mnt_parent;
2187 		goto resume;
2188 	}
2189 	return found;
2190 }
2191 
2192 /*
2193  * process a list of expirable mountpoints with the intent of discarding any
2194  * submounts of a specific parent mountpoint
2195  *
2196  * vfsmount_lock must be held for write
2197  */
2198 static void shrink_submounts(struct mount *mnt)
2199 {
2200 	LIST_HEAD(graveyard);
2201 	struct mount *m;
2202 
2203 	/* extract submounts of 'mountpoint' from the expiration list */
2204 	while (select_submounts(mnt, &graveyard)) {
2205 		while (!list_empty(&graveyard)) {
2206 			m = list_first_entry(&graveyard, struct mount,
2207 						mnt_expire);
2208 			touch_mnt_namespace(m->mnt_ns);
2209 			umount_tree(m, 1);
2210 		}
2211 	}
2212 }
2213 
2214 /*
2215  * Some copy_from_user() implementations do not return the exact number of
2216  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2217  * Note that this function differs from copy_from_user() in that it will oops
2218  * on bad values of `to', rather than returning a short copy.
2219  */
2220 static long exact_copy_from_user(void *to, const void __user * from,
2221 				 unsigned long n)
2222 {
2223 	char *t = to;
2224 	const char __user *f = from;
2225 	char c;
2226 
2227 	if (!access_ok(VERIFY_READ, from, n))
2228 		return n;
2229 
2230 	while (n) {
2231 		if (__get_user(c, f)) {
2232 			memset(t, 0, n);
2233 			break;
2234 		}
2235 		*t++ = c;
2236 		f++;
2237 		n--;
2238 	}
2239 	return n;
2240 }
2241 
2242 int copy_mount_options(const void __user * data, unsigned long *where)
2243 {
2244 	int i;
2245 	unsigned long page;
2246 	unsigned long size;
2247 
2248 	*where = 0;
2249 	if (!data)
2250 		return 0;
2251 
2252 	if (!(page = __get_free_page(GFP_KERNEL)))
2253 		return -ENOMEM;
2254 
2255 	/* We only care that *some* data at the address the user
2256 	 * gave us is valid.  Just in case, we'll zero
2257 	 * the remainder of the page.
2258 	 */
2259 	/* copy_from_user cannot cross TASK_SIZE ! */
2260 	size = TASK_SIZE - (unsigned long)data;
2261 	if (size > PAGE_SIZE)
2262 		size = PAGE_SIZE;
2263 
2264 	i = size - exact_copy_from_user((void *)page, data, size);
2265 	if (!i) {
2266 		free_page(page);
2267 		return -EFAULT;
2268 	}
2269 	if (i != PAGE_SIZE)
2270 		memset((char *)page + i, 0, PAGE_SIZE - i);
2271 	*where = page;
2272 	return 0;
2273 }
2274 
2275 int copy_mount_string(const void __user *data, char **where)
2276 {
2277 	char *tmp;
2278 
2279 	if (!data) {
2280 		*where = NULL;
2281 		return 0;
2282 	}
2283 
2284 	tmp = strndup_user(data, PAGE_SIZE);
2285 	if (IS_ERR(tmp))
2286 		return PTR_ERR(tmp);
2287 
2288 	*where = tmp;
2289 	return 0;
2290 }
2291 
2292 /*
2293  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2294  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2295  *
2296  * data is a (void *) that can point to any structure up to
2297  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2298  * information (or be NULL).
2299  *
2300  * Pre-0.97 versions of mount() didn't have a flags word.
2301  * When the flags word was introduced its top half was required
2302  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2303  * Therefore, if this magic number is present, it carries no information
2304  * and must be discarded.
