xref: /linux/fs/namespace.c (revision cc4589ebfae6f8dbb5cf880a0a67eedab3416492)
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/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/acct.h>
18 #include <linux/capability.h>
19 #include <linux/cpumask.h>
20 #include <linux/module.h>
21 #include <linux/sysfs.h>
22 #include <linux/seq_file.h>
23 #include <linux/mnt_namespace.h>
24 #include <linux/namei.h>
25 #include <linux/nsproxy.h>
26 #include <linux/security.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/log2.h>
30 #include <linux/idr.h>
31 #include <linux/fs_struct.h>
32 #include <linux/fsnotify.h>
33 #include <asm/uaccess.h>
34 #include <asm/unistd.h>
35 #include "pnode.h"
36 #include "internal.h"
37 
38 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
39 #define HASH_SIZE (1UL << HASH_SHIFT)
40 
41 /* spinlock for vfsmount related operations, inplace of dcache_lock */
42 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
43 
44 static int event;
45 static DEFINE_IDA(mnt_id_ida);
46 static DEFINE_IDA(mnt_group_ida);
47 static int mnt_id_start = 0;
48 static int mnt_group_start = 1;
49 
50 static struct list_head *mount_hashtable __read_mostly;
51 static struct kmem_cache *mnt_cache __read_mostly;
52 static struct rw_semaphore namespace_sem;
53 
54 /* /sys/fs */
55 struct kobject *fs_kobj;
56 EXPORT_SYMBOL_GPL(fs_kobj);
57 
58 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
59 {
60 	unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
61 	tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
62 	tmp = tmp + (tmp >> HASH_SHIFT);
63 	return tmp & (HASH_SIZE - 1);
64 }
65 
66 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
67 
68 /* allocation is serialized by namespace_sem */
69 static int mnt_alloc_id(struct vfsmount *mnt)
70 {
71 	int res;
72 
73 retry:
74 	ida_pre_get(&mnt_id_ida, GFP_KERNEL);
75 	spin_lock(&vfsmount_lock);
76 	res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
77 	if (!res)
78 		mnt_id_start = mnt->mnt_id + 1;
79 	spin_unlock(&vfsmount_lock);
80 	if (res == -EAGAIN)
81 		goto retry;
82 
83 	return res;
84 }
85 
86 static void mnt_free_id(struct vfsmount *mnt)
87 {
88 	int id = mnt->mnt_id;
89 	spin_lock(&vfsmount_lock);
90 	ida_remove(&mnt_id_ida, id);
91 	if (mnt_id_start > id)
92 		mnt_id_start = id;
93 	spin_unlock(&vfsmount_lock);
94 }
95 
96 /*
97  * Allocate a new peer group ID
98  *
99  * mnt_group_ida is protected by namespace_sem
100  */
101 static int mnt_alloc_group_id(struct vfsmount *mnt)
102 {
103 	int res;
104 
105 	if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
106 		return -ENOMEM;
107 
108 	res = ida_get_new_above(&mnt_group_ida,
109 				mnt_group_start,
110 				&mnt->mnt_group_id);
111 	if (!res)
112 		mnt_group_start = mnt->mnt_group_id + 1;
113 
114 	return res;
115 }
116 
117 /*
118  * Release a peer group ID
119  */
120 void mnt_release_group_id(struct vfsmount *mnt)
121 {
122 	int id = mnt->mnt_group_id;
123 	ida_remove(&mnt_group_ida, id);
124 	if (mnt_group_start > id)
125 		mnt_group_start = id;
126 	mnt->mnt_group_id = 0;
127 }
128 
129 struct vfsmount *alloc_vfsmnt(const char *name)
130 {
131 	struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
132 	if (mnt) {
133 		int err;
134 
135 		err = mnt_alloc_id(mnt);
136 		if (err)
137 			goto out_free_cache;
138 
139 		if (name) {
140 			mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
141 			if (!mnt->mnt_devname)
142 				goto out_free_id;
143 		}
144 
145 		atomic_set(&mnt->mnt_count, 1);
146 		INIT_LIST_HEAD(&mnt->mnt_hash);
147 		INIT_LIST_HEAD(&mnt->mnt_child);
148 		INIT_LIST_HEAD(&mnt->mnt_mounts);
149 		INIT_LIST_HEAD(&mnt->mnt_list);
150 		INIT_LIST_HEAD(&mnt->mnt_expire);
151 		INIT_LIST_HEAD(&mnt->mnt_share);
152 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
153 		INIT_LIST_HEAD(&mnt->mnt_slave);
154 #ifdef CONFIG_FSNOTIFY
155 		INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
156 #endif
157 #ifdef CONFIG_SMP
158 		mnt->mnt_writers = alloc_percpu(int);
159 		if (!mnt->mnt_writers)
160 			goto out_free_devname;
161 #else
162 		mnt->mnt_writers = 0;
163 #endif
164 	}
165 	return mnt;
166 
167 #ifdef CONFIG_SMP
168 out_free_devname:
169 	kfree(mnt->mnt_devname);
170 #endif
171 out_free_id:
172 	mnt_free_id(mnt);
173 out_free_cache:
174 	kmem_cache_free(mnt_cache, mnt);
175 	return NULL;
176 }
177 
178 /*
179  * Most r/o checks on a fs are for operations that take
180  * discrete amounts of time, like a write() or unlink().
181  * We must keep track of when those operations start
182  * (for permission checks) and when they end, so that
183  * we can determine when writes are able to occur to
184  * a filesystem.
185  */
186 /*
187  * __mnt_is_readonly: check whether a mount is read-only
188  * @mnt: the mount to check for its write status
189  *
190  * This shouldn't be used directly ouside of the VFS.
191  * It does not guarantee that the filesystem will stay
192  * r/w, just that it is right *now*.  This can not and
193  * should not be used in place of IS_RDONLY(inode).
194  * mnt_want/drop_write() will _keep_ the filesystem
195  * r/w.
196  */
197 int __mnt_is_readonly(struct vfsmount *mnt)
198 {
199 	if (mnt->mnt_flags & MNT_READONLY)
200 		return 1;
201 	if (mnt->mnt_sb->s_flags & MS_RDONLY)
202 		return 1;
203 	return 0;
204 }
205 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
206 
207 static inline void inc_mnt_writers(struct vfsmount *mnt)
208 {
209 #ifdef CONFIG_SMP
210 	(*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
211 #else
212 	mnt->mnt_writers++;
213 #endif
214 }
215 
216 static inline void dec_mnt_writers(struct vfsmount *mnt)
217 {
218 #ifdef CONFIG_SMP
219 	(*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
220 #else
221 	mnt->mnt_writers--;
222 #endif
223 }
224 
225 static unsigned int count_mnt_writers(struct vfsmount *mnt)
226 {
227 #ifdef CONFIG_SMP
228 	unsigned int count = 0;
229 	int cpu;
230 
231 	for_each_possible_cpu(cpu) {
232 		count += *per_cpu_ptr(mnt->mnt_writers, cpu);
233 	}
234 
235 	return count;
236 #else
237 	return mnt->mnt_writers;
238 #endif
239 }
240 
241 /*
242  * Most r/o checks on a fs are for operations that take
243  * discrete amounts of time, like a write() or unlink().
244  * We must keep track of when those operations start
245  * (for permission checks) and when they end, so that
246  * we can determine when writes are able to occur to
247  * a filesystem.
248  */
249 /**
250  * mnt_want_write - get write access to a mount
251  * @mnt: the mount on which to take a write
252  *
253  * This tells the low-level filesystem that a write is
254  * about to be performed to it, and makes sure that
255  * writes are allowed before returning success.  When
256  * the write operation is finished, mnt_drop_write()
257  * must be called.  This is effectively a refcount.
258  */
259 int mnt_want_write(struct vfsmount *mnt)
260 {
261 	int ret = 0;
262 
263 	preempt_disable();
264 	inc_mnt_writers(mnt);
265 	/*
266 	 * The store to inc_mnt_writers must be visible before we pass
267 	 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
268 	 * incremented count after it has set MNT_WRITE_HOLD.
269 	 */
270 	smp_mb();
271 	while (mnt->mnt_flags & MNT_WRITE_HOLD)
272 		cpu_relax();
273 	/*
274 	 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
275 	 * be set to match its requirements. So we must not load that until
276 	 * MNT_WRITE_HOLD is cleared.
277 	 */
278 	smp_rmb();
279 	if (__mnt_is_readonly(mnt)) {
280 		dec_mnt_writers(mnt);
281 		ret = -EROFS;
282 		goto out;
283 	}
284 out:
285 	preempt_enable();
286 	return ret;
287 }
288 EXPORT_SYMBOL_GPL(mnt_want_write);
289 
290 /**
291  * mnt_clone_write - get write access to a mount
292  * @mnt: the mount on which to take a write
293  *
294  * This is effectively like mnt_want_write, except
295  * it must only be used to take an extra write reference
296  * on a mountpoint that we already know has a write reference
297  * on it. This allows some optimisation.
298  *
299  * After finished, mnt_drop_write must be called as usual to
300  * drop the reference.
