xref: /openbmc/linux/fs/namespace.c (revision d7a3d85e)
1 /*
2  *  linux/fs/namespace.c
3  *
4  * (C) Copyright Al Viro 2000, 2001
5  *	Released under GPL v2.
6  *
7  * Based on code from fs/super.c, copyright Linus Torvalds and others.
8  * Heavily rewritten.
9  */
10 
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/init.h>		/* init_rootfs */
20 #include <linux/fs_struct.h>	/* get_fs_root et.al. */
21 #include <linux/fsnotify.h>	/* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
27 #include "pnode.h"
28 #include "internal.h"
29 
30 static unsigned int m_hash_mask __read_mostly;
31 static unsigned int m_hash_shift __read_mostly;
32 static unsigned int mp_hash_mask __read_mostly;
33 static unsigned int mp_hash_shift __read_mostly;
34 
35 static __initdata unsigned long mhash_entries;
36 static int __init set_mhash_entries(char *str)
37 {
38 	if (!str)
39 		return 0;
40 	mhash_entries = simple_strtoul(str, &str, 0);
41 	return 1;
42 }
43 __setup("mhash_entries=", set_mhash_entries);
44 
45 static __initdata unsigned long mphash_entries;
46 static int __init set_mphash_entries(char *str)
47 {
48 	if (!str)
49 		return 0;
50 	mphash_entries = simple_strtoul(str, &str, 0);
51 	return 1;
52 }
53 __setup("mphash_entries=", set_mphash_entries);
54 
55 static u64 event;
56 static DEFINE_IDA(mnt_id_ida);
57 static DEFINE_IDA(mnt_group_ida);
58 static DEFINE_SPINLOCK(mnt_id_lock);
59 static int mnt_id_start = 0;
60 static int mnt_group_start = 1;
61 
62 static struct hlist_head *mount_hashtable __read_mostly;
63 static struct hlist_head *mountpoint_hashtable __read_mostly;
64 static struct kmem_cache *mnt_cache __read_mostly;
65 static DECLARE_RWSEM(namespace_sem);
66 
67 /* /sys/fs */
68 struct kobject *fs_kobj;
69 EXPORT_SYMBOL_GPL(fs_kobj);
70 
71 /*
72  * vfsmount lock may be taken for read to prevent changes to the
73  * vfsmount hash, ie. during mountpoint lookups or walking back
74  * up the tree.
75  *
76  * It should be taken for write in all cases where the vfsmount
77  * tree or hash is modified or when a vfsmount structure is modified.
78  */
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
80 
81 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
82 {
83 	unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 	tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
85 	tmp = tmp + (tmp >> m_hash_shift);
86 	return &mount_hashtable[tmp & m_hash_mask];
87 }
88 
89 static inline struct hlist_head *mp_hash(struct dentry *dentry)
90 {
91 	unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 	tmp = tmp + (tmp >> mp_hash_shift);
93 	return &mountpoint_hashtable[tmp & mp_hash_mask];
94 }
95 
96 /*
97  * allocation is serialized by namespace_sem, but we need the spinlock to
98  * serialize with freeing.
99  */
100 static int mnt_alloc_id(struct mount *mnt)
101 {
102 	int res;
103 
104 retry:
105 	ida_pre_get(&mnt_id_ida, GFP_KERNEL);
106 	spin_lock(&mnt_id_lock);
107 	res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
108 	if (!res)
109 		mnt_id_start = mnt->mnt_id + 1;
110 	spin_unlock(&mnt_id_lock);
111 	if (res == -EAGAIN)
112 		goto retry;
113 
114 	return res;
115 }
116 
117 static void mnt_free_id(struct mount *mnt)
118 {
119 	int id = mnt->mnt_id;
120 	spin_lock(&mnt_id_lock);
121 	ida_remove(&mnt_id_ida, id);
122 	if (mnt_id_start > id)
123 		mnt_id_start = id;
124 	spin_unlock(&mnt_id_lock);
125 }
126 
127 /*
128  * Allocate a new peer group ID
129  *
130  * mnt_group_ida is protected by namespace_sem
131  */
132 static int mnt_alloc_group_id(struct mount *mnt)
133 {
134 	int res;
135 
136 	if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
137 		return -ENOMEM;
138 
139 	res = ida_get_new_above(&mnt_group_ida,
140 				mnt_group_start,
141 				&mnt->mnt_group_id);
142 	if (!res)
143 		mnt_group_start = mnt->mnt_group_id + 1;
144 
145 	return res;
146 }
147 
148 /*
149  * Release a peer group ID
150  */
151 void mnt_release_group_id(struct mount *mnt)
152 {
153 	int id = mnt->mnt_group_id;
154 	ida_remove(&mnt_group_ida, id);
155 	if (mnt_group_start > id)
156 		mnt_group_start = id;
157 	mnt->mnt_group_id = 0;
158 }
159 
160 /*
161  * vfsmount lock must be held for read
162  */
163 static inline void mnt_add_count(struct mount *mnt, int n)
164 {
165 #ifdef CONFIG_SMP
166 	this_cpu_add(mnt->mnt_pcp->mnt_count, n);
167 #else
168 	preempt_disable();
169 	mnt->mnt_count += n;
170 	preempt_enable();
171 #endif
172 }
173 
174 /*
175  * vfsmount lock must be held for write
176  */
177 unsigned int mnt_get_count(struct mount *mnt)
178 {
179 #ifdef CONFIG_SMP
180 	unsigned int count = 0;
181 	int cpu;
182 
183 	for_each_possible_cpu(cpu) {
184 		count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
185 	}
186 
187 	return count;
188 #else
189 	return mnt->mnt_count;
190 #endif
191 }
192 
193 static void drop_mountpoint(struct fs_pin *p)
194 {
195 	struct mount *m = container_of(p, struct mount, mnt_umount);
196 	dput(m->mnt_ex_mountpoint);
197 	pin_remove(p);
198 	mntput(&m->mnt);
199 }
200 
201 static struct mount *alloc_vfsmnt(const char *name)
202 {
203 	struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
204 	if (mnt) {
205 		int err;
206 
207 		err = mnt_alloc_id(mnt);
208 		if (err)
209 			goto out_free_cache;
210 
211 		if (name) {
212 			mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
213 			if (!mnt->mnt_devname)
214 				goto out_free_id;
215 		}
216 
217 #ifdef CONFIG_SMP
218 		mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
219 		if (!mnt->mnt_pcp)
220 			goto out_free_devname;
221 
222 		this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
223 #else
224 		mnt->mnt_count = 1;
225 		mnt->mnt_writers = 0;
226 #endif
227 
228 		INIT_HLIST_NODE(&mnt->mnt_hash);
229 		INIT_LIST_HEAD(&mnt->mnt_child);
230 		INIT_LIST_HEAD(&mnt->mnt_mounts);
231 		INIT_LIST_HEAD(&mnt->mnt_list);
232 		INIT_LIST_HEAD(&mnt->mnt_expire);
233 		INIT_LIST_HEAD(&mnt->mnt_share);
234 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
235 		INIT_LIST_HEAD(&mnt->mnt_slave);
236 		INIT_HLIST_NODE(&mnt->mnt_mp_list);
237 #ifdef CONFIG_FSNOTIFY
238 		INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
239 #endif
240 		init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
241 	}
242 	return mnt;
243 
244 #ifdef CONFIG_SMP
245 out_free_devname:
246 	kfree_const(mnt->mnt_devname);
247 #endif
248 out_free_id:
249 	mnt_free_id(mnt);
250 out_free_cache:
251 	kmem_cache_free(mnt_cache, mnt);
252 	return NULL;
253 }
254 
255 /*
256  * Most r/o checks on a fs are for operations that take
257  * discrete amounts of time, like a write() or unlink().
258  * We must keep track of when those operations start
259  * (for permission checks) and when they end, so that
260  * we can determine when writes are able to occur to
261  * a filesystem.
262  */
263 /*
264  * __mnt_is_readonly: check whether a mount is read-only
265  * @mnt: the mount to check for its write status
266  *
267  * This shouldn't be used directly ouside of the VFS.
268  * It does not guarantee that the filesystem will stay
269  * r/w, just that it is right *now*.  This can not and
270  * should not be used in place of IS_RDONLY(inode).
271  * mnt_want/drop_write() will _keep_ the filesystem
272  * r/w.
273  */
274 int __mnt_is_readonly(struct vfsmount *mnt)
275 {
276 	if (mnt->mnt_flags & MNT_READONLY)
277 		return 1;
278 	if (mnt->mnt_sb->s_flags & MS_RDONLY)
279 		return 1;
280 	return 0;
281 }
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283 
284 static inline void mnt_inc_writers(struct mount *mnt)
285 {
286 #ifdef CONFIG_SMP
287 	this_cpu_inc(mnt->mnt_pcp->mnt_writers);
288 #else
289 	mnt->mnt_writers++;
290 #endif
291 }
292 
293 static inline void mnt_dec_writers(struct mount *mnt)
294 {
295 #ifdef CONFIG_SMP
296 	this_cpu_dec(mnt->mnt_pcp->mnt_writers);
297 #else
298 	mnt->mnt_writers--;
299 #endif
300 }
301 
302 static unsigned int mnt_get_writers(struct mount *mnt)
303 {
304 #ifdef CONFIG_SMP
305 	unsigned int count = 0;
306 	int cpu;
307 
308 	for_each_possible_cpu(cpu) {
309 		count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
310 	}
311 
312 	return count;
313 #else
314 	return mnt->mnt_writers;
315 #endif
316 }
317 
318 static int mnt_is_readonly(struct vfsmount *mnt)
319 {
320 	if (mnt->mnt_sb->s_readonly_remount)
321 		return 1;
322 	/* Order wrt setting s_flags/s_readonly_remount in do_remount() */
323 	smp_rmb();
324 	return __mnt_is_readonly(mnt);
325 }
326 
327 /*
328  * Most r/o & frozen checks on a fs are for operations that take discrete
329  * amounts of time, like a write() or unlink().  We must keep track of when
330  * those operations start (for permission checks) and when they end, so that we
331  * can determine when writes are able to occur to a filesystem.
332  */
333 /**
334  * __mnt_want_write - get write access to a mount without freeze protection
335  * @m: the mount on which to take a write
336  *
337  * This tells the low-level filesystem that a write is about to be performed to
338  * it, and makes sure that writes are allowed (mnt it read-write) before
339  * returning success. This operation does not protect against filesystem being
340  * frozen. When the write operation is finished, __mnt_drop_write() must be
341  * called. This is effectively a refcount.
342  */
343 int __mnt_want_write(struct vfsmount *m)
344 {
345 	struct mount *mnt = real_mount(m);
346 	int ret = 0;
347 
348 	preempt_disable();
349 	mnt_inc_writers(mnt);
350 	/*
351 	 * The store to mnt_inc_writers must be visible before we pass
352 	 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 	 * incremented count after it has set MNT_WRITE_HOLD.
354 	 */
355 	smp_mb();
356 	while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
357 		cpu_relax();
358 	/*
359 	 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 	 * be set to match its requirements. So we must not load that until
361 	 * MNT_WRITE_HOLD is cleared.
362 	 */
363 	smp_rmb();
364 	if (mnt_is_readonly(m)) {
365 		mnt_dec_writers(mnt);
366 		ret = -EROFS;
367 	}
368 	preempt_enable();
369 
370 	return ret;
371 }
372 
373 /**
374  * mnt_want_write - get write access to a mount
375  * @m: the mount on which to take a write
376  *
377  * This tells the low-level filesystem that a write is about to be performed to
378  * it, and makes sure that writes are allowed (mount is read-write, filesystem
379  * is not frozen) before returning success.  When the write operation is
380  * finished, mnt_drop_write() must be called.  This is effectively a refcount.
381  */
382 int mnt_want_write(struct vfsmount *m)
383 {
384 	int ret;
385 
386 	sb_start_write(m->mnt_sb);
387 	ret = __mnt_want_write(m);
388 	if (ret)
389 		sb_end_write(m->mnt_sb);
390 	return ret;
391 }
392 EXPORT_SYMBOL_GPL(mnt_want_write);
393 
394 /**
395  * mnt_clone_write - get write access to a mount
396  * @mnt: the mount on which to take a write
397  *
398  * This is effectively like mnt_want_write, except
399  * it must only be used to take an extra write reference
400  * on a mountpoint that we already know has a write reference
401  * on it. This allows some optimisation.
402  *
403  * After finished, mnt_drop_write must be called as usual to
404  * drop the reference.
