xref: /openbmc/linux/fs/namespace.c (revision ae3473231e77a3f1909d48cd144cebe5e1d049b3)
1 /*
2  *  linux/fs/namespace.c
3  *
4  * (C) Copyright Al Viro 2000, 2001
5  *	Released under GPL v2.
6  *
7  * Based on code from fs/super.c, copyright Linus Torvalds and others.
8  * Heavily rewritten.
9  */
10 
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/init.h>		/* init_rootfs */
20 #include <linux/fs_struct.h>	/* get_fs_root et.al. */
21 #include <linux/fsnotify.h>	/* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
27 #include "pnode.h"
28 #include "internal.h"
29 
30 /* Maximum number of mounts in a mount namespace */
31 unsigned int sysctl_mount_max __read_mostly = 100000;
32 
33 static unsigned int m_hash_mask __read_mostly;
34 static unsigned int m_hash_shift __read_mostly;
35 static unsigned int mp_hash_mask __read_mostly;
36 static unsigned int mp_hash_shift __read_mostly;
37 
38 static __initdata unsigned long mhash_entries;
39 static int __init set_mhash_entries(char *str)
40 {
41 	if (!str)
42 		return 0;
43 	mhash_entries = simple_strtoul(str, &str, 0);
44 	return 1;
45 }
46 __setup("mhash_entries=", set_mhash_entries);
47 
48 static __initdata unsigned long mphash_entries;
49 static int __init set_mphash_entries(char *str)
50 {
51 	if (!str)
52 		return 0;
53 	mphash_entries = simple_strtoul(str, &str, 0);
54 	return 1;
55 }
56 __setup("mphash_entries=", set_mphash_entries);
57 
58 static u64 event;
59 static DEFINE_IDA(mnt_id_ida);
60 static DEFINE_IDA(mnt_group_ida);
61 static DEFINE_SPINLOCK(mnt_id_lock);
62 static int mnt_id_start = 0;
63 static int mnt_group_start = 1;
64 
65 static struct hlist_head *mount_hashtable __read_mostly;
66 static struct hlist_head *mountpoint_hashtable __read_mostly;
67 static struct kmem_cache *mnt_cache __read_mostly;
68 static DECLARE_RWSEM(namespace_sem);
69 
70 /* /sys/fs */
71 struct kobject *fs_kobj;
72 EXPORT_SYMBOL_GPL(fs_kobj);
73 
74 /*
75  * vfsmount lock may be taken for read to prevent changes to the
76  * vfsmount hash, ie. during mountpoint lookups or walking back
77  * up the tree.
78  *
79  * It should be taken for write in all cases where the vfsmount
80  * tree or hash is modified or when a vfsmount structure is modified.
81  */
82 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
83 
84 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
85 {
86 	unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
87 	tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
88 	tmp = tmp + (tmp >> m_hash_shift);
89 	return &mount_hashtable[tmp & m_hash_mask];
90 }
91 
92 static inline struct hlist_head *mp_hash(struct dentry *dentry)
93 {
94 	unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
95 	tmp = tmp + (tmp >> mp_hash_shift);
96 	return &mountpoint_hashtable[tmp & mp_hash_mask];
97 }
98 
99 static int mnt_alloc_id(struct mount *mnt)
100 {
101 	int res;
102 
103 retry:
104 	ida_pre_get(&mnt_id_ida, GFP_KERNEL);
105 	spin_lock(&mnt_id_lock);
106 	res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
107 	if (!res)
108 		mnt_id_start = mnt->mnt_id + 1;
109 	spin_unlock(&mnt_id_lock);
110 	if (res == -EAGAIN)
111 		goto retry;
112 
113 	return res;
114 }
115 
116 static void mnt_free_id(struct mount *mnt)
117 {
118 	int id = mnt->mnt_id;
119 	spin_lock(&mnt_id_lock);
120 	ida_remove(&mnt_id_ida, id);
121 	if (mnt_id_start > id)
122 		mnt_id_start = id;
123 	spin_unlock(&mnt_id_lock);
124 }
125 
126 /*
127  * Allocate a new peer group ID
128  *
129  * mnt_group_ida is protected by namespace_sem
130  */
131 static int mnt_alloc_group_id(struct mount *mnt)
132 {
133 	int res;
134 
135 	if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
136 		return -ENOMEM;
137 
138 	res = ida_get_new_above(&mnt_group_ida,
139 				mnt_group_start,
140 				&mnt->mnt_group_id);
141 	if (!res)
142 		mnt_group_start = mnt->mnt_group_id + 1;
143 
144 	return res;
145 }
146 
147 /*
148  * Release a peer group ID
149  */
150 void mnt_release_group_id(struct mount *mnt)
151 {
152 	int id = mnt->mnt_group_id;
153 	ida_remove(&mnt_group_ida, id);
154 	if (mnt_group_start > id)
155 		mnt_group_start = id;
156 	mnt->mnt_group_id = 0;
157 }
158 
159 /*
160  * vfsmount lock must be held for read
161  */
162 static inline void mnt_add_count(struct mount *mnt, int n)
163 {
164 #ifdef CONFIG_SMP
165 	this_cpu_add(mnt->mnt_pcp->mnt_count, n);
166 #else
167 	preempt_disable();
168 	mnt->mnt_count += n;
169 	preempt_enable();
170 #endif
171 }
172 
173 /*
174  * vfsmount lock must be held for write
175  */
176 unsigned int mnt_get_count(struct mount *mnt)
177 {
178 #ifdef CONFIG_SMP
179 	unsigned int count = 0;
180 	int cpu;
181 
182 	for_each_possible_cpu(cpu) {
183 		count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
184 	}
185 
186 	return count;
187 #else
188 	return mnt->mnt_count;
189 #endif
190 }
191 
192 static void drop_mountpoint(struct fs_pin *p)
193 {
194 	struct mount *m = container_of(p, struct mount, mnt_umount);
195 	dput(m->mnt_ex_mountpoint);
196 	pin_remove(p);
197 	mntput(&m->mnt);
198 }
199 
200 static struct mount *alloc_vfsmnt(const char *name)
201 {
202 	struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
203 	if (mnt) {
204 		int err;
205 
206 		err = mnt_alloc_id(mnt);
207 		if (err)
208 			goto out_free_cache;
209 
210 		if (name) {
211 			mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
212 			if (!mnt->mnt_devname)
213 				goto out_free_id;
214 		}
215 
216 #ifdef CONFIG_SMP
217 		mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
218 		if (!mnt->mnt_pcp)
219 			goto out_free_devname;
220 
221 		this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
222 #else
223 		mnt->mnt_count = 1;
224 		mnt->mnt_writers = 0;
225 #endif
226 
227 		INIT_HLIST_NODE(&mnt->mnt_hash);
228 		INIT_LIST_HEAD(&mnt->mnt_child);
229 		INIT_LIST_HEAD(&mnt->mnt_mounts);
230 		INIT_LIST_HEAD(&mnt->mnt_list);
231 		INIT_LIST_HEAD(&mnt->mnt_expire);
232 		INIT_LIST_HEAD(&mnt->mnt_share);
233 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
234 		INIT_LIST_HEAD(&mnt->mnt_slave);
235 		INIT_HLIST_NODE(&mnt->mnt_mp_list);
236 #ifdef CONFIG_FSNOTIFY
237 		INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
238 #endif
239 		init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
240 	}
241 	return mnt;
242 
243 #ifdef CONFIG_SMP
244 out_free_devname:
245 	kfree_const(mnt->mnt_devname);
246 #endif
247 out_free_id:
248 	mnt_free_id(mnt);
249 out_free_cache:
250 	kmem_cache_free(mnt_cache, mnt);
251 	return NULL;
252 }
253 
254 /*
255  * Most r/o checks on a fs are for operations that take
256  * discrete amounts of time, like a write() or unlink().
257  * We must keep track of when those operations start
258  * (for permission checks) and when they end, so that
259  * we can determine when writes are able to occur to
260  * a filesystem.
261  */
262 /*
263  * __mnt_is_readonly: check whether a mount is read-only
264  * @mnt: the mount to check for its write status
265  *
266  * This shouldn't be used directly ouside of the VFS.
267  * It does not guarantee that the filesystem will stay
268  * r/w, just that it is right *now*.  This can not and
269  * should not be used in place of IS_RDONLY(inode).
270  * mnt_want/drop_write() will _keep_ the filesystem
271  * r/w.
272  */
273 int __mnt_is_readonly(struct vfsmount *mnt)
274 {
275 	if (mnt->mnt_flags & MNT_READONLY)
276 		return 1;
277 	if (mnt->mnt_sb->s_flags & MS_RDONLY)
278 		return 1;
279 	return 0;
280 }
281 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
282 
283 static inline void mnt_inc_writers(struct mount *mnt)
284 {
285 #ifdef CONFIG_SMP
286 	this_cpu_inc(mnt->mnt_pcp->mnt_writers);
287 #else
288 	mnt->mnt_writers++;
289 #endif
290 }
291 
292 static inline void mnt_dec_writers(struct mount *mnt)
293 {
294 #ifdef CONFIG_SMP
295 	this_cpu_dec(mnt->mnt_pcp->mnt_writers);
296 #else
297 	mnt->mnt_writers--;
298 #endif
299 }
300 
301 static unsigned int mnt_get_writers(struct mount *mnt)
302 {
303 #ifdef CONFIG_SMP
304 	unsigned int count = 0;
305 	int cpu;
306 
307 	for_each_possible_cpu(cpu) {
308 		count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
309 	}
310 
311 	return count;
312 #else
313 	return mnt->mnt_writers;
314 #endif
315 }
316 
317 static int mnt_is_readonly(struct vfsmount *mnt)
318 {
319 	if (mnt->mnt_sb->s_readonly_remount)
320 		return 1;
321 	/* Order wrt setting s_flags/s_readonly_remount in do_remount() */
322 	smp_rmb();
323 	return __mnt_is_readonly(mnt);
324 }
325 
326 /*
327  * Most r/o & frozen checks on a fs are for operations that take discrete
328  * amounts of time, like a write() or unlink().  We must keep track of when
329  * those operations start (for permission checks) and when they end, so that we
330  * can determine when writes are able to occur to a filesystem.
331  */
332 /**
333  * __mnt_want_write - get write access to a mount without freeze protection
334  * @m: the mount on which to take a write
335  *
336  * This tells the low-level filesystem that a write is about to be performed to
337  * it, and makes sure that writes are allowed (mnt it read-write) before
338  * returning success. This operation does not protect against filesystem being
339  * frozen. When the write operation is finished, __mnt_drop_write() must be
340  * called. This is effectively a refcount.
341  */
342 int __mnt_want_write(struct vfsmount *m)
343 {
344 	struct mount *mnt = real_mount(m);
345 	int ret = 0;
346 
347 	preempt_disable();
348 	mnt_inc_writers(mnt);
349 	/*
350 	 * The store to mnt_inc_writers must be visible before we pass
351 	 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
352 	 * incremented count after it has set MNT_WRITE_HOLD.
353 	 */
354 	smp_mb();
355 	while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
356 		cpu_relax();
357 	/*
358 	 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
359 	 * be set to match its requirements. So we must not load that until
360 	 * MNT_WRITE_HOLD is cleared.
361 	 */
362 	smp_rmb();
363 	if (mnt_is_readonly(m)) {
364 		mnt_dec_writers(mnt);
365 		ret = -EROFS;
366 	}
367 	preempt_enable();
368 
369 	return ret;
370 }
371 
372 /**
373  * mnt_want_write - get write access to a mount
374  * @m: the mount on which to take a write
375  *
376  * This tells the low-level filesystem that a write is about to be performed to
377  * it, and makes sure that writes are allowed (mount is read-write, filesystem
378  * is not frozen) before returning success.  When the write operation is
379  * finished, mnt_drop_write() must be called.  This is effectively a refcount.
380  */
381 int mnt_want_write(struct vfsmount *m)
382 {
383 	int ret;
384 
385 	sb_start_write(m->mnt_sb);
386 	ret = __mnt_want_write(m);
387 	if (ret)
388 		sb_end_write(m->mnt_sb);
389 	return ret;
390 }
391 EXPORT_SYMBOL_GPL(mnt_want_write);
392 
393 /**
394  * mnt_clone_write - get write access to a mount
395  * @mnt: the mount on which to take a write
396  *
397  * This is effectively like mnt_want_write, except
398  * it must only be used to take an extra write reference
399  * on a mountpoint that we already know has a write reference
400  * on it. This allows some optimisation.
401  *
402  * After finished, mnt_drop_write must be called as usual to
403  * drop the reference.
404  */
405 int mnt_clone_write(struct vfsmount *mnt)
406 {
407 	/* superblock may be r/o */
408 	if (__mnt_is_readonly(mnt))
409 		return -EROFS;
410 	preempt_disable();
411 	mnt_inc_writers(real_mount(mnt));
412 	preempt_enable();
413 	return 0;
414 }
415 EXPORT_SYMBOL_GPL(mnt_clone_write);
416 
417 /**
418  * __mnt_want_write_file - get write access to a file's mount
419  * @file: the file who's mount on which to take a write
420  *
421  * This is like __mnt_want_write, but it takes a file and can
422  * do some optimisations if the file is open for write already
423  */
424 int __mnt_want_write_file(struct file *file)
425 {
426 	if (!(file->f_mode & FMODE_WRITER))
427 		return __mnt_want_write(file->f_path.mnt);
428 	else
429 		return mnt_clone_write(file->f_path.mnt);
430 }
431 
432 /**
433  * mnt_want_write_file - get write access to a file's mount
434  * @file: the file who's mount on which to take a write
435  *
436  * This is like mnt_want_write, but it takes a file and can
437  * do some optimisations if the file is open for write already
438  */
439 int mnt_want_write_file(struct file *file)
440 {
441 	int ret;
442 
443 	sb_start_write(file->f_path.mnt->mnt_sb);
444 	ret = __mnt_want_write_file(file);
445 	if (ret)
446 		sb_end_write(file->f_path.mnt->mnt_sb);
447 	return ret;
448 }
449 EXPORT_SYMBOL_GPL(mnt_want_write_file);
450 
451 /**
452  * __mnt_drop_write - give up write access to a mount
453  * @mnt: the mount on which to give up write access
454  *
455  * Tells the low-level filesystem that we are done
456  * performing writes to it.  Must be matched with
457  * __mnt_want_write() call above.