301  */
302 int mnt_clone_write(struct vfsmount *mnt)
303 {
304 	/* superblock may be r/o */
305 	if (__mnt_is_readonly(mnt))
306 		return -EROFS;
307 	preempt_disable();
308 	inc_mnt_writers(mnt);
309 	preempt_enable();
310 	return 0;
311 }
312 EXPORT_SYMBOL_GPL(mnt_clone_write);
313 
314 /**
315  * mnt_want_write_file - get write access to a file's mount
316  * @file: the file who's mount on which to take a write
317  *
318  * This is like mnt_want_write, but it takes a file and can
319  * do some optimisations if the file is open for write already
320  */
321 int mnt_want_write_file(struct file *file)
322 {
323 	struct inode *inode = file->f_dentry->d_inode;
324 	if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
325 		return mnt_want_write(file->f_path.mnt);
326 	else
327 		return mnt_clone_write(file->f_path.mnt);
328 }
329 EXPORT_SYMBOL_GPL(mnt_want_write_file);
330 
331 /**
332  * mnt_drop_write - give up write access to a mount
333  * @mnt: the mount on which to give up write access
334  *
335  * Tells the low-level filesystem that we are done
336  * performing writes to it.  Must be matched with
337  * mnt_want_write() call above.
338  */
339 void mnt_drop_write(struct vfsmount *mnt)
340 {
341 	preempt_disable();
342 	dec_mnt_writers(mnt);
343 	preempt_enable();
344 }
345 EXPORT_SYMBOL_GPL(mnt_drop_write);
346 
347 static int mnt_make_readonly(struct vfsmount *mnt)
348 {
349 	int ret = 0;
350 
351 	spin_lock(&vfsmount_lock);
352 	mnt->mnt_flags |= MNT_WRITE_HOLD;
353 	/*
354 	 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
355 	 * should be visible before we do.
356 	 */
357 	smp_mb();
358 
359 	/*
360 	 * With writers on hold, if this value is zero, then there are
361 	 * definitely no active writers (although held writers may subsequently
362 	 * increment the count, they'll have to wait, and decrement it after
363 	 * seeing MNT_READONLY).
364 	 *
365 	 * It is OK to have counter incremented on one CPU and decremented on
366 	 * another: the sum will add up correctly. The danger would be when we
367 	 * sum up each counter, if we read a counter before it is incremented,
368 	 * but then read another CPU's count which it has been subsequently
369 	 * decremented from -- we would see more decrements than we should.
370 	 * MNT_WRITE_HOLD protects against this scenario, because
371 	 * mnt_want_write first increments count, then smp_mb, then spins on
372 	 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
373 	 * we're counting up here.
374 	 */
375 	if (count_mnt_writers(mnt) > 0)
376 		ret = -EBUSY;
377 	else
378 		mnt->mnt_flags |= MNT_READONLY;
379 	/*
380 	 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
381 	 * that become unheld will see MNT_READONLY.
382 	 */
383 	smp_wmb();
384 	mnt->mnt_flags &= ~MNT_WRITE_HOLD;
385 	spin_unlock(&vfsmount_lock);
386 	return ret;
387 }
388 
389 static void __mnt_unmake_readonly(struct vfsmount *mnt)
390 {
391 	spin_lock(&vfsmount_lock);
392 	mnt->mnt_flags &= ~MNT_READONLY;
393 	spin_unlock(&vfsmount_lock);
394 }
395 
396 void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
397 {
398 	mnt->mnt_sb = sb;
399 	mnt->mnt_root = dget(sb->s_root);
400 }
401 
402 EXPORT_SYMBOL(simple_set_mnt);
403 
404 void free_vfsmnt(struct vfsmount *mnt)
405 {
406 	kfree(mnt->mnt_devname);
407 	mnt_free_id(mnt);
408 #ifdef CONFIG_SMP
409 	free_percpu(mnt->mnt_writers);
410 #endif
411 	kmem_cache_free(mnt_cache, mnt);
412 }
413 
414 /*
415  * find the first or last mount at @dentry on vfsmount @mnt depending on
416  * @dir. If @dir is set return the first mount else return the last mount.
417  */
418 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
419 			      int dir)
420 {
421 	struct list_head *head = mount_hashtable + hash(mnt, dentry);
422 	struct list_head *tmp = head;
423 	struct vfsmount *p, *found = NULL;
424 
425 	for (;;) {
426 		tmp = dir ? tmp->next : tmp->prev;
427 		p = NULL;
428 		if (tmp == head)
429 			break;
430 		p = list_entry(tmp, struct vfsmount, mnt_hash);
431 		if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
432 			found = p;
433 			break;
434 		}
435 	}
436 	return found;
437 }
438 
439 /*
440  * lookup_mnt increments the ref count before returning
441  * the vfsmount struct.
442  */
443 struct vfsmount *lookup_mnt(struct path *path)
444 {
445 	struct vfsmount *child_mnt;
446 	spin_lock(&vfsmount_lock);
447 	if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
448 		mntget(child_mnt);
449 	spin_unlock(&vfsmount_lock);
450 	return child_mnt;
451 }
452 
453 static inline int check_mnt(struct vfsmount *mnt)
454 {
455 	return mnt->mnt_ns == current->nsproxy->mnt_ns;
456 }
457 
458 static void touch_mnt_namespace(struct mnt_namespace *ns)
459 {
460 	if (ns) {
461 		ns->event = ++event;
462 		wake_up_interruptible(&ns->poll);
463 	}
464 }
465 
466 static void __touch_mnt_namespace(struct mnt_namespace *ns)
467 {
468 	if (ns && ns->event != event) {
469 		ns->event = event;
470 		wake_up_interruptible(&ns->poll);
471 	}
472 }
473 
474 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
475 {
476 	old_path->dentry = mnt->mnt_mountpoint;
477 	old_path->mnt = mnt->mnt_parent;
478 	mnt->mnt_parent = mnt;
479 	mnt->mnt_mountpoint = mnt->mnt_root;
480 	list_del_init(&mnt->mnt_child);
481 	list_del_init(&mnt->mnt_hash);
482 	old_path->dentry->d_mounted--;
483 }
484 
485 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
486 			struct vfsmount *child_mnt)
487 {
488 	child_mnt->mnt_parent = mntget(mnt);
489 	child_mnt->mnt_mountpoint = dget(dentry);
490 	dentry->d_mounted++;
491 }
492 
493 static void attach_mnt(struct vfsmount *mnt, struct path *path)
494 {
495 	mnt_set_mountpoint(path->mnt, path->dentry, mnt);
496 	list_add_tail(&mnt->mnt_hash, mount_hashtable +
497 			hash(path->mnt, path->dentry));
498 	list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
499 }
500 
501 /*
502  * the caller must hold vfsmount_lock
503  */
504 static void commit_tree(struct vfsmount *mnt)
505 {
506 	struct vfsmount *parent = mnt->mnt_parent;
507 	struct vfsmount *m;
508 	LIST_HEAD(head);
509 	struct mnt_namespace *n = parent->mnt_ns;
510 
511 	BUG_ON(parent == mnt);
512 
513 	list_add_tail(&head, &mnt->mnt_list);
514 	list_for_each_entry(m, &head, mnt_list)
515 		m->mnt_ns = n;
516 	list_splice(&head, n->list.prev);
517 
518 	list_add_tail(&mnt->mnt_hash, mount_hashtable +
519 				hash(parent, mnt->mnt_mountpoint));
520 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
521 	touch_mnt_namespace(n);
522 }
523 
524 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
525 {
526 	struct list_head *next = p->mnt_mounts.next;
527 	if (next == &p->mnt_mounts) {
528 		while (1) {
529 			if (p == root)
530 				return NULL;
531 			next = p->mnt_child.next;
532 			if (next != &p->mnt_parent->mnt_mounts)
533 				break;
534 			p = p->mnt_parent;
535 		}
536 	}
537 	return list_entry(next, struct vfsmount, mnt_child);
538 }
539 
540 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
541 {
542 	struct list_head *prev = p->mnt_mounts.prev;
543 	while (prev != &p->mnt_mounts) {
544 		p = list_entry(prev, struct vfsmount, mnt_child);
545 		prev = p->mnt_mounts.prev;
546 	}
547 	return p;
548 }
549 
550 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
551 					int flag)
552 {
553 	struct super_block *sb = old->mnt_sb;
554 	struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
555 
556 	if (mnt) {
557 		if (flag & (CL_SLAVE | CL_PRIVATE))
558 			mnt->mnt_group_id = 0; /* not a peer of original */
559 		else
560 			mnt->mnt_group_id = old->mnt_group_id;
561 
562 		if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
563 			int err = mnt_alloc_group_id(mnt);
564 			if (err)
565 				goto out_free;
566 		}
567 
568 		mnt->mnt_flags = old->mnt_flags;
569 		atomic_inc(&sb->s_active);
570 		mnt->mnt_sb = sb;
571 		mnt->mnt_root = dget(root);
572 		mnt->mnt_mountpoint = mnt->mnt_root;
573 		mnt->mnt_parent = mnt;
574 
575 		if (flag & CL_SLAVE) {
576 			list_add(&mnt->mnt_slave, &old->mnt_slave_list);
577 			mnt->mnt_master = old;
578 			CLEAR_MNT_SHARED(mnt);
579 		} else if (!(flag & CL_PRIVATE)) {
580 			if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
581 				list_add(&mnt->mnt_share, &old->mnt_share);
582 			if (IS_MNT_SLAVE(old))
583 				list_add(&mnt->mnt_slave, &old->mnt_slave);
584 			mnt->mnt_master = old->mnt_master;
585 		}
586 		if (flag & CL_MAKE_SHARED)
587 			set_mnt_shared(mnt);
588 
589 		/* stick the duplicate mount on the same expiry list
590 		 * as the original if that was on one */
591 		if (flag & CL_EXPIRE) {
592 			if (!list_empty(&old->mnt_expire))
593 				list_add(&mnt->mnt_expire, &old->mnt_expire);
594 		}
595 	}
596 	return mnt;
597 
598  out_free:
599 	free_vfsmnt(mnt);
600 	return NULL;
601 }
602 
603 static inline void __mntput(struct vfsmount *mnt)
604 {
605 	struct super_block *sb = mnt->mnt_sb;
606 	/*
607 	 * This probably indicates that somebody messed
608 	 * up a mnt_want/drop_write() pair.  If this
609 	 * happens, the filesystem was probably unable
610 	 * to make r/w->r/o transitions.