405  */
406 int mnt_clone_write(struct vfsmount *mnt)
407 {
408 	/* superblock may be r/o */
409 	if (__mnt_is_readonly(mnt))
410 		return -EROFS;
411 	preempt_disable();
412 	mnt_inc_writers(real_mount(mnt));
413 	preempt_enable();
414 	return 0;
415 }
416 EXPORT_SYMBOL_GPL(mnt_clone_write);
417 
418 /**
419  * __mnt_want_write_file - get write access to a file's mount
420  * @file: the file who's mount on which to take a write
421  *
422  * This is like __mnt_want_write, but it takes a file and can
423  * do some optimisations if the file is open for write already
424  */
425 int __mnt_want_write_file(struct file *file)
426 {
427 	if (!(file->f_mode & FMODE_WRITER))
428 		return __mnt_want_write(file->f_path.mnt);
429 	else
430 		return mnt_clone_write(file->f_path.mnt);
431 }
432 
433 /**
434  * mnt_want_write_file - get write access to a file's mount
435  * @file: the file who's mount on which to take a write
436  *
437  * This is like mnt_want_write, but it takes a file and can
438  * do some optimisations if the file is open for write already
439  */
440 int mnt_want_write_file(struct file *file)
441 {
442 	int ret;
443 
444 	sb_start_write(file->f_path.mnt->mnt_sb);
445 	ret = __mnt_want_write_file(file);
446 	if (ret)
447 		sb_end_write(file->f_path.mnt->mnt_sb);
448 	return ret;
449 }
450 EXPORT_SYMBOL_GPL(mnt_want_write_file);
451 
452 /**
453  * __mnt_drop_write - give up write access to a mount
454  * @mnt: the mount on which to give up write access
455  *
456  * Tells the low-level filesystem that we are done
457  * performing writes to it.  Must be matched with
458  * __mnt_want_write() call above.
459  */
460 void __mnt_drop_write(struct vfsmount *mnt)
461 {
462 	preempt_disable();
463 	mnt_dec_writers(real_mount(mnt));
464 	preempt_enable();
465 }
466 
467 /**
468  * mnt_drop_write - give up write access to a mount
469  * @mnt: the mount on which to give up write access
470  *
471  * Tells the low-level filesystem that we are done performing writes to it and
472  * also allows filesystem to be frozen again.  Must be matched with
473  * mnt_want_write() call above.
474  */
475 void mnt_drop_write(struct vfsmount *mnt)
476 {
477 	__mnt_drop_write(mnt);
478 	sb_end_write(mnt->mnt_sb);
479 }
480 EXPORT_SYMBOL_GPL(mnt_drop_write);
481 
482 void __mnt_drop_write_file(struct file *file)
483 {
484 	__mnt_drop_write(file->f_path.mnt);
485 }
486 
487 void mnt_drop_write_file(struct file *file)
488 {
489 	mnt_drop_write(file->f_path.mnt);
490 }
491 EXPORT_SYMBOL(mnt_drop_write_file);
492 
493 static int mnt_make_readonly(struct mount *mnt)
494 {
495 	int ret = 0;
496 
497 	lock_mount_hash();
498 	mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
499 	/*
500 	 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 	 * should be visible before we do.
502 	 */
503 	smp_mb();
504 
505 	/*
506 	 * With writers on hold, if this value is zero, then there are
507 	 * definitely no active writers (although held writers may subsequently
508 	 * increment the count, they'll have to wait, and decrement it after
509 	 * seeing MNT_READONLY).
510 	 *
511 	 * It is OK to have counter incremented on one CPU and decremented on
512 	 * another: the sum will add up correctly. The danger would be when we
513 	 * sum up each counter, if we read a counter before it is incremented,
514 	 * but then read another CPU's count which it has been subsequently
515 	 * decremented from -- we would see more decrements than we should.
516 	 * MNT_WRITE_HOLD protects against this scenario, because
517 	 * mnt_want_write first increments count, then smp_mb, then spins on
518 	 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 	 * we're counting up here.
520 	 */
521 	if (mnt_get_writers(mnt) > 0)
522 		ret = -EBUSY;
523 	else
524 		mnt->mnt.mnt_flags |= MNT_READONLY;
525 	/*
526 	 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 	 * that become unheld will see MNT_READONLY.
528 	 */
529 	smp_wmb();
530 	mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
531 	unlock_mount_hash();
532 	return ret;
533 }
534 
535 static void __mnt_unmake_readonly(struct mount *mnt)
536 {
537 	lock_mount_hash();
538 	mnt->mnt.mnt_flags &= ~MNT_READONLY;
539 	unlock_mount_hash();
540 }
541 
542 int sb_prepare_remount_readonly(struct super_block *sb)
543 {
544 	struct mount *mnt;
545 	int err = 0;
546 
547 	/* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
548 	if (atomic_long_read(&sb->s_remove_count))
549 		return -EBUSY;
550 
551 	lock_mount_hash();
552 	list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 		if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 			mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
555 			smp_mb();
556 			if (mnt_get_writers(mnt) > 0) {
557 				err = -EBUSY;
558 				break;
559 			}
560 		}
561 	}
562 	if (!err && atomic_long_read(&sb->s_remove_count))
563 		err = -EBUSY;
564 
565 	if (!err) {
566 		sb->s_readonly_remount = 1;
567 		smp_wmb();
568 	}
569 	list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 		if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 			mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
572 	}
573 	unlock_mount_hash();
574 
575 	return err;
576 }
577 
578 static void free_vfsmnt(struct mount *mnt)
579 {
580 	kfree_const(mnt->mnt_devname);
581 #ifdef CONFIG_SMP
582 	free_percpu(mnt->mnt_pcp);
583 #endif
584 	kmem_cache_free(mnt_cache, mnt);
585 }
586 
587 static void delayed_free_vfsmnt(struct rcu_head *head)
588 {
589 	free_vfsmnt(container_of(head, struct mount, mnt_rcu));
590 }
591 
592 /* call under rcu_read_lock */
593 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
594 {
595 	struct mount *mnt;
596 	if (read_seqretry(&mount_lock, seq))
597 		return false;
598 	if (bastard == NULL)
599 		return true;
600 	mnt = real_mount(bastard);
601 	mnt_add_count(mnt, 1);
602 	if (likely(!read_seqretry(&mount_lock, seq)))
603 		return true;
604 	if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 		mnt_add_count(mnt, -1);
606 		return false;
607 	}
608 	rcu_read_unlock();
609 	mntput(bastard);
610 	rcu_read_lock();
611 	return false;
612 }
613 
614 /*
615  * find the first mount at @dentry on vfsmount @mnt.
616  * call under rcu_read_lock()
617  */
618 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
619 {
620 	struct hlist_head *head = m_hash(mnt, dentry);
621 	struct mount *p;
622 
623 	hlist_for_each_entry_rcu(p, head, mnt_hash)
624 		if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
625 			return p;
626 	return NULL;
627 }
628 
629 /*
630  * find the last mount at @dentry on vfsmount @mnt.
631  * mount_lock must be held.
632  */
633 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
634 {
635 	struct mount *p, *res = NULL;
636 	p = __lookup_mnt(mnt, dentry);
637 	if (!p)
638 		goto out;
639 	if (!(p->mnt.mnt_flags & MNT_UMOUNT))
640 		res = p;
641 	hlist_for_each_entry_continue(p, mnt_hash) {
642 		if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
643 			break;
644 		if (!(p->mnt.mnt_flags & MNT_UMOUNT))
645 			res = p;
646 	}
647 out:
648 	return res;
649 }
650 
651 /*
652  * lookup_mnt - Return the first child mount mounted at path
653  *
654  * "First" means first mounted chronologically.  If you create the
655  * following mounts:
656  *
657  * mount /dev/sda1 /mnt
658  * mount /dev/sda2 /mnt
659  * mount /dev/sda3 /mnt
660  *
661  * Then lookup_mnt() on the base /mnt dentry in the root mount will
662  * return successively the root dentry and vfsmount of /dev/sda1, then
663  * /dev/sda2, then /dev/sda3, then NULL.
664  *
665  * lookup_mnt takes a reference to the found vfsmount.
666  */
667 struct vfsmount *lookup_mnt(struct path *path)
668 {
669 	struct mount *child_mnt;
670 	struct vfsmount *m;
671 	unsigned seq;
672 
673 	rcu_read_lock();
674 	do {
675 		seq = read_seqbegin(&mount_lock);
676 		child_mnt = __lookup_mnt(path->mnt, path->dentry);
677 		m = child_mnt ? &child_mnt->mnt : NULL;
678 	} while (!legitimize_mnt(m, seq));
679 	rcu_read_unlock();
680 	return m;
681 }
682 
683 /*
684  * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
685  *                         current mount namespace.
686  *
687  * The common case is dentries are not mountpoints at all and that
688  * test is handled inline.  For the slow case when we are actually
689  * dealing with a mountpoint of some kind, walk through all of the
690  * mounts in the current mount namespace and test to see if the dentry
691  * is a mountpoint.
692  *
693  * The mount_hashtable is not usable in the context because we
694  * need to identify all mounts that may be in the current mount
695  * namespace not just a mount that happens to have some specified
696  * parent mount.
697  */
698 bool __is_local_mountpoint(struct dentry *dentry)
699 {
700 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
701 	struct mount *mnt;
702 	bool is_covered = false;
703 
704 	if (!d_mountpoint(dentry))
705 		goto out;
706 
707 	down_read(&namespace_sem);
708 	list_for_each_entry(mnt, &ns->list, mnt_list) {
709 		is_covered = (mnt->mnt_mountpoint == dentry);
710 		if (is_covered)
711 			break;
712 	}
713 	up_read(&namespace_sem);
714 out:
715 	return is_covered;
716 }
717 
718 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
719 {
720 	struct hlist_head *chain = mp_hash(dentry);
721 	struct mountpoint *mp;
722 
723 	hlist_for_each_entry(mp, chain, m_hash) {
724 		if (mp->m_dentry == dentry) {
725 			/* might be worth a WARN_ON() */
726 			if (d_unlinked(dentry))
727 				return ERR_PTR(-ENOENT);
728 			mp->m_count++;
729 			return mp;
730 		}
731 	}
732 	return NULL;
733 }
734 
735 static struct mountpoint *new_mountpoint(struct dentry *dentry)
736 {
737 	struct hlist_head *chain = mp_hash(dentry);
738 	struct mountpoint *mp;
739 	int ret;
740 
741 	mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
742 	if (!mp)
743 		return ERR_PTR(-ENOMEM);
744 
745 	ret = d_set_mounted(dentry);
746 	if (ret) {
747 		kfree(mp);
748 		return ERR_PTR(ret);
749 	}
750 
751 	mp->m_dentry = dentry;
752 	mp->m_count = 1;
753 	hlist_add_head(&mp->m_hash, chain);
754 	INIT_HLIST_HEAD(&mp->m_list);
755 	return mp;
756 }
757 
758 static void put_mountpoint(struct mountpoint *mp)
759 {
760 	if (!--mp->m_count) {
761 		struct dentry *dentry = mp->m_dentry;
762 		BUG_ON(!hlist_empty(&mp->m_list));
763 		spin_lock(&dentry->d_lock);
764 		dentry->d_flags &= ~DCACHE_MOUNTED;
765 		spin_unlock(&dentry->d_lock);
766 		hlist_del(&mp->m_hash);
767 		kfree(mp);
768 	}
769 }
770 
771 static inline int check_mnt(struct mount *mnt)
772 {
773 	return mnt->mnt_ns == current->nsproxy->mnt_ns;
774 }
775 
776 /*
777  * vfsmount lock must be held for write
778  */
779 static void touch_mnt_namespace(struct mnt_namespace *ns)
780 {
781 	if (ns) {
782 		ns->event = ++event;
783 		wake_up_interruptible(&ns->poll);
784 	}
785 }
786 
787 /*
788  * vfsmount lock must be held for write
789  */
790 static void __touch_mnt_namespace(struct mnt_namespace *ns)
791 {
792 	if (ns && ns->event != event) {
793 		ns->event = event;
794 		wake_up_interruptible(&ns->poll);
795 	}
796 }
797 
798 /*
799  * vfsmount lock must be held for write
800  */
801 static void unhash_mnt(struct mount *mnt)
802 {
803 	mnt->mnt_parent = mnt;
804 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
805 	list_del_init(&mnt->mnt_child);
806 	hlist_del_init_rcu(&mnt->mnt_hash);
807 	hlist_del_init(&mnt->mnt_mp_list);
808 	put_mountpoint(mnt->mnt_mp);
809 	mnt->mnt_mp = NULL;
810 }
811 
812 /*
813  * vfsmount lock must be held for write
814  */
815 static void detach_mnt(struct mount *mnt, struct path *old_path)
816 {
817 	old_path->dentry = mnt->mnt_mountpoint;
818 	old_path->mnt = &mnt->mnt_parent->mnt;
819 	unhash_mnt(mnt);
820 }
821 
822 /*
823  * vfsmount lock must be held for write
824  */
825 static void umount_mnt(struct mount *mnt)
826 {
827 	/* old mountpoint will be dropped when we can do that */
828 	mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
829 	unhash_mnt(mnt);
830 }
831 
832 /*
833  * vfsmount lock must be held for write
834  */
835 void mnt_set_mountpoint(struct mount *mnt,
836 			struct mountpoint *mp,
837 			struct mount *child_mnt)
838 {
839 	mp->m_count++;
840 	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
841 	child_mnt->mnt_mountpoint = dget(mp->m_dentry);
842 	child_mnt->mnt_parent = mnt;
843 	child_mnt->mnt_mp = mp;
844 	hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
845 }
846 
847 /*