2305  */
2306 long do_mount(const char *dev_name, const char *dir_name,
2307 		const char *type_page, unsigned long flags, void *data_page)
2308 {
2309 	struct path path;
2310 	int retval = 0;
2311 	int mnt_flags = 0;
2312 
2313 	/* Discard magic */
2314 	if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2315 		flags &= ~MS_MGC_MSK;
2316 
2317 	/* Basic sanity checks */
2318 
2319 	if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2320 		return -EINVAL;
2321 
2322 	if (data_page)
2323 		((char *)data_page)[PAGE_SIZE - 1] = 0;
2324 
2325 	/* ... and get the mountpoint */
2326 	retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2327 	if (retval)
2328 		return retval;
2329 
2330 	retval = security_sb_mount(dev_name, &path,
2331 				   type_page, flags, data_page);
2332 	if (!retval && !may_mount())
2333 		retval = -EPERM;
2334 	if (retval)
2335 		goto dput_out;
2336 
2337 	/* Default to relatime unless overriden */
2338 	if (!(flags & MS_NOATIME))
2339 		mnt_flags |= MNT_RELATIME;
2340 
2341 	/* Separate the per-mountpoint flags */
2342 	if (flags & MS_NOSUID)
2343 		mnt_flags |= MNT_NOSUID;
2344 	if (flags & MS_NODEV)
2345 		mnt_flags |= MNT_NODEV;
2346 	if (flags & MS_NOEXEC)
2347 		mnt_flags |= MNT_NOEXEC;
2348 	if (flags & MS_NOATIME)
2349 		mnt_flags |= MNT_NOATIME;
2350 	if (flags & MS_NODIRATIME)
2351 		mnt_flags |= MNT_NODIRATIME;
2352 	if (flags & MS_STRICTATIME)
2353 		mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2354 	if (flags & MS_RDONLY)
2355 		mnt_flags |= MNT_READONLY;
2356 
2357 	flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2358 		   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2359 		   MS_STRICTATIME);
2360 
2361 	if (flags & MS_REMOUNT)
2362 		retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2363 				    data_page);
2364 	else if (flags & MS_BIND)
2365 		retval = do_loopback(&path, dev_name, flags & MS_REC);
2366 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2367 		retval = do_change_type(&path, flags);
2368 	else if (flags & MS_MOVE)
2369 		retval = do_move_mount(&path, dev_name);
2370 	else
2371 		retval = do_new_mount(&path, type_page, flags, mnt_flags,
2372 				      dev_name, data_page);
2373 dput_out:
2374 	path_put(&path);
2375 	return retval;
2376 }
2377 
2378 static void free_mnt_ns(struct mnt_namespace *ns)
2379 {
2380 	proc_free_inum(ns->proc_inum);
2381 	put_user_ns(ns->user_ns);
2382 	kfree(ns);
2383 }
2384 
2385 /*
2386  * Assign a sequence number so we can detect when we attempt to bind
2387  * mount a reference to an older mount namespace into the current
2388  * mount namespace, preventing reference counting loops.  A 64bit
2389  * number incrementing at 10Ghz will take 12,427 years to wrap which
2390  * is effectively never, so we can ignore the possibility.
2391  */
2392 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2393 
2394 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2395 {
2396 	struct mnt_namespace *new_ns;
2397 	int ret;
2398 
2399 	new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2400 	if (!new_ns)
2401 		return ERR_PTR(-ENOMEM);
2402 	ret = proc_alloc_inum(&new_ns->proc_inum);
2403 	if (ret) {
2404 		kfree(new_ns);
2405 		return ERR_PTR(ret);
2406 	}
2407 	new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2408 	atomic_set(&new_ns->count, 1);
2409 	new_ns->root = NULL;
2410 	INIT_LIST_HEAD(&new_ns->list);
2411 	init_waitqueue_head(&new_ns->poll);
2412 	new_ns->event = 0;
2413 	new_ns->user_ns = get_user_ns(user_ns);
2414 	return new_ns;
2415 }
2416 
2417 /*
2418  * Allocate a new namespace structure and populate it with contents
2419  * copied from the namespace of the passed in task structure.
2420  */
2421 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2422 		struct user_namespace *user_ns, struct fs_struct *fs)
2423 {
2424 	struct mnt_namespace *new_ns;
2425 	struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2426 	struct mount *p, *q;
2427 	struct mount *old = mnt_ns->root;
2428 	struct mount *new;
2429 	int copy_flags;
2430 
2431 	new_ns = alloc_mnt_ns(user_ns);
2432 	if (IS_ERR(new_ns))
2433 		return new_ns;
2434 
2435 	namespace_lock();
2436 	/* First pass: copy the tree topology */
2437 	copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2438 	if (user_ns != mnt_ns->user_ns)
2439 		copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2440 	new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2441 	if (IS_ERR(new)) {
2442 		namespace_unlock();
2443 		free_mnt_ns(new_ns);
2444 		return ERR_CAST(new);
2445 	}
2446 	new_ns->root = new;
2447 	br_write_lock(&vfsmount_lock);
2448 	list_add_tail(&new_ns->list, &new->mnt_list);
2449 	br_write_unlock(&vfsmount_lock);
2450 
2451 	/*
2452 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2453 	 * as belonging to new namespace.  We have already acquired a private
2454 	 * fs_struct, so tsk->fs->lock is not needed.