611 	 */
612 	/*
613 	 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
614 	 * provides barriers, so count_mnt_writers() below is safe.  AV
615 	 */
616 	WARN_ON(count_mnt_writers(mnt));
617 	fsnotify_vfsmount_delete(mnt);
618 	dput(mnt->mnt_root);
619 	free_vfsmnt(mnt);
620 	deactivate_super(sb);
621 }
622 
623 void mntput_no_expire(struct vfsmount *mnt)
624 {
625 repeat:
626 	if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
627 		if (likely(!mnt->mnt_pinned)) {
628 			spin_unlock(&vfsmount_lock);
629 			__mntput(mnt);
630 			return;
631 		}
632 		atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
633 		mnt->mnt_pinned = 0;
634 		spin_unlock(&vfsmount_lock);
635 		acct_auto_close_mnt(mnt);
636 		goto repeat;
637 	}
638 }
639 
640 EXPORT_SYMBOL(mntput_no_expire);
641 
642 void mnt_pin(struct vfsmount *mnt)
643 {
644 	spin_lock(&vfsmount_lock);
645 	mnt->mnt_pinned++;
646 	spin_unlock(&vfsmount_lock);
647 }
648 
649 EXPORT_SYMBOL(mnt_pin);
650 
651 void mnt_unpin(struct vfsmount *mnt)
652 {
653 	spin_lock(&vfsmount_lock);
654 	if (mnt->mnt_pinned) {
655 		atomic_inc(&mnt->mnt_count);
656 		mnt->mnt_pinned--;
657 	}
658 	spin_unlock(&vfsmount_lock);
659 }
660 
661 EXPORT_SYMBOL(mnt_unpin);
662 
663 static inline void mangle(struct seq_file *m, const char *s)
664 {
665 	seq_escape(m, s, " \t\n\\");
666 }
667 
668 /*
669  * Simple .show_options callback for filesystems which don't want to
670  * implement more complex mount option showing.
671  *
672  * See also save_mount_options().
673  */
674 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
675 {
676 	const char *options;
677 
678 	rcu_read_lock();
679 	options = rcu_dereference(mnt->mnt_sb->s_options);
680 
681 	if (options != NULL && options[0]) {
682 		seq_putc(m, ',');
683 		mangle(m, options);
684 	}
685 	rcu_read_unlock();
686 
687 	return 0;
688 }
689 EXPORT_SYMBOL(generic_show_options);
690 
691 /*
692  * If filesystem uses generic_show_options(), this function should be
693  * called from the fill_super() callback.
694  *
695  * The .remount_fs callback usually needs to be handled in a special
696  * way, to make sure, that previous options are not overwritten if the
697  * remount fails.
698  *
699  * Also note, that if the filesystem's .remount_fs function doesn't
700  * reset all options to their default value, but changes only newly
701  * given options, then the displayed options will not reflect reality
702  * any more.
703  */
704 void save_mount_options(struct super_block *sb, char *options)
705 {
706 	BUG_ON(sb->s_options);
707 	rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
708 }
709 EXPORT_SYMBOL(save_mount_options);
710 
711 void replace_mount_options(struct super_block *sb, char *options)
712 {
713 	char *old = sb->s_options;
714 	rcu_assign_pointer(sb->s_options, options);
715 	if (old) {
716 		synchronize_rcu();
717 		kfree(old);
718 	}
719 }
720 EXPORT_SYMBOL(replace_mount_options);
721 
722 #ifdef CONFIG_PROC_FS
723 /* iterator */
724 static void *m_start(struct seq_file *m, loff_t *pos)
725 {
726 	struct proc_mounts *p = m->private;
727 
728 	down_read(&namespace_sem);
729 	return seq_list_start(&p->ns->list, *pos);
730 }
731 
732 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
733 {
734 	struct proc_mounts *p = m->private;
735 
736 	return seq_list_next(v, &p->ns->list, pos);
737 }
738 
739 static void m_stop(struct seq_file *m, void *v)
740 {
741 	up_read(&namespace_sem);
742 }
743 
744 int mnt_had_events(struct proc_mounts *p)
745 {
746 	struct mnt_namespace *ns = p->ns;
747 	int res = 0;
748 
749 	spin_lock(&vfsmount_lock);
750 	if (p->event != ns->event) {
751 		p->event = ns->event;
752 		res = 1;
753 	}
754 	spin_unlock(&vfsmount_lock);
755 
756 	return res;
757 }
758 
759 struct proc_fs_info {
760 	int flag;
761 	const char *str;
762 };
763 
764 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
765 {
766 	static const struct proc_fs_info fs_info[] = {
767 		{ MS_SYNCHRONOUS, ",sync" },
768 		{ MS_DIRSYNC, ",dirsync" },
769 		{ MS_MANDLOCK, ",mand" },
770 		{ 0, NULL }
771 	};
772 	const struct proc_fs_info *fs_infop;
773 
774 	for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
775 		if (sb->s_flags & fs_infop->flag)
776 			seq_puts(m, fs_infop->str);
777 	}
778 
779 	return security_sb_show_options(m, sb);
780 }
781 
782 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
783 {
784 	static const struct proc_fs_info mnt_info[] = {
785 		{ MNT_NOSUID, ",nosuid" },
786 		{ MNT_NODEV, ",nodev" },
787 		{ MNT_NOEXEC, ",noexec" },
788 		{ MNT_NOATIME, ",noatime" },
789 		{ MNT_NODIRATIME, ",nodiratime" },
790 		{ MNT_RELATIME, ",relatime" },
791 		{ MNT_STRICTATIME, ",strictatime" },
792 		{ 0, NULL }
793 	};
794 	const struct proc_fs_info *fs_infop;
795 
796 	for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
797 		if (mnt->mnt_flags & fs_infop->flag)
798 			seq_puts(m, fs_infop->str);
799 	}
800 }
801 
802 static void show_type(struct seq_file *m, struct super_block *sb)
803 {
804 	mangle(m, sb->s_type->name);
805 	if (sb->s_subtype && sb->s_subtype[0]) {
806 		seq_putc(m, '.');
807 		mangle(m, sb->s_subtype);
808 	}
809 }
810 
811 static int show_vfsmnt(struct seq_file *m, void *v)
812 {
813 	struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
814 	int err = 0;
815 	struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
816 
817 	mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
818 	seq_putc(m, ' ');
819 	seq_path(m, &mnt_path, " \t\n\\");
820 	seq_putc(m, ' ');
821 	show_type(m, mnt->mnt_sb);
822 	seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
823 	err = show_sb_opts(m, mnt->mnt_sb);
824 	if (err)
825 		goto out;
826 	show_mnt_opts(m, mnt);
827 	if (mnt->mnt_sb->s_op->show_options)
828 		err = mnt->mnt_sb->s_op->show_options(m, mnt);
829 	seq_puts(m, " 0 0\n");
830 out:
831 	return err;
832 }
833 
834 const struct seq_operations mounts_op = {
835 	.start	= m_start,
836 	.next	= m_next,
837 	.stop	= m_stop,
838 	.show	= show_vfsmnt
839 };
840 
841 static int show_mountinfo(struct seq_file *m, void *v)
842 {
843 	struct proc_mounts *p = m->private;
844 	struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
845 	struct super_block *sb = mnt->mnt_sb;
846 	struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
847 	struct path root = p->root;
848 	int err = 0;
849 
850 	seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
851 		   MAJOR(sb->s_dev), MINOR(sb->s_dev));
852 	seq_dentry(m, mnt->mnt_root, " \t\n\\");
853 	seq_putc(m, ' ');
854 	seq_path_root(m, &mnt_path, &root, " \t\n\\");
855 	if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
856 		/*
857 		 * Mountpoint is outside root, discard that one.  Ugly,
858 		 * but less so than trying to do that in iterator in a
859 		 * race-free way (due to renames).
860 		 */
861 		return SEQ_SKIP;
862 	}
863 	seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
864 	show_mnt_opts(m, mnt);
865 
866 	/* Tagged fields ("foo:X" or "bar") */
867 	if (IS_MNT_SHARED(mnt))
868 		seq_printf(m, " shared:%i", mnt->mnt_group_id);
869 	if (IS_MNT_SLAVE(mnt)) {
870 		int master = mnt->mnt_master->mnt_group_id;
871 		int dom = get_dominating_id(mnt, &p->root);
872 		seq_printf(m, " master:%i", master);
873 		if (dom && dom != master)
874 			seq_printf(m, " propagate_from:%i", dom);
875 	}
876 	if (IS_MNT_UNBINDABLE(mnt))
877 		seq_puts(m, " unbindable");
878 
879 	/* Filesystem specific data */
880 	seq_puts(m, " - ");
881 	show_type(m, sb);
882 	seq_putc(m, ' ');
883 	mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
884 	seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
885 	err = show_sb_opts(m, sb);
886 	if (err)
887 		goto out;
888 	if (sb->s_op->show_options)
889 		err = sb->s_op->show_options(m, mnt);
890 	seq_putc(m, '\n');
891 out:
892 	return err;
893 }
894 
895 const struct seq_operations mountinfo_op = {
896 	.start	= m_start,
897 	.next	= m_next,
898 	.stop	= m_stop,
899 	.show	= show_mountinfo,
900 };
901 
902 static int show_vfsstat(struct seq_file *m, void *v)
903 {
904 	struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
905 	struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
906 	int err = 0;
907 
908 	/* device */
909 	if (mnt->mnt_devname) {
910 		seq_puts(m, "device ");
911 		mangle(m, mnt->mnt_devname);
912 	} else
913 		seq_puts(m, "no device");
914 
915 	/* mount point */
916 	seq_puts(m, " mounted on ");
917 	seq_path(m, &mnt_path, " \t\n\\");
918 	seq_putc(m, ' ');
919 
920 	/* file system type */
921 	seq_puts(m, "with fstype ");
922 	show_type(m, mnt->mnt_sb);
923 
924 	/* optional statistics */
925 	if (mnt->mnt_sb->s_op->show_stats) {
926 		seq_putc(m, ' ');
927 		err = mnt->mnt_sb->s_op->show_stats(m, mnt);
928 	}
929 
930 	seq_putc(m, '\n');
931 	return err;
932 }
933 
934 const struct seq_operations mountstats_op = {
935 	.start	= m_start,
936 	.next	= m_next,
937 	.stop	= m_stop,
938 	.show	= show_vfsstat,
939 };
940 #endif  /* CONFIG_PROC_FS */
941 
942 /**
943  * may_umount_tree - check if a mount tree is busy
944  * @mnt: root of mount tree
945  *
946  * This is called to check if a tree of mounts has any
947  * open files, pwds, chroots or sub mounts that are
948  * busy.