458  */
459 void __mnt_drop_write(struct vfsmount *mnt)
460 {
461 	preempt_disable();
462 	mnt_dec_writers(real_mount(mnt));
463 	preempt_enable();
464 }
465 
466 /**
467  * mnt_drop_write - give up write access to a mount
468  * @mnt: the mount on which to give up write access
469  *
470  * Tells the low-level filesystem that we are done performing writes to it and
471  * also allows filesystem to be frozen again.  Must be matched with
472  * mnt_want_write() call above.
473  */
474 void mnt_drop_write(struct vfsmount *mnt)
475 {
476 	__mnt_drop_write(mnt);
477 	sb_end_write(mnt->mnt_sb);
478 }
479 EXPORT_SYMBOL_GPL(mnt_drop_write);
480 
481 void __mnt_drop_write_file(struct file *file)
482 {
483 	__mnt_drop_write(file->f_path.mnt);
484 }
485 
486 void mnt_drop_write_file(struct file *file)
487 {
488 	mnt_drop_write(file->f_path.mnt);
489 }
490 EXPORT_SYMBOL(mnt_drop_write_file);
491 
492 static int mnt_make_readonly(struct mount *mnt)
493 {
494 	int ret = 0;
495 
496 	lock_mount_hash();
497 	mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
498 	/*
499 	 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
500 	 * should be visible before we do.
501 	 */
502 	smp_mb();
503 
504 	/*
505 	 * With writers on hold, if this value is zero, then there are
506 	 * definitely no active writers (although held writers may subsequently
507 	 * increment the count, they'll have to wait, and decrement it after
508 	 * seeing MNT_READONLY).
509 	 *
510 	 * It is OK to have counter incremented on one CPU and decremented on
511 	 * another: the sum will add up correctly. The danger would be when we
512 	 * sum up each counter, if we read a counter before it is incremented,
513 	 * but then read another CPU's count which it has been subsequently
514 	 * decremented from -- we would see more decrements than we should.
515 	 * MNT_WRITE_HOLD protects against this scenario, because
516 	 * mnt_want_write first increments count, then smp_mb, then spins on
517 	 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
518 	 * we're counting up here.
519 	 */
520 	if (mnt_get_writers(mnt) > 0)
521 		ret = -EBUSY;
522 	else
523 		mnt->mnt.mnt_flags |= MNT_READONLY;
524 	/*
525 	 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
526 	 * that become unheld will see MNT_READONLY.
527 	 */
528 	smp_wmb();
529 	mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
530 	unlock_mount_hash();
531 	return ret;
532 }
533 
534 static void __mnt_unmake_readonly(struct mount *mnt)
535 {
536 	lock_mount_hash();
537 	mnt->mnt.mnt_flags &= ~MNT_READONLY;
538 	unlock_mount_hash();
539 }
540 
541 int sb_prepare_remount_readonly(struct super_block *sb)
542 {
543 	struct mount *mnt;
544 	int err = 0;
545 
546 	/* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
547 	if (atomic_long_read(&sb->s_remove_count))
548 		return -EBUSY;
549 
550 	lock_mount_hash();
551 	list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
552 		if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
553 			mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
554 			smp_mb();
555 			if (mnt_get_writers(mnt) > 0) {
556 				err = -EBUSY;
557 				break;
558 			}
559 		}
560 	}
561 	if (!err && atomic_long_read(&sb->s_remove_count))
562 		err = -EBUSY;
563 
564 	if (!err) {
565 		sb->s_readonly_remount = 1;
566 		smp_wmb();
567 	}
568 	list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
569 		if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
570 			mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
571 	}
572 	unlock_mount_hash();
573 
574 	return err;
575 }
576 
577 static void free_vfsmnt(struct mount *mnt)
578 {
579 	kfree_const(mnt->mnt_devname);
580 #ifdef CONFIG_SMP
581 	free_percpu(mnt->mnt_pcp);
582 #endif
583 	kmem_cache_free(mnt_cache, mnt);
584 }
585 
586 static void delayed_free_vfsmnt(struct rcu_head *head)
587 {
588 	free_vfsmnt(container_of(head, struct mount, mnt_rcu));
589 }
590 
591 /* call under rcu_read_lock */
592 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
593 {
594 	struct mount *mnt;
595 	if (read_seqretry(&mount_lock, seq))
596 		return 1;
597 	if (bastard == NULL)
598 		return 0;
599 	mnt = real_mount(bastard);
600 	mnt_add_count(mnt, 1);
601 	if (likely(!read_seqretry(&mount_lock, seq)))
602 		return 0;
603 	if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
604 		mnt_add_count(mnt, -1);
605 		return 1;
606 	}
607 	return -1;
608 }
609 
610 /* call under rcu_read_lock */
611 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
612 {
613 	int res = __legitimize_mnt(bastard, seq);
614 	if (likely(!res))
615 		return true;
616 	if (unlikely(res < 0)) {
617 		rcu_read_unlock();
618 		mntput(bastard);
619 		rcu_read_lock();
620 	}
621 	return false;
622 }
623 
624 /*
625  * find the first mount at @dentry on vfsmount @mnt.
626  * call under rcu_read_lock()
627  */
628 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
629 {
630 	struct hlist_head *head = m_hash(mnt, dentry);
631 	struct mount *p;
632 
633 	hlist_for_each_entry_rcu(p, head, mnt_hash)
634 		if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
635 			return p;
636 	return NULL;
637 }
638 
639 /*
640  * find the last mount at @dentry on vfsmount @mnt.
641  * mount_lock must be held.
642  */
643 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
644 {
645 	struct mount *p, *res = NULL;
646 	p = __lookup_mnt(mnt, dentry);
647 	if (!p)
648 		goto out;
649 	if (!(p->mnt.mnt_flags & MNT_UMOUNT))
650 		res = p;
651 	hlist_for_each_entry_continue(p, mnt_hash) {
652 		if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
653 			break;
654 		if (!(p->mnt.mnt_flags & MNT_UMOUNT))
655 			res = p;
656 	}
657 out:
658 	return res;
659 }
660 
661 /*
662  * lookup_mnt - Return the first child mount mounted at path
663  *
664  * "First" means first mounted chronologically.  If you create the
665  * following mounts:
666  *
667  * mount /dev/sda1 /mnt
668  * mount /dev/sda2 /mnt
669  * mount /dev/sda3 /mnt
670  *
671  * Then lookup_mnt() on the base /mnt dentry in the root mount will
672  * return successively the root dentry and vfsmount of /dev/sda1, then
673  * /dev/sda2, then /dev/sda3, then NULL.
674  *
675  * lookup_mnt takes a reference to the found vfsmount.
676  */
677 struct vfsmount *lookup_mnt(const struct path *path)
678 {
679 	struct mount *child_mnt;
680 	struct vfsmount *m;
681 	unsigned seq;
682 
683 	rcu_read_lock();
684 	do {
685 		seq = read_seqbegin(&mount_lock);
686 		child_mnt = __lookup_mnt(path->mnt, path->dentry);
687 		m = child_mnt ? &child_mnt->mnt : NULL;
688 	} while (!legitimize_mnt(m, seq));
689 	rcu_read_unlock();
690 	return m;
691 }
692 
693 /*
694  * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
695  *                         current mount namespace.
696  *
697  * The common case is dentries are not mountpoints at all and that
698  * test is handled inline.  For the slow case when we are actually
699  * dealing with a mountpoint of some kind, walk through all of the
700  * mounts in the current mount namespace and test to see if the dentry
701  * is a mountpoint.
702  *
703  * The mount_hashtable is not usable in the context because we
704  * need to identify all mounts that may be in the current mount
705  * namespace not just a mount that happens to have some specified
706  * parent mount.
707  */
708 bool __is_local_mountpoint(struct dentry *dentry)
709 {
710 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
711 	struct mount *mnt;
712 	bool is_covered = false;
713 
714 	if (!d_mountpoint(dentry))
715 		goto out;
716 
717 	down_read(&namespace_sem);
718 	list_for_each_entry(mnt, &ns->list, mnt_list) {
719 		is_covered = (mnt->mnt_mountpoint == dentry);
720 		if (is_covered)
721 			break;
722 	}
723 	up_read(&namespace_sem);
724 out:
725 	return is_covered;
726 }
727 
728 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
729 {
730 	struct hlist_head *chain = mp_hash(dentry);
731 	struct mountpoint *mp;
732 
733 	hlist_for_each_entry(mp, chain, m_hash) {
734 		if (mp->m_dentry == dentry) {
735 			/* might be worth a WARN_ON() */
736 			if (d_unlinked(dentry))
737 				return ERR_PTR(-ENOENT);
738 			mp->m_count++;
739 			return mp;
740 		}
741 	}
742 	return NULL;
743 }
744 
745 static struct mountpoint *new_mountpoint(struct dentry *dentry)
746 {
747 	struct hlist_head *chain = mp_hash(dentry);
748 	struct mountpoint *mp;
749 	int ret;
750 
751 	mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
752 	if (!mp)
753 		return ERR_PTR(-ENOMEM);
754 
755 	ret = d_set_mounted(dentry);
756 	if (ret) {
757 		kfree(mp);
758 		return ERR_PTR(ret);
759 	}
760 
761 	mp->m_dentry = dentry;
762 	mp->m_count = 1;
763 	hlist_add_head(&mp->m_hash, chain);
764 	INIT_HLIST_HEAD(&mp->m_list);
765 	return mp;
766 }
767 
768 static void put_mountpoint(struct mountpoint *mp)
769 {
770 	if (!--mp->m_count) {
771 		struct dentry *dentry = mp->m_dentry;
772 		BUG_ON(!hlist_empty(&mp->m_list));
773 		spin_lock(&dentry->d_lock);
774 		dentry->d_flags &= ~DCACHE_MOUNTED;
775 		spin_unlock(&dentry->d_lock);
776 		hlist_del(&mp->m_hash);
777 		kfree(mp);
778 	}
779 }
780 
781 static inline int check_mnt(struct mount *mnt)
782 {
783 	return mnt->mnt_ns == current->nsproxy->mnt_ns;
784 }
785 
786 /*
787  * vfsmount lock must be held for write
788  */
789 static void touch_mnt_namespace(struct mnt_namespace *ns)
790 {
791 	if (ns) {
792 		ns->event = ++event;
793 		wake_up_interruptible(&ns->poll);
794 	}
795 }
796 
797 /*
798  * vfsmount lock must be held for write
799  */
800 static void __touch_mnt_namespace(struct mnt_namespace *ns)
801 {
802 	if (ns && ns->event != event) {
803 		ns->event = event;
804 		wake_up_interruptible(&ns->poll);
805 	}
806 }
807 
808 /*
809  * vfsmount lock must be held for write
810  */
811 static void unhash_mnt(struct mount *mnt)
812 {
813 	mnt->mnt_parent = mnt;
814 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
815 	list_del_init(&mnt->mnt_child);
816 	hlist_del_init_rcu(&mnt->mnt_hash);
817 	hlist_del_init(&mnt->mnt_mp_list);
818 	put_mountpoint(mnt->mnt_mp);
819 	mnt->mnt_mp = NULL;
820 }
821 
822 /*
823  * vfsmount lock must be held for write
824  */
825 static void detach_mnt(struct mount *mnt, struct path *old_path)
826 {
827 	old_path->dentry = mnt->mnt_mountpoint;
828 	old_path->mnt = &mnt->mnt_parent->mnt;
829 	unhash_mnt(mnt);
830 }
831 
832 /*
833  * vfsmount lock must be held for write
834  */
835 static void umount_mnt(struct mount *mnt)
836 {
837 	/* old mountpoint will be dropped when we can do that */
838 	mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
839 	unhash_mnt(mnt);
840 }
841 
842 /*
843  * vfsmount lock must be held for write
844  */
845 void mnt_set_mountpoint(struct mount *mnt,
846 			struct mountpoint *mp,
847 			struct mount *child_mnt)
848 {
849 	mp->m_count++;
850 	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
851 	child_mnt->mnt_mountpoint = dget(mp->m_dentry);
852 	child_mnt->mnt_parent = mnt;
853 	child_mnt->mnt_mp = mp;
854 	hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
855 }
856 
857 /*
858  * vfsmount lock must be held for write
859  */
860 static void attach_mnt(struct mount *mnt,
861 			struct mount *parent,
862 			struct mountpoint *mp)
863 {
864 	mnt_set_mountpoint(parent, mp, mnt);
865 	hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
866 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
867 }
868 
869 static void attach_shadowed(struct mount *mnt,
870 			struct mount *parent,
871 			struct mount *shadows)
872 {
873 	if (shadows) {
874 		hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
875 		list_add(&mnt->mnt_child, &shadows->mnt_child);
876 	} else {
877 		hlist_add_head_rcu(&mnt->mnt_hash,
878 				m_hash(&parent->mnt, mnt->mnt_mountpoint));
879 		list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
880 	}
881 }
882 
883 /*
884  * vfsmount lock must be held for write
885  */
886 static void commit_tree(struct mount *mnt, struct mount *shadows)
887 {
888 	struct mount *parent = mnt->mnt_parent;
889 	struct mount *m;
890 	LIST_HEAD(head);
891 	struct mnt_namespace *n = parent->mnt_ns;
892 
893 	BUG_ON(parent == mnt);
894 
895 	list_add_tail(&head, &mnt->mnt_list);
896 	list_for_each_entry(m, &head, mnt_list)
897 		m->mnt_ns = n;
898 
899 	list_splice(&head, n->list.prev);
900 