848  * vfsmount lock must be held for write
849  */
850 static void attach_mnt(struct mount *mnt,
851 			struct mount *parent,
852 			struct mountpoint *mp)
853 {
854 	mnt_set_mountpoint(parent, mp, mnt);
855 	hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
856 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
857 }
858 
859 static void attach_shadowed(struct mount *mnt,
860 			struct mount *parent,
861 			struct mount *shadows)
862 {
863 	if (shadows) {
864 		hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
865 		list_add(&mnt->mnt_child, &shadows->mnt_child);
866 	} else {
867 		hlist_add_head_rcu(&mnt->mnt_hash,
868 				m_hash(&parent->mnt, mnt->mnt_mountpoint));
869 		list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
870 	}
871 }
872 
873 /*
874  * vfsmount lock must be held for write
875  */
876 static void commit_tree(struct mount *mnt, struct mount *shadows)
877 {
878 	struct mount *parent = mnt->mnt_parent;
879 	struct mount *m;
880 	LIST_HEAD(head);
881 	struct mnt_namespace *n = parent->mnt_ns;
882 
883 	BUG_ON(parent == mnt);
884 
885 	list_add_tail(&head, &mnt->mnt_list);
886 	list_for_each_entry(m, &head, mnt_list)
887 		m->mnt_ns = n;
888 
889 	list_splice(&head, n->list.prev);
890 
891 	attach_shadowed(mnt, parent, shadows);
892 	touch_mnt_namespace(n);
893 }
894 
895 static struct mount *next_mnt(struct mount *p, struct mount *root)
896 {
897 	struct list_head *next = p->mnt_mounts.next;
898 	if (next == &p->mnt_mounts) {
899 		while (1) {
900 			if (p == root)
901 				return NULL;
902 			next = p->mnt_child.next;
903 			if (next != &p->mnt_parent->mnt_mounts)
904 				break;
905 			p = p->mnt_parent;
906 		}
907 	}
908 	return list_entry(next, struct mount, mnt_child);
909 }
910 
911 static struct mount *skip_mnt_tree(struct mount *p)
912 {
913 	struct list_head *prev = p->mnt_mounts.prev;
914 	while (prev != &p->mnt_mounts) {
915 		p = list_entry(prev, struct mount, mnt_child);
916 		prev = p->mnt_mounts.prev;
917 	}
918 	return p;
919 }
920 
921 struct vfsmount *
922 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
923 {
924 	struct mount *mnt;
925 	struct dentry *root;
926 
927 	if (!type)
928 		return ERR_PTR(-ENODEV);
929 
930 	mnt = alloc_vfsmnt(name);
931 	if (!mnt)
932 		return ERR_PTR(-ENOMEM);
933 
934 	if (flags & MS_KERNMOUNT)
935 		mnt->mnt.mnt_flags = MNT_INTERNAL;
936 
937 	root = mount_fs(type, flags, name, data);
938 	if (IS_ERR(root)) {
939 		mnt_free_id(mnt);
940 		free_vfsmnt(mnt);
941 		return ERR_CAST(root);
942 	}
943 
944 	mnt->mnt.mnt_root = root;
945 	mnt->mnt.mnt_sb = root->d_sb;
946 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
947 	mnt->mnt_parent = mnt;
948 	lock_mount_hash();
949 	list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
950 	unlock_mount_hash();
951 	return &mnt->mnt;
952 }
953 EXPORT_SYMBOL_GPL(vfs_kern_mount);
954 
955 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
956 					int flag)
957 {
958 	struct super_block *sb = old->mnt.mnt_sb;
959 	struct mount *mnt;
960 	int err;
961 
962 	mnt = alloc_vfsmnt(old->mnt_devname);
963 	if (!mnt)
964 		return ERR_PTR(-ENOMEM);
965 
966 	if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
967 		mnt->mnt_group_id = 0; /* not a peer of original */
968 	else
969 		mnt->mnt_group_id = old->mnt_group_id;
970 
971 	if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
972 		err = mnt_alloc_group_id(mnt);
973 		if (err)
974 			goto out_free;
975 	}
976 
977 	mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
978 	/* Don't allow unprivileged users to change mount flags */
979 	if (flag & CL_UNPRIVILEGED) {
980 		mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
981 
982 		if (mnt->mnt.mnt_flags & MNT_READONLY)
983 			mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
984 
985 		if (mnt->mnt.mnt_flags & MNT_NODEV)
986 			mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
987 
988 		if (mnt->mnt.mnt_flags & MNT_NOSUID)
989 			mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
990 
991 		if (mnt->mnt.mnt_flags & MNT_NOEXEC)
992 			mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
993 	}
994 
995 	/* Don't allow unprivileged users to reveal what is under a mount */
996 	if ((flag & CL_UNPRIVILEGED) &&
997 	    (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
998 		mnt->mnt.mnt_flags |= MNT_LOCKED;
999 
1000 	atomic_inc(&sb->s_active);
1001 	mnt->mnt.mnt_sb = sb;
1002 	mnt->mnt.mnt_root = dget(root);
1003 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1004 	mnt->mnt_parent = mnt;
1005 	lock_mount_hash();
1006 	list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1007 	unlock_mount_hash();
1008 
1009 	if ((flag & CL_SLAVE) ||
1010 	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1011 		list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1012 		mnt->mnt_master = old;
1013 		CLEAR_MNT_SHARED(mnt);
1014 	} else if (!(flag & CL_PRIVATE)) {
1015 		if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1016 			list_add(&mnt->mnt_share, &old->mnt_share);
1017 		if (IS_MNT_SLAVE(old))
1018 			list_add(&mnt->mnt_slave, &old->mnt_slave);
1019 		mnt->mnt_master = old->mnt_master;
1020 	}
1021 	if (flag & CL_MAKE_SHARED)
1022 		set_mnt_shared(mnt);
1023 
1024 	/* stick the duplicate mount on the same expiry list
1025 	 * as the original if that was on one */
1026 	if (flag & CL_EXPIRE) {
1027 		if (!list_empty(&old->mnt_expire))
1028 			list_add(&mnt->mnt_expire, &old->mnt_expire);
1029 	}
1030 
1031 	return mnt;
1032 
1033  out_free:
1034 	mnt_free_id(mnt);
1035 	free_vfsmnt(mnt);
1036 	return ERR_PTR(err);
1037 }
1038 
1039 static void cleanup_mnt(struct mount *mnt)
1040 {
1041 	/*
1042 	 * This probably indicates that somebody messed
1043 	 * up a mnt_want/drop_write() pair.  If this
1044 	 * happens, the filesystem was probably unable
1045 	 * to make r/w->r/o transitions.
1046 	 */
1047 	/*
1048 	 * The locking used to deal with mnt_count decrement provides barriers,
1049 	 * so mnt_get_writers() below is safe.
1050 	 */
1051 	WARN_ON(mnt_get_writers(mnt));
1052 	if (unlikely(mnt->mnt_pins.first))
1053 		mnt_pin_kill(mnt);
1054 	fsnotify_vfsmount_delete(&mnt->mnt);
1055 	dput(mnt->mnt.mnt_root);
1056 	deactivate_super(mnt->mnt.mnt_sb);
1057 	mnt_free_id(mnt);
1058 	call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1059 }
1060 
1061 static void __cleanup_mnt(struct rcu_head *head)
1062 {
1063 	cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1064 }
1065 
1066 static LLIST_HEAD(delayed_mntput_list);
1067 static void delayed_mntput(struct work_struct *unused)
1068 {
1069 	struct llist_node *node = llist_del_all(&delayed_mntput_list);
1070 	struct llist_node *next;
1071 
1072 	for (; node; node = next) {
1073 		next = llist_next(node);
1074 		cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1075 	}
1076 }
1077 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1078 
1079 static void mntput_no_expire(struct mount *mnt)
1080 {
1081 	rcu_read_lock();
1082 	mnt_add_count(mnt, -1);
1083 	if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1084 		rcu_read_unlock();
1085 		return;
1086 	}
1087 	lock_mount_hash();
1088 	if (mnt_get_count(mnt)) {
1089 		rcu_read_unlock();
1090 		unlock_mount_hash();
1091 		return;
1092 	}
1093 	if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1094 		rcu_read_unlock();
1095 		unlock_mount_hash();
1096 		return;
1097 	}
1098 	mnt->mnt.mnt_flags |= MNT_DOOMED;
1099 	rcu_read_unlock();
1100 
1101 	list_del(&mnt->mnt_instance);
1102 
1103 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1104 		struct mount *p, *tmp;
1105 		list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1106 			umount_mnt(p);
1107 		}
1108 	}
1109 	unlock_mount_hash();
1110 
1111 	if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1112 		struct task_struct *task = current;
1113 		if (likely(!(task->flags & PF_KTHREAD))) {
1114 			init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1115 			if (!task_work_add(task, &mnt->mnt_rcu, true))
1116 				return;
1117 		}
1118 		if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1119 			schedule_delayed_work(&delayed_mntput_work, 1);
1120 		return;
1121 	}
1122 	cleanup_mnt(mnt);
1123 }
1124 
1125 void mntput(struct vfsmount *mnt)
1126 {
1127 	if (mnt) {
1128 		struct mount *m = real_mount(mnt);
1129 		/* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1130 		if (unlikely(m->mnt_expiry_mark))
1131 			m->mnt_expiry_mark = 0;
1132 		mntput_no_expire(m);
1133 	}
1134 }
1135 EXPORT_SYMBOL(mntput);
1136 
1137 struct vfsmount *mntget(struct vfsmount *mnt)
1138 {
1139 	if (mnt)
1140 		mnt_add_count(real_mount(mnt), 1);
1141 	return mnt;
1142 }
1143 EXPORT_SYMBOL(mntget);
1144 
1145 struct vfsmount *mnt_clone_internal(struct path *path)
1146 {
1147 	struct mount *p;
1148 	p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1149 	if (IS_ERR(p))
1150 		return ERR_CAST(p);
1151 	p->mnt.mnt_flags |= MNT_INTERNAL;
1152 	return &p->mnt;
1153 }
1154 
1155 static inline void mangle(struct seq_file *m, const char *s)
1156 {
1157 	seq_escape(m, s, " \t\n\\");
1158 }
1159 
1160 /*
1161  * Simple .show_options callback for filesystems which don't want to
1162  * implement more complex mount option showing.
1163  *
1164  * See also save_mount_options().
1165  */
1166 int generic_show_options(struct seq_file *m, struct dentry *root)
1167 {
1168 	const char *options;
1169 
1170 	rcu_read_lock();
1171 	options = rcu_dereference(root->d_sb->s_options);
1172 
1173 	if (options != NULL && options[0]) {
1174 		seq_putc(m, ',');
1175 		mangle(m, options);
1176 	}
1177 	rcu_read_unlock();
1178 
1179 	return 0;
1180 }
1181 EXPORT_SYMBOL(generic_show_options);
1182 
1183 /*
1184  * If filesystem uses generic_show_options(), this function should be
1185  * called from the fill_super() callback.
1186  *
1187  * The .remount_fs callback usually needs to be handled in a special
1188  * way, to make sure, that previous options are not overwritten if the
1189  * remount fails.
1190  *
1191  * Also note, that if the filesystem's .remount_fs function doesn't
1192  * reset all options to their default value, but changes only newly
1193  * given options, then the displayed options will not reflect reality
1194  * any more.
1195  */
1196 void save_mount_options(struct super_block *sb, char *options)
1197 {
1198 	BUG_ON(sb->s_options);
1199 	rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1200 }
1201 EXPORT_SYMBOL(save_mount_options);
1202 
1203 void replace_mount_options(struct super_block *sb, char *options)
1204 {
1205 	char *old = sb->s_options;
1206 	rcu_assign_pointer(sb->s_options, options);
1207 	if (old) {
1208 		synchronize_rcu();
1209 		kfree(old);
1210 	}
1211 }
1212 EXPORT_SYMBOL(replace_mount_options);
1213 
1214 #ifdef CONFIG_PROC_FS
1215 /* iterator; we want it to have access to namespace_sem, thus here... */
1216 static void *m_start(struct seq_file *m, loff_t *pos)
1217 {
1218 	struct proc_mounts *p = proc_mounts(m);
1219 
1220 	down_read(&namespace_sem);
1221 	if (p->cached_event == p->ns->event) {
1222 		void *v = p->cached_mount;
1223 		if (*pos == p->cached_index)
1224 			return v;
1225 		if (*pos == p->cached_index + 1) {
1226 			v = seq_list_next(v, &p->ns->list, &p->cached_index);
1227 			return p->cached_mount = v;
1228 		}
1229 	}
1230 
1231 	p->cached_event = p->ns->event;
1232 	p->cached_mount = seq_list_start(&p->ns->list, *pos);
1233 	p->cached_index = *pos;
1234 	return p->cached_mount;
1235 }
1236 
1237 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1238 {
1239 	struct proc_mounts *p = proc_mounts(m);
1240 
1241 	p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1242 	p->cached_index = *pos;
1243 	return p->cached_mount;
1244 }
1245 
1246 static void m_stop(struct seq_file *m, void *v)
1247 {
1248 	up_read(&namespace_sem);
1249 }
1250 
1251 static int m_show(struct seq_file *m, void *v)
1252 {
1253 	struct proc_mounts *p = proc_mounts(m);
1254 	struct mount *r = list_entry(v, struct mount, mnt_list);
1255 	return p->show(m, &r->mnt);
1256 }
1257 
1258 const struct seq_operations mounts_op = {
1259 	.start	= m_start,
1260 	.next	= m_next,
1261 	.stop	= m_stop,
1262 	.show	= m_show,
1263 };
1264 #endif  /* CONFIG_PROC_FS */
1265 
1266 /**
1267  * may_umount_tree - check if a mount tree is busy
1268  * @mnt: root of mount tree
1269  *
1270  * This is called to check if a tree of mounts has any
1271  * open files, pwds, chroots or sub mounts that are
1272  * busy.