2455 	 */
2456 	p = old;
2457 	q = new;
2458 	while (p) {
2459 		q->mnt_ns = new_ns;
2460 		if (fs) {
2461 			if (&p->mnt == fs->root.mnt) {
2462 				fs->root.mnt = mntget(&q->mnt);
2463 				rootmnt = &p->mnt;
2464 			}
2465 			if (&p->mnt == fs->pwd.mnt) {
2466 				fs->pwd.mnt = mntget(&q->mnt);
2467 				pwdmnt = &p->mnt;
2468 			}
2469 		}
2470 		p = next_mnt(p, old);
2471 		q = next_mnt(q, new);
2472 		if (!q)
2473 			break;
2474 		while (p->mnt.mnt_root != q->mnt.mnt_root)
2475 			p = next_mnt(p, old);
2476 	}
2477 	namespace_unlock();
2478 
2479 	if (rootmnt)
2480 		mntput(rootmnt);
2481 	if (pwdmnt)
2482 		mntput(pwdmnt);
2483 
2484 	return new_ns;
2485 }
2486 
2487 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2488 		struct user_namespace *user_ns, struct fs_struct *new_fs)
2489 {
2490 	struct mnt_namespace *new_ns;
2491 
2492 	BUG_ON(!ns);
2493 	get_mnt_ns(ns);
2494 
2495 	if (!(flags & CLONE_NEWNS))
2496 		return ns;
2497 
2498 	new_ns = dup_mnt_ns(ns, user_ns, new_fs);
2499 
2500 	put_mnt_ns(ns);
2501 	return new_ns;
2502 }
2503 
2504 /**
2505  * create_mnt_ns - creates a private namespace and adds a root filesystem
2506  * @mnt: pointer to the new root filesystem mountpoint
2507  */
2508 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2509 {
2510 	struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2511 	if (!IS_ERR(new_ns)) {
2512 		struct mount *mnt = real_mount(m);
2513 		mnt->mnt_ns = new_ns;
2514 		new_ns->root = mnt;
2515 		list_add(&mnt->mnt_list, &new_ns->list);
2516 	} else {
2517 		mntput(m);
2518 	}
2519 	return new_ns;
2520 }
2521 
2522 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2523 {
2524 	struct mnt_namespace *ns;
2525 	struct super_block *s;
2526 	struct path path;
2527 	int err;
2528 
2529 	ns = create_mnt_ns(mnt);
2530 	if (IS_ERR(ns))
2531 		return ERR_CAST(ns);
2532 
2533 	err = vfs_path_lookup(mnt->mnt_root, mnt,
2534 			name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2535 
2536 	put_mnt_ns(ns);
2537 
2538 	if (err)
2539 		return ERR_PTR(err);
2540 
2541 	/* trade a vfsmount reference for active sb one */
2542 	s = path.mnt->mnt_sb;
2543 	atomic_inc(&s->s_active);
2544 	mntput(path.mnt);
2545 	/* lock the sucker */
2546 	down_write(&s->s_umount);
2547 	/* ... and return the root of (sub)tree on it */
2548 	return path.dentry;
2549 }
2550 EXPORT_SYMBOL(mount_subtree);
2551 
2552 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2553 		char __user *, type, unsigned long, flags, void __user *, data)
2554 {
2555 	int ret;
2556 	char *kernel_type;
2557 	struct filename *kernel_dir;
2558 	char *kernel_dev;
2559 	unsigned long data_page;
2560 
2561 	ret = copy_mount_string(type, &kernel_type);
2562 	if (ret < 0)
2563 		goto out_type;
2564 
2565 	kernel_dir = getname(dir_name);
2566 	if (IS_ERR(kernel_dir)) {
2567 		ret = PTR_ERR(kernel_dir);
2568 		goto out_dir;
2569 	}
2570 
2571 	ret = copy_mount_string(dev_name, &kernel_dev);
2572 	if (ret < 0)
2573 		goto out_dev;
2574 
2575 	ret = copy_mount_options(data, &data_page);
2576 	if (ret < 0)
2577 		goto out_data;
2578 
2579 	ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2580 		(void *) data_page);
2581 
2582 	free_page(data_page);
2583 out_data:
2584 	kfree(kernel_dev);
2585 out_dev:
2586 	putname(kernel_dir);
2587 out_dir:
2588 	kfree(kernel_type);
2589 out_type:
2590 	return ret;
2591 }
2592 
2593 /*
2594  * Return true if path is reachable from root
2595  *
2596  * namespace_sem or vfsmount_lock is held
2597  */
2598 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2599 			 const struct path *root)
2600 {
2601 	while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2602 		dentry = mnt->mnt_mountpoint;
2603 		mnt = mnt->mnt_parent;
2604 	}
2605 	return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2606 }
2607 
2608 int path_is_under(struct path *path1, struct path *path2)
2609 {
2610 	int res;
2611 	br_read_lock(&vfsmount_lock);
2612 	res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2613 	br_read_unlock(&vfsmount_lock);
2614 	return res;
2615 }
2616 EXPORT_SYMBOL(path_is_under);
2617 
2618 /*
2619  * pivot_root Semantics:
2620  * Moves the root file system of the current process to the directory put_old,
2621  * makes new_root as the new root file system of the current process, and sets
2622  * root/cwd of all processes which had them on the current root to new_root.