901 	n->mounts += n->pending_mounts;
902 	n->pending_mounts = 0;
903 
904 	attach_shadowed(mnt, parent, shadows);
905 	touch_mnt_namespace(n);
906 }
907 
908 static struct mount *next_mnt(struct mount *p, struct mount *root)
909 {
910 	struct list_head *next = p->mnt_mounts.next;
911 	if (next == &p->mnt_mounts) {
912 		while (1) {
913 			if (p == root)
914 				return NULL;
915 			next = p->mnt_child.next;
916 			if (next != &p->mnt_parent->mnt_mounts)
917 				break;
918 			p = p->mnt_parent;
919 		}
920 	}
921 	return list_entry(next, struct mount, mnt_child);
922 }
923 
924 static struct mount *skip_mnt_tree(struct mount *p)
925 {
926 	struct list_head *prev = p->mnt_mounts.prev;
927 	while (prev != &p->mnt_mounts) {
928 		p = list_entry(prev, struct mount, mnt_child);
929 		prev = p->mnt_mounts.prev;
930 	}
931 	return p;
932 }
933 
934 struct vfsmount *
935 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
936 {
937 	struct mount *mnt;
938 	struct dentry *root;
939 
940 	if (!type)
941 		return ERR_PTR(-ENODEV);
942 
943 	mnt = alloc_vfsmnt(name);
944 	if (!mnt)
945 		return ERR_PTR(-ENOMEM);
946 
947 	if (flags & MS_KERNMOUNT)
948 		mnt->mnt.mnt_flags = MNT_INTERNAL;
949 
950 	root = mount_fs(type, flags, name, data);
951 	if (IS_ERR(root)) {
952 		mnt_free_id(mnt);
953 		free_vfsmnt(mnt);
954 		return ERR_CAST(root);
955 	}
956 
957 	mnt->mnt.mnt_root = root;
958 	mnt->mnt.mnt_sb = root->d_sb;
959 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
960 	mnt->mnt_parent = mnt;
961 	lock_mount_hash();
962 	list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
963 	unlock_mount_hash();
964 	return &mnt->mnt;
965 }
966 EXPORT_SYMBOL_GPL(vfs_kern_mount);
967 
968 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
969 					int flag)
970 {
971 	struct super_block *sb = old->mnt.mnt_sb;
972 	struct mount *mnt;
973 	int err;
974 
975 	mnt = alloc_vfsmnt(old->mnt_devname);
976 	if (!mnt)
977 		return ERR_PTR(-ENOMEM);
978 
979 	if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
980 		mnt->mnt_group_id = 0; /* not a peer of original */
981 	else
982 		mnt->mnt_group_id = old->mnt_group_id;
983 
984 	if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
985 		err = mnt_alloc_group_id(mnt);
986 		if (err)
987 			goto out_free;
988 	}
989 
990 	mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
991 	/* Don't allow unprivileged users to change mount flags */
992 	if (flag & CL_UNPRIVILEGED) {
993 		mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
994 
995 		if (mnt->mnt.mnt_flags & MNT_READONLY)
996 			mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
997 
998 		if (mnt->mnt.mnt_flags & MNT_NODEV)
999 			mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1000 
1001 		if (mnt->mnt.mnt_flags & MNT_NOSUID)
1002 			mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1003 
1004 		if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1005 			mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1006 	}
1007 
1008 	/* Don't allow unprivileged users to reveal what is under a mount */
1009 	if ((flag & CL_UNPRIVILEGED) &&
1010 	    (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1011 		mnt->mnt.mnt_flags |= MNT_LOCKED;
1012 
1013 	atomic_inc(&sb->s_active);
1014 	mnt->mnt.mnt_sb = sb;
1015 	mnt->mnt.mnt_root = dget(root);
1016 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1017 	mnt->mnt_parent = mnt;
1018 	lock_mount_hash();
1019 	list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1020 	unlock_mount_hash();
1021 
1022 	if ((flag & CL_SLAVE) ||
1023 	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1024 		list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1025 		mnt->mnt_master = old;
1026 		CLEAR_MNT_SHARED(mnt);
1027 	} else if (!(flag & CL_PRIVATE)) {
1028 		if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1029 			list_add(&mnt->mnt_share, &old->mnt_share);
1030 		if (IS_MNT_SLAVE(old))
1031 			list_add(&mnt->mnt_slave, &old->mnt_slave);
1032 		mnt->mnt_master = old->mnt_master;
1033 	} else {
1034 		CLEAR_MNT_SHARED(mnt);
1035 	}
1036 	if (flag & CL_MAKE_SHARED)
1037 		set_mnt_shared(mnt);
1038 
1039 	/* stick the duplicate mount on the same expiry list
1040 	 * as the original if that was on one */
1041 	if (flag & CL_EXPIRE) {
1042 		if (!list_empty(&old->mnt_expire))
1043 			list_add(&mnt->mnt_expire, &old->mnt_expire);
1044 	}
1045 
1046 	return mnt;
1047 
1048  out_free:
1049 	mnt_free_id(mnt);
1050 	free_vfsmnt(mnt);
1051 	return ERR_PTR(err);
1052 }
1053 
1054 static void cleanup_mnt(struct mount *mnt)
1055 {
1056 	/*
1057 	 * This probably indicates that somebody messed
1058 	 * up a mnt_want/drop_write() pair.  If this
1059 	 * happens, the filesystem was probably unable
1060 	 * to make r/w->r/o transitions.
1061 	 */
1062 	/*
1063 	 * The locking used to deal with mnt_count decrement provides barriers,
1064 	 * so mnt_get_writers() below is safe.
1065 	 */
1066 	WARN_ON(mnt_get_writers(mnt));
1067 	if (unlikely(mnt->mnt_pins.first))
1068 		mnt_pin_kill(mnt);
1069 	fsnotify_vfsmount_delete(&mnt->mnt);
1070 	dput(mnt->mnt.mnt_root);
1071 	deactivate_super(mnt->mnt.mnt_sb);
1072 	mnt_free_id(mnt);
1073 	call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1074 }
1075 
1076 static void __cleanup_mnt(struct rcu_head *head)
1077 {
1078 	cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1079 }
1080 
1081 static LLIST_HEAD(delayed_mntput_list);
1082 static void delayed_mntput(struct work_struct *unused)
1083 {
1084 	struct llist_node *node = llist_del_all(&delayed_mntput_list);
1085 	struct llist_node *next;
1086 
1087 	for (; node; node = next) {
1088 		next = llist_next(node);
1089 		cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1090 	}
1091 }
1092 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1093 
1094 static void mntput_no_expire(struct mount *mnt)
1095 {
1096 	rcu_read_lock();
1097 	mnt_add_count(mnt, -1);
1098 	if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1099 		rcu_read_unlock();
1100 		return;
1101 	}
1102 	lock_mount_hash();
1103 	if (mnt_get_count(mnt)) {
1104 		rcu_read_unlock();
1105 		unlock_mount_hash();
1106 		return;
1107 	}
1108 	if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1109 		rcu_read_unlock();
1110 		unlock_mount_hash();
1111 		return;
1112 	}
1113 	mnt->mnt.mnt_flags |= MNT_DOOMED;
1114 	rcu_read_unlock();
1115 
1116 	list_del(&mnt->mnt_instance);
1117 
1118 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1119 		struct mount *p, *tmp;
1120 		list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1121 			umount_mnt(p);
1122 		}
1123 	}
1124 	unlock_mount_hash();
1125 
1126 	if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1127 		struct task_struct *task = current;
1128 		if (likely(!(task->flags & PF_KTHREAD))) {
1129 			init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1130 			if (!task_work_add(task, &mnt->mnt_rcu, true))
1131 				return;
1132 		}
1133 		if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1134 			schedule_delayed_work(&delayed_mntput_work, 1);
1135 		return;
1136 	}
1137 	cleanup_mnt(mnt);
1138 }
1139 
1140 void mntput(struct vfsmount *mnt)
1141 {
1142 	if (mnt) {
1143 		struct mount *m = real_mount(mnt);
1144 		/* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1145 		if (unlikely(m->mnt_expiry_mark))
1146 			m->mnt_expiry_mark = 0;
1147 		mntput_no_expire(m);
1148 	}
1149 }
1150 EXPORT_SYMBOL(mntput);
1151 
1152 struct vfsmount *mntget(struct vfsmount *mnt)
1153 {
1154 	if (mnt)
1155 		mnt_add_count(real_mount(mnt), 1);
1156 	return mnt;
1157 }
1158 EXPORT_SYMBOL(mntget);
1159 
1160 /* path_is_mountpoint() - Check if path is a mount in the current
1161  *                          namespace.
1162  *
1163  *  d_mountpoint() can only be used reliably to establish if a dentry is
1164  *  not mounted in any namespace and that common case is handled inline.
1165  *  d_mountpoint() isn't aware of the possibility there may be multiple
1166  *  mounts using a given dentry in a different namespace. This function
1167  *  checks if the passed in path is a mountpoint rather than the dentry
1168  *  alone.
1169  */
1170 bool path_is_mountpoint(const struct path *path)
1171 {
1172 	unsigned seq;
1173 	bool res;
1174 
1175 	if (!d_mountpoint(path->dentry))
1176 		return false;
1177 
1178 	rcu_read_lock();
1179 	do {
1180 		seq = read_seqbegin(&mount_lock);
1181 		res = __path_is_mountpoint(path);
1182 	} while (read_seqretry(&mount_lock, seq));
1183 	rcu_read_unlock();
1184 
1185 	return res;
1186 }
1187 EXPORT_SYMBOL(path_is_mountpoint);
1188 
1189 struct vfsmount *mnt_clone_internal(const struct path *path)
1190 {
1191 	struct mount *p;
1192 	p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1193 	if (IS_ERR(p))
1194 		return ERR_CAST(p);
1195 	p->mnt.mnt_flags |= MNT_INTERNAL;
1196 	return &p->mnt;
1197 }
1198 
1199 static inline void mangle(struct seq_file *m, const char *s)
1200 {
1201 	seq_escape(m, s, " \t\n\\");
1202 }
1203 
1204 /*
1205  * Simple .show_options callback for filesystems which don't want to
1206  * implement more complex mount option showing.
1207  *
1208  * See also save_mount_options().
1209  */
1210 int generic_show_options(struct seq_file *m, struct dentry *root)
1211 {
1212 	const char *options;
1213 
1214 	rcu_read_lock();
1215 	options = rcu_dereference(root->d_sb->s_options);
1216 
1217 	if (options != NULL && options[0]) {
1218 		seq_putc(m, ',');
1219 		mangle(m, options);
1220 	}
1221 	rcu_read_unlock();
1222 
1223 	return 0;
1224 }
1225 EXPORT_SYMBOL(generic_show_options);
1226 
1227 /*
1228  * If filesystem uses generic_show_options(), this function should be
1229  * called from the fill_super() callback.
1230  *
1231  * The .remount_fs callback usually needs to be handled in a special
1232  * way, to make sure, that previous options are not overwritten if the
1233  * remount fails.
1234  *
1235  * Also note, that if the filesystem's .remount_fs function doesn't
1236  * reset all options to their default value, but changes only newly
1237  * given options, then the displayed options will not reflect reality
1238  * any more.
1239  */
1240 void save_mount_options(struct super_block *sb, char *options)
1241 {
1242 	BUG_ON(sb->s_options);
1243 	rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1244 }
1245 EXPORT_SYMBOL(save_mount_options);
1246 
1247 void replace_mount_options(struct super_block *sb, char *options)
1248 {
1249 	char *old = sb->s_options;
1250 	rcu_assign_pointer(sb->s_options, options);
1251 	if (old) {
1252 		synchronize_rcu();
1253 		kfree(old);
1254 	}
1255 }
1256 EXPORT_SYMBOL(replace_mount_options);
1257 
1258 #ifdef CONFIG_PROC_FS
1259 /* iterator; we want it to have access to namespace_sem, thus here... */
1260 static void *m_start(struct seq_file *m, loff_t *pos)
1261 {
1262 	struct proc_mounts *p = m->private;
1263 
1264 	down_read(&namespace_sem);
1265 	if (p->cached_event == p->ns->event) {
1266 		void *v = p->cached_mount;
1267 		if (*pos == p->cached_index)
1268 			return v;
1269 		if (*pos == p->cached_index + 1) {
1270 			v = seq_list_next(v, &p->ns->list, &p->cached_index);
1271 			return p->cached_mount = v;
1272 		}
1273 	}
1274 
1275 	p->cached_event = p->ns->event;
1276 	p->cached_mount = seq_list_start(&p->ns->list, *pos);
1277 	p->cached_index = *pos;
1278 	return p->cached_mount;
1279 }
1280 
1281 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1282 {
1283 	struct proc_mounts *p = m->private;
1284 
1285 	p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1286 	p->cached_index = *pos;
1287 	return p->cached_mount;
1288 }
1289 
1290 static void m_stop(struct seq_file *m, void *v)
1291 {
1292 	up_read(&namespace_sem);
1293 }
1294 
1295 static int m_show(struct seq_file *m, void *v)
1296 {
1297 	struct proc_mounts *p = m->private;
1298 	struct mount *r = list_entry(v, struct mount, mnt_list);
1299 	return p->show(m, &r->mnt);
1300 }
1301 
1302 const struct seq_operations mounts_op = {
1303 	.start	= m_start,
1304 	.next	= m_next,
1305 	.stop	= m_stop,
1306 	.show	= m_show,
1307 };
1308 #endif  /* CONFIG_PROC_FS */
1309 
1310 /**
1311  * may_umount_tree - check if a mount tree is busy
1312  * @mnt: root of mount tree
1313  *
1314  * This is called to check if a tree of mounts has any
1315  * open files, pwds, chroots or sub mounts that are
1316  * busy.