949  */
950 int may_umount_tree(struct vfsmount *mnt)
951 {
952 	int actual_refs = 0;
953 	int minimum_refs = 0;
954 	struct vfsmount *p;
955 
956 	spin_lock(&vfsmount_lock);
957 	for (p = mnt; p; p = next_mnt(p, mnt)) {
958 		actual_refs += atomic_read(&p->mnt_count);
959 		minimum_refs += 2;
960 	}
961 	spin_unlock(&vfsmount_lock);
962 
963 	if (actual_refs > minimum_refs)
964 		return 0;
965 
966 	return 1;
967 }
968 
969 EXPORT_SYMBOL(may_umount_tree);
970 
971 /**
972  * may_umount - check if a mount point is busy
973  * @mnt: root of mount
974  *
975  * This is called to check if a mount point has any
976  * open files, pwds, chroots or sub mounts. If the
977  * mount has sub mounts this will return busy
978  * regardless of whether the sub mounts are busy.
979  *
980  * Doesn't take quota and stuff into account. IOW, in some cases it will
981  * give false negatives. The main reason why it's here is that we need
982  * a non-destructive way to look for easily umountable filesystems.
983  */
984 int may_umount(struct vfsmount *mnt)
985 {
986 	int ret = 1;
987 	down_read(&namespace_sem);
988 	spin_lock(&vfsmount_lock);
989 	if (propagate_mount_busy(mnt, 2))
990 		ret = 0;
991 	spin_unlock(&vfsmount_lock);
992 	up_read(&namespace_sem);
993 	return ret;
994 }
995 
996 EXPORT_SYMBOL(may_umount);
997 
998 void release_mounts(struct list_head *head)
999 {
1000 	struct vfsmount *mnt;
1001 	while (!list_empty(head)) {
1002 		mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1003 		list_del_init(&mnt->mnt_hash);
1004 		if (mnt->mnt_parent != mnt) {
1005 			struct dentry *dentry;
1006 			struct vfsmount *m;
1007 			spin_lock(&vfsmount_lock);
1008 			dentry = mnt->mnt_mountpoint;
1009 			m = mnt->mnt_parent;
1010 			mnt->mnt_mountpoint = mnt->mnt_root;
1011 			mnt->mnt_parent = mnt;
1012 			m->mnt_ghosts--;
1013 			spin_unlock(&vfsmount_lock);
1014 			dput(dentry);
1015 			mntput(m);
1016 		}
1017 		mntput(mnt);
1018 	}
1019 }
1020 
1021 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1022 {
1023 	struct vfsmount *p;
1024 
1025 	for (p = mnt; p; p = next_mnt(p, mnt))
1026 		list_move(&p->mnt_hash, kill);
1027 
1028 	if (propagate)
1029 		propagate_umount(kill);
1030 
1031 	list_for_each_entry(p, kill, mnt_hash) {
1032 		list_del_init(&p->mnt_expire);
1033 		list_del_init(&p->mnt_list);
1034 		__touch_mnt_namespace(p->mnt_ns);
1035 		p->mnt_ns = NULL;
1036 		list_del_init(&p->mnt_child);
1037 		if (p->mnt_parent != p) {
1038 			p->mnt_parent->mnt_ghosts++;
1039 			p->mnt_mountpoint->d_mounted--;
1040 		}
1041 		change_mnt_propagation(p, MS_PRIVATE);
1042 	}
1043 }
1044 
1045 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1046 
1047 static int do_umount(struct vfsmount *mnt, int flags)
1048 {
1049 	struct super_block *sb = mnt->mnt_sb;
1050 	int retval;
1051 	LIST_HEAD(umount_list);
1052 
1053 	retval = security_sb_umount(mnt, flags);
1054 	if (retval)
1055 		return retval;
1056 
1057 	/*
1058 	 * Allow userspace to request a mountpoint be expired rather than
1059 	 * unmounting unconditionally. Unmount only happens if:
1060 	 *  (1) the mark is already set (the mark is cleared by mntput())
1061 	 *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1062 	 */
1063 	if (flags & MNT_EXPIRE) {
1064 		if (mnt == current->fs->root.mnt ||
1065 		    flags & (MNT_FORCE | MNT_DETACH))
1066 			return -EINVAL;
1067 
1068 		if (atomic_read(&mnt->mnt_count) != 2)
1069 			return -EBUSY;
1070 
1071 		if (!xchg(&mnt->mnt_expiry_mark, 1))
1072 			return -EAGAIN;
1073 	}
1074 
1075 	/*
1076 	 * If we may have to abort operations to get out of this
1077 	 * mount, and they will themselves hold resources we must
1078 	 * allow the fs to do things. In the Unix tradition of
1079 	 * 'Gee thats tricky lets do it in userspace' the umount_begin
1080 	 * might fail to complete on the first run through as other tasks
1081 	 * must return, and the like. Thats for the mount program to worry
1082 	 * about for the moment.
1083 	 */
1084 
1085 	if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1086 		sb->s_op->umount_begin(sb);
1087 	}
1088 
1089 	/*
1090 	 * No sense to grab the lock for this test, but test itself looks
1091 	 * somewhat bogus. Suggestions for better replacement?
1092 	 * Ho-hum... In principle, we might treat that as umount + switch
1093 	 * to rootfs. GC would eventually take care of the old vfsmount.
1094 	 * Actually it makes sense, especially if rootfs would contain a
1095 	 * /reboot - static binary that would close all descriptors and
1096 	 * call reboot(9). Then init(8) could umount root and exec /reboot.
1097 	 */
1098 	if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1099 		/*
1100 		 * Special case for "unmounting" root ...
1101 		 * we just try to remount it readonly.
1102 		 */
1103 		down_write(&sb->s_umount);
1104 		if (!(sb->s_flags & MS_RDONLY))
1105 			retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1106 		up_write(&sb->s_umount);
1107 		return retval;
1108 	}
1109 
1110 	down_write(&namespace_sem);
1111 	spin_lock(&vfsmount_lock);
1112 	event++;
1113 
1114 	if (!(flags & MNT_DETACH))
1115 		shrink_submounts(mnt, &umount_list);
1116 
1117 	retval = -EBUSY;
1118 	if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1119 		if (!list_empty(&mnt->mnt_list))
1120 			umount_tree(mnt, 1, &umount_list);
1121 		retval = 0;
1122 	}
1123 	spin_unlock(&vfsmount_lock);
1124 	up_write(&namespace_sem);
1125 	release_mounts(&umount_list);
1126 	return retval;
1127 }
1128 
1129 /*
1130  * Now umount can handle mount points as well as block devices.
1131  * This is important for filesystems which use unnamed block devices.
1132  *
1133  * We now support a flag for forced unmount like the other 'big iron'
1134  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1135  */
1136 
1137 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1138 {
1139 	struct path path;
1140 	int retval;
1141 	int lookup_flags = 0;
1142 
1143 	if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1144 		return -EINVAL;
1145 
1146 	if (!(flags & UMOUNT_NOFOLLOW))
1147 		lookup_flags |= LOOKUP_FOLLOW;
1148 
1149 	retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1150 	if (retval)
1151 		goto out;
1152 	retval = -EINVAL;
1153 	if (path.dentry != path.mnt->mnt_root)
1154 		goto dput_and_out;
1155 	if (!check_mnt(path.mnt))
1156 		goto dput_and_out;
1157 
1158 	retval = -EPERM;
1159 	if (!capable(CAP_SYS_ADMIN))
1160 		goto dput_and_out;
1161 
1162 	retval = do_umount(path.mnt, flags);
1163 dput_and_out:
1164 	/* we mustn't call path_put() as that would clear mnt_expiry_mark */
1165 	dput(path.dentry);
1166 	mntput_no_expire(path.mnt);
1167 out:
1168 	return retval;
1169 }
1170 
1171 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1172 
1173 /*
1174  *	The 2.0 compatible umount. No flags.