2623  *
2624  * Restrictions:
2625  * The new_root and put_old must be directories, and  must not be on the
2626  * same file  system as the current process root. The put_old  must  be
2627  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2628  * pointed to by put_old must yield the same directory as new_root. No other
2629  * file system may be mounted on put_old. After all, new_root is a mountpoint.
2630  *
2631  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2632  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2633  * in this situation.
2634  *
2635  * Notes:
2636  *  - we don't move root/cwd if they are not at the root (reason: if something
2637  *    cared enough to change them, it's probably wrong to force them elsewhere)
2638  *  - it's okay to pick a root that isn't the root of a file system, e.g.
2639  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2640  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2641  *    first.
2642  */
2643 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2644 		const char __user *, put_old)
2645 {
2646 	struct path new, old, parent_path, root_parent, root;
2647 	struct mount *new_mnt, *root_mnt, *old_mnt;
2648 	struct mountpoint *old_mp, *root_mp;
2649 	int error;
2650 
2651 	if (!may_mount())
2652 		return -EPERM;
2653 
2654 	error = user_path_dir(new_root, &new);
2655 	if (error)
2656 		goto out0;
2657 
2658 	error = user_path_dir(put_old, &old);
2659 	if (error)
2660 		goto out1;
2661 
2662 	error = security_sb_pivotroot(&old, &new);
2663 	if (error)
2664 		goto out2;
2665 
2666 	get_fs_root(current->fs, &root);
2667 	old_mp = lock_mount(&old);
2668 	error = PTR_ERR(old_mp);
2669 	if (IS_ERR(old_mp))
2670 		goto out3;
2671 
2672 	error = -EINVAL;
2673 	new_mnt = real_mount(new.mnt);
2674 	root_mnt = real_mount(root.mnt);
2675 	old_mnt = real_mount(old.mnt);
2676 	if (IS_MNT_SHARED(old_mnt) ||
2677 		IS_MNT_SHARED(new_mnt->mnt_parent) ||
2678 		IS_MNT_SHARED(root_mnt->mnt_parent))
2679 		goto out4;
2680 	if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2681 		goto out4;
2682 	if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2683 		goto out4;
2684 	error = -ENOENT;
2685 	if (d_unlinked(new.dentry))
2686 		goto out4;
2687 	error = -EBUSY;
2688 	if (new_mnt == root_mnt || old_mnt == root_mnt)
2689 		goto out4; /* loop, on the same file system  */
2690 	error = -EINVAL;
2691 	if (root.mnt->mnt_root != root.dentry)
2692 		goto out4; /* not a mountpoint */
2693 	if (!mnt_has_parent(root_mnt))
2694 		goto out4; /* not attached */
2695 	root_mp = root_mnt->mnt_mp;
2696 	if (new.mnt->mnt_root != new.dentry)
2697 		goto out4; /* not a mountpoint */
2698 	if (!mnt_has_parent(new_mnt))
2699 		goto out4; /* not attached */
2700 	/* make sure we can reach put_old from new_root */
2701 	if (!is_path_reachable(old_mnt, old.dentry, &new))
2702 		goto out4;
2703 	root_mp->m_count++; /* pin it so it won't go away */
2704 	br_write_lock(&vfsmount_lock);
2705 	detach_mnt(new_mnt, &parent_path);
2706 	detach_mnt(root_mnt, &root_parent);
2707 	if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2708 		new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2709 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2710 	}
2711 	/* mount old root on put_old */
2712 	attach_mnt(root_mnt, old_mnt, old_mp);
2713 	/* mount new_root on / */