1317  */
1318 int may_umount_tree(struct vfsmount *m)
1319 {
1320 	struct mount *mnt = real_mount(m);
1321 	int actual_refs = 0;
1322 	int minimum_refs = 0;
1323 	struct mount *p;
1324 	BUG_ON(!m);
1325 
1326 	/* write lock needed for mnt_get_count */
1327 	lock_mount_hash();
1328 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1329 		actual_refs += mnt_get_count(p);
1330 		minimum_refs += 2;
1331 	}
1332 	unlock_mount_hash();
1333 
1334 	if (actual_refs > minimum_refs)
1335 		return 0;
1336 
1337 	return 1;
1338 }
1339 
1340 EXPORT_SYMBOL(may_umount_tree);
1341 
1342 /**
1343  * may_umount - check if a mount point is busy
1344  * @mnt: root of mount
1345  *
1346  * This is called to check if a mount point has any
1347  * open files, pwds, chroots or sub mounts. If the
1348  * mount has sub mounts this will return busy
1349  * regardless of whether the sub mounts are busy.
1350  *
1351  * Doesn't take quota and stuff into account. IOW, in some cases it will
1352  * give false negatives. The main reason why it's here is that we need
1353  * a non-destructive way to look for easily umountable filesystems.
1354  */
1355 int may_umount(struct vfsmount *mnt)
1356 {
1357 	int ret = 1;
1358 	down_read(&namespace_sem);
1359 	lock_mount_hash();
1360 	if (propagate_mount_busy(real_mount(mnt), 2))
1361 		ret = 0;
1362 	unlock_mount_hash();
1363 	up_read(&namespace_sem);
1364 	return ret;
1365 }
1366 
1367 EXPORT_SYMBOL(may_umount);
1368 
1369 static HLIST_HEAD(unmounted);	/* protected by namespace_sem */
1370 
1371 static void namespace_unlock(void)
1372 {
1373 	struct hlist_head head;
1374 
1375 	hlist_move_list(&unmounted, &head);
1376 
1377 	up_write(&namespace_sem);
1378 
1379 	if (likely(hlist_empty(&head)))
1380 		return;
1381 
1382 	synchronize_rcu();
1383 
1384 	group_pin_kill(&head);
1385 }
1386 
1387 static inline void namespace_lock(void)
1388 {
1389 	down_write(&namespace_sem);
1390 }
1391 
1392 enum umount_tree_flags {
1393 	UMOUNT_SYNC = 1,
1394 	UMOUNT_PROPAGATE = 2,
1395 	UMOUNT_CONNECTED = 4,
1396 };
1397 
1398 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1399 {
1400 	/* Leaving mounts connected is only valid for lazy umounts */
1401 	if (how & UMOUNT_SYNC)
1402 		return true;
1403 
1404 	/* A mount without a parent has nothing to be connected to */
1405 	if (!mnt_has_parent(mnt))
1406 		return true;
1407 
1408 	/* Because the reference counting rules change when mounts are
1409 	 * unmounted and connected, umounted mounts may not be
1410 	 * connected to mounted mounts.
1411 	 */
1412 	if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1413 		return true;
1414 
1415 	/* Has it been requested that the mount remain connected? */
1416 	if (how & UMOUNT_CONNECTED)
1417 		return false;
1418 
1419 	/* Is the mount locked such that it needs to remain connected? */
1420 	if (IS_MNT_LOCKED(mnt))
1421 		return false;
1422 
1423 	/* By default disconnect the mount */
1424 	return true;
1425 }
1426 
1427 /*
1428  * mount_lock must be held
1429  * namespace_sem must be held for write
1430  */
1431 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1432 {
1433 	LIST_HEAD(tmp_list);
1434 	struct mount *p;
1435 
1436 	if (how & UMOUNT_PROPAGATE)
1437 		propagate_mount_unlock(mnt);
1438 
1439 	/* Gather the mounts to umount */
1440 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1441 		p->mnt.mnt_flags |= MNT_UMOUNT;
1442 		list_move(&p->mnt_list, &tmp_list);
1443 	}
1444 
1445 	/* Hide the mounts from mnt_mounts */
1446 	list_for_each_entry(p, &tmp_list, mnt_list) {
1447 		list_del_init(&p->mnt_child);
1448 	}
1449 
1450 	/* Add propogated mounts to the tmp_list */
1451 	if (how & UMOUNT_PROPAGATE)
1452 		propagate_umount(&tmp_list);
1453 
1454 	while (!list_empty(&tmp_list)) {
1455 		struct mnt_namespace *ns;
1456 		bool disconnect;
1457 		p = list_first_entry(&tmp_list, struct mount, mnt_list);
1458 		list_del_init(&p->mnt_expire);
1459 		list_del_init(&p->mnt_list);
1460 		ns = p->mnt_ns;
1461 		if (ns) {
1462 			ns->mounts--;
1463 			__touch_mnt_namespace(ns);
1464 		}
1465 		p->mnt_ns = NULL;
1466 		if (how & UMOUNT_SYNC)
1467 			p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1468 
1469 		disconnect = disconnect_mount(p, how);
1470 
1471 		pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1472 				 disconnect ? &unmounted : NULL);
1473 		if (mnt_has_parent(p)) {
1474 			mnt_add_count(p->mnt_parent, -1);
1475 			if (!disconnect) {
1476 				/* Don't forget about p */
1477 				list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1478 			} else {
1479 				umount_mnt(p);
1480 			}
1481 		}
1482 		change_mnt_propagation(p, MS_PRIVATE);
1483 	}
1484 }
1485 
1486 static void shrink_submounts(struct mount *mnt);
1487 
1488 static int do_umount(struct mount *mnt, int flags)
1489 {
1490 	struct super_block *sb = mnt->mnt.mnt_sb;
1491 	int retval;
1492 
1493 	retval = security_sb_umount(&mnt->mnt, flags);
1494 	if (retval)
1495 		return retval;
1496 
1497 	/*
1498 	 * Allow userspace to request a mountpoint be expired rather than
1499 	 * unmounting unconditionally. Unmount only happens if:
1500 	 *  (1) the mark is already set (the mark is cleared by mntput())
1501 	 *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1502 	 */
1503 	if (flags & MNT_EXPIRE) {
1504 		if (&mnt->mnt == current->fs->root.mnt ||
1505 		    flags & (MNT_FORCE | MNT_DETACH))
1506 			return -EINVAL;
1507 
1508 		/*
1509 		 * probably don't strictly need the lock here if we examined
1510 		 * all race cases, but it's a slowpath.
1511 		 */
1512 		lock_mount_hash();
1513 		if (mnt_get_count(mnt) != 2) {
1514 			unlock_mount_hash();
1515 			return -EBUSY;
1516 		}
1517 		unlock_mount_hash();
1518 
1519 		if (!xchg(&mnt->mnt_expiry_mark, 1))
1520 			return -EAGAIN;
1521 	}
1522 
1523 	/*
1524 	 * If we may have to abort operations to get out of this
1525 	 * mount, and they will themselves hold resources we must
1526 	 * allow the fs to do things. In the Unix tradition of
1527 	 * 'Gee thats tricky lets do it in userspace' the umount_begin
1528 	 * might fail to complete on the first run through as other tasks
1529 	 * must return, and the like. Thats for the mount program to worry
1530 	 * about for the moment.
1531 	 */
1532 
1533 	if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1534 		sb->s_op->umount_begin(sb);
1535 	}
1536 
1537 	/*
1538 	 * No sense to grab the lock for this test, but test itself looks
1539 	 * somewhat bogus. Suggestions for better replacement?
1540 	 * Ho-hum... In principle, we might treat that as umount + switch
1541 	 * to rootfs. GC would eventually take care of the old vfsmount.
1542 	 * Actually it makes sense, especially if rootfs would contain a
1543 	 * /reboot - static binary that would close all descriptors and
1544 	 * call reboot(9). Then init(8) could umount root and exec /reboot.
1545 	 */
1546 	if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1547 		/*
1548 		 * Special case for "unmounting" root ...
1549 		 * we just try to remount it readonly.
1550 		 */
1551 		if (!capable(CAP_SYS_ADMIN))
1552 			return -EPERM;
1553 		down_write(&sb->s_umount);
1554 		if (!(sb->s_flags & MS_RDONLY))
1555 			retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1556 		up_write(&sb->s_umount);
1557 		return retval;
1558 	}
1559 
1560 	namespace_lock();
1561 	lock_mount_hash();
1562 	event++;
1563 
1564 	if (flags & MNT_DETACH) {
1565 		if (!list_empty(&mnt->mnt_list))
1566 			umount_tree(mnt, UMOUNT_PROPAGATE);
1567 		retval = 0;
1568 	} else {
1569 		shrink_submounts(mnt);
1570 		retval = -EBUSY;
1571 		if (!propagate_mount_busy(mnt, 2)) {
1572 			if (!list_empty(&mnt->mnt_list))
1573 				umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1574 			retval = 0;
1575 		}
1576 	}
1577 	unlock_mount_hash();
1578 	namespace_unlock();
1579 	return retval;
1580 }
1581 
1582 /*
1583  * __detach_mounts - lazily unmount all mounts on the specified dentry
1584  *
1585  * During unlink, rmdir, and d_drop it is possible to loose the path
1586  * to an existing mountpoint, and wind up leaking the mount.
1587  * detach_mounts allows lazily unmounting those mounts instead of
1588  * leaking them.
1589  *
1590  * The caller may hold dentry->d_inode->i_mutex.
1591  */
1592 void __detach_mounts(struct dentry *dentry)
1593 {
1594 	struct mountpoint *mp;
1595 	struct mount *mnt;
1596 
1597 	namespace_lock();
1598 	mp = lookup_mountpoint(dentry);
1599 	if (IS_ERR_OR_NULL(mp))
1600 		goto out_unlock;
1601 
1602 	lock_mount_hash();
1603 	event++;
1604 	while (!hlist_empty(&mp->m_list)) {
1605 		mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1606 		if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1607 			hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1608 			umount_mnt(mnt);
1609 		}
1610 		else umount_tree(mnt, UMOUNT_CONNECTED);
1611 	}
1612 	unlock_mount_hash();
1613 	put_mountpoint(mp);
1614 out_unlock:
1615 	namespace_unlock();
1616 }
1617 
1618 /*
1619  * Is the caller allowed to modify his namespace?
1620  */
1621 static inline bool may_mount(void)
1622 {
1623 	return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1624 }
1625 
1626 static inline bool may_mandlock(void)
1627 {
1628 #ifndef	CONFIG_MANDATORY_FILE_LOCKING
1629 	return false;
1630 #endif
1631 	return capable(CAP_SYS_ADMIN);
1632 }
1633 
1634 /*
1635  * Now umount can handle mount points as well as block devices.
1636  * This is important for filesystems which use unnamed block devices.
1637  *
1638  * We now support a flag for forced unmount like the other 'big iron'
1639  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1640  */
1641 
1642 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1643 {
1644 	struct path path;
1645 	struct mount *mnt;
1646 	int retval;
1647 	int lookup_flags = 0;
1648 
1649 	if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1650 		return -EINVAL;
1651 
1652 	if (!may_mount())
1653 		return -EPERM;
1654 
1655 	if (!(flags & UMOUNT_NOFOLLOW))
1656 		lookup_flags |= LOOKUP_FOLLOW;
1657 
1658 	retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1659 	if (retval)
1660 		goto out;
1661 	mnt = real_mount(path.mnt);
1662 	retval = -EINVAL;
1663 	if (path.dentry != path.mnt->mnt_root)
1664 		goto dput_and_out;
1665 	if (!check_mnt(mnt))
1666 		goto dput_and_out;
1667 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
1668 		goto dput_and_out;
1669 	retval = -EPERM;
1670 	if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1671 		goto dput_and_out;
1672 
1673 	retval = do_umount(mnt, flags);
1674 dput_and_out:
1675 	/* we mustn't call path_put() as that would clear mnt_expiry_mark */
1676 	dput(path.dentry);
1677 	mntput_no_expire(mnt);
1678 out:
1679 	return retval;
1680 }
1681 
1682 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1683 
1684 /*
1685  *	The 2.0 compatible umount. No flags.
1686  */
1687 SYSCALL_DEFINE1(oldumount, char __user *, name)
1688 {
1689 	return sys_umount(name, 0);
1690 }
1691 
1692 #endif
1693 
1694 static bool is_mnt_ns_file(struct dentry *dentry)
1695 {
1696 	/* Is this a proxy for a mount namespace? */
1697 	return dentry->d_op == &ns_dentry_operations &&
1698 	       dentry->d_fsdata == &mntns_operations;
1699 }
1700 
1701 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1702 {
1703 	return container_of(ns, struct mnt_namespace, ns);
1704 }
1705 
1706 static bool mnt_ns_loop(struct dentry *dentry)
1707 {
1708 	/* Could bind mounting the mount namespace inode cause a
1709 	 * mount namespace loop?