1273  */
1274 int may_umount_tree(struct vfsmount *m)
1275 {
1276 	struct mount *mnt = real_mount(m);
1277 	int actual_refs = 0;
1278 	int minimum_refs = 0;
1279 	struct mount *p;
1280 	BUG_ON(!m);
1281 
1282 	/* write lock needed for mnt_get_count */
1283 	lock_mount_hash();
1284 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1285 		actual_refs += mnt_get_count(p);
1286 		minimum_refs += 2;
1287 	}
1288 	unlock_mount_hash();
1289 
1290 	if (actual_refs > minimum_refs)
1291 		return 0;
1292 
1293 	return 1;
1294 }
1295 
1296 EXPORT_SYMBOL(may_umount_tree);
1297 
1298 /**
1299  * may_umount - check if a mount point is busy
1300  * @mnt: root of mount
1301  *
1302  * This is called to check if a mount point has any
1303  * open files, pwds, chroots or sub mounts. If the
1304  * mount has sub mounts this will return busy
1305  * regardless of whether the sub mounts are busy.
1306  *
1307  * Doesn't take quota and stuff into account. IOW, in some cases it will
1308  * give false negatives. The main reason why it's here is that we need
1309  * a non-destructive way to look for easily umountable filesystems.
1310  */
1311 int may_umount(struct vfsmount *mnt)
1312 {
1313 	int ret = 1;
1314 	down_read(&namespace_sem);
1315 	lock_mount_hash();
1316 	if (propagate_mount_busy(real_mount(mnt), 2))
1317 		ret = 0;
1318 	unlock_mount_hash();
1319 	up_read(&namespace_sem);
1320 	return ret;
1321 }
1322 
1323 EXPORT_SYMBOL(may_umount);
1324 
1325 static HLIST_HEAD(unmounted);	/* protected by namespace_sem */
1326 
1327 static void namespace_unlock(void)
1328 {
1329 	struct hlist_head head;
1330 
1331 	hlist_move_list(&unmounted, &head);
1332 
1333 	up_write(&namespace_sem);
1334 
1335 	if (likely(hlist_empty(&head)))
1336 		return;
1337 
1338 	synchronize_rcu();
1339 
1340 	group_pin_kill(&head);
1341 }
1342 
1343 static inline void namespace_lock(void)
1344 {
1345 	down_write(&namespace_sem);
1346 }
1347 
1348 enum umount_tree_flags {
1349 	UMOUNT_SYNC = 1,
1350 	UMOUNT_PROPAGATE = 2,
1351 	UMOUNT_CONNECTED = 4,
1352 };
1353 /*
1354  * mount_lock must be held
1355  * namespace_sem must be held for write
1356  */
1357 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1358 {
1359 	LIST_HEAD(tmp_list);
1360 	struct mount *p;
1361 
1362 	if (how & UMOUNT_PROPAGATE)
1363 		propagate_mount_unlock(mnt);
1364 
1365 	/* Gather the mounts to umount */
1366 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1367 		p->mnt.mnt_flags |= MNT_UMOUNT;
1368 		list_move(&p->mnt_list, &tmp_list);
1369 	}
1370 
1371 	/* Hide the mounts from mnt_mounts */
1372 	list_for_each_entry(p, &tmp_list, mnt_list) {
1373 		list_del_init(&p->mnt_child);
1374 	}
1375 
1376 	/* Add propogated mounts to the tmp_list */
1377 	if (how & UMOUNT_PROPAGATE)
1378 		propagate_umount(&tmp_list);
1379 
1380 	while (!list_empty(&tmp_list)) {
1381 		bool disconnect;
1382 		p = list_first_entry(&tmp_list, struct mount, mnt_list);
1383 		list_del_init(&p->mnt_expire);
1384 		list_del_init(&p->mnt_list);
1385 		__touch_mnt_namespace(p->mnt_ns);
1386 		p->mnt_ns = NULL;
1387 		if (how & UMOUNT_SYNC)
1388 			p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1389 
1390 		disconnect = !(((how & UMOUNT_CONNECTED) &&
1391 				mnt_has_parent(p) &&
1392 				(p->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) ||
1393 			       IS_MNT_LOCKED_AND_LAZY(p));
1394 
1395 		pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1396 				 disconnect ? &unmounted : NULL);
1397 		if (mnt_has_parent(p)) {
1398 			mnt_add_count(p->mnt_parent, -1);
1399 			if (!disconnect) {
1400 				/* Don't forget about p */
1401 				list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1402 			} else {
1403 				umount_mnt(p);
1404 			}
1405 		}
1406 		change_mnt_propagation(p, MS_PRIVATE);
1407 	}
1408 }
1409 
1410 static void shrink_submounts(struct mount *mnt);
1411 
1412 static int do_umount(struct mount *mnt, int flags)
1413 {
1414 	struct super_block *sb = mnt->mnt.mnt_sb;
1415 	int retval;
1416 
1417 	retval = security_sb_umount(&mnt->mnt, flags);
1418 	if (retval)
1419 		return retval;
1420 
1421 	/*
1422 	 * Allow userspace to request a mountpoint be expired rather than
1423 	 * unmounting unconditionally. Unmount only happens if:
1424 	 *  (1) the mark is already set (the mark is cleared by mntput())
1425 	 *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1426 	 */
1427 	if (flags & MNT_EXPIRE) {
1428 		if (&mnt->mnt == current->fs->root.mnt ||
1429 		    flags & (MNT_FORCE | MNT_DETACH))
1430 			return -EINVAL;
1431 
1432 		/*
1433 		 * probably don't strictly need the lock here if we examined
1434 		 * all race cases, but it's a slowpath.
1435 		 */
1436 		lock_mount_hash();
1437 		if (mnt_get_count(mnt) != 2) {
1438 			unlock_mount_hash();
1439 			return -EBUSY;
1440 		}
1441 		unlock_mount_hash();
1442 
1443 		if (!xchg(&mnt->mnt_expiry_mark, 1))
1444 			return -EAGAIN;
1445 	}
1446 
1447 	/*
1448 	 * If we may have to abort operations to get out of this
1449 	 * mount, and they will themselves hold resources we must
1450 	 * allow the fs to do things. In the Unix tradition of
1451 	 * 'Gee thats tricky lets do it in userspace' the umount_begin
1452 	 * might fail to complete on the first run through as other tasks
1453 	 * must return, and the like. Thats for the mount program to worry
1454 	 * about for the moment.
1455 	 */
1456 
1457 	if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1458 		sb->s_op->umount_begin(sb);
1459 	}
1460 
1461 	/*
1462 	 * No sense to grab the lock for this test, but test itself looks
1463 	 * somewhat bogus. Suggestions for better replacement?
1464 	 * Ho-hum... In principle, we might treat that as umount + switch
1465 	 * to rootfs. GC would eventually take care of the old vfsmount.
1466 	 * Actually it makes sense, especially if rootfs would contain a
1467 	 * /reboot - static binary that would close all descriptors and
1468 	 * call reboot(9). Then init(8) could umount root and exec /reboot.
1469 	 */
1470 	if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1471 		/*
1472 		 * Special case for "unmounting" root ...
1473 		 * we just try to remount it readonly.
1474 		 */
1475 		if (!capable(CAP_SYS_ADMIN))
1476 			return -EPERM;
1477 		down_write(&sb->s_umount);
1478 		if (!(sb->s_flags & MS_RDONLY))
1479 			retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1480 		up_write(&sb->s_umount);
1481 		return retval;
1482 	}
1483 
1484 	namespace_lock();
1485 	lock_mount_hash();
1486 	event++;
1487 
1488 	if (flags & MNT_DETACH) {
1489 		if (!list_empty(&mnt->mnt_list))
1490 			umount_tree(mnt, UMOUNT_PROPAGATE);
1491 		retval = 0;
1492 	} else {
1493 		shrink_submounts(mnt);
1494 		retval = -EBUSY;
1495 		if (!propagate_mount_busy(mnt, 2)) {
1496 			if (!list_empty(&mnt->mnt_list))
1497 				umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1498 			retval = 0;
1499 		}
1500 	}
1501 	unlock_mount_hash();
1502 	namespace_unlock();
1503 	return retval;
1504 }
1505 
1506 /*
1507  * __detach_mounts - lazily unmount all mounts on the specified dentry
1508  *
1509  * During unlink, rmdir, and d_drop it is possible to loose the path
1510  * to an existing mountpoint, and wind up leaking the mount.
1511  * detach_mounts allows lazily unmounting those mounts instead of
1512  * leaking them.
1513  *
1514  * The caller may hold dentry->d_inode->i_mutex.
1515  */
1516 void __detach_mounts(struct dentry *dentry)
1517 {
1518 	struct mountpoint *mp;
1519 	struct mount *mnt;
1520 
1521 	namespace_lock();
1522 	mp = lookup_mountpoint(dentry);
1523 	if (IS_ERR_OR_NULL(mp))
1524 		goto out_unlock;
1525 
1526 	lock_mount_hash();
1527 	while (!hlist_empty(&mp->m_list)) {
1528 		mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1529 		if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1530 			struct mount *p, *tmp;
1531 			list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1532 				hlist_add_head(&p->mnt_umount.s_list, &unmounted);
1533 				umount_mnt(p);
1534 			}
1535 		}
1536 		else umount_tree(mnt, UMOUNT_CONNECTED);
1537 	}
1538 	unlock_mount_hash();
1539 	put_mountpoint(mp);
1540 out_unlock:
1541 	namespace_unlock();
1542 }
1543 
1544 /*
1545  * Is the caller allowed to modify his namespace?
1546  */
1547 static inline bool may_mount(void)
1548 {
1549 	return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1550 }
1551 
1552 /*
1553  * Now umount can handle mount points as well as block devices.
1554  * This is important for filesystems which use unnamed block devices.
1555  *
1556  * We now support a flag for forced unmount like the other 'big iron'
1557  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1558  */
1559 
1560 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1561 {
1562 	struct path path;
1563 	struct mount *mnt;
1564 	int retval;
1565 	int lookup_flags = 0;
1566 
1567 	if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1568 		return -EINVAL;
1569 
1570 	if (!may_mount())
1571 		return -EPERM;
1572 
1573 	if (!(flags & UMOUNT_NOFOLLOW))
1574 		lookup_flags |= LOOKUP_FOLLOW;
1575 
1576 	retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1577 	if (retval)
1578 		goto out;
1579 	mnt = real_mount(path.mnt);
1580 	retval = -EINVAL;
1581 	if (path.dentry != path.mnt->mnt_root)
1582 		goto dput_and_out;
1583 	if (!check_mnt(mnt))
1584 		goto dput_and_out;
1585 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
1586 		goto dput_and_out;
1587 	retval = -EPERM;
1588 	if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1589 		goto dput_and_out;
1590 
1591 	retval = do_umount(mnt, flags);
1592 dput_and_out:
1593 	/* we mustn't call path_put() as that would clear mnt_expiry_mark */
1594 	dput(path.dentry);
1595 	mntput_no_expire(mnt);
1596 out:
1597 	return retval;
1598 }
1599 
1600 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1601 
1602 /*
1603  *	The 2.0 compatible umount. No flags.
1604  */
1605 SYSCALL_DEFINE1(oldumount, char __user *, name)
1606 {
1607 	return sys_umount(name, 0);
1608 }
1609 
1610 #endif
1611 
1612 static bool is_mnt_ns_file(struct dentry *dentry)
1613 {
1614 	/* Is this a proxy for a mount namespace? */
1615 	return dentry->d_op == &ns_dentry_operations &&
1616 	       dentry->d_fsdata == &mntns_operations;
1617 }
1618 
1619 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1620 {
1621 	return container_of(ns, struct mnt_namespace, ns);
1622 }
1623 
1624 static bool mnt_ns_loop(struct dentry *dentry)
1625 {
1626 	/* Could bind mounting the mount namespace inode cause a
1627 	 * mount namespace loop?