1175  */
1176 SYSCALL_DEFINE1(oldumount, char __user *, name)
1177 {
1178 	return sys_umount(name, 0);
1179 }
1180 
1181 #endif
1182 
1183 static int mount_is_safe(struct path *path)
1184 {
1185 	if (capable(CAP_SYS_ADMIN))
1186 		return 0;
1187 	return -EPERM;
1188 #ifdef notyet
1189 	if (S_ISLNK(path->dentry->d_inode->i_mode))
1190 		return -EPERM;
1191 	if (path->dentry->d_inode->i_mode & S_ISVTX) {
1192 		if (current_uid() != path->dentry->d_inode->i_uid)
1193 			return -EPERM;
1194 	}
1195 	if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1196 		return -EPERM;
1197 	return 0;
1198 #endif
1199 }
1200 
1201 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1202 					int flag)
1203 {
1204 	struct vfsmount *res, *p, *q, *r, *s;
1205 	struct path path;
1206 
1207 	if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1208 		return NULL;
1209 
1210 	res = q = clone_mnt(mnt, dentry, flag);
1211 	if (!q)
1212 		goto Enomem;
1213 	q->mnt_mountpoint = mnt->mnt_mountpoint;
1214 
1215 	p = mnt;
1216 	list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1217 		if (!is_subdir(r->mnt_mountpoint, dentry))
1218 			continue;
1219 
1220 		for (s = r; s; s = next_mnt(s, r)) {
1221 			if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1222 				s = skip_mnt_tree(s);
1223 				continue;
1224 			}
1225 			while (p != s->mnt_parent) {
1226 				p = p->mnt_parent;
1227 				q = q->mnt_parent;
1228 			}
1229 			p = s;
1230 			path.mnt = q;
1231 			path.dentry = p->mnt_mountpoint;
1232 			q = clone_mnt(p, p->mnt_root, flag);
1233 			if (!q)
1234 				goto Enomem;
1235 			spin_lock(&vfsmount_lock);
1236 			list_add_tail(&q->mnt_list, &res->mnt_list);
1237 			attach_mnt(q, &path);
1238 			spin_unlock(&vfsmount_lock);
1239 		}
1240 	}
1241 	return res;
1242 Enomem:
1243 	if (res) {
1244 		LIST_HEAD(umount_list);
1245 		spin_lock(&vfsmount_lock);
1246 		umount_tree(res, 0, &umount_list);
1247 		spin_unlock(&vfsmount_lock);
1248 		release_mounts(&umount_list);
1249 	}
1250 	return NULL;
1251 }
1252 
1253 struct vfsmount *collect_mounts(struct path *path)
1254 {
1255 	struct vfsmount *tree;
1256 	down_write(&namespace_sem);
1257 	tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1258 	up_write(&namespace_sem);
1259 	return tree;
1260 }
1261 
1262 void drop_collected_mounts(struct vfsmount *mnt)
1263 {
1264 	LIST_HEAD(umount_list);
1265 	down_write(&namespace_sem);
1266 	spin_lock(&vfsmount_lock);
1267 	umount_tree(mnt, 0, &umount_list);
1268 	spin_unlock(&vfsmount_lock);
1269 	up_write(&namespace_sem);
1270 	release_mounts(&umount_list);
1271 }
1272 
1273 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1274 		   struct vfsmount *root)
1275 {
1276 	struct vfsmount *mnt;
1277 	int res = f(root, arg);
1278 	if (res)
1279 		return res;
1280 	list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1281 		res = f(mnt, arg);
1282 		if (res)
1283 			return res;
1284 	}
1285 	return 0;
1286 }
1287 
1288 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1289 {
1290 	struct vfsmount *p;
1291 
1292 	for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1293 		if (p->mnt_group_id && !IS_MNT_SHARED(p))
1294 			mnt_release_group_id(p);
1295 	}
1296 }
1297 
1298 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1299 {
1300 	struct vfsmount *p;
1301 
1302 	for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1303 		if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1304 			int err = mnt_alloc_group_id(p);
1305 			if (err) {
1306 				cleanup_group_ids(mnt, p);
1307 				return err;
1308 			}
1309 		}
1310 	}
1311 
1312 	return 0;
1313 }
1314 
1315 /*
1316  *  @source_mnt : mount tree to be attached
1317  *  @nd         : place the mount tree @source_mnt is attached
1318  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1319  *  		   store the parent mount and mountpoint dentry.
1320  *  		   (done when source_mnt is moved)
1321  *
1322  *  NOTE: in the table below explains the semantics when a source mount
1323  *  of a given type is attached to a destination mount of a given type.
1324  * ---------------------------------------------------------------------------
1325  * |         BIND MOUNT OPERATION                                            |
1326  * |**************************************************************************
1327  * | source-->| shared        |       private  |       slave    | unbindable |
1328  * | dest     |               |                |                |            |
1329  * |   |      |               |                |                |            |
1330  * |   v      |               |                |                |            |
1331  * |**************************************************************************
1332  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1333  * |          |               |                |                |            |
1334  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1335  * ***************************************************************************
1336  * A bind operation clones the source mount and mounts the clone on the
1337  * destination mount.
1338  *
1339  * (++)  the cloned mount is propagated to all the mounts in the propagation
1340  * 	 tree of the destination mount and the cloned mount is added to
1341  * 	 the peer group of the source mount.
1342  * (+)   the cloned mount is created under the destination mount and is marked
1343  *       as shared. The cloned mount is added to the peer group of the source
1344  *       mount.
1345  * (+++) the mount is propagated to all the mounts in the propagation tree
1346  *       of the destination mount and the cloned mount is made slave
1347  *       of the same master as that of the source mount. The cloned mount
1348  *       is marked as 'shared and slave'.
1349  * (*)   the cloned mount is made a slave of the same master as that of the
1350  * 	 source mount.
1351  *
1352  * ---------------------------------------------------------------------------
1353  * |         		MOVE MOUNT OPERATION                                 |
1354  * |**************************************************************************
1355  * | source-->| shared        |       private  |       slave    | unbindable |
1356  * | dest     |               |                |                |            |
1357  * |   |      |               |                |                |            |
1358  * |   v      |               |                |                |            |
1359  * |**************************************************************************
1360  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1361  * |          |               |                |                |            |
1362  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1363  * ***************************************************************************
1364  *
1365  * (+)  the mount is moved to the destination. And is then propagated to
1366  * 	all the mounts in the propagation tree of the destination mount.
1367  * (+*)  the mount is moved to the destination.
1368  * (+++)  the mount is moved to the destination and is then propagated to
1369  * 	all the mounts belonging to the destination mount's propagation tree.
1370  * 	the mount is marked as 'shared and slave'.
1371  * (*)	the mount continues to be a slave at the new location.
1372  *
1373  * if the source mount is a tree, the operations explained above is
1374  * applied to each mount in the tree.
1375  * Must be called without spinlocks held, since this function can sleep
1376  * in allocations.
1377  */
1378 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1379 			struct path *path, struct path *parent_path)
1380 {
1381 	LIST_HEAD(tree_list);
1382 	struct vfsmount *dest_mnt = path->mnt;
1383 	struct dentry *dest_dentry = path->dentry;
1384 	struct vfsmount *child, *p;
1385 	int err;
1386 
1387 	if (IS_MNT_SHARED(dest_mnt)) {
1388 		err = invent_group_ids(source_mnt, true);
1389 		if (err)
1390 			goto out;
1391 	}
1392 	err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1393 	if (err)
1394 		goto out_cleanup_ids;
1395 
1396 	spin_lock(&vfsmount_lock);
1397 
1398 	if (IS_MNT_SHARED(dest_mnt)) {
1399 		for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1400 			set_mnt_shared(p);
1401 	}
1402 	if (parent_path) {
1403 		detach_mnt(source_mnt, parent_path);
1404 		attach_mnt(source_mnt, path);
1405 		touch_mnt_namespace(parent_path->mnt->mnt_ns);
1406 	} else {
1407 		mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1408 		commit_tree(source_mnt);
1409 	}
1410 
1411 	list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1412 		list_del_init(&child->mnt_hash);
1413 		commit_tree(child);
1414 	}
1415 	spin_unlock(&vfsmount_lock);
1416 	return 0;
1417 
1418  out_cleanup_ids:
1419 	if (IS_MNT_SHARED(dest_mnt))
1420 		cleanup_group_ids(source_mnt, NULL);
1421  out:
1422 	return err;
1423 }
1424 
1425 static int graft_tree(struct vfsmount *mnt, struct path *path)
1426 {
1427 	int err;
1428 	if (mnt->mnt_sb->s_flags & MS_NOUSER)
1429 		return -EINVAL;
1430 
1431 	if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1432 	      S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1433 		return -ENOTDIR;
1434 
1435 	err = -ENOENT;
1436 	mutex_lock(&path->dentry->d_inode->i_mutex);
1437 	if (cant_mount(path->dentry))
1438 		goto out_unlock;
1439 
1440 	if (!d_unlinked(path->dentry))
1441 		err = attach_recursive_mnt(mnt, path, NULL);
1442 out_unlock:
1443 	mutex_unlock(&path->dentry->d_inode->i_mutex);
1444 	return err;
1445 }
1446 
1447 /*
1448  * recursively change the type of the mountpoint.
1449  */
1450 static int do_change_type(struct path *path, int flag)
1451 {
1452 	struct vfsmount *m, *mnt = path->mnt;
1453 	int recurse = flag & MS_REC;
1454 	int type = flag & ~MS_REC;
1455 	int err = 0;
1456 
1457 	if (!capable(CAP_SYS_ADMIN))
1458 		return -EPERM;
1459 
1460 	if (path->dentry != path->mnt->mnt_root)
1461 		return -EINVAL;
1462 
1463 	down_write(&namespace_sem);
1464 	if (type == MS_SHARED) {
1465 		err = invent_group_ids(mnt, recurse);
1466 		if (err)
1467 			goto out_unlock;
1468 	}
1469 
1470 	spin_lock(&vfsmount_lock);
1471 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1472 		change_mnt_propagation(m, type);
1473 	spin_unlock(&vfsmount_lock);
1474 
1475  out_unlock:
1476 	up_write(&namespace_sem);
1477 	return err;
1478 }
1479 
1480 /*
1481  * do loopback mount.