1710 	 */
1711 	struct mnt_namespace *mnt_ns;
1712 	if (!is_mnt_ns_file(dentry))
1713 		return false;
1714 
1715 	mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1716 	return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1717 }
1718 
1719 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1720 					int flag)
1721 {
1722 	struct mount *res, *p, *q, *r, *parent;
1723 
1724 	if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1725 		return ERR_PTR(-EINVAL);
1726 
1727 	if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1728 		return ERR_PTR(-EINVAL);
1729 
1730 	res = q = clone_mnt(mnt, dentry, flag);
1731 	if (IS_ERR(q))
1732 		return q;
1733 
1734 	q->mnt_mountpoint = mnt->mnt_mountpoint;
1735 
1736 	p = mnt;
1737 	list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1738 		struct mount *s;
1739 		if (!is_subdir(r->mnt_mountpoint, dentry))
1740 			continue;
1741 
1742 		for (s = r; s; s = next_mnt(s, r)) {
1743 			struct mount *t = NULL;
1744 			if (!(flag & CL_COPY_UNBINDABLE) &&
1745 			    IS_MNT_UNBINDABLE(s)) {
1746 				s = skip_mnt_tree(s);
1747 				continue;
1748 			}
1749 			if (!(flag & CL_COPY_MNT_NS_FILE) &&
1750 			    is_mnt_ns_file(s->mnt.mnt_root)) {
1751 				s = skip_mnt_tree(s);
1752 				continue;
1753 			}
1754 			while (p != s->mnt_parent) {
1755 				p = p->mnt_parent;
1756 				q = q->mnt_parent;
1757 			}
1758 			p = s;
1759 			parent = q;
1760 			q = clone_mnt(p, p->mnt.mnt_root, flag);
1761 			if (IS_ERR(q))
1762 				goto out;
1763 			lock_mount_hash();
1764 			list_add_tail(&q->mnt_list, &res->mnt_list);
1765 			mnt_set_mountpoint(parent, p->mnt_mp, q);
1766 			if (!list_empty(&parent->mnt_mounts)) {
1767 				t = list_last_entry(&parent->mnt_mounts,
1768 					struct mount, mnt_child);
1769 				if (t->mnt_mp != p->mnt_mp)
1770 					t = NULL;
1771 			}
1772 			attach_shadowed(q, parent, t);
1773 			unlock_mount_hash();
1774 		}
1775 	}
1776 	return res;
1777 out:
1778 	if (res) {
1779 		lock_mount_hash();
1780 		umount_tree(res, UMOUNT_SYNC);
1781 		unlock_mount_hash();
1782 	}
1783 	return q;
1784 }
1785 
1786 /* Caller should check returned pointer for errors */
1787 
1788 struct vfsmount *collect_mounts(const struct path *path)
1789 {
1790 	struct mount *tree;
1791 	namespace_lock();
1792 	if (!check_mnt(real_mount(path->mnt)))
1793 		tree = ERR_PTR(-EINVAL);
1794 	else
1795 		tree = copy_tree(real_mount(path->mnt), path->dentry,
1796 				 CL_COPY_ALL | CL_PRIVATE);
1797 	namespace_unlock();
1798 	if (IS_ERR(tree))
1799 		return ERR_CAST(tree);
1800 	return &tree->mnt;
1801 }
1802 
1803 void drop_collected_mounts(struct vfsmount *mnt)
1804 {
1805 	namespace_lock();
1806 	lock_mount_hash();
1807 	umount_tree(real_mount(mnt), UMOUNT_SYNC);
1808 	unlock_mount_hash();
1809 	namespace_unlock();
1810 }
1811 
1812 /**
1813  * clone_private_mount - create a private clone of a path
1814  *
1815  * This creates a new vfsmount, which will be the clone of @path.  The new will
1816  * not be attached anywhere in the namespace and will be private (i.e. changes
1817  * to the originating mount won't be propagated into this).
1818  *
1819  * Release with mntput().
1820  */
1821 struct vfsmount *clone_private_mount(const struct path *path)
1822 {
1823 	struct mount *old_mnt = real_mount(path->mnt);
1824 	struct mount *new_mnt;
1825 
1826 	if (IS_MNT_UNBINDABLE(old_mnt))
1827 		return ERR_PTR(-EINVAL);
1828 
1829 	new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1830 	if (IS_ERR(new_mnt))
1831 		return ERR_CAST(new_mnt);
1832 
1833 	return &new_mnt->mnt;
1834 }
1835 EXPORT_SYMBOL_GPL(clone_private_mount);
1836 
1837 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1838 		   struct vfsmount *root)
1839 {
1840 	struct mount *mnt;
1841 	int res = f(root, arg);
1842 	if (res)
1843 		return res;
1844 	list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1845 		res = f(&mnt->mnt, arg);
1846 		if (res)
1847 			return res;
1848 	}
1849 	return 0;
1850 }
1851 
1852 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1853 {
1854 	struct mount *p;
1855 
1856 	for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1857 		if (p->mnt_group_id && !IS_MNT_SHARED(p))
1858 			mnt_release_group_id(p);
1859 	}
1860 }
1861 
1862 static int invent_group_ids(struct mount *mnt, bool recurse)
1863 {
1864 	struct mount *p;
1865 
1866 	for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1867 		if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1868 			int err = mnt_alloc_group_id(p);
1869 			if (err) {
1870 				cleanup_group_ids(mnt, p);
1871 				return err;
1872 			}
1873 		}
1874 	}
1875 
1876 	return 0;
1877 }
1878 
1879 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1880 {
1881 	unsigned int max = READ_ONCE(sysctl_mount_max);
1882 	unsigned int mounts = 0, old, pending, sum;
1883 	struct mount *p;
1884 
1885 	for (p = mnt; p; p = next_mnt(p, mnt))
1886 		mounts++;
1887 
1888 	old = ns->mounts;
1889 	pending = ns->pending_mounts;
1890 	sum = old + pending;
1891 	if ((old > sum) ||
1892 	    (pending > sum) ||
1893 	    (max < sum) ||
1894 	    (mounts > (max - sum)))
1895 		return -ENOSPC;
1896 
1897 	ns->pending_mounts = pending + mounts;
1898 	return 0;
1899 }
1900 
1901 /*
1902  *  @source_mnt : mount tree to be attached
1903  *  @nd         : place the mount tree @source_mnt is attached
1904  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1905  *  		   store the parent mount and mountpoint dentry.
1906  *  		   (done when source_mnt is moved)
1907  *
1908  *  NOTE: in the table below explains the semantics when a source mount
1909  *  of a given type is attached to a destination mount of a given type.
1910  * ---------------------------------------------------------------------------
1911  * |         BIND MOUNT OPERATION                                            |
1912  * |**************************************************************************
1913  * | source-->| shared        |       private  |       slave    | unbindable |
1914  * | dest     |               |                |                |            |
1915  * |   |      |               |                |                |            |
1916  * |   v      |               |                |                |            |
1917  * |**************************************************************************
1918  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1919  * |          |               |                |                |            |
1920  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1921  * ***************************************************************************
1922  * A bind operation clones the source mount and mounts the clone on the
1923  * destination mount.
1924  *
1925  * (++)  the cloned mount is propagated to all the mounts in the propagation
1926  * 	 tree of the destination mount and the cloned mount is added to
1927  * 	 the peer group of the source mount.
1928  * (+)   the cloned mount is created under the destination mount and is marked
1929  *       as shared. The cloned mount is added to the peer group of the source
1930  *       mount.
1931  * (+++) the mount is propagated to all the mounts in the propagation tree
1932  *       of the destination mount and the cloned mount is made slave
1933  *       of the same master as that of the source mount. The cloned mount
1934  *       is marked as 'shared and slave'.
1935  * (*)   the cloned mount is made a slave of the same master as that of the
1936  * 	 source mount.
1937  *
1938  * ---------------------------------------------------------------------------
1939  * |         		MOVE MOUNT OPERATION                                 |
1940  * |**************************************************************************
1941  * | source-->| shared        |       private  |       slave    | unbindable |
1942  * | dest     |               |                |                |            |
1943  * |   |      |               |                |                |            |
1944  * |   v      |               |                |                |            |
1945  * |**************************************************************************
1946  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1947  * |          |               |                |                |            |
1948  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1949  * ***************************************************************************
1950  *
1951  * (+)  the mount is moved to the destination. And is then propagated to
1952  * 	all the mounts in the propagation tree of the destination mount.
1953  * (+*)  the mount is moved to the destination.
1954  * (+++)  the mount is moved to the destination and is then propagated to
1955  * 	all the mounts belonging to the destination mount's propagation tree.
1956  * 	the mount is marked as 'shared and slave'.
1957  * (*)	the mount continues to be a slave at the new location.
1958  *
1959  * if the source mount is a tree, the operations explained above is
1960  * applied to each mount in the tree.
1961  * Must be called without spinlocks held, since this function can sleep
1962  * in allocations.
1963  */
1964 static int attach_recursive_mnt(struct mount *source_mnt,
1965 			struct mount *dest_mnt,
1966 			struct mountpoint *dest_mp,
1967 			struct path *parent_path)
1968 {
1969 	HLIST_HEAD(tree_list);
1970 	struct mnt_namespace *ns = dest_mnt->mnt_ns;
1971 	struct mount *child, *p;
1972 	struct hlist_node *n;
1973 	int err;
1974 
1975 	/* Is there space to add these mounts to the mount namespace? */
1976 	if (!parent_path) {
1977 		err = count_mounts(ns, source_mnt);
1978 		if (err)
1979 			goto out;
1980 	}
1981 
1982 	if (IS_MNT_SHARED(dest_mnt)) {
1983 		err = invent_group_ids(source_mnt, true);
1984 		if (err)
1985 			goto out;
1986 		err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1987 		lock_mount_hash();
1988 		if (err)
1989 			goto out_cleanup_ids;
1990 		for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1991 			set_mnt_shared(p);
1992 	} else {
1993 		lock_mount_hash();
1994 	}
1995 	if (parent_path) {
1996 		detach_mnt(source_mnt, parent_path);
1997 		attach_mnt(source_mnt, dest_mnt, dest_mp);
1998 		touch_mnt_namespace(source_mnt->mnt_ns);
1999 	} else {
2000 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2001 		commit_tree(source_mnt, NULL);
2002 	}
2003 
2004 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2005 		struct mount *q;
2006 		hlist_del_init(&child->mnt_hash);
2007 		q = __lookup_mnt_last(&child->mnt_parent->mnt,
2008 				      child->mnt_mountpoint);
2009 		commit_tree(child, q);
2010 	}
2011 	unlock_mount_hash();
2012 
2013 	return 0;
2014 
2015  out_cleanup_ids:
2016 	while (!hlist_empty(&tree_list)) {
2017 		child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2018 		child->mnt_parent->mnt_ns->pending_mounts = 0;
2019 		umount_tree(child, UMOUNT_SYNC);
2020 	}
2021 	unlock_mount_hash();
2022 	cleanup_group_ids(source_mnt, NULL);
2023  out:
2024 	ns->pending_mounts = 0;
2025 	return err;
2026 }
2027 
2028 static struct mountpoint *lock_mount(struct path *path)
2029 {
2030 	struct vfsmount *mnt;
2031 	struct dentry *dentry = path->dentry;
2032 retry:
2033 	inode_lock(dentry->d_inode);
2034 	if (unlikely(cant_mount(dentry))) {
2035 		inode_unlock(dentry->d_inode);
2036 		return ERR_PTR(-ENOENT);
2037 	}
2038 	namespace_lock();
2039 	mnt = lookup_mnt(path);
2040 	if (likely(!mnt)) {
2041 		struct mountpoint *mp = lookup_mountpoint(dentry);
2042 		if (!mp)
2043 			mp = new_mountpoint(dentry);
2044 		if (IS_ERR(mp)) {
2045 			namespace_unlock();
2046 			inode_unlock(dentry->d_inode);
2047 			return mp;
2048 		}
2049 		return mp;
2050 	}
2051 	namespace_unlock();
2052 	inode_unlock(path->dentry->d_inode);
2053 	path_put(path);
2054 	path->mnt = mnt;
2055 	dentry = path->dentry = dget(mnt->mnt_root);
2056 	goto retry;
2057 }
2058 
2059 static void unlock_mount(struct mountpoint *where)
2060 {
2061 	struct dentry *dentry = where->m_dentry;
2062 	put_mountpoint(where);
2063 	namespace_unlock();
2064 	inode_unlock(dentry->d_inode);
2065 }
2066 
2067 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2068 {
2069 	if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2070 		return -EINVAL;
2071 
2072 	if (d_is_dir(mp->m_dentry) !=
2073 	      d_is_dir(mnt->mnt.mnt_root))
2074 		return -ENOTDIR;
2075 
2076 	return attach_recursive_mnt(mnt, p, mp, NULL);
2077 }
2078 
2079 /*
2080  * Sanity check the flags to change_mnt_propagation.
2081  */
2082 
2083 static int flags_to_propagation_type(int flags)
2084 {
2085 	int type = flags & ~(MS_REC | MS_SILENT);
2086 
2087 	/* Fail if any non-propagation flags are set */
2088 	if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2089 		return 0;
2090 	/* Only one propagation flag should be set */
2091 	if (!is_power_of_2(type))
2092 		return 0;
2093 	return type;
2094 }
2095 
2096 /*
2097  * recursively change the type of the mountpoint.
2098  */
2099 static int do_change_type(struct path *path, int flag)
2100 {
2101 	struct mount *m;
2102 	struct mount *mnt = real_mount(path->mnt);
2103 	int recurse = flag & MS_REC;
2104 	int type;
2105 	int err = 0;
2106 
2107 	if (path->dentry != path->mnt->mnt_root)
2108 		return -EINVAL;
2109 
2110 	type = flags_to_propagation_type(flag);
2111 	if (!type)
2112 		return -EINVAL;
2113 
2114 	namespace_lock();
2115 	if (type == MS_SHARED) {
2116 		err = invent_group_ids(mnt, recurse);
2117 		if (err)
2118 			goto out_unlock;
2119 	}
2120 
2121 	lock_mount_hash();
2122 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2123 		change_mnt_propagation(m, type);
2124 	unlock_mount_hash();
2125 
2126  out_unlock:
2127 	namespace_unlock();
2128 	return err;
2129 }
2130 
2131 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2132 {
2133 	struct mount *child;
2134 	list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2135 		if (!is_subdir(child->mnt_mountpoint, dentry))
2136 			continue;
2137 
2138 		if (child->mnt.mnt_flags & MNT_LOCKED)
2139 			return true;
2140 	}
2141 	return false;
2142 }
2143 
2144 /*
2145  * do loopback mount.