1628 	 */
1629 	struct mnt_namespace *mnt_ns;
1630 	if (!is_mnt_ns_file(dentry))
1631 		return false;
1632 
1633 	mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1634 	return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1635 }
1636 
1637 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1638 					int flag)
1639 {
1640 	struct mount *res, *p, *q, *r, *parent;
1641 
1642 	if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1643 		return ERR_PTR(-EINVAL);
1644 
1645 	if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1646 		return ERR_PTR(-EINVAL);
1647 
1648 	res = q = clone_mnt(mnt, dentry, flag);
1649 	if (IS_ERR(q))
1650 		return q;
1651 
1652 	q->mnt_mountpoint = mnt->mnt_mountpoint;
1653 
1654 	p = mnt;
1655 	list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1656 		struct mount *s;
1657 		if (!is_subdir(r->mnt_mountpoint, dentry))
1658 			continue;
1659 
1660 		for (s = r; s; s = next_mnt(s, r)) {
1661 			struct mount *t = NULL;
1662 			if (!(flag & CL_COPY_UNBINDABLE) &&
1663 			    IS_MNT_UNBINDABLE(s)) {
1664 				s = skip_mnt_tree(s);
1665 				continue;
1666 			}
1667 			if (!(flag & CL_COPY_MNT_NS_FILE) &&
1668 			    is_mnt_ns_file(s->mnt.mnt_root)) {
1669 				s = skip_mnt_tree(s);
1670 				continue;
1671 			}
1672 			while (p != s->mnt_parent) {
1673 				p = p->mnt_parent;
1674 				q = q->mnt_parent;
1675 			}
1676 			p = s;
1677 			parent = q;
1678 			q = clone_mnt(p, p->mnt.mnt_root, flag);
1679 			if (IS_ERR(q))
1680 				goto out;
1681 			lock_mount_hash();
1682 			list_add_tail(&q->mnt_list, &res->mnt_list);
1683 			mnt_set_mountpoint(parent, p->mnt_mp, q);
1684 			if (!list_empty(&parent->mnt_mounts)) {
1685 				t = list_last_entry(&parent->mnt_mounts,
1686 					struct mount, mnt_child);
1687 				if (t->mnt_mp != p->mnt_mp)
1688 					t = NULL;
1689 			}
1690 			attach_shadowed(q, parent, t);
1691 			unlock_mount_hash();
1692 		}
1693 	}
1694 	return res;
1695 out:
1696 	if (res) {
1697 		lock_mount_hash();
1698 		umount_tree(res, UMOUNT_SYNC);
1699 		unlock_mount_hash();
1700 	}
1701 	return q;
1702 }
1703 
1704 /* Caller should check returned pointer for errors */
1705 
1706 struct vfsmount *collect_mounts(struct path *path)
1707 {
1708 	struct mount *tree;
1709 	namespace_lock();
1710 	if (!check_mnt(real_mount(path->mnt)))
1711 		tree = ERR_PTR(-EINVAL);
1712 	else
1713 		tree = copy_tree(real_mount(path->mnt), path->dentry,
1714 				 CL_COPY_ALL | CL_PRIVATE);
1715 	namespace_unlock();
1716 	if (IS_ERR(tree))
1717 		return ERR_CAST(tree);
1718 	return &tree->mnt;
1719 }
1720 
1721 void drop_collected_mounts(struct vfsmount *mnt)
1722 {
1723 	namespace_lock();
1724 	lock_mount_hash();
1725 	umount_tree(real_mount(mnt), UMOUNT_SYNC);
1726 	unlock_mount_hash();
1727 	namespace_unlock();
1728 }
1729 
1730 /**
1731  * clone_private_mount - create a private clone of a path
1732  *
1733  * This creates a new vfsmount, which will be the clone of @path.  The new will
1734  * not be attached anywhere in the namespace and will be private (i.e. changes
1735  * to the originating mount won't be propagated into this).
1736  *
1737  * Release with mntput().
1738  */
1739 struct vfsmount *clone_private_mount(struct path *path)
1740 {
1741 	struct mount *old_mnt = real_mount(path->mnt);
1742 	struct mount *new_mnt;
1743 
1744 	if (IS_MNT_UNBINDABLE(old_mnt))
1745 		return ERR_PTR(-EINVAL);
1746 
1747 	down_read(&namespace_sem);
1748 	new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1749 	up_read(&namespace_sem);
1750 	if (IS_ERR(new_mnt))
1751 		return ERR_CAST(new_mnt);
1752 
1753 	return &new_mnt->mnt;
1754 }
1755 EXPORT_SYMBOL_GPL(clone_private_mount);
1756 
1757 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1758 		   struct vfsmount *root)
1759 {
1760 	struct mount *mnt;
1761 	int res = f(root, arg);
1762 	if (res)
1763 		return res;
1764 	list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1765 		res = f(&mnt->mnt, arg);
1766 		if (res)
1767 			return res;
1768 	}
1769 	return 0;
1770 }
1771 
1772 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1773 {
1774 	struct mount *p;
1775 
1776 	for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1777 		if (p->mnt_group_id && !IS_MNT_SHARED(p))
1778 			mnt_release_group_id(p);
1779 	}
1780 }
1781 
1782 static int invent_group_ids(struct mount *mnt, bool recurse)
1783 {
1784 	struct mount *p;
1785 
1786 	for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1787 		if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1788 			int err = mnt_alloc_group_id(p);
1789 			if (err) {
1790 				cleanup_group_ids(mnt, p);
1791 				return err;
1792 			}
1793 		}
1794 	}
1795 
1796 	return 0;
1797 }
1798 
1799 /*
1800  *  @source_mnt : mount tree to be attached
1801  *  @nd         : place the mount tree @source_mnt is attached
1802  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1803  *  		   store the parent mount and mountpoint dentry.
1804  *  		   (done when source_mnt is moved)
1805  *
1806  *  NOTE: in the table below explains the semantics when a source mount
1807  *  of a given type is attached to a destination mount of a given type.
1808  * ---------------------------------------------------------------------------
1809  * |         BIND MOUNT OPERATION                                            |
1810  * |**************************************************************************
1811  * | source-->| shared        |       private  |       slave    | unbindable |
1812  * | dest     |               |                |                |            |
1813  * |   |      |               |                |                |            |
1814  * |   v      |               |                |                |            |
1815  * |**************************************************************************
1816  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1817  * |          |               |                |                |            |
1818  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1819  * ***************************************************************************
1820  * A bind operation clones the source mount and mounts the clone on the
1821  * destination mount.
1822  *
1823  * (++)  the cloned mount is propagated to all the mounts in the propagation
1824  * 	 tree of the destination mount and the cloned mount is added to
1825  * 	 the peer group of the source mount.
1826  * (+)   the cloned mount is created under the destination mount and is marked
1827  *       as shared. The cloned mount is added to the peer group of the source
1828  *       mount.
1829  * (+++) the mount is propagated to all the mounts in the propagation tree
1830  *       of the destination mount and the cloned mount is made slave
1831  *       of the same master as that of the source mount. The cloned mount
1832  *       is marked as 'shared and slave'.
1833  * (*)   the cloned mount is made a slave of the same master as that of the
1834  * 	 source mount.
1835  *
1836  * ---------------------------------------------------------------------------
1837  * |         		MOVE MOUNT OPERATION                                 |
1838  * |**************************************************************************
1839  * | source-->| shared        |       private  |       slave    | unbindable |
1840  * | dest     |               |                |                |            |
1841  * |   |      |               |                |                |            |
1842  * |   v      |               |                |                |            |
1843  * |**************************************************************************
1844  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1845  * |          |               |                |                |            |
1846  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1847  * ***************************************************************************
1848  *
1849  * (+)  the mount is moved to the destination. And is then propagated to
1850  * 	all the mounts in the propagation tree of the destination mount.
1851  * (+*)  the mount is moved to the destination.
1852  * (+++)  the mount is moved to the destination and is then propagated to
1853  * 	all the mounts belonging to the destination mount's propagation tree.
1854  * 	the mount is marked as 'shared and slave'.
1855  * (*)	the mount continues to be a slave at the new location.
1856  *
1857  * if the source mount is a tree, the operations explained above is
1858  * applied to each mount in the tree.
1859  * Must be called without spinlocks held, since this function can sleep
1860  * in allocations.
1861  */
1862 static int attach_recursive_mnt(struct mount *source_mnt,
1863 			struct mount *dest_mnt,
1864 			struct mountpoint *dest_mp,
1865 			struct path *parent_path)
1866 {
1867 	HLIST_HEAD(tree_list);
1868 	struct mount *child, *p;
1869 	struct hlist_node *n;
1870 	int err;
1871 
1872 	if (IS_MNT_SHARED(dest_mnt)) {
1873 		err = invent_group_ids(source_mnt, true);
1874 		if (err)
1875 			goto out;
1876 		err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1877 		lock_mount_hash();
1878 		if (err)
1879 			goto out_cleanup_ids;
1880 		for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1881 			set_mnt_shared(p);
1882 	} else {
1883 		lock_mount_hash();
1884 	}
1885 	if (parent_path) {
1886 		detach_mnt(source_mnt, parent_path);
1887 		attach_mnt(source_mnt, dest_mnt, dest_mp);
1888 		touch_mnt_namespace(source_mnt->mnt_ns);
1889 	} else {
1890 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1891 		commit_tree(source_mnt, NULL);
1892 	}
1893 
1894 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1895 		struct mount *q;
1896 		hlist_del_init(&child->mnt_hash);
1897 		q = __lookup_mnt_last(&child->mnt_parent->mnt,
1898 				      child->mnt_mountpoint);
1899 		commit_tree(child, q);
1900 	}
1901 	unlock_mount_hash();
1902 
1903 	return 0;
1904 
1905  out_cleanup_ids:
1906 	while (!hlist_empty(&tree_list)) {
1907 		child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1908 		umount_tree(child, UMOUNT_SYNC);
1909 	}
1910 	unlock_mount_hash();
1911 	cleanup_group_ids(source_mnt, NULL);
1912  out:
1913 	return err;
1914 }
1915 
1916 static struct mountpoint *lock_mount(struct path *path)
1917 {
1918 	struct vfsmount *mnt;
1919 	struct dentry *dentry = path->dentry;
1920 retry:
1921 	mutex_lock(&dentry->d_inode->i_mutex);
1922 	if (unlikely(cant_mount(dentry))) {
1923 		mutex_unlock(&dentry->d_inode->i_mutex);
1924 		return ERR_PTR(-ENOENT);
1925 	}
1926 	namespace_lock();
1927 	mnt = lookup_mnt(path);
1928 	if (likely(!mnt)) {
1929 		struct mountpoint *mp = lookup_mountpoint(dentry);
1930 		if (!mp)
1931 			mp = new_mountpoint(dentry);
1932 		if (IS_ERR(mp)) {
1933 			namespace_unlock();
1934 			mutex_unlock(&dentry->d_inode->i_mutex);
1935 			return mp;
1936 		}
1937 		return mp;
1938 	}
1939 	namespace_unlock();
1940 	mutex_unlock(&path->dentry->d_inode->i_mutex);
1941 	path_put(path);
1942 	path->mnt = mnt;
1943 	dentry = path->dentry = dget(mnt->mnt_root);
1944 	goto retry;
1945 }
1946 
1947 static void unlock_mount(struct mountpoint *where)
1948 {
1949 	struct dentry *dentry = where->m_dentry;
1950 	put_mountpoint(where);
1951 	namespace_unlock();
1952 	mutex_unlock(&dentry->d_inode->i_mutex);
1953 }
1954 
1955 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1956 {
1957 	if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1958 		return -EINVAL;
1959 
1960 	if (d_is_dir(mp->m_dentry) !=
1961 	      d_is_dir(mnt->mnt.mnt_root))
1962 		return -ENOTDIR;
1963 
1964 	return attach_recursive_mnt(mnt, p, mp, NULL);
1965 }
1966 
1967 /*
1968  * Sanity check the flags to change_mnt_propagation.
1969  */
1970 
1971 static int flags_to_propagation_type(int flags)
1972 {
1973 	int type = flags & ~(MS_REC | MS_SILENT);
1974 
1975 	/* Fail if any non-propagation flags are set */
1976 	if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1977 		return 0;
1978 	/* Only one propagation flag should be set */
1979 	if (!is_power_of_2(type))
1980 		return 0;
1981 	return type;
1982 }
1983 
1984 /*
1985  * recursively change the type of the mountpoint.
1986  */
1987 static int do_change_type(struct path *path, int flag)
1988 {
1989 	struct mount *m;
1990 	struct mount *mnt = real_mount(path->mnt);
1991 	int recurse = flag & MS_REC;
1992 	int type;
1993 	int err = 0;
1994 
1995 	if (path->dentry != path->mnt->mnt_root)
1996 		return -EINVAL;
1997 
1998 	type = flags_to_propagation_type(flag);
1999 	if (!type)
2000 		return -EINVAL;
2001 
2002 	namespace_lock();
2003 	if (type == MS_SHARED) {
2004 		err = invent_group_ids(mnt, recurse);
2005 		if (err)
2006 			goto out_unlock;
2007 	}
2008 
2009 	lock_mount_hash();
2010 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2011 		change_mnt_propagation(m, type);
2012 	unlock_mount_hash();
2013 
2014  out_unlock:
2015 	namespace_unlock();
2016 	return err;
2017 }
2018 
2019 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2020 {
2021 	struct mount *child;
2022 	list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2023 		if (!is_subdir(child->mnt_mountpoint, dentry))
2024 			continue;
2025 
2026 		if (child->mnt.mnt_flags & MNT_LOCKED)
2027 			return true;
2028 	}
2029 	return false;
2030 }
2031 
2032 /*
2033  * do loopback mount.