1482  */
1483 static int do_loopback(struct path *path, char *old_name,
1484 				int recurse)
1485 {
1486 	struct path old_path;
1487 	struct vfsmount *mnt = NULL;
1488 	int err = mount_is_safe(path);
1489 	if (err)
1490 		return err;
1491 	if (!old_name || !*old_name)
1492 		return -EINVAL;
1493 	err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1494 	if (err)
1495 		return err;
1496 
1497 	down_write(&namespace_sem);
1498 	err = -EINVAL;
1499 	if (IS_MNT_UNBINDABLE(old_path.mnt))
1500 		goto out;
1501 
1502 	if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1503 		goto out;
1504 
1505 	err = -ENOMEM;
1506 	if (recurse)
1507 		mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1508 	else
1509 		mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1510 
1511 	if (!mnt)
1512 		goto out;
1513 
1514 	err = graft_tree(mnt, path);
1515 	if (err) {
1516 		LIST_HEAD(umount_list);
1517 		spin_lock(&vfsmount_lock);
1518 		umount_tree(mnt, 0, &umount_list);
1519 		spin_unlock(&vfsmount_lock);
1520 		release_mounts(&umount_list);
1521 	}
1522 
1523 out:
1524 	up_write(&namespace_sem);
1525 	path_put(&old_path);
1526 	return err;
1527 }
1528 
1529 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1530 {
1531 	int error = 0;
1532 	int readonly_request = 0;
1533 
1534 	if (ms_flags & MS_RDONLY)
1535 		readonly_request = 1;
1536 	if (readonly_request == __mnt_is_readonly(mnt))
1537 		return 0;
1538 
1539 	if (readonly_request)
1540 		error = mnt_make_readonly(mnt);
1541 	else
1542 		__mnt_unmake_readonly(mnt);
1543 	return error;
1544 }
1545 
1546 /*
1547  * change filesystem flags. dir should be a physical root of filesystem.
1548  * If you've mounted a non-root directory somewhere and want to do remount
1549  * on it - tough luck.
1550  */
1551 static int do_remount(struct path *path, int flags, int mnt_flags,
1552 		      void *data)
1553 {
1554 	int err;
1555 	struct super_block *sb = path->mnt->mnt_sb;
1556 
1557 	if (!capable(CAP_SYS_ADMIN))
1558 		return -EPERM;
1559 
1560 	if (!check_mnt(path->mnt))
1561 		return -EINVAL;
1562 
1563 	if (path->dentry != path->mnt->mnt_root)
1564 		return -EINVAL;
1565 
1566 	down_write(&sb->s_umount);
1567 	if (flags & MS_BIND)
1568 		err = change_mount_flags(path->mnt, flags);
1569 	else
1570 		err = do_remount_sb(sb, flags, data, 0);
1571 	if (!err) {
1572 		spin_lock(&vfsmount_lock);
1573 		mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1574 		path->mnt->mnt_flags = mnt_flags;
1575 		spin_unlock(&vfsmount_lock);
1576 	}
1577 	up_write(&sb->s_umount);
1578 	if (!err) {
1579 		spin_lock(&vfsmount_lock);
1580 		touch_mnt_namespace(path->mnt->mnt_ns);
1581 		spin_unlock(&vfsmount_lock);
1582 	}
1583 	return err;
1584 }
1585 
1586 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1587 {
1588 	struct vfsmount *p;
1589 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1590 		if (IS_MNT_UNBINDABLE(p))
1591 			return 1;
1592 	}
1593 	return 0;
1594 }
1595 
1596 static int do_move_mount(struct path *path, char *old_name)
1597 {
1598 	struct path old_path, parent_path;
1599 	struct vfsmount *p;
1600 	int err = 0;
1601 	if (!capable(CAP_SYS_ADMIN))
1602 		return -EPERM;
1603 	if (!old_name || !*old_name)
1604 		return -EINVAL;
1605 	err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1606 	if (err)
1607 		return err;
1608 
1609 	down_write(&namespace_sem);
1610 	while (d_mountpoint(path->dentry) &&
1611 	       follow_down(path))
1612 		;
1613 	err = -EINVAL;
1614 	if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1615 		goto out;
1616 
1617 	err = -ENOENT;
1618 	mutex_lock(&path->dentry->d_inode->i_mutex);
1619 	if (cant_mount(path->dentry))
1620 		goto out1;
1621 
1622 	if (d_unlinked(path->dentry))
1623 		goto out1;
1624 
1625 	err = -EINVAL;
1626 	if (old_path.dentry != old_path.mnt->mnt_root)
1627 		goto out1;
1628 
1629 	if (old_path.mnt == old_path.mnt->mnt_parent)
1630 		goto out1;
1631 
1632 	if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1633 	      S_ISDIR(old_path.dentry->d_inode->i_mode))
1634 		goto out1;
1635 	/*
1636 	 * Don't move a mount residing in a shared parent.
1637 	 */
1638 	if (old_path.mnt->mnt_parent &&
1639 	    IS_MNT_SHARED(old_path.mnt->mnt_parent))
1640 		goto out1;
1641 	/*
1642 	 * Don't move a mount tree containing unbindable mounts to a destination
1643 	 * mount which is shared.
1644 	 */
1645 	if (IS_MNT_SHARED(path->mnt) &&
1646 	    tree_contains_unbindable(old_path.mnt))
1647 		goto out1;
1648 	err = -ELOOP;
1649 	for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1650 		if (p == old_path.mnt)
1651 			goto out1;
1652 
1653 	err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1654 	if (err)
1655 		goto out1;
1656 
1657 	/* if the mount is moved, it should no longer be expire
1658 	 * automatically */
1659 	list_del_init(&old_path.mnt->mnt_expire);
1660 out1:
1661 	mutex_unlock(&path->dentry->d_inode->i_mutex);
1662 out:
1663 	up_write(&namespace_sem);
1664 	if (!err)
1665 		path_put(&parent_path);
1666 	path_put(&old_path);
1667 	return err;
1668 }
1669 
1670 /*
1671  * create a new mount for userspace and request it to be added into the
1672  * namespace's tree
1673  */
1674 static int do_new_mount(struct path *path, char *type, int flags,
1675 			int mnt_flags, char *name, void *data)
1676 {
1677 	struct vfsmount *mnt;
1678 
1679 	if (!type)
1680 		return -EINVAL;
1681 
1682 	/* we need capabilities... */
1683 	if (!capable(CAP_SYS_ADMIN))
1684 		return -EPERM;
1685 
1686 	lock_kernel();
1687 	mnt = do_kern_mount(type, flags, name, data);
1688 	unlock_kernel();
1689 	if (IS_ERR(mnt))
1690 		return PTR_ERR(mnt);
1691 
1692 	return do_add_mount(mnt, path, mnt_flags, NULL);
1693 }
1694 
1695 /*
1696  * add a mount into a namespace's mount tree
1697  * - provide the option of adding the new mount to an expiration list
1698  */
1699 int do_add_mount(struct vfsmount *newmnt, struct path *path,
1700 		 int mnt_flags, struct list_head *fslist)
1701 {
1702 	int err;
1703 
1704 	mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1705 
1706 	down_write(&namespace_sem);
1707 	/* Something was mounted here while we slept */
1708 	while (d_mountpoint(path->dentry) &&
1709 	       follow_down(path))
1710 		;
1711 	err = -EINVAL;
1712 	if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1713 		goto unlock;
1714 
1715 	/* Refuse the same filesystem on the same mount point */
1716 	err = -EBUSY;
1717 	if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1718 	    path->mnt->mnt_root == path->dentry)
1719 		goto unlock;
1720 
1721 	err = -EINVAL;
1722 	if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1723 		goto unlock;
1724 
1725 	newmnt->mnt_flags = mnt_flags;
1726 	if ((err = graft_tree(newmnt, path)))
1727 		goto unlock;
1728 
1729 	if (fslist) /* add to the specified expiration list */
1730 		list_add_tail(&newmnt->mnt_expire, fslist);
1731 
1732 	up_write(&namespace_sem);
1733 	return 0;
1734 
1735 unlock:
1736 	up_write(&namespace_sem);
1737 	mntput(newmnt);
1738 	return err;
1739 }
1740 
1741 EXPORT_SYMBOL_GPL(do_add_mount);
1742 
1743 /*
1744  * process a list of expirable mountpoints with the intent of discarding any
1745  * mountpoints that aren't in use and haven't been touched since last we came
1746  * here
1747  */
1748 void mark_mounts_for_expiry(struct list_head *mounts)
1749 {
1750 	struct vfsmount *mnt, *next;
1751 	LIST_HEAD(graveyard);
1752 	LIST_HEAD(umounts);
1753 
1754 	if (list_empty(mounts))
1755 		return;
1756 
1757 	down_write(&namespace_sem);
1758 	spin_lock(&vfsmount_lock);
1759 
1760 	/* extract from the expiration list every vfsmount that matches the
1761 	 * following criteria:
1762 	 * - only referenced by its parent vfsmount
1763 	 * - still marked for expiry (marked on the last call here; marks are
1764 	 *   cleared by mntput())
1765 	 */
1766 	list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1767 		if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1768 			propagate_mount_busy(mnt, 1))
1769 			continue;
1770 		list_move(&mnt->mnt_expire, &graveyard);
1771 	}
1772 	while (!list_empty(&graveyard)) {
1773 		mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1774 		touch_mnt_namespace(mnt->mnt_ns);
1775 		umount_tree(mnt, 1, &umounts);
1776 	}
1777 	spin_unlock(&vfsmount_lock);
1778 	up_write(&namespace_sem);
1779 
1780 	release_mounts(&umounts);
1781 }
1782 
1783 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1784 
1785 /*
1786  * Ripoff of 'select_parent()'
1787  *
1788  * search the list of submounts for a given mountpoint, and move any
1789  * shrinkable submounts to the 'graveyard' list.