2146  */
2147 static int do_loopback(struct path *path, const char *old_name,
2148 				int recurse)
2149 {
2150 	struct path old_path;
2151 	struct mount *mnt = NULL, *old, *parent;
2152 	struct mountpoint *mp;
2153 	int err;
2154 	if (!old_name || !*old_name)
2155 		return -EINVAL;
2156 	err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2157 	if (err)
2158 		return err;
2159 
2160 	err = -EINVAL;
2161 	if (mnt_ns_loop(old_path.dentry))
2162 		goto out;
2163 
2164 	mp = lock_mount(path);
2165 	err = PTR_ERR(mp);
2166 	if (IS_ERR(mp))
2167 		goto out;
2168 
2169 	old = real_mount(old_path.mnt);
2170 	parent = real_mount(path->mnt);
2171 
2172 	err = -EINVAL;
2173 	if (IS_MNT_UNBINDABLE(old))
2174 		goto out2;
2175 
2176 	if (!check_mnt(parent))
2177 		goto out2;
2178 
2179 	if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2180 		goto out2;
2181 
2182 	if (!recurse && has_locked_children(old, old_path.dentry))
2183 		goto out2;
2184 
2185 	if (recurse)
2186 		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2187 	else
2188 		mnt = clone_mnt(old, old_path.dentry, 0);
2189 
2190 	if (IS_ERR(mnt)) {
2191 		err = PTR_ERR(mnt);
2192 		goto out2;
2193 	}
2194 
2195 	mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2196 
2197 	err = graft_tree(mnt, parent, mp);
2198 	if (err) {
2199 		lock_mount_hash();
2200 		umount_tree(mnt, UMOUNT_SYNC);
2201 		unlock_mount_hash();
2202 	}
2203 out2:
2204 	unlock_mount(mp);
2205 out:
2206 	path_put(&old_path);
2207 	return err;
2208 }
2209 
2210 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2211 {
2212 	int error = 0;
2213 	int readonly_request = 0;
2214 
2215 	if (ms_flags & MS_RDONLY)
2216 		readonly_request = 1;
2217 	if (readonly_request == __mnt_is_readonly(mnt))
2218 		return 0;
2219 
2220 	if (readonly_request)
2221 		error = mnt_make_readonly(real_mount(mnt));
2222 	else
2223 		__mnt_unmake_readonly(real_mount(mnt));
2224 	return error;
2225 }
2226 
2227 /*
2228  * change filesystem flags. dir should be a physical root of filesystem.
2229  * If you've mounted a non-root directory somewhere and want to do remount
2230  * on it - tough luck.
2231  */
2232 static int do_remount(struct path *path, int flags, int mnt_flags,
2233 		      void *data)
2234 {
2235 	int err;
2236 	struct super_block *sb = path->mnt->mnt_sb;
2237 	struct mount *mnt = real_mount(path->mnt);
2238 
2239 	if (!check_mnt(mnt))
2240 		return -EINVAL;
2241 
2242 	if (path->dentry != path->mnt->mnt_root)
2243 		return -EINVAL;
2244 
2245 	/* Don't allow changing of locked mnt flags.
2246 	 *
2247 	 * No locks need to be held here while testing the various
2248 	 * MNT_LOCK flags because those flags can never be cleared
2249 	 * once they are set.
2250 	 */
2251 	if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2252 	    !(mnt_flags & MNT_READONLY)) {
2253 		return -EPERM;
2254 	}
2255 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2256 	    !(mnt_flags & MNT_NODEV)) {
2257 		return -EPERM;
2258 	}
2259 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2260 	    !(mnt_flags & MNT_NOSUID)) {
2261 		return -EPERM;
2262 	}
2263 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2264 	    !(mnt_flags & MNT_NOEXEC)) {
2265 		return -EPERM;
2266 	}
2267 	if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2268 	    ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2269 		return -EPERM;
2270 	}
2271 
2272 	err = security_sb_remount(sb, data);
2273 	if (err)
2274 		return err;
2275 
2276 	down_write(&sb->s_umount);
2277 	if (flags & MS_BIND)
2278 		err = change_mount_flags(path->mnt, flags);
2279 	else if (!capable(CAP_SYS_ADMIN))
2280 		err = -EPERM;
2281 	else
2282 		err = do_remount_sb(sb, flags, data, 0);
2283 	if (!err) {
2284 		lock_mount_hash();
2285 		mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2286 		mnt->mnt.mnt_flags = mnt_flags;
2287 		touch_mnt_namespace(mnt->mnt_ns);
2288 		unlock_mount_hash();
2289 	}
2290 	up_write(&sb->s_umount);
2291 	return err;
2292 }
2293 
2294 static inline int tree_contains_unbindable(struct mount *mnt)
2295 {
2296 	struct mount *p;
2297 	for (p = mnt; p; p = next_mnt(p, mnt)) {
2298 		if (IS_MNT_UNBINDABLE(p))
2299 			return 1;
2300 	}
2301 	return 0;
2302 }
2303 
2304 static int do_move_mount(struct path *path, const char *old_name)
2305 {
2306 	struct path old_path, parent_path;
2307 	struct mount *p;
2308 	struct mount *old;
2309 	struct mountpoint *mp;
2310 	int err;
2311 	if (!old_name || !*old_name)
2312 		return -EINVAL;
2313 	err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2314 	if (err)
2315 		return err;
2316 
2317 	mp = lock_mount(path);
2318 	err = PTR_ERR(mp);
2319 	if (IS_ERR(mp))
2320 		goto out;
2321 
2322 	old = real_mount(old_path.mnt);
2323 	p = real_mount(path->mnt);
2324 
2325 	err = -EINVAL;
2326 	if (!check_mnt(p) || !check_mnt(old))
2327 		goto out1;
2328 
2329 	if (old->mnt.mnt_flags & MNT_LOCKED)
2330 		goto out1;
2331 
2332 	err = -EINVAL;
2333 	if (old_path.dentry != old_path.mnt->mnt_root)
2334 		goto out1;
2335 
2336 	if (!mnt_has_parent(old))
2337 		goto out1;
2338 
2339 	if (d_is_dir(path->dentry) !=
2340 	      d_is_dir(old_path.dentry))
2341 		goto out1;
2342 	/*
2343 	 * Don't move a mount residing in a shared parent.
2344 	 */
2345 	if (IS_MNT_SHARED(old->mnt_parent))
2346 		goto out1;
2347 	/*
2348 	 * Don't move a mount tree containing unbindable mounts to a destination
2349 	 * mount which is shared.
2350 	 */
2351 	if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2352 		goto out1;
2353 	err = -ELOOP;
2354 	for (; mnt_has_parent(p); p = p->mnt_parent)
2355 		if (p == old)
2356 			goto out1;
2357 
2358 	err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2359 	if (err)
2360 		goto out1;
2361 
2362 	/* if the mount is moved, it should no longer be expire
2363 	 * automatically */
2364 	list_del_init(&old->mnt_expire);
2365 out1:
2366 	unlock_mount(mp);
2367 out:
2368 	if (!err)
2369 		path_put(&parent_path);
2370 	path_put(&old_path);
2371 	return err;
2372 }
2373 
2374 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2375 {
2376 	int err;
2377 	const char *subtype = strchr(fstype, '.');
2378 	if (subtype) {
2379 		subtype++;
2380 		err = -EINVAL;
2381 		if (!subtype[0])
2382 			goto err;
2383 	} else
2384 		subtype = "";
2385 
2386 	mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2387 	err = -ENOMEM;
2388 	if (!mnt->mnt_sb->s_subtype)
2389 		goto err;
2390 	return mnt;
2391 
2392  err:
2393 	mntput(mnt);
2394 	return ERR_PTR(err);
2395 }
2396 
2397 /*
2398  * add a mount into a namespace's mount tree
2399  */
2400 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2401 {
2402 	struct mountpoint *mp;
2403 	struct mount *parent;
2404 	int err;
2405 
2406 	mnt_flags &= ~MNT_INTERNAL_FLAGS;
2407 
2408 	mp = lock_mount(path);
2409 	if (IS_ERR(mp))
2410 		return PTR_ERR(mp);
2411 
2412 	parent = real_mount(path->mnt);
2413 	err = -EINVAL;
2414 	if (unlikely(!check_mnt(parent))) {
2415 		/* that's acceptable only for automounts done in private ns */
2416 		if (!(mnt_flags & MNT_SHRINKABLE))
2417 			goto unlock;
2418 		/* ... and for those we'd better have mountpoint still alive */
2419 		if (!parent->mnt_ns)
2420 			goto unlock;
2421 	}
2422 
2423 	/* Refuse the same filesystem on the same mount point */
2424 	err = -EBUSY;
2425 	if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2426 	    path->mnt->mnt_root == path->dentry)
2427 		goto unlock;
2428 
2429 	err = -EINVAL;
2430 	if (d_is_symlink(newmnt->mnt.mnt_root))
2431 		goto unlock;
2432 
2433 	newmnt->mnt.mnt_flags = mnt_flags;
2434 	err = graft_tree(newmnt, parent, mp);
2435 
2436 unlock:
2437 	unlock_mount(mp);
2438 	return err;
2439 }
2440 
2441 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2442 
2443 /*
2444  * create a new mount for userspace and request it to be added into the
2445  * namespace's tree
2446  */
2447 static int do_new_mount(struct path *path, const char *fstype, int flags,
2448 			int mnt_flags, const char *name, void *data)
2449 {
2450 	struct file_system_type *type;
2451 	struct vfsmount *mnt;
2452 	int err;
2453 
2454 	if (!fstype)
2455 		return -EINVAL;
2456 
2457 	type = get_fs_type(fstype);
2458 	if (!type)
2459 		return -ENODEV;
2460 
2461 	mnt = vfs_kern_mount(type, flags, name, data);
2462 	if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2463 	    !mnt->mnt_sb->s_subtype)
2464 		mnt = fs_set_subtype(mnt, fstype);
2465 
2466 	put_filesystem(type);
2467 	if (IS_ERR(mnt))
2468 		return PTR_ERR(mnt);
2469 
2470 	if (mount_too_revealing(mnt, &mnt_flags)) {
2471 		mntput(mnt);
2472 		return -EPERM;
2473 	}
2474 
2475 	err = do_add_mount(real_mount(mnt), path, mnt_flags);
2476 	if (err)
2477 		mntput(mnt);
2478 	return err;
2479 }
2480 
2481 int finish_automount(struct vfsmount *m, struct path *path)
2482 {
2483 	struct mount *mnt = real_mount(m);
2484 	int err;
2485 	/* The new mount record should have at least 2 refs to prevent it being
2486 	 * expired before we get a chance to add it
2487 	 */
2488 	BUG_ON(mnt_get_count(mnt) < 2);
2489 
2490 	if (m->mnt_sb == path->mnt->mnt_sb &&
2491 	    m->mnt_root == path->dentry) {
2492 		err = -ELOOP;
2493 		goto fail;
2494 	}
2495 
2496 	err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2497 	if (!err)
2498 		return 0;
2499 fail:
2500 	/* remove m from any expiration list it may be on */
2501 	if (!list_empty(&mnt->mnt_expire)) {
2502 		namespace_lock();
2503 		list_del_init(&mnt->mnt_expire);
2504 		namespace_unlock();
2505 	}
2506 	mntput(m);
2507 	mntput(m);
2508 	return err;
2509 }
2510 
2511 /**
2512  * mnt_set_expiry - Put a mount on an expiration list
2513  * @mnt: The mount to list.
2514  * @expiry_list: The list to add the mount to.
2515  */
2516 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2517 {
2518 	namespace_lock();
2519 
2520 	list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2521 
2522 	namespace_unlock();
2523 }
2524 EXPORT_SYMBOL(mnt_set_expiry);
2525 
2526 /*
2527  * process a list of expirable mountpoints with the intent of discarding any
2528  * mountpoints that aren't in use and haven't been touched since last we came
2529  * here
2530  */
2531 void mark_mounts_for_expiry(struct list_head *mounts)
2532 {
2533 	struct mount *mnt, *next;
2534 	LIST_HEAD(graveyard);
2535 
2536 	if (list_empty(mounts))
2537 		return;
2538 
2539 	namespace_lock();
2540 	lock_mount_hash();
2541 
2542 	/* extract from the expiration list every vfsmount that matches the
2543 	 * following criteria:
2544 	 * - only referenced by its parent vfsmount
2545 	 * - still marked for expiry (marked on the last call here; marks are
2546 	 *   cleared by mntput())
2547 	 */
2548 	list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2549 		if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2550 			propagate_mount_busy(mnt, 1))
2551 			continue;
2552 		list_move(&mnt->mnt_expire, &graveyard);
2553 	}
2554 	while (!list_empty(&graveyard)) {
2555 		mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2556 		touch_mnt_namespace(mnt->mnt_ns);
2557 		umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2558 	}
2559 	unlock_mount_hash();
2560 	namespace_unlock();
2561 }
2562 
2563 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2564 
2565 /*
2566  * Ripoff of 'select_parent()'
2567  *
2568  * search the list of submounts for a given mountpoint, and move any
2569  * shrinkable submounts to the 'graveyard' list.
2570  */
2571 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2572 {
2573 	struct mount *this_parent = parent;
2574 	struct list_head *next;
2575 	int found = 0;
2576 
2577 repeat:
2578 	next = this_parent->mnt_mounts.next;
2579 resume:
2580 	while (next != &this_parent->mnt_mounts) {
2581 		struct list_head *tmp = next;
2582 		struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2583 
2584 		next = tmp->next;
2585 		if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2586 			continue;
2587 		/*
2588 		 * Descend a level if the d_mounts list is non-empty.
2589 		 */
2590 		if (!list_empty(&mnt->mnt_mounts)) {
2591 			this_parent = mnt;
2592 			goto repeat;
2593 		}
2594 
2595 		if (!propagate_mount_busy(mnt, 1)) {
2596 			list_move_tail(&mnt->mnt_expire, graveyard);
2597 			found++;
2598 		}
2599 	}
2600 	/*
2601 	 * All done at this level ... ascend and resume the search
2602 	 */
2603 	if (this_parent != parent) {
2604 		next = this_parent->mnt_child.next;
2605 		this_parent = this_parent->mnt_parent;
2606 		goto resume;
2607 	}
2608 	return found;
2609 }
2610 
2611 /*
2612  * process a list of expirable mountpoints with the intent of discarding any
2613  * submounts of a specific parent mountpoint
2614  *
2615  * mount_lock must be held for write
2616  */
2617 static void shrink_submounts(struct mount *mnt)
2618 {
2619 	LIST_HEAD(graveyard);
2620 	struct mount *m;
2621 
2622 	/* extract submounts of 'mountpoint' from the expiration list */
2623 	while (select_submounts(mnt, &graveyard)) {
2624 		while (!list_empty(&graveyard)) {
2625 			m = list_first_entry(&graveyard, struct mount,
2626 						mnt_expire);
2627 			touch_mnt_namespace(m->mnt_ns);
2628 			umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2629 		}
2630 	}
2631 }
2632 
2633 /*
2634  * Some copy_from_user() implementations do not return the exact number of
2635  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2636  * Note that this function differs from copy_from_user() in that it will oops
2637  * on bad values of `to', rather than returning a short copy.