2034  */
2035 static int do_loopback(struct path *path, const char *old_name,
2036 				int recurse)
2037 {
2038 	struct path old_path;
2039 	struct mount *mnt = NULL, *old, *parent;
2040 	struct mountpoint *mp;
2041 	int err;
2042 	if (!old_name || !*old_name)
2043 		return -EINVAL;
2044 	err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2045 	if (err)
2046 		return err;
2047 
2048 	err = -EINVAL;
2049 	if (mnt_ns_loop(old_path.dentry))
2050 		goto out;
2051 
2052 	mp = lock_mount(path);
2053 	err = PTR_ERR(mp);
2054 	if (IS_ERR(mp))
2055 		goto out;
2056 
2057 	old = real_mount(old_path.mnt);
2058 	parent = real_mount(path->mnt);
2059 
2060 	err = -EINVAL;
2061 	if (IS_MNT_UNBINDABLE(old))
2062 		goto out2;
2063 
2064 	if (!check_mnt(parent))
2065 		goto out2;
2066 
2067 	if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2068 		goto out2;
2069 
2070 	if (!recurse && has_locked_children(old, old_path.dentry))
2071 		goto out2;
2072 
2073 	if (recurse)
2074 		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2075 	else
2076 		mnt = clone_mnt(old, old_path.dentry, 0);
2077 
2078 	if (IS_ERR(mnt)) {
2079 		err = PTR_ERR(mnt);
2080 		goto out2;
2081 	}
2082 
2083 	mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2084 
2085 	err = graft_tree(mnt, parent, mp);
2086 	if (err) {
2087 		lock_mount_hash();
2088 		umount_tree(mnt, UMOUNT_SYNC);
2089 		unlock_mount_hash();
2090 	}
2091 out2:
2092 	unlock_mount(mp);
2093 out:
2094 	path_put(&old_path);
2095 	return err;
2096 }
2097 
2098 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2099 {
2100 	int error = 0;
2101 	int readonly_request = 0;
2102 
2103 	if (ms_flags & MS_RDONLY)
2104 		readonly_request = 1;
2105 	if (readonly_request == __mnt_is_readonly(mnt))
2106 		return 0;
2107 
2108 	if (readonly_request)
2109 		error = mnt_make_readonly(real_mount(mnt));
2110 	else
2111 		__mnt_unmake_readonly(real_mount(mnt));
2112 	return error;
2113 }
2114 
2115 /*
2116  * change filesystem flags. dir should be a physical root of filesystem.
2117  * If you've mounted a non-root directory somewhere and want to do remount
2118  * on it - tough luck.
2119  */
2120 static int do_remount(struct path *path, int flags, int mnt_flags,
2121 		      void *data)
2122 {
2123 	int err;
2124 	struct super_block *sb = path->mnt->mnt_sb;
2125 	struct mount *mnt = real_mount(path->mnt);
2126 
2127 	if (!check_mnt(mnt))
2128 		return -EINVAL;
2129 
2130 	if (path->dentry != path->mnt->mnt_root)
2131 		return -EINVAL;
2132 
2133 	/* Don't allow changing of locked mnt flags.
2134 	 *
2135 	 * No locks need to be held here while testing the various
2136 	 * MNT_LOCK flags because those flags can never be cleared
2137 	 * once they are set.
2138 	 */
2139 	if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2140 	    !(mnt_flags & MNT_READONLY)) {
2141 		return -EPERM;
2142 	}
2143 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2144 	    !(mnt_flags & MNT_NODEV)) {
2145 		/* Was the nodev implicitly added in mount? */
2146 		if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2147 		    !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2148 			mnt_flags |= MNT_NODEV;
2149 		} else {
2150 			return -EPERM;
2151 		}
2152 	}
2153 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2154 	    !(mnt_flags & MNT_NOSUID)) {
2155 		return -EPERM;
2156 	}
2157 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2158 	    !(mnt_flags & MNT_NOEXEC)) {
2159 		return -EPERM;
2160 	}
2161 	if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2162 	    ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2163 		return -EPERM;
2164 	}
2165 
2166 	err = security_sb_remount(sb, data);
2167 	if (err)
2168 		return err;
2169 
2170 	down_write(&sb->s_umount);
2171 	if (flags & MS_BIND)
2172 		err = change_mount_flags(path->mnt, flags);
2173 	else if (!capable(CAP_SYS_ADMIN))
2174 		err = -EPERM;
2175 	else
2176 		err = do_remount_sb(sb, flags, data, 0);
2177 	if (!err) {
2178 		lock_mount_hash();
2179 		mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2180 		mnt->mnt.mnt_flags = mnt_flags;
2181 		touch_mnt_namespace(mnt->mnt_ns);
2182 		unlock_mount_hash();
2183 	}
2184 	up_write(&sb->s_umount);
2185 	return err;
2186 }
2187 
2188 static inline int tree_contains_unbindable(struct mount *mnt)
2189 {
2190 	struct mount *p;
2191 	for (p = mnt; p; p = next_mnt(p, mnt)) {
2192 		if (IS_MNT_UNBINDABLE(p))
2193 			return 1;
2194 	}
2195 	return 0;
2196 }
2197 
2198 static int do_move_mount(struct path *path, const char *old_name)
2199 {
2200 	struct path old_path, parent_path;
2201 	struct mount *p;
2202 	struct mount *old;
2203 	struct mountpoint *mp;
2204 	int err;
2205 	if (!old_name || !*old_name)
2206 		return -EINVAL;
2207 	err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2208 	if (err)
2209 		return err;
2210 
2211 	mp = lock_mount(path);
2212 	err = PTR_ERR(mp);
2213 	if (IS_ERR(mp))
2214 		goto out;
2215 
2216 	old = real_mount(old_path.mnt);
2217 	p = real_mount(path->mnt);
2218 
2219 	err = -EINVAL;
2220 	if (!check_mnt(p) || !check_mnt(old))
2221 		goto out1;
2222 
2223 	if (old->mnt.mnt_flags & MNT_LOCKED)
2224 		goto out1;
2225 
2226 	err = -EINVAL;
2227 	if (old_path.dentry != old_path.mnt->mnt_root)
2228 		goto out1;
2229 
2230 	if (!mnt_has_parent(old))
2231 		goto out1;
2232 
2233 	if (d_is_dir(path->dentry) !=
2234 	      d_is_dir(old_path.dentry))
2235 		goto out1;
2236 	/*
2237 	 * Don't move a mount residing in a shared parent.
2238 	 */
2239 	if (IS_MNT_SHARED(old->mnt_parent))
2240 		goto out1;
2241 	/*
2242 	 * Don't move a mount tree containing unbindable mounts to a destination
2243 	 * mount which is shared.
2244 	 */
2245 	if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2246 		goto out1;
2247 	err = -ELOOP;
2248 	for (; mnt_has_parent(p); p = p->mnt_parent)
2249 		if (p == old)
2250 			goto out1;
2251 
2252 	err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2253 	if (err)
2254 		goto out1;
2255 
2256 	/* if the mount is moved, it should no longer be expire
2257 	 * automatically */
2258 	list_del_init(&old->mnt_expire);
2259 out1:
2260 	unlock_mount(mp);
2261 out:
2262 	if (!err)
2263 		path_put(&parent_path);
2264 	path_put(&old_path);
2265 	return err;
2266 }
2267 
2268 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2269 {
2270 	int err;
2271 	const char *subtype = strchr(fstype, '.');
2272 	if (subtype) {
2273 		subtype++;
2274 		err = -EINVAL;
2275 		if (!subtype[0])
2276 			goto err;
2277 	} else
2278 		subtype = "";
2279 
2280 	mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2281 	err = -ENOMEM;
2282 	if (!mnt->mnt_sb->s_subtype)
2283 		goto err;
2284 	return mnt;
2285 
2286  err:
2287 	mntput(mnt);
2288 	return ERR_PTR(err);
2289 }
2290 
2291 /*
2292  * add a mount into a namespace's mount tree
2293  */
2294 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2295 {
2296 	struct mountpoint *mp;
2297 	struct mount *parent;
2298 	int err;
2299 
2300 	mnt_flags &= ~MNT_INTERNAL_FLAGS;
2301 
2302 	mp = lock_mount(path);
2303 	if (IS_ERR(mp))
2304 		return PTR_ERR(mp);
2305 
2306 	parent = real_mount(path->mnt);
2307 	err = -EINVAL;
2308 	if (unlikely(!check_mnt(parent))) {
2309 		/* that's acceptable only for automounts done in private ns */
2310 		if (!(mnt_flags & MNT_SHRINKABLE))
2311 			goto unlock;
2312 		/* ... and for those we'd better have mountpoint still alive */
2313 		if (!parent->mnt_ns)
2314 			goto unlock;
2315 	}
2316 
2317 	/* Refuse the same filesystem on the same mount point */
2318 	err = -EBUSY;
2319 	if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2320 	    path->mnt->mnt_root == path->dentry)
2321 		goto unlock;
2322 
2323 	err = -EINVAL;
2324 	if (d_is_symlink(newmnt->mnt.mnt_root))
2325 		goto unlock;
2326 
2327 	newmnt->mnt.mnt_flags = mnt_flags;
2328 	err = graft_tree(newmnt, parent, mp);
2329 
2330 unlock:
2331 	unlock_mount(mp);
2332 	return err;
2333 }
2334 
2335 /*
2336  * create a new mount for userspace and request it to be added into the
2337  * namespace's tree
2338  */
2339 static int do_new_mount(struct path *path, const char *fstype, int flags,
2340 			int mnt_flags, const char *name, void *data)
2341 {
2342 	struct file_system_type *type;
2343 	struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2344 	struct vfsmount *mnt;
2345 	int err;
2346 
2347 	if (!fstype)
2348 		return -EINVAL;
2349 
2350 	type = get_fs_type(fstype);
2351 	if (!type)
2352 		return -ENODEV;
2353 
2354 	if (user_ns != &init_user_ns) {
2355 		if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2356 			put_filesystem(type);
2357 			return -EPERM;
2358 		}
2359 		/* Only in special cases allow devices from mounts
2360 		 * created outside the initial user namespace.
2361 		 */
2362 		if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2363 			flags |= MS_NODEV;
2364 			mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2365 		}
2366 	}
2367 
2368 	mnt = vfs_kern_mount(type, flags, name, data);
2369 	if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2370 	    !mnt->mnt_sb->s_subtype)
2371 		mnt = fs_set_subtype(mnt, fstype);
2372 
2373 	put_filesystem(type);
2374 	if (IS_ERR(mnt))
2375 		return PTR_ERR(mnt);
2376 
2377 	err = do_add_mount(real_mount(mnt), path, mnt_flags);
2378 	if (err)
2379 		mntput(mnt);
2380 	return err;
2381 }
2382 
2383 int finish_automount(struct vfsmount *m, struct path *path)
2384 {
2385 	struct mount *mnt = real_mount(m);
2386 	int err;
2387 	/* The new mount record should have at least 2 refs to prevent it being
2388 	 * expired before we get a chance to add it
2389 	 */
2390 	BUG_ON(mnt_get_count(mnt) < 2);
2391 
2392 	if (m->mnt_sb == path->mnt->mnt_sb &&
2393 	    m->mnt_root == path->dentry) {
2394 		err = -ELOOP;
2395 		goto fail;
2396 	}
2397 
2398 	err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2399 	if (!err)
2400 		return 0;
2401 fail:
2402 	/* remove m from any expiration list it may be on */
2403 	if (!list_empty(&mnt->mnt_expire)) {
2404 		namespace_lock();
2405 		list_del_init(&mnt->mnt_expire);
2406 		namespace_unlock();
2407 	}
2408 	mntput(m);
2409 	mntput(m);
2410 	return err;
2411 }
2412 
2413 /**
2414  * mnt_set_expiry - Put a mount on an expiration list
2415  * @mnt: The mount to list.
2416  * @expiry_list: The list to add the mount to.
2417  */
2418 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2419 {
2420 	namespace_lock();
2421 
2422 	list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2423 
2424 	namespace_unlock();
2425 }
2426 EXPORT_SYMBOL(mnt_set_expiry);
2427 
2428 /*
2429  * process a list of expirable mountpoints with the intent of discarding any
2430  * mountpoints that aren't in use and haven't been touched since last we came
2431  * here
2432  */
2433 void mark_mounts_for_expiry(struct list_head *mounts)
2434 {
2435 	struct mount *mnt, *next;
2436 	LIST_HEAD(graveyard);
2437 
2438 	if (list_empty(mounts))
2439 		return;
2440 
2441 	namespace_lock();
2442 	lock_mount_hash();
2443 
2444 	/* extract from the expiration list every vfsmount that matches the
2445 	 * following criteria:
2446 	 * - only referenced by its parent vfsmount
2447 	 * - still marked for expiry (marked on the last call here; marks are
2448 	 *   cleared by mntput())
2449 	 */
2450 	list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2451 		if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2452 			propagate_mount_busy(mnt, 1))
2453 			continue;
2454 		list_move(&mnt->mnt_expire, &graveyard);
2455 	}
2456 	while (!list_empty(&graveyard)) {
2457 		mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2458 		touch_mnt_namespace(mnt->mnt_ns);
2459 		umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2460 	}
2461 	unlock_mount_hash();
2462 	namespace_unlock();
2463 }
2464 
2465 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2466 
2467 /*
2468  * Ripoff of 'select_parent()'
2469  *
2470  * search the list of submounts for a given mountpoint, and move any
2471  * shrinkable submounts to the 'graveyard' list.