1790  */
1791 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1792 {
1793 	struct vfsmount *this_parent = parent;
1794 	struct list_head *next;
1795 	int found = 0;
1796 
1797 repeat:
1798 	next = this_parent->mnt_mounts.next;
1799 resume:
1800 	while (next != &this_parent->mnt_mounts) {
1801 		struct list_head *tmp = next;
1802 		struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1803 
1804 		next = tmp->next;
1805 		if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1806 			continue;
1807 		/*
1808 		 * Descend a level if the d_mounts list is non-empty.
1809 		 */
1810 		if (!list_empty(&mnt->mnt_mounts)) {
1811 			this_parent = mnt;
1812 			goto repeat;
1813 		}
1814 
1815 		if (!propagate_mount_busy(mnt, 1)) {
1816 			list_move_tail(&mnt->mnt_expire, graveyard);
1817 			found++;
1818 		}
1819 	}
1820 	/*
1821 	 * All done at this level ... ascend and resume the search
1822 	 */
1823 	if (this_parent != parent) {
1824 		next = this_parent->mnt_child.next;
1825 		this_parent = this_parent->mnt_parent;
1826 		goto resume;
1827 	}
1828 	return found;
1829 }
1830 
1831 /*
1832  * process a list of expirable mountpoints with the intent of discarding any
1833  * submounts of a specific parent mountpoint
1834  */
1835 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1836 {
1837 	LIST_HEAD(graveyard);
1838 	struct vfsmount *m;
1839 
1840 	/* extract submounts of 'mountpoint' from the expiration list */
1841 	while (select_submounts(mnt, &graveyard)) {
1842 		while (!list_empty(&graveyard)) {
1843 			m = list_first_entry(&graveyard, struct vfsmount,
1844 						mnt_expire);
1845 			touch_mnt_namespace(m->mnt_ns);
1846 			umount_tree(m, 1, umounts);
1847 		}
1848 	}
1849 }
1850 
1851 /*
1852  * Some copy_from_user() implementations do not return the exact number of
1853  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
1854  * Note that this function differs from copy_from_user() in that it will oops
1855  * on bad values of `to', rather than returning a short copy.
1856  */
1857 static long exact_copy_from_user(void *to, const void __user * from,
1858 				 unsigned long n)
1859 {
1860 	char *t = to;
1861 	const char __user *f = from;
1862 	char c;
1863 
1864 	if (!access_ok(VERIFY_READ, from, n))
1865 		return n;
1866 
1867 	while (n) {
1868 		if (__get_user(c, f)) {
1869 			memset(t, 0, n);
1870 			break;
1871 		}
1872 		*t++ = c;
1873 		f++;
1874 		n--;
1875 	}
1876 	return n;
1877 }
1878 
1879 int copy_mount_options(const void __user * data, unsigned long *where)
1880 {
1881 	int i;
1882 	unsigned long page;
1883 	unsigned long size;
1884 
1885 	*where = 0;
1886 	if (!data)
1887 		return 0;
1888 
1889 	if (!(page = __get_free_page(GFP_KERNEL)))
1890 		return -ENOMEM;
1891 
1892 	/* We only care that *some* data at the address the user
1893 	 * gave us is valid.  Just in case, we'll zero
1894 	 * the remainder of the page.
1895 	 */
1896 	/* copy_from_user cannot cross TASK_SIZE ! */
1897 	size = TASK_SIZE - (unsigned long)data;
1898 	if (size > PAGE_SIZE)
1899 		size = PAGE_SIZE;
1900 
1901 	i = size - exact_copy_from_user((void *)page, data, size);
1902 	if (!i) {
1903 		free_page(page);
1904 		return -EFAULT;
1905 	}
1906 	if (i != PAGE_SIZE)
1907 		memset((char *)page + i, 0, PAGE_SIZE - i);
1908 	*where = page;
1909 	return 0;
1910 }
1911 
1912 int copy_mount_string(const void __user *data, char **where)
1913 {
1914 	char *tmp;
1915 
1916 	if (!data) {
1917 		*where = NULL;
1918 		return 0;
1919 	}
1920 
1921 	tmp = strndup_user(data, PAGE_SIZE);
1922 	if (IS_ERR(tmp))
1923 		return PTR_ERR(tmp);
1924 
1925 	*where = tmp;
1926 	return 0;
1927 }
1928 
1929 /*
1930  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1931  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1932  *
1933  * data is a (void *) that can point to any structure up to
1934  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1935  * information (or be NULL).
1936  *
1937  * Pre-0.97 versions of mount() didn't have a flags word.
1938  * When the flags word was introduced its top half was required
1939  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1940  * Therefore, if this magic number is present, it carries no information
1941  * and must be discarded.
1942  */
1943 long do_mount(char *dev_name, char *dir_name, char *type_page,
1944 		  unsigned long flags, void *data_page)
1945 {
1946 	struct path path;
1947 	int retval = 0;
1948 	int mnt_flags = 0;
1949 
1950 	/* Discard magic */
1951 	if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1952 		flags &= ~MS_MGC_MSK;
1953 
1954 	/* Basic sanity checks */
1955 
1956 	if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1957 		return -EINVAL;
1958 
1959 	if (data_page)
1960 		((char *)data_page)[PAGE_SIZE - 1] = 0;
1961 
1962 	/* ... and get the mountpoint */
1963 	retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1964 	if (retval)
1965 		return retval;
1966 
1967 	retval = security_sb_mount(dev_name, &path,
1968 				   type_page, flags, data_page);
1969 	if (retval)
1970 		goto dput_out;
1971 
1972 	/* Default to relatime unless overriden */
1973 	if (!(flags & MS_NOATIME))
1974 		mnt_flags |= MNT_RELATIME;
1975 
1976 	/* Separate the per-mountpoint flags */
1977 	if (flags & MS_NOSUID)
1978 		mnt_flags |= MNT_NOSUID;
1979 	if (flags & MS_NODEV)
1980 		mnt_flags |= MNT_NODEV;
1981 	if (flags & MS_NOEXEC)
1982 		mnt_flags |= MNT_NOEXEC;
1983 	if (flags & MS_NOATIME)
1984 		mnt_flags |= MNT_NOATIME;
1985 	if (flags & MS_NODIRATIME)
1986 		mnt_flags |= MNT_NODIRATIME;
1987 	if (flags & MS_STRICTATIME)
1988 		mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
1989 	if (flags & MS_RDONLY)
1990 		mnt_flags |= MNT_READONLY;
1991 
1992 	flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
1993 		   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1994 		   MS_STRICTATIME);
1995 
1996 	if (flags & MS_REMOUNT)
1997 		retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1998 				    data_page);
1999 	else if (flags & MS_BIND)
2000 		retval = do_loopback(&path, dev_name, flags & MS_REC);
2001 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2002 		retval = do_change_type(&path, flags);
2003 	else if (flags & MS_MOVE)
2004 		retval = do_move_mount(&path, dev_name);
2005 	else
2006 		retval = do_new_mount(&path, type_page, flags, mnt_flags,
2007 				      dev_name, data_page);
2008 dput_out:
2009 	path_put(&path);
2010 	return retval;
2011 }
2012 
2013 static struct mnt_namespace *alloc_mnt_ns(void)
2014 {
2015 	struct mnt_namespace *new_ns;
2016 
2017 	new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2018 	if (!new_ns)
2019 		return ERR_PTR(-ENOMEM);
2020 	atomic_set(&new_ns->count, 1);
2021 	new_ns->root = NULL;
2022 	INIT_LIST_HEAD(&new_ns->list);
2023 	init_waitqueue_head(&new_ns->poll);
2024 	new_ns->event = 0;
2025 	return new_ns;
2026 }
2027 
2028 /*
2029  * Allocate a new namespace structure and populate it with contents
2030  * copied from the namespace of the passed in task structure.
2031  */
2032 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2033 		struct fs_struct *fs)
2034 {
2035 	struct mnt_namespace *new_ns;
2036 	struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2037 	struct vfsmount *p, *q;
2038 
2039 	new_ns = alloc_mnt_ns();
2040 	if (IS_ERR(new_ns))
2041 		return new_ns;
2042 
2043 	down_write(&namespace_sem);
2044 	/* First pass: copy the tree topology */
2045 	new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2046 					CL_COPY_ALL | CL_EXPIRE);
2047 	if (!new_ns->root) {
2048 		up_write(&namespace_sem);
2049 		kfree(new_ns);
2050 		return ERR_PTR(-ENOMEM);
2051 	}
2052 	spin_lock(&vfsmount_lock);
2053 	list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2054 	spin_unlock(&vfsmount_lock);
2055 
2056 	/*
2057 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2058 	 * as belonging to new namespace.  We have already acquired a private
2059 	 * fs_struct, so tsk->fs->lock is not needed.