2638  */
2639 static long exact_copy_from_user(void *to, const void __user * from,
2640 				 unsigned long n)
2641 {
2642 	char *t = to;
2643 	const char __user *f = from;
2644 	char c;
2645 
2646 	if (!access_ok(VERIFY_READ, from, n))
2647 		return n;
2648 
2649 	while (n) {
2650 		if (__get_user(c, f)) {
2651 			memset(t, 0, n);
2652 			break;
2653 		}
2654 		*t++ = c;
2655 		f++;
2656 		n--;
2657 	}
2658 	return n;
2659 }
2660 
2661 void *copy_mount_options(const void __user * data)
2662 {
2663 	int i;
2664 	unsigned long size;
2665 	char *copy;
2666 
2667 	if (!data)
2668 		return NULL;
2669 
2670 	copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2671 	if (!copy)
2672 		return ERR_PTR(-ENOMEM);
2673 
2674 	/* We only care that *some* data at the address the user
2675 	 * gave us is valid.  Just in case, we'll zero
2676 	 * the remainder of the page.
2677 	 */
2678 	/* copy_from_user cannot cross TASK_SIZE ! */
2679 	size = TASK_SIZE - (unsigned long)data;
2680 	if (size > PAGE_SIZE)
2681 		size = PAGE_SIZE;
2682 
2683 	i = size - exact_copy_from_user(copy, data, size);
2684 	if (!i) {
2685 		kfree(copy);
2686 		return ERR_PTR(-EFAULT);
2687 	}
2688 	if (i != PAGE_SIZE)
2689 		memset(copy + i, 0, PAGE_SIZE - i);
2690 	return copy;
2691 }
2692 
2693 char *copy_mount_string(const void __user *data)
2694 {
2695 	return data ? strndup_user(data, PAGE_SIZE) : NULL;
2696 }
2697 
2698 /*
2699  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2700  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2701  *
2702  * data is a (void *) that can point to any structure up to
2703  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2704  * information (or be NULL).
2705  *
2706  * Pre-0.97 versions of mount() didn't have a flags word.
2707  * When the flags word was introduced its top half was required
2708  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2709  * Therefore, if this magic number is present, it carries no information
2710  * and must be discarded.
2711  */
2712 long do_mount(const char *dev_name, const char __user *dir_name,
2713 		const char *type_page, unsigned long flags, void *data_page)
2714 {
2715 	struct path path;
2716 	int retval = 0;
2717 	int mnt_flags = 0;
2718 
2719 	/* Discard magic */
2720 	if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2721 		flags &= ~MS_MGC_MSK;
2722 
2723 	/* Basic sanity checks */
2724 	if (data_page)
2725 		((char *)data_page)[PAGE_SIZE - 1] = 0;
2726 
2727 	/* ... and get the mountpoint */
2728 	retval = user_path(dir_name, &path);
2729 	if (retval)
2730 		return retval;
2731 
2732 	retval = security_sb_mount(dev_name, &path,
2733 				   type_page, flags, data_page);
2734 	if (!retval && !may_mount())
2735 		retval = -EPERM;
2736 	if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
2737 		retval = -EPERM;
2738 	if (retval)
2739 		goto dput_out;
2740 
2741 	/* Default to relatime unless overriden */
2742 	if (!(flags & MS_NOATIME))
2743 		mnt_flags |= MNT_RELATIME;
2744 
2745 	/* Separate the per-mountpoint flags */
2746 	if (flags & MS_NOSUID)
2747 		mnt_flags |= MNT_NOSUID;
2748 	if (flags & MS_NODEV)
2749 		mnt_flags |= MNT_NODEV;
2750 	if (flags & MS_NOEXEC)
2751 		mnt_flags |= MNT_NOEXEC;
2752 	if (flags & MS_NOATIME)
2753 		mnt_flags |= MNT_NOATIME;
2754 	if (flags & MS_NODIRATIME)
2755 		mnt_flags |= MNT_NODIRATIME;
2756 	if (flags & MS_STRICTATIME)
2757 		mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2758 	if (flags & MS_RDONLY)
2759 		mnt_flags |= MNT_READONLY;
2760 
2761 	/* The default atime for remount is preservation */
2762 	if ((flags & MS_REMOUNT) &&
2763 	    ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2764 		       MS_STRICTATIME)) == 0)) {
2765 		mnt_flags &= ~MNT_ATIME_MASK;
2766 		mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2767 	}
2768 
2769 	flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2770 		   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2771 		   MS_STRICTATIME | MS_NOREMOTELOCK);
2772 
2773 	if (flags & MS_REMOUNT)
2774 		retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2775 				    data_page);
2776 	else if (flags & MS_BIND)
2777 		retval = do_loopback(&path, dev_name, flags & MS_REC);
2778 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2779 		retval = do_change_type(&path, flags);
2780 	else if (flags & MS_MOVE)
2781 		retval = do_move_mount(&path, dev_name);
2782 	else
2783 		retval = do_new_mount(&path, type_page, flags, mnt_flags,
2784 				      dev_name, data_page);
2785 dput_out:
2786 	path_put(&path);
2787 	return retval;
2788 }
2789 
2790 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2791 {
2792 	return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2793 }
2794 
2795 static void dec_mnt_namespaces(struct ucounts *ucounts)
2796 {
2797 	dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2798 }
2799 
2800 static void free_mnt_ns(struct mnt_namespace *ns)
2801 {
2802 	ns_free_inum(&ns->ns);
2803 	dec_mnt_namespaces(ns->ucounts);
2804 	put_user_ns(ns->user_ns);
2805 	kfree(ns);
2806 }
2807 
2808 /*
2809  * Assign a sequence number so we can detect when we attempt to bind
2810  * mount a reference to an older mount namespace into the current
2811  * mount namespace, preventing reference counting loops.  A 64bit
2812  * number incrementing at 10Ghz will take 12,427 years to wrap which
2813  * is effectively never, so we can ignore the possibility.
2814  */
2815 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2816 
2817 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2818 {
2819 	struct mnt_namespace *new_ns;
2820 	struct ucounts *ucounts;
2821 	int ret;
2822 
2823 	ucounts = inc_mnt_namespaces(user_ns);
2824 	if (!ucounts)
2825 		return ERR_PTR(-ENOSPC);
2826 
2827 	new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2828 	if (!new_ns) {
2829 		dec_mnt_namespaces(ucounts);
2830 		return ERR_PTR(-ENOMEM);
2831 	}
2832 	ret = ns_alloc_inum(&new_ns->ns);
2833 	if (ret) {
2834 		kfree(new_ns);
2835 		dec_mnt_namespaces(ucounts);
2836 		return ERR_PTR(ret);
2837 	}
2838 	new_ns->ns.ops = &mntns_operations;
2839 	new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2840 	atomic_set(&new_ns->count, 1);
2841 	new_ns->root = NULL;
2842 	INIT_LIST_HEAD(&new_ns->list);
2843 	init_waitqueue_head(&new_ns->poll);
2844 	new_ns->event = 0;
2845 	new_ns->user_ns = get_user_ns(user_ns);
2846 	new_ns->ucounts = ucounts;
2847 	new_ns->mounts = 0;
2848 	new_ns->pending_mounts = 0;
2849 	return new_ns;
2850 }
2851 
2852 __latent_entropy
2853 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2854 		struct user_namespace *user_ns, struct fs_struct *new_fs)
2855 {
2856 	struct mnt_namespace *new_ns;
2857 	struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2858 	struct mount *p, *q;
2859 	struct mount *old;
2860 	struct mount *new;
2861 	int copy_flags;
2862 
2863 	BUG_ON(!ns);
2864 
2865 	if (likely(!(flags & CLONE_NEWNS))) {
2866 		get_mnt_ns(ns);
2867 		return ns;
2868 	}
2869 
2870 	old = ns->root;
2871 
2872 	new_ns = alloc_mnt_ns(user_ns);
2873 	if (IS_ERR(new_ns))
2874 		return new_ns;
2875 
2876 	namespace_lock();
2877 	/* First pass: copy the tree topology */
2878 	copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2879 	if (user_ns != ns->user_ns)
2880 		copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2881 	new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2882 	if (IS_ERR(new)) {
2883 		namespace_unlock();
2884 		free_mnt_ns(new_ns);
2885 		return ERR_CAST(new);
2886 	}
2887 	new_ns->root = new;
2888 	list_add_tail(&new_ns->list, &new->mnt_list);
2889 
2890 	/*
2891 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2892 	 * as belonging to new namespace.  We have already acquired a private
2893 	 * fs_struct, so tsk->fs->lock is not needed.
2894 	 */
2895 	p = old;
2896 	q = new;
2897 	while (p) {
2898 		q->mnt_ns = new_ns;
2899 		new_ns->mounts++;
2900 		if (new_fs) {
2901 			if (&p->mnt == new_fs->root.mnt) {
2902 				new_fs->root.mnt = mntget(&q->mnt);
2903 				rootmnt = &p->mnt;
2904 			}
2905 			if (&p->mnt == new_fs->pwd.mnt) {
2906 				new_fs->pwd.mnt = mntget(&q->mnt);
2907 				pwdmnt = &p->mnt;
2908 			}
2909 		}
2910 		p = next_mnt(p, old);
2911 		q = next_mnt(q, new);
2912 		if (!q)
2913 			break;
2914 		while (p->mnt.mnt_root != q->mnt.mnt_root)
2915 			p = next_mnt(p, old);
2916 	}
2917 	namespace_unlock();
2918 
2919 	if (rootmnt)
2920 		mntput(rootmnt);
2921 	if (pwdmnt)
2922 		mntput(pwdmnt);
2923 
2924 	return new_ns;
2925 }
2926 
2927 /**
2928  * create_mnt_ns - creates a private namespace and adds a root filesystem
2929  * @mnt: pointer to the new root filesystem mountpoint
2930  */
2931 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2932 {
2933 	struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2934 	if (!IS_ERR(new_ns)) {
2935 		struct mount *mnt = real_mount(m);
2936 		mnt->mnt_ns = new_ns;
2937 		new_ns->root = mnt;
2938 		new_ns->mounts++;
2939 		list_add(&mnt->mnt_list, &new_ns->list);
2940 	} else {
2941 		mntput(m);
2942 	}
2943 	return new_ns;
2944 }
2945 
2946 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2947 {
2948 	struct mnt_namespace *ns;
2949 	struct super_block *s;
2950 	struct path path;
2951 	int err;
2952 
2953 	ns = create_mnt_ns(mnt);
2954 	if (IS_ERR(ns))
2955 		return ERR_CAST(ns);
2956 
2957 	err = vfs_path_lookup(mnt->mnt_root, mnt,
2958 			name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2959 
2960 	put_mnt_ns(ns);
2961 
2962 	if (err)
2963 		return ERR_PTR(err);
2964 
2965 	/* trade a vfsmount reference for active sb one */
2966 	s = path.mnt->mnt_sb;
2967 	atomic_inc(&s->s_active);
2968 	mntput(path.mnt);
2969 	/* lock the sucker */
2970 	down_write(&s->s_umount);
2971 	/* ... and return the root of (sub)tree on it */
2972 	return path.dentry;
2973 }
2974 EXPORT_SYMBOL(mount_subtree);
2975 
2976 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2977 		char __user *, type, unsigned long, flags, void __user *, data)
2978 {
2979 	int ret;
2980 	char *kernel_type;
2981 	char *kernel_dev;
2982 	void *options;
2983 
2984 	kernel_type = copy_mount_string(type);
2985 	ret = PTR_ERR(kernel_type);
2986 	if (IS_ERR(kernel_type))
2987 		goto out_type;
2988 
2989 	kernel_dev = copy_mount_string(dev_name);
2990 	ret = PTR_ERR(kernel_dev);
2991 	if (IS_ERR(kernel_dev))
2992 		goto out_dev;
2993 
2994 	options = copy_mount_options(data);
2995 	ret = PTR_ERR(options);
2996 	if (IS_ERR(options))
2997 		goto out_data;
2998 
2999 	ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3000 
3001 	kfree(options);
3002 out_data:
3003 	kfree(kernel_dev);
3004 out_dev:
3005 	kfree(kernel_type);
3006 out_type:
3007 	return ret;
3008 }
3009 
3010 /*
3011  * Return true if path is reachable from root
3012  *
3013  * namespace_sem or mount_lock is held
3014  */
3015 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3016 			 const struct path *root)
3017 {
3018 	while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3019 		dentry = mnt->mnt_mountpoint;
3020 		mnt = mnt->mnt_parent;
3021 	}
3022 	return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3023 }
3024 
3025 bool path_is_under(const struct path *path1, const struct path *path2)
3026 {
3027 	bool res;
3028 	read_seqlock_excl(&mount_lock);
3029 	res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3030 	read_sequnlock_excl(&mount_lock);
3031 	return res;
3032 }
3033 EXPORT_SYMBOL(path_is_under);
3034 
3035 /*
3036  * pivot_root Semantics:
3037  * Moves the root file system of the current process to the directory put_old,
3038  * makes new_root as the new root file system of the current process, and sets
3039  * root/cwd of all processes which had them on the current root to new_root.