2472  */
2473 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2474 {
2475 	struct mount *this_parent = parent;
2476 	struct list_head *next;
2477 	int found = 0;
2478 
2479 repeat:
2480 	next = this_parent->mnt_mounts.next;
2481 resume:
2482 	while (next != &this_parent->mnt_mounts) {
2483 		struct list_head *tmp = next;
2484 		struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2485 
2486 		next = tmp->next;
2487 		if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2488 			continue;
2489 		/*
2490 		 * Descend a level if the d_mounts list is non-empty.
2491 		 */
2492 		if (!list_empty(&mnt->mnt_mounts)) {
2493 			this_parent = mnt;
2494 			goto repeat;
2495 		}
2496 
2497 		if (!propagate_mount_busy(mnt, 1)) {
2498 			list_move_tail(&mnt->mnt_expire, graveyard);
2499 			found++;
2500 		}
2501 	}
2502 	/*
2503 	 * All done at this level ... ascend and resume the search
2504 	 */
2505 	if (this_parent != parent) {
2506 		next = this_parent->mnt_child.next;
2507 		this_parent = this_parent->mnt_parent;
2508 		goto resume;
2509 	}
2510 	return found;
2511 }
2512 
2513 /*
2514  * process a list of expirable mountpoints with the intent of discarding any
2515  * submounts of a specific parent mountpoint
2516  *
2517  * mount_lock must be held for write
2518  */
2519 static void shrink_submounts(struct mount *mnt)
2520 {
2521 	LIST_HEAD(graveyard);
2522 	struct mount *m;
2523 
2524 	/* extract submounts of 'mountpoint' from the expiration list */
2525 	while (select_submounts(mnt, &graveyard)) {
2526 		while (!list_empty(&graveyard)) {
2527 			m = list_first_entry(&graveyard, struct mount,
2528 						mnt_expire);
2529 			touch_mnt_namespace(m->mnt_ns);
2530 			umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2531 		}
2532 	}
2533 }
2534 
2535 /*
2536  * Some copy_from_user() implementations do not return the exact number of
2537  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2538  * Note that this function differs from copy_from_user() in that it will oops
2539  * on bad values of `to', rather than returning a short copy.
2540  */
2541 static long exact_copy_from_user(void *to, const void __user * from,
2542 				 unsigned long n)
2543 {
2544 	char *t = to;
2545 	const char __user *f = from;
2546 	char c;
2547 
2548 	if (!access_ok(VERIFY_READ, from, n))
2549 		return n;
2550 
2551 	while (n) {
2552 		if (__get_user(c, f)) {
2553 			memset(t, 0, n);
2554 			break;
2555 		}
2556 		*t++ = c;
2557 		f++;
2558 		n--;
2559 	}
2560 	return n;
2561 }
2562 
2563 int copy_mount_options(const void __user * data, unsigned long *where)
2564 {
2565 	int i;
2566 	unsigned long page;
2567 	unsigned long size;
2568 
2569 	*where = 0;
2570 	if (!data)
2571 		return 0;
2572 
2573 	if (!(page = __get_free_page(GFP_KERNEL)))
2574 		return -ENOMEM;
2575 
2576 	/* We only care that *some* data at the address the user
2577 	 * gave us is valid.  Just in case, we'll zero
2578 	 * the remainder of the page.
2579 	 */
2580 	/* copy_from_user cannot cross TASK_SIZE ! */
2581 	size = TASK_SIZE - (unsigned long)data;
2582 	if (size > PAGE_SIZE)
2583 		size = PAGE_SIZE;
2584 
2585 	i = size - exact_copy_from_user((void *)page, data, size);
2586 	if (!i) {
2587 		free_page(page);
2588 		return -EFAULT;
2589 	}
2590 	if (i != PAGE_SIZE)
2591 		memset((char *)page + i, 0, PAGE_SIZE - i);
2592 	*where = page;
2593 	return 0;
2594 }
2595 
2596 char *copy_mount_string(const void __user *data)
2597 {
2598 	return data ? strndup_user(data, PAGE_SIZE) : NULL;
2599 }
2600 
2601 /*
2602  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2603  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2604  *
2605  * data is a (void *) that can point to any structure up to
2606  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2607  * information (or be NULL).
2608  *
2609  * Pre-0.97 versions of mount() didn't have a flags word.
2610  * When the flags word was introduced its top half was required
2611  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2612  * Therefore, if this magic number is present, it carries no information
2613  * and must be discarded.
2614  */
2615 long do_mount(const char *dev_name, const char __user *dir_name,
2616 		const char *type_page, unsigned long flags, void *data_page)
2617 {
2618 	struct path path;
2619 	int retval = 0;
2620 	int mnt_flags = 0;
2621 
2622 	/* Discard magic */
2623 	if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2624 		flags &= ~MS_MGC_MSK;
2625 
2626 	/* Basic sanity checks */
2627 	if (data_page)
2628 		((char *)data_page)[PAGE_SIZE - 1] = 0;
2629 
2630 	/* ... and get the mountpoint */
2631 	retval = user_path(dir_name, &path);
2632 	if (retval)
2633 		return retval;
2634 
2635 	retval = security_sb_mount(dev_name, &path,
2636 				   type_page, flags, data_page);
2637 	if (!retval && !may_mount())
2638 		retval = -EPERM;
2639 	if (retval)
2640 		goto dput_out;
2641 
2642 	/* Default to relatime unless overriden */
2643 	if (!(flags & MS_NOATIME))
2644 		mnt_flags |= MNT_RELATIME;
2645 
2646 	/* Separate the per-mountpoint flags */
2647 	if (flags & MS_NOSUID)
2648 		mnt_flags |= MNT_NOSUID;
2649 	if (flags & MS_NODEV)
2650 		mnt_flags |= MNT_NODEV;
2651 	if (flags & MS_NOEXEC)
2652 		mnt_flags |= MNT_NOEXEC;
2653 	if (flags & MS_NOATIME)
2654 		mnt_flags |= MNT_NOATIME;
2655 	if (flags & MS_NODIRATIME)
2656 		mnt_flags |= MNT_NODIRATIME;
2657 	if (flags & MS_STRICTATIME)
2658 		mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2659 	if (flags & MS_RDONLY)
2660 		mnt_flags |= MNT_READONLY;
2661 
2662 	/* The default atime for remount is preservation */
2663 	if ((flags & MS_REMOUNT) &&
2664 	    ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2665 		       MS_STRICTATIME)) == 0)) {
2666 		mnt_flags &= ~MNT_ATIME_MASK;
2667 		mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2668 	}
2669 
2670 	flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2671 		   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2672 		   MS_STRICTATIME);
2673 
2674 	if (flags & MS_REMOUNT)
2675 		retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2676 				    data_page);
2677 	else if (flags & MS_BIND)
2678 		retval = do_loopback(&path, dev_name, flags & MS_REC);
2679 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2680 		retval = do_change_type(&path, flags);
2681 	else if (flags & MS_MOVE)
2682 		retval = do_move_mount(&path, dev_name);
2683 	else
2684 		retval = do_new_mount(&path, type_page, flags, mnt_flags,
2685 				      dev_name, data_page);
2686 dput_out:
2687 	path_put(&path);
2688 	return retval;
2689 }
2690 
2691 static void free_mnt_ns(struct mnt_namespace *ns)
2692 {
2693 	ns_free_inum(&ns->ns);
2694 	put_user_ns(ns->user_ns);
2695 	kfree(ns);
2696 }
2697 
2698 /*
2699  * Assign a sequence number so we can detect when we attempt to bind
2700  * mount a reference to an older mount namespace into the current
2701  * mount namespace, preventing reference counting loops.  A 64bit
2702  * number incrementing at 10Ghz will take 12,427 years to wrap which
2703  * is effectively never, so we can ignore the possibility.
2704  */
2705 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2706 
2707 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2708 {
2709 	struct mnt_namespace *new_ns;
2710 	int ret;
2711 
2712 	new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2713 	if (!new_ns)
2714 		return ERR_PTR(-ENOMEM);
2715 	ret = ns_alloc_inum(&new_ns->ns);
2716 	if (ret) {
2717 		kfree(new_ns);
2718 		return ERR_PTR(ret);
2719 	}
2720 	new_ns->ns.ops = &mntns_operations;
2721 	new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2722 	atomic_set(&new_ns->count, 1);
2723 	new_ns->root = NULL;
2724 	INIT_LIST_HEAD(&new_ns->list);
2725 	init_waitqueue_head(&new_ns->poll);
2726 	new_ns->event = 0;
2727 	new_ns->user_ns = get_user_ns(user_ns);
2728 	return new_ns;
2729 }
2730 
2731 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2732 		struct user_namespace *user_ns, struct fs_struct *new_fs)
2733 {
2734 	struct mnt_namespace *new_ns;
2735 	struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2736 	struct mount *p, *q;
2737 	struct mount *old;
2738 	struct mount *new;
2739 	int copy_flags;
2740 
2741 	BUG_ON(!ns);
2742 
2743 	if (likely(!(flags & CLONE_NEWNS))) {
2744 		get_mnt_ns(ns);
2745 		return ns;
2746 	}
2747 
2748 	old = ns->root;
2749 
2750 	new_ns = alloc_mnt_ns(user_ns);
2751 	if (IS_ERR(new_ns))
2752 		return new_ns;
2753 
2754 	namespace_lock();
2755 	/* First pass: copy the tree topology */
2756 	copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2757 	if (user_ns != ns->user_ns)
2758 		copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2759 	new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2760 	if (IS_ERR(new)) {
2761 		namespace_unlock();
2762 		free_mnt_ns(new_ns);
2763 		return ERR_CAST(new);
2764 	}
2765 	new_ns->root = new;
2766 	list_add_tail(&new_ns->list, &new->mnt_list);
2767 
2768 	/*
2769 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2770 	 * as belonging to new namespace.  We have already acquired a private
2771 	 * fs_struct, so tsk->fs->lock is not needed.
2772 	 */
2773 	p = old;
2774 	q = new;
2775 	while (p) {
2776 		q->mnt_ns = new_ns;
2777 		if (new_fs) {
2778 			if (&p->mnt == new_fs->root.mnt) {
2779 				new_fs->root.mnt = mntget(&q->mnt);
2780 				rootmnt = &p->mnt;
2781 			}
2782 			if (&p->mnt == new_fs->pwd.mnt) {
2783 				new_fs->pwd.mnt = mntget(&q->mnt);
2784 				pwdmnt = &p->mnt;
2785 			}
2786 		}
2787 		p = next_mnt(p, old);
2788 		q = next_mnt(q, new);
2789 		if (!q)
2790 			break;
2791 		while (p->mnt.mnt_root != q->mnt.mnt_root)
2792 			p = next_mnt(p, old);
2793 	}
2794 	namespace_unlock();
2795 
2796 	if (rootmnt)
2797 		mntput(rootmnt);
2798 	if (pwdmnt)
2799 		mntput(pwdmnt);
2800 
2801 	return new_ns;
2802 }
2803 
2804 /**
2805  * create_mnt_ns - creates a private namespace and adds a root filesystem
2806  * @mnt: pointer to the new root filesystem mountpoint
2807  */
2808 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2809 {
2810 	struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2811 	if (!IS_ERR(new_ns)) {
2812 		struct mount *mnt = real_mount(m);
2813 		mnt->mnt_ns = new_ns;
2814 		new_ns->root = mnt;
2815 		list_add(&mnt->mnt_list, &new_ns->list);
2816 	} else {
2817 		mntput(m);
2818 	}
2819 	return new_ns;
2820 }
2821 
2822 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2823 {
2824 	struct mnt_namespace *ns;
2825 	struct super_block *s;
2826 	struct path path;
2827 	int err;
2828 
2829 	ns = create_mnt_ns(mnt);
2830 	if (IS_ERR(ns))
2831 		return ERR_CAST(ns);
2832 
2833 	err = vfs_path_lookup(mnt->mnt_root, mnt,
2834 			name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2835 
2836 	put_mnt_ns(ns);
2837 
2838 	if (err)
2839 		return ERR_PTR(err);
2840 
2841 	/* trade a vfsmount reference for active sb one */
2842 	s = path.mnt->mnt_sb;
2843 	atomic_inc(&s->s_active);
2844 	mntput(path.mnt);
2845 	/* lock the sucker */
2846 	down_write(&s->s_umount);
2847 	/* ... and return the root of (sub)tree on it */
2848 	return path.dentry;
2849 }
2850 EXPORT_SYMBOL(mount_subtree);
2851 
2852 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2853 		char __user *, type, unsigned long, flags, void __user *, data)
2854 {
2855 	int ret;
2856 	char *kernel_type;
2857 	char *kernel_dev;
2858 	unsigned long data_page;
2859 
2860 	kernel_type = copy_mount_string(type);
2861 	ret = PTR_ERR(kernel_type);
2862 	if (IS_ERR(kernel_type))
2863 		goto out_type;
2864 
2865 	kernel_dev = copy_mount_string(dev_name);
2866 	ret = PTR_ERR(kernel_dev);
2867 	if (IS_ERR(kernel_dev))
2868 		goto out_dev;
2869 
2870 	ret = copy_mount_options(data, &data_page);
2871 	if (ret < 0)
2872 		goto out_data;
2873 
2874 	ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2875 		(void *) data_page);
2876 
2877 	free_page(data_page);
2878 out_data:
2879 	kfree(kernel_dev);
2880 out_dev:
2881 	kfree(kernel_type);
2882 out_type:
2883 	return ret;
2884 }
2885 
2886 /*
2887  * Return true if path is reachable from root
2888  *
2889  * namespace_sem or mount_lock is held
2890  */
2891 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2892 			 const struct path *root)
2893 {
2894 	while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2895 		dentry = mnt->mnt_mountpoint;
2896 		mnt = mnt->mnt_parent;
2897 	}
2898 	return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2899 }
2900 
2901 int path_is_under(struct path *path1, struct path *path2)
2902 {
2903 	int res;
2904 	read_seqlock_excl(&mount_lock);
2905 	res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2906 	read_sequnlock_excl(&mount_lock);
2907 	return res;
2908 }
2909 EXPORT_SYMBOL(path_is_under);
2910 
2911 /*
2912  * pivot_root Semantics:
2913  * Moves the root file system of the current process to the directory put_old,
2914  * makes new_root as the new root file system of the current process, and sets
2915  * root/cwd of all processes which had them on the current root to new_root.