2714 	attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2715 	touch_mnt_namespace(current->nsproxy->mnt_ns);
2716 	br_write_unlock(&vfsmount_lock);
2717 	chroot_fs_refs(&root, &new);
2718 	put_mountpoint(root_mp);
2719 	error = 0;
2720 out4:
2721 	unlock_mount(old_mp);
2722 	if (!error) {
2723 		path_put(&root_parent);
2724 		path_put(&parent_path);
2725 	}
2726 out3:
2727 	path_put(&root);
2728 out2:
2729 	path_put(&old);
2730 out1:
2731 	path_put(&new);
2732 out0:
2733 	return error;
2734 }
2735 
2736 static void __init init_mount_tree(void)
2737 {
2738 	struct vfsmount *mnt;
2739 	struct mnt_namespace *ns;
2740 	struct path root;
2741 	struct file_system_type *type;
2742 
2743 	type = get_fs_type("rootfs");
2744 	if (!type)
2745 		panic("Can't find rootfs type");
2746 	mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2747 	put_filesystem(type);
2748 	if (IS_ERR(mnt))
2749 		panic("Can't create rootfs");
2750 
2751 	ns = create_mnt_ns(mnt);
2752 	if (IS_ERR(ns))
2753 		panic("Can't allocate initial namespace");
2754 
2755 	init_task.nsproxy->mnt_ns = ns;
2756 	get_mnt_ns(ns);
2757 
2758 	root.mnt = mnt;
2759 	root.dentry = mnt->mnt_root;
2760 
2761 	set_fs_pwd(current->fs, &root);
2762 	set_fs_root(current->fs, &root);
2763 }
2764 
2765 void __init mnt_init(void)
2766 {
2767 	unsigned u;
2768 	int err;
2769 
2770 	init_rwsem(&namespace_sem);
2771 
2772 	mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2773 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2774 
2775 	mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2776 	mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2777 
2778 	if (!mount_hashtable || !mountpoint_hashtable)
2779 		panic("Failed to allocate mount hash table\n");
2780 
2781 	printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2782 
2783 	for (u = 0; u < HASH_SIZE; u++)
2784 		INIT_LIST_HEAD(&mount_hashtable[u]);
2785 	for (u = 0; u < HASH_SIZE; u++)
2786 		INIT_LIST_HEAD(&mountpoint_hashtable[u]);
2787 
2788 	br_lock_init(&vfsmount_lock);
2789 
2790 	err = sysfs_init();
2791 	if (err)
2792 		printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2793 			__func__, err);
2794 	fs_kobj = kobject_create_and_add("fs", NULL);
2795 	if (!fs_kobj)
2796 		printk(KERN_WARNING "%s: kobj create error\n", __func__);
2797 	init_rootfs();
2798 	init_mount_tree();
2799 }
2800 
2801 void put_mnt_ns(struct mnt_namespace *ns)
2802 {
2803 	if (!atomic_dec_and_test(&ns->count))
2804 		return;
2805 	namespace_lock();
2806 	br_write_lock(&vfsmount_lock);
2807 	umount_tree(ns->root, 0);
2808 	br_write_unlock(&vfsmount_lock);
2809 	namespace_unlock();
2810 	free_mnt_ns(ns);
2811 }
2812 
2813 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2814 {
2815 	struct vfsmount *mnt;
2816 	mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2817 	if (!IS_ERR(mnt)) {
2818 		/*
2819 		 * it is a longterm mount, don't release mnt until
2820 		 * we unmount before file sys is unregistered
2821 		*/
2822 		real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2823 	}
2824 	return mnt;
2825 }
2826 EXPORT_SYMBOL_GPL(kern_mount_data);
2827 
2828 void kern_unmount(struct vfsmount *mnt)
2829 {
2830 	/* release long term mount so mount point can be released */
2831 	if (!IS_ERR_OR_NULL(mnt)) {
2832 		br_write_lock(&vfsmount_lock);
2833 		real_mount(mnt)->mnt_ns = NULL;
2834 		br_write_unlock(&vfsmount_lock);
2835 		mntput(mnt);
2836 	}
2837 }