3040  *
3041  * Restrictions:
3042  * The new_root and put_old must be directories, and  must not be on the
3043  * same file  system as the current process root. The put_old  must  be
3044  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3045  * pointed to by put_old must yield the same directory as new_root. No other
3046  * file system may be mounted on put_old. After all, new_root is a mountpoint.
3047  *
3048  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3049  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3050  * in this situation.
3051  *
3052  * Notes:
3053  *  - we don't move root/cwd if they are not at the root (reason: if something
3054  *    cared enough to change them, it's probably wrong to force them elsewhere)
3055  *  - it's okay to pick a root that isn't the root of a file system, e.g.
3056  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3057  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3058  *    first.
3059  */
3060 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3061 		const char __user *, put_old)
3062 {
3063 	struct path new, old, parent_path, root_parent, root;
3064 	struct mount *new_mnt, *root_mnt, *old_mnt;
3065 	struct mountpoint *old_mp, *root_mp;
3066 	int error;
3067 
3068 	if (!may_mount())
3069 		return -EPERM;
3070 
3071 	error = user_path_dir(new_root, &new);
3072 	if (error)
3073 		goto out0;
3074 
3075 	error = user_path_dir(put_old, &old);
3076 	if (error)
3077 		goto out1;
3078 
3079 	error = security_sb_pivotroot(&old, &new);
3080 	if (error)
3081 		goto out2;
3082 
3083 	get_fs_root(current->fs, &root);
3084 	old_mp = lock_mount(&old);
3085 	error = PTR_ERR(old_mp);
3086 	if (IS_ERR(old_mp))
3087 		goto out3;
3088 
3089 	error = -EINVAL;
3090 	new_mnt = real_mount(new.mnt);
3091 	root_mnt = real_mount(root.mnt);
3092 	old_mnt = real_mount(old.mnt);
3093 	if (IS_MNT_SHARED(old_mnt) ||
3094 		IS_MNT_SHARED(new_mnt->mnt_parent) ||
3095 		IS_MNT_SHARED(root_mnt->mnt_parent))
3096 		goto out4;
3097 	if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3098 		goto out4;
3099 	if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3100 		goto out4;
3101 	error = -ENOENT;
3102 	if (d_unlinked(new.dentry))
3103 		goto out4;
3104 	error = -EBUSY;
3105 	if (new_mnt == root_mnt || old_mnt == root_mnt)
3106 		goto out4; /* loop, on the same file system  */
3107 	error = -EINVAL;
3108 	if (root.mnt->mnt_root != root.dentry)
3109 		goto out4; /* not a mountpoint */
3110 	if (!mnt_has_parent(root_mnt))
3111 		goto out4; /* not attached */
3112 	root_mp = root_mnt->mnt_mp;
3113 	if (new.mnt->mnt_root != new.dentry)
3114 		goto out4; /* not a mountpoint */
3115 	if (!mnt_has_parent(new_mnt))
3116 		goto out4; /* not attached */
3117 	/* make sure we can reach put_old from new_root */
3118 	if (!is_path_reachable(old_mnt, old.dentry, &new))
3119 		goto out4;
3120 	/* make certain new is below the root */
3121 	if (!is_path_reachable(new_mnt, new.dentry, &root))
3122 		goto out4;
3123 	root_mp->m_count++; /* pin it so it won't go away */
3124 	lock_mount_hash();
3125 	detach_mnt(new_mnt, &parent_path);
3126 	detach_mnt(root_mnt, &root_parent);
3127 	if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3128 		new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3129 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3130 	}
3131 	/* mount old root on put_old */
3132 	attach_mnt(root_mnt, old_mnt, old_mp);
3133 	/* mount new_root on / */
3134 	attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3135 	touch_mnt_namespace(current->nsproxy->mnt_ns);
3136 	/* A moved mount should not expire automatically */
3137 	list_del_init(&new_mnt->mnt_expire);
3138 	unlock_mount_hash();
3139 	chroot_fs_refs(&root, &new);
3140 	put_mountpoint(root_mp);
3141 	error = 0;
3142 out4:
3143 	unlock_mount(old_mp);
3144 	if (!error) {
3145 		path_put(&root_parent);
3146 		path_put(&parent_path);
3147 	}
3148 out3:
3149 	path_put(&root);
3150 out2:
3151 	path_put(&old);
3152 out1:
3153 	path_put(&new);
3154 out0:
3155 	return error;
3156 }
3157 
3158 static void __init init_mount_tree(void)
3159 {
3160 	struct vfsmount *mnt;
3161 	struct mnt_namespace *ns;
3162 	struct path root;
3163 	struct file_system_type *type;
3164 
3165 	type = get_fs_type("rootfs");
3166 	if (!type)
3167 		panic("Can't find rootfs type");
3168 	mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3169 	put_filesystem(type);
3170 	if (IS_ERR(mnt))
3171 		panic("Can't create rootfs");
3172 
3173 	ns = create_mnt_ns(mnt);
3174 	if (IS_ERR(ns))
3175 		panic("Can't allocate initial namespace");
3176 
3177 	init_task.nsproxy->mnt_ns = ns;
3178 	get_mnt_ns(ns);
3179 
3180 	root.mnt = mnt;
3181 	root.dentry = mnt->mnt_root;
3182 	mnt->mnt_flags |= MNT_LOCKED;
3183 
3184 	set_fs_pwd(current->fs, &root);
3185 	set_fs_root(current->fs, &root);
3186 }
3187 
3188 void __init mnt_init(void)
3189 {
3190 	unsigned u;
3191 	int err;
3192 
3193 	mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3194 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3195 
3196 	mount_hashtable = alloc_large_system_hash("Mount-cache",
3197 				sizeof(struct hlist_head),
3198 				mhash_entries, 19,
3199 				0,
3200 				&m_hash_shift, &m_hash_mask, 0, 0);
3201 	mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3202 				sizeof(struct hlist_head),
3203 				mphash_entries, 19,
3204 				0,
3205 				&mp_hash_shift, &mp_hash_mask, 0, 0);
3206 
3207 	if (!mount_hashtable || !mountpoint_hashtable)
3208 		panic("Failed to allocate mount hash table\n");
3209 
3210 	for (u = 0; u <= m_hash_mask; u++)
3211 		INIT_HLIST_HEAD(&mount_hashtable[u]);
3212 	for (u = 0; u <= mp_hash_mask; u++)
3213 		INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3214 
3215 	kernfs_init();
3216 
3217 	err = sysfs_init();
3218 	if (err)
3219 		printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3220 			__func__, err);
3221 	fs_kobj = kobject_create_and_add("fs", NULL);
3222 	if (!fs_kobj)
3223 		printk(KERN_WARNING "%s: kobj create error\n", __func__);
3224 	init_rootfs();
3225 	init_mount_tree();
3226 }
3227 
3228 void put_mnt_ns(struct mnt_namespace *ns)
3229 {
3230 	if (!atomic_dec_and_test(&ns->count))
3231 		return;
3232 	drop_collected_mounts(&ns->root->mnt);
3233 	free_mnt_ns(ns);
3234 }
3235 
3236 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3237 {
3238 	struct vfsmount *mnt;
3239 	mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3240 	if (!IS_ERR(mnt)) {
3241 		/*
3242 		 * it is a longterm mount, don't release mnt until
3243 		 * we unmount before file sys is unregistered
3244 		*/
3245 		real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3246 	}
3247 	return mnt;
3248 }
3249 EXPORT_SYMBOL_GPL(kern_mount_data);
3250 
3251 void kern_unmount(struct vfsmount *mnt)
3252 {
3253 	/* release long term mount so mount point can be released */
3254 	if (!IS_ERR_OR_NULL(mnt)) {
3255 		real_mount(mnt)->mnt_ns = NULL;
3256 		synchronize_rcu();	/* yecchhh... */
3257 		mntput(mnt);
3258 	}
3259 }
3260 EXPORT_SYMBOL(kern_unmount);
3261 
3262 bool our_mnt(struct vfsmount *mnt)
3263 {
3264 	return check_mnt(real_mount(mnt));
3265 }
3266 
3267 bool current_chrooted(void)
3268 {
3269 	/* Does the current process have a non-standard root */
3270 	struct path ns_root;
3271 	struct path fs_root;
3272 	bool chrooted;
3273 
3274 	/* Find the namespace root */
3275 	ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3276 	ns_root.dentry = ns_root.mnt->mnt_root;
3277 	path_get(&ns_root);
3278 	while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3279 		;
3280 
3281 	get_fs_root(current->fs, &fs_root);
3282 
3283 	chrooted = !path_equal(&fs_root, &ns_root);
3284 
3285 	path_put(&fs_root);
3286 	path_put(&ns_root);
3287 
3288 	return chrooted;
3289 }
3290 
3291 static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3292 				int *new_mnt_flags)
3293 {
3294 	int new_flags = *new_mnt_flags;
3295 	struct mount *mnt;
3296 	bool visible = false;
3297 
3298 	down_read(&namespace_sem);
3299 	list_for_each_entry(mnt, &ns->list, mnt_list) {
3300 		struct mount *child;
3301 		int mnt_flags;
3302 
3303 		if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3304 			continue;
3305 
3306 		/* This mount is not fully visible if it's root directory
3307 		 * is not the root directory of the filesystem.
3308 		 */
3309 		if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3310 			continue;
3311 
3312 		/* A local view of the mount flags */
3313 		mnt_flags = mnt->mnt.mnt_flags;
3314 
3315 		/* Don't miss readonly hidden in the superblock flags */
3316 		if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3317 			mnt_flags |= MNT_LOCK_READONLY;
3318 
3319 		/* Verify the mount flags are equal to or more permissive
3320 		 * than the proposed new mount.
3321 		 */
3322 		if ((mnt_flags & MNT_LOCK_READONLY) &&
3323 		    !(new_flags & MNT_READONLY))
3324 			continue;
3325 		if ((mnt_flags & MNT_LOCK_ATIME) &&
3326 		    ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3327 			continue;
3328 
3329 		/* This mount is not fully visible if there are any
3330 		 * locked child mounts that cover anything except for
3331 		 * empty directories.
3332 		 */
3333 		list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3334 			struct inode *inode = child->mnt_mountpoint->d_inode;
3335 			/* Only worry about locked mounts */
3336 			if (!(child->mnt.mnt_flags & MNT_LOCKED))
3337 				continue;
3338 			/* Is the directory permanetly empty? */
3339 			if (!is_empty_dir_inode(inode))
3340 				goto next;
3341 		}
3342 		/* Preserve the locked attributes */
3343 		*new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3344 					       MNT_LOCK_ATIME);
3345 		visible = true;
3346 		goto found;
3347 	next:	;
3348 	}
3349 found:
3350 	up_read(&namespace_sem);
3351 	return visible;
3352 }
3353 
3354 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3355 {
3356 	const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3357 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3358 	unsigned long s_iflags;
3359 
3360 	if (ns->user_ns == &init_user_ns)
3361 		return false;
3362 
3363 	/* Can this filesystem be too revealing? */
3364 	s_iflags = mnt->mnt_sb->s_iflags;
3365 	if (!(s_iflags & SB_I_USERNS_VISIBLE))
3366 		return false;
3367 
3368 	if ((s_iflags & required_iflags) != required_iflags) {
3369 		WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3370 			  required_iflags);
3371 		return true;
3372 	}
3373 
3374 	return !mnt_already_visible(ns, mnt, new_mnt_flags);
3375 }
3376 
3377 bool mnt_may_suid(struct vfsmount *mnt)
3378 {
3379 	/*
3380 	 * Foreign mounts (accessed via fchdir or through /proc
3381 	 * symlinks) are always treated as if they are nosuid.  This
3382 	 * prevents namespaces from trusting potentially unsafe
3383 	 * suid/sgid bits, file caps, or security labels that originate
3384 	 * in other namespaces.
3385 	 */
3386 	return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3387 	       current_in_userns(mnt->mnt_sb->s_user_ns);
3388 }
3389 
3390 static struct ns_common *mntns_get(struct task_struct *task)
3391 {
3392 	struct ns_common *ns = NULL;
3393 	struct nsproxy *nsproxy;
3394 
3395 	task_lock(task);
3396 	nsproxy = task->nsproxy;
3397 	if (nsproxy) {
3398 		ns = &nsproxy->mnt_ns->ns;
3399 		get_mnt_ns(to_mnt_ns(ns));
3400 	}
3401 	task_unlock(task);
3402 
3403 	return ns;
3404 }
3405 
3406 static void mntns_put(struct ns_common *ns)
3407 {
3408 	put_mnt_ns(to_mnt_ns(ns));
3409 }
3410 
3411 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3412 {
3413 	struct fs_struct *fs = current->fs;
3414 	struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3415 	struct path root;
3416 
3417 	if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3418 	    !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3419 	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3420 		return -EPERM;
3421 
3422 	if (fs->users != 1)
3423 		return -EINVAL;
3424 
3425 	get_mnt_ns(mnt_ns);
3426 	put_mnt_ns(nsproxy->mnt_ns);
3427 	nsproxy->mnt_ns = mnt_ns;
3428 
3429 	/* Find the root */
3430 	root.mnt    = &mnt_ns->root->mnt;
3431 	root.dentry = mnt_ns->root->mnt.mnt_root;
3432 	path_get(&root);
3433 	while(d_mountpoint(root.dentry) && follow_down_one(&root))
3434 		;
3435 
3436 	/* Update the pwd and root */
3437 	set_fs_pwd(fs, &root);
3438 	set_fs_root(fs, &root);
3439 
3440 	path_put(&root);
3441 	return 0;
3442 }
3443 
3444 static struct user_namespace *mntns_owner(struct ns_common *ns)
3445 {
3446 	return to_mnt_ns(ns)->user_ns;
3447 }
3448 
3449 const struct proc_ns_operations mntns_operations = {
3450 	.name		= "mnt",
3451 	.type		= CLONE_NEWNS,
3452 	.get		= mntns_get,
3453 	.put		= mntns_put,
3454 	.install	= mntns_install,
3455 	.owner		= mntns_owner,
3456 };
3457