2916  *
2917  * Restrictions:
2918  * The new_root and put_old must be directories, and  must not be on the
2919  * same file  system as the current process root. The put_old  must  be
2920  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2921  * pointed to by put_old must yield the same directory as new_root. No other
2922  * file system may be mounted on put_old. After all, new_root is a mountpoint.
2923  *
2924  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2925  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2926  * in this situation.
2927  *
2928  * Notes:
2929  *  - we don't move root/cwd if they are not at the root (reason: if something
2930  *    cared enough to change them, it's probably wrong to force them elsewhere)
2931  *  - it's okay to pick a root that isn't the root of a file system, e.g.
2932  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2933  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2934  *    first.
2935  */
2936 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2937 		const char __user *, put_old)
2938 {
2939 	struct path new, old, parent_path, root_parent, root;
2940 	struct mount *new_mnt, *root_mnt, *old_mnt;
2941 	struct mountpoint *old_mp, *root_mp;
2942 	int error;
2943 
2944 	if (!may_mount())
2945 		return -EPERM;
2946 
2947 	error = user_path_dir(new_root, &new);
2948 	if (error)
2949 		goto out0;
2950 
2951 	error = user_path_dir(put_old, &old);
2952 	if (error)
2953 		goto out1;
2954 
2955 	error = security_sb_pivotroot(&old, &new);
2956 	if (error)
2957 		goto out2;
2958 
2959 	get_fs_root(current->fs, &root);
2960 	old_mp = lock_mount(&old);
2961 	error = PTR_ERR(old_mp);
2962 	if (IS_ERR(old_mp))
2963 		goto out3;
2964 
2965 	error = -EINVAL;
2966 	new_mnt = real_mount(new.mnt);
2967 	root_mnt = real_mount(root.mnt);
2968 	old_mnt = real_mount(old.mnt);
2969 	if (IS_MNT_SHARED(old_mnt) ||
2970 		IS_MNT_SHARED(new_mnt->mnt_parent) ||
2971 		IS_MNT_SHARED(root_mnt->mnt_parent))
2972 		goto out4;
2973 	if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2974 		goto out4;
2975 	if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2976 		goto out4;
2977 	error = -ENOENT;
2978 	if (d_unlinked(new.dentry))
2979 		goto out4;
2980 	error = -EBUSY;
2981 	if (new_mnt == root_mnt || old_mnt == root_mnt)
2982 		goto out4; /* loop, on the same file system  */
2983 	error = -EINVAL;
2984 	if (root.mnt->mnt_root != root.dentry)
2985 		goto out4; /* not a mountpoint */
2986 	if (!mnt_has_parent(root_mnt))
2987 		goto out4; /* not attached */
2988 	root_mp = root_mnt->mnt_mp;
2989 	if (new.mnt->mnt_root != new.dentry)
2990 		goto out4; /* not a mountpoint */
2991 	if (!mnt_has_parent(new_mnt))
2992 		goto out4; /* not attached */
2993 	/* make sure we can reach put_old from new_root */
2994 	if (!is_path_reachable(old_mnt, old.dentry, &new))
2995 		goto out4;
2996 	/* make certain new is below the root */
2997 	if (!is_path_reachable(new_mnt, new.dentry, &root))
2998 		goto out4;
2999 	root_mp->m_count++; /* pin it so it won't go away */
3000 	lock_mount_hash();
3001 	detach_mnt(new_mnt, &parent_path);
3002 	detach_mnt(root_mnt, &root_parent);
3003 	if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3004 		new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3005 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3006 	}
3007 	/* mount old root on put_old */
3008 	attach_mnt(root_mnt, old_mnt, old_mp);
3009 	/* mount new_root on / */
3010 	attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3011 	touch_mnt_namespace(current->nsproxy->mnt_ns);
3012 	/* A moved mount should not expire automatically */
3013 	list_del_init(&new_mnt->mnt_expire);
3014 	unlock_mount_hash();
3015 	chroot_fs_refs(&root, &new);
3016 	put_mountpoint(root_mp);
3017 	error = 0;
3018 out4:
3019 	unlock_mount(old_mp);
3020 	if (!error) {
3021 		path_put(&root_parent);
3022 		path_put(&parent_path);
3023 	}
3024 out3:
3025 	path_put(&root);
3026 out2:
3027 	path_put(&old);
3028 out1:
3029 	path_put(&new);
3030 out0:
3031 	return error;
3032 }
3033 
3034 static void __init init_mount_tree(void)
3035 {
3036 	struct vfsmount *mnt;
3037 	struct mnt_namespace *ns;
3038 	struct path root;
3039 	struct file_system_type *type;
3040 
3041 	type = get_fs_type("rootfs");
3042 	if (!type)
3043 		panic("Can't find rootfs type");
3044 	mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3045 	put_filesystem(type);
3046 	if (IS_ERR(mnt))
3047 		panic("Can't create rootfs");
3048 
3049 	ns = create_mnt_ns(mnt);
3050 	if (IS_ERR(ns))
3051 		panic("Can't allocate initial namespace");
3052 
3053 	init_task.nsproxy->mnt_ns = ns;
3054 	get_mnt_ns(ns);
3055 
3056 	root.mnt = mnt;
3057 	root.dentry = mnt->mnt_root;
3058 	mnt->mnt_flags |= MNT_LOCKED;
3059 
3060 	set_fs_pwd(current->fs, &root);
3061 	set_fs_root(current->fs, &root);
3062 }
3063 
3064 void __init mnt_init(void)
3065 {
3066 	unsigned u;
3067 	int err;
3068 
3069 	mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3070 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3071 
3072 	mount_hashtable = alloc_large_system_hash("Mount-cache",
3073 				sizeof(struct hlist_head),
3074 				mhash_entries, 19,
3075 				0,
3076 				&m_hash_shift, &m_hash_mask, 0, 0);
3077 	mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3078 				sizeof(struct hlist_head),
3079 				mphash_entries, 19,
3080 				0,
3081 				&mp_hash_shift, &mp_hash_mask, 0, 0);
3082 
3083 	if (!mount_hashtable || !mountpoint_hashtable)
3084 		panic("Failed to allocate mount hash table\n");
3085 
3086 	for (u = 0; u <= m_hash_mask; u++)
3087 		INIT_HLIST_HEAD(&mount_hashtable[u]);
3088 	for (u = 0; u <= mp_hash_mask; u++)
3089 		INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3090 
3091 	kernfs_init();
3092 
3093 	err = sysfs_init();
3094 	if (err)
3095 		printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3096 			__func__, err);
3097 	fs_kobj = kobject_create_and_add("fs", NULL);
3098 	if (!fs_kobj)
3099 		printk(KERN_WARNING "%s: kobj create error\n", __func__);
3100 	init_rootfs();
3101 	init_mount_tree();
3102 }
3103 
3104 void put_mnt_ns(struct mnt_namespace *ns)
3105 {
3106 	if (!atomic_dec_and_test(&ns->count))
3107 		return;
3108 	drop_collected_mounts(&ns->root->mnt);
3109 	free_mnt_ns(ns);
3110 }
3111 
3112 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3113 {
3114 	struct vfsmount *mnt;
3115 	mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3116 	if (!IS_ERR(mnt)) {
3117 		/*
3118 		 * it is a longterm mount, don't release mnt until
3119 		 * we unmount before file sys is unregistered
3120 		*/
3121 		real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3122 	}
3123 	return mnt;
3124 }
3125 EXPORT_SYMBOL_GPL(kern_mount_data);
3126 
3127 void kern_unmount(struct vfsmount *mnt)
3128 {
3129 	/* release long term mount so mount point can be released */
3130 	if (!IS_ERR_OR_NULL(mnt)) {
3131 		real_mount(mnt)->mnt_ns = NULL;
3132 		synchronize_rcu();	/* yecchhh... */
3133 		mntput(mnt);
3134 	}
3135 }
3136 EXPORT_SYMBOL(kern_unmount);
3137 
3138 bool our_mnt(struct vfsmount *mnt)
3139 {
3140 	return check_mnt(real_mount(mnt));
3141 }
3142 
3143 bool current_chrooted(void)
3144 {
3145 	/* Does the current process have a non-standard root */
3146 	struct path ns_root;
3147 	struct path fs_root;
3148 	bool chrooted;
3149 
3150 	/* Find the namespace root */
3151 	ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3152 	ns_root.dentry = ns_root.mnt->mnt_root;
3153 	path_get(&ns_root);
3154 	while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3155 		;
3156 
3157 	get_fs_root(current->fs, &fs_root);
3158 
3159 	chrooted = !path_equal(&fs_root, &ns_root);
3160 
3161 	path_put(&fs_root);
3162 	path_put(&ns_root);
3163 
3164 	return chrooted;
3165 }
3166 
3167 bool fs_fully_visible(struct file_system_type *type)
3168 {
3169 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3170 	struct mount *mnt;
3171 	bool visible = false;
3172 
3173 	if (unlikely(!ns))
3174 		return false;
3175 
3176 	down_read(&namespace_sem);
3177 	list_for_each_entry(mnt, &ns->list, mnt_list) {
3178 		struct mount *child;
3179 		if (mnt->mnt.mnt_sb->s_type != type)
3180 			continue;
3181 
3182 		/* This mount is not fully visible if it's root directory
3183 		 * is not the root directory of the filesystem.
3184 		 */
3185 		if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3186 			continue;
3187 
3188 		/* This mount is not fully visible if there are any child mounts
3189 		 * that cover anything except for empty directories.
3190 		 */
3191 		list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3192 			struct inode *inode = child->mnt_mountpoint->d_inode;
3193 			if (!S_ISDIR(inode->i_mode))
3194 				goto next;
3195 			if (inode->i_nlink > 2)
3196 				goto next;
3197 		}
3198 		visible = true;
3199 		goto found;
3200 	next:	;
3201 	}
3202 found:
3203 	up_read(&namespace_sem);
3204 	return visible;
3205 }
3206 
3207 static struct ns_common *mntns_get(struct task_struct *task)
3208 {
3209 	struct ns_common *ns = NULL;
3210 	struct nsproxy *nsproxy;
3211 
3212 	task_lock(task);
3213 	nsproxy = task->nsproxy;
3214 	if (nsproxy) {
3215 		ns = &nsproxy->mnt_ns->ns;
3216 		get_mnt_ns(to_mnt_ns(ns));
3217 	}
3218 	task_unlock(task);
3219 
3220 	return ns;
3221 }
3222 
3223 static void mntns_put(struct ns_common *ns)
3224 {
3225 	put_mnt_ns(to_mnt_ns(ns));
3226 }
3227 
3228 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3229 {
3230 	struct fs_struct *fs = current->fs;
3231 	struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3232 	struct path root;
3233 
3234 	if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3235 	    !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3236 	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3237 		return -EPERM;
3238 
3239 	if (fs->users != 1)
3240 		return -EINVAL;
3241 
3242 	get_mnt_ns(mnt_ns);
3243 	put_mnt_ns(nsproxy->mnt_ns);
3244 	nsproxy->mnt_ns = mnt_ns;
3245 
3246 	/* Find the root */
3247 	root.mnt    = &mnt_ns->root->mnt;
3248 	root.dentry = mnt_ns->root->mnt.mnt_root;
3249 	path_get(&root);
3250 	while(d_mountpoint(root.dentry) && follow_down_one(&root))
3251 		;
3252 
3253 	/* Update the pwd and root */
3254 	set_fs_pwd(fs, &root);
3255 	set_fs_root(fs, &root);
3256 
3257 	path_put(&root);
3258 	return 0;
3259 }
3260 
3261 const struct proc_ns_operations mntns_operations = {
3262 	.name		= "mnt",
3263 	.type		= CLONE_NEWNS,
3264 	.get		= mntns_get,
3265 	.put		= mntns_put,
3266 	.install	= mntns_install,
3267 };
3268