2060 	 */
2061 	p = mnt_ns->root;
2062 	q = new_ns->root;
2063 	while (p) {
2064 		q->mnt_ns = new_ns;
2065 		if (fs) {
2066 			if (p == fs->root.mnt) {
2067 				rootmnt = p;
2068 				fs->root.mnt = mntget(q);
2069 			}
2070 			if (p == fs->pwd.mnt) {
2071 				pwdmnt = p;
2072 				fs->pwd.mnt = mntget(q);
2073 			}
2074 		}
2075 		p = next_mnt(p, mnt_ns->root);
2076 		q = next_mnt(q, new_ns->root);
2077 	}
2078 	up_write(&namespace_sem);
2079 
2080 	if (rootmnt)
2081 		mntput(rootmnt);
2082 	if (pwdmnt)
2083 		mntput(pwdmnt);
2084 
2085 	return new_ns;
2086 }
2087 
2088 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2089 		struct fs_struct *new_fs)
2090 {
2091 	struct mnt_namespace *new_ns;
2092 
2093 	BUG_ON(!ns);
2094 	get_mnt_ns(ns);
2095 
2096 	if (!(flags & CLONE_NEWNS))
2097 		return ns;
2098 
2099 	new_ns = dup_mnt_ns(ns, new_fs);
2100 
2101 	put_mnt_ns(ns);
2102 	return new_ns;
2103 }
2104 
2105 /**
2106  * create_mnt_ns - creates a private namespace and adds a root filesystem
2107  * @mnt: pointer to the new root filesystem mountpoint
2108  */
2109 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2110 {
2111 	struct mnt_namespace *new_ns;
2112 
2113 	new_ns = alloc_mnt_ns();
2114 	if (!IS_ERR(new_ns)) {
2115 		mnt->mnt_ns = new_ns;
2116 		new_ns->root = mnt;
2117 		list_add(&new_ns->list, &new_ns->root->mnt_list);
2118 	}
2119 	return new_ns;
2120 }
2121 EXPORT_SYMBOL(create_mnt_ns);
2122 
2123 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2124 		char __user *, type, unsigned long, flags, void __user *, data)
2125 {
2126 	int ret;
2127 	char *kernel_type;
2128 	char *kernel_dir;
2129 	char *kernel_dev;
2130 	unsigned long data_page;
2131 
2132 	ret = copy_mount_string(type, &kernel_type);
2133 	if (ret < 0)
2134 		goto out_type;
2135 
2136 	kernel_dir = getname(dir_name);
2137 	if (IS_ERR(kernel_dir)) {
2138 		ret = PTR_ERR(kernel_dir);
2139 		goto out_dir;
2140 	}
2141 
2142 	ret = copy_mount_string(dev_name, &kernel_dev);
2143 	if (ret < 0)
2144 		goto out_dev;
2145 
2146 	ret = copy_mount_options(data, &data_page);
2147 	if (ret < 0)
2148 		goto out_data;
2149 
2150 	ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2151 		(void *) data_page);
2152 
2153 	free_page(data_page);
2154 out_data:
2155 	kfree(kernel_dev);
2156 out_dev:
2157 	putname(kernel_dir);
2158 out_dir:
2159 	kfree(kernel_type);
2160 out_type:
2161 	return ret;
2162 }
2163 
2164 /*
2165  * pivot_root Semantics:
2166  * Moves the root file system of the current process to the directory put_old,
2167  * makes new_root as the new root file system of the current process, and sets
2168  * root/cwd of all processes which had them on the current root to new_root.
2169  *
2170  * Restrictions:
2171  * The new_root and put_old must be directories, and  must not be on the
2172  * same file  system as the current process root. The put_old  must  be
2173  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2174  * pointed to by put_old must yield the same directory as new_root. No other
2175  * file system may be mounted on put_old. After all, new_root is a mountpoint.
2176  *
2177  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2178  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2179  * in this situation.
2180  *
2181  * Notes:
2182  *  - we don't move root/cwd if they are not at the root (reason: if something
2183  *    cared enough to change them, it's probably wrong to force them elsewhere)
2184  *  - it's okay to pick a root that isn't the root of a file system, e.g.
2185  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2186  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2187  *    first.
2188  */
2189 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2190 		const char __user *, put_old)
2191 {
2192 	struct vfsmount *tmp;
2193 	struct path new, old, parent_path, root_parent, root;
2194 	int error;
2195 
2196 	if (!capable(CAP_SYS_ADMIN))
2197 		return -EPERM;
2198 
2199 	error = user_path_dir(new_root, &new);
2200 	if (error)
2201 		goto out0;
2202 	error = -EINVAL;
2203 	if (!check_mnt(new.mnt))
2204 		goto out1;
2205 
2206 	error = user_path_dir(put_old, &old);
2207 	if (error)
2208 		goto out1;
2209 
2210 	error = security_sb_pivotroot(&old, &new);
2211 	if (error) {
2212 		path_put(&old);
2213 		goto out1;
2214 	}
2215 
2216 	read_lock(&current->fs->lock);
2217 	root = current->fs->root;
2218 	path_get(&current->fs->root);
2219 	read_unlock(&current->fs->lock);
2220 	down_write(&namespace_sem);
2221 	mutex_lock(&old.dentry->d_inode->i_mutex);
2222 	error = -EINVAL;
2223 	if (IS_MNT_SHARED(old.mnt) ||
2224 		IS_MNT_SHARED(new.mnt->mnt_parent) ||
2225 		IS_MNT_SHARED(root.mnt->mnt_parent))
2226 		goto out2;
2227 	if (!check_mnt(root.mnt))
2228 		goto out2;
2229 	error = -ENOENT;
2230 	if (cant_mount(old.dentry))
2231 		goto out2;
2232 	if (d_unlinked(new.dentry))
2233 		goto out2;
2234 	if (d_unlinked(old.dentry))
2235 		goto out2;
2236 	error = -EBUSY;
2237 	if (new.mnt == root.mnt ||
2238 	    old.mnt == root.mnt)
2239 		goto out2; /* loop, on the same file system  */
2240 	error = -EINVAL;
2241 	if (root.mnt->mnt_root != root.dentry)
2242 		goto out2; /* not a mountpoint */
2243 	if (root.mnt->mnt_parent == root.mnt)
2244 		goto out2; /* not attached */
2245 	if (new.mnt->mnt_root != new.dentry)
2246 		goto out2; /* not a mountpoint */
2247 	if (new.mnt->mnt_parent == new.mnt)
2248 		goto out2; /* not attached */
2249 	/* make sure we can reach put_old from new_root */
2250 	tmp = old.mnt;
2251 	spin_lock(&vfsmount_lock);
2252 	if (tmp != new.mnt) {
2253 		for (;;) {
2254 			if (tmp->mnt_parent == tmp)
2255 				goto out3; /* already mounted on put_old */
2256 			if (tmp->mnt_parent == new.mnt)
2257 				break;
2258 			tmp = tmp->mnt_parent;
2259 		}
2260 		if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2261 			goto out3;
2262 	} else if (!is_subdir(old.dentry, new.dentry))
2263 		goto out3;
2264 	detach_mnt(new.mnt, &parent_path);
2265 	detach_mnt(root.mnt, &root_parent);
2266 	/* mount old root on put_old */
2267 	attach_mnt(root.mnt, &old);
2268 	/* mount new_root on / */
2269 	attach_mnt(new.mnt, &root_parent);
2270 	touch_mnt_namespace(current->nsproxy->mnt_ns);
2271 	spin_unlock(&vfsmount_lock);
2272 	chroot_fs_refs(&root, &new);
2273 	error = 0;
2274 	path_put(&root_parent);
2275 	path_put(&parent_path);
2276 out2:
2277 	mutex_unlock(&old.dentry->d_inode->i_mutex);
2278 	up_write(&namespace_sem);
2279 	path_put(&root);
2280 	path_put(&old);
2281 out1:
2282 	path_put(&new);
2283 out0:
2284 	return error;
2285 out3:
2286 	spin_unlock(&vfsmount_lock);
2287 	goto out2;
2288 }
2289 
2290 static void __init init_mount_tree(void)
2291 {
2292 	struct vfsmount *mnt;
2293 	struct mnt_namespace *ns;
2294 	struct path root;
2295 
2296 	mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2297 	if (IS_ERR(mnt))
2298 		panic("Can't create rootfs");
2299 	ns = create_mnt_ns(mnt);
2300 	if (IS_ERR(ns))
2301 		panic("Can't allocate initial namespace");
2302 
2303 	init_task.nsproxy->mnt_ns = ns;
2304 	get_mnt_ns(ns);
2305 
2306 	root.mnt = ns->root;
2307 	root.dentry = ns->root->mnt_root;
2308 
2309 	set_fs_pwd(current->fs, &root);
2310 	set_fs_root(current->fs, &root);
2311 }
2312 
2313 void __init mnt_init(void)
2314 {
2315 	unsigned u;
2316 	int err;
2317 
2318 	init_rwsem(&namespace_sem);
2319 
2320 	mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2321 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2322 
2323 	mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2324 
2325 	if (!mount_hashtable)
2326 		panic("Failed to allocate mount hash table\n");
2327 
2328 	printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2329 
2330 	for (u = 0; u < HASH_SIZE; u++)
2331 		INIT_LIST_HEAD(&mount_hashtable[u]);
2332 
2333 	err = sysfs_init();
2334 	if (err)
2335 		printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2336 			__func__, err);
2337 	fs_kobj = kobject_create_and_add("fs", NULL);
2338 	if (!fs_kobj)
2339 		printk(KERN_WARNING "%s: kobj create error\n", __func__);
2340 	init_rootfs();
2341 	init_mount_tree();
2342 }
2343 
2344 void put_mnt_ns(struct mnt_namespace *ns)
2345 {
2346 	LIST_HEAD(umount_list);
2347 
2348 	if (!atomic_dec_and_test(&ns->count))
2349 		return;
2350 	down_write(&namespace_sem);
2351 	spin_lock(&vfsmount_lock);
2352 	umount_tree(ns->root, 0, &umount_list);
2353 	spin_unlock(&vfsmount_lock);
2354 	up_write(&namespace_sem);
2355 	release_mounts(&umount_list);
2356 	kfree(ns);
2357 }
2358 EXPORT_SYMBOL(put_mnt_ns);
2359