2838 EXPORT_SYMBOL(kern_unmount);
2839 
2840 bool our_mnt(struct vfsmount *mnt)
2841 {
2842 	return check_mnt(real_mount(mnt));
2843 }
2844 
2845 bool current_chrooted(void)
2846 {
2847 	/* Does the current process have a non-standard root */
2848 	struct path ns_root;
2849 	struct path fs_root;
2850 	bool chrooted;
2851 
2852 	/* Find the namespace root */
2853 	ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2854 	ns_root.dentry = ns_root.mnt->mnt_root;
2855 	path_get(&ns_root);
2856 	while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2857 		;
2858 
2859 	get_fs_root(current->fs, &fs_root);
2860 
2861 	chrooted = !path_equal(&fs_root, &ns_root);
2862 
2863 	path_put(&fs_root);
2864 	path_put(&ns_root);
2865 
2866 	return chrooted;
2867 }
2868 
2869 bool fs_fully_visible(struct file_system_type *type)
2870 {
2871 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2872 	struct mount *mnt;
2873 	bool visible = false;
2874 
2875 	if (unlikely(!ns))
2876 		return false;
2877 
2878 	namespace_lock();
2879 	list_for_each_entry(mnt, &ns->list, mnt_list) {
2880 		struct mount *child;
2881 		if (mnt->mnt.mnt_sb->s_type != type)
2882 			continue;
2883 
2884 		/* This mount is not fully visible if there are any child mounts
2885 		 * that cover anything except for empty directories.
2886 		 */
2887 		list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2888 			struct inode *inode = child->mnt_mountpoint->d_inode;
2889 			if (!S_ISDIR(inode->i_mode))
2890 				goto next;
2891 			if (inode->i_nlink != 2)
2892 				goto next;
2893 		}
2894 		visible = true;
2895 		goto found;
2896 	next:	;
2897 	}
2898 found:
2899 	namespace_unlock();
2900 	return visible;
2901 }
2902 
2903 static void *mntns_get(struct task_struct *task)
2904 {
2905 	struct mnt_namespace *ns = NULL;
2906 	struct nsproxy *nsproxy;
2907 
2908 	rcu_read_lock();
2909 	nsproxy = task_nsproxy(task);
2910 	if (nsproxy) {
2911 		ns = nsproxy->mnt_ns;
2912 		get_mnt_ns(ns);
2913 	}
2914 	rcu_read_unlock();
2915 
2916 	return ns;
2917 }
2918 
2919 static void mntns_put(void *ns)
2920 {
2921 	put_mnt_ns(ns);
2922 }
2923 
2924 static int mntns_install(struct nsproxy *nsproxy, void *ns)
2925 {
2926 	struct fs_struct *fs = current->fs;
2927 	struct mnt_namespace *mnt_ns = ns;
2928 	struct path root;
2929 
2930 	if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2931 	    !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
2932 	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
2933 		return -EPERM;
2934 
2935 	if (fs->users != 1)
2936 		return -EINVAL;
2937 
2938 	get_mnt_ns(mnt_ns);
2939 	put_mnt_ns(nsproxy->mnt_ns);
2940 	nsproxy->mnt_ns = mnt_ns;
2941 
2942 	/* Find the root */
2943 	root.mnt    = &mnt_ns->root->mnt;
2944 	root.dentry = mnt_ns->root->mnt.mnt_root;
2945 	path_get(&root);
2946 	while(d_mountpoint(root.dentry) && follow_down_one(&root))
2947 		;
2948 
2949 	/* Update the pwd and root */
2950 	set_fs_pwd(fs, &root);
2951 	set_fs_root(fs, &root);
2952 
2953 	path_put(&root);
2954 	return 0;
2955 }
2956 
2957 static unsigned int mntns_inum(void *ns)
2958 {
2959 	struct mnt_namespace *mnt_ns = ns;
2960 	return mnt_ns->proc_inum;
2961 }
2962 
2963 const struct proc_ns_operations mntns_operations = {
2964 	.name		= "mnt",
2965 	.type		= CLONE_NEWNS,
2966 	.get		= mntns_get,
2967 	.put		= mntns_put,
2968 	.install	= mntns_install,
2969 	.inum		= mntns_inum,
2970 };
2971