xref: /openbmc/linux/fs/namespace.c (revision c21b37f6)
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/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.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/security.h>
26 #include <linux/mount.h>
27 #include <linux/ramfs.h>
28 #include <asm/uaccess.h>
29 #include <asm/unistd.h>
30 #include "pnode.h"
31 #include "internal.h"
32 
33 /* spinlock for vfsmount related operations, inplace of dcache_lock */
34 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
35 
36 static int event;
37 
38 static struct list_head *mount_hashtable __read_mostly;
39 static int hash_mask __read_mostly, hash_bits __read_mostly;
40 static struct kmem_cache *mnt_cache __read_mostly;
41 static struct rw_semaphore namespace_sem;
42 
43 /* /sys/fs */
44 decl_subsys(fs, NULL, NULL);
45 EXPORT_SYMBOL_GPL(fs_subsys);
46 
47 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
48 {
49 	unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
50 	tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
51 	tmp = tmp + (tmp >> hash_bits);
52 	return tmp & hash_mask;
53 }
54 
55 struct vfsmount *alloc_vfsmnt(const char *name)
56 {
57 	struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
58 	if (mnt) {
59 		atomic_set(&mnt->mnt_count, 1);
60 		INIT_LIST_HEAD(&mnt->mnt_hash);
61 		INIT_LIST_HEAD(&mnt->mnt_child);
62 		INIT_LIST_HEAD(&mnt->mnt_mounts);
63 		INIT_LIST_HEAD(&mnt->mnt_list);
64 		INIT_LIST_HEAD(&mnt->mnt_expire);
65 		INIT_LIST_HEAD(&mnt->mnt_share);
66 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
67 		INIT_LIST_HEAD(&mnt->mnt_slave);
68 		if (name) {
69 			int size = strlen(name) + 1;
70 			char *newname = kmalloc(size, GFP_KERNEL);
71 			if (newname) {
72 				memcpy(newname, name, size);
73 				mnt->mnt_devname = newname;
74 			}
75 		}
76 	}
77 	return mnt;
78 }
79 
80 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
81 {
82 	mnt->mnt_sb = sb;
83 	mnt->mnt_root = dget(sb->s_root);
84 	return 0;
85 }
86 
87 EXPORT_SYMBOL(simple_set_mnt);
88 
89 void free_vfsmnt(struct vfsmount *mnt)
90 {
91 	kfree(mnt->mnt_devname);
92 	kmem_cache_free(mnt_cache, mnt);
93 }
94 
95 /*
96  * find the first or last mount at @dentry on vfsmount @mnt depending on
97  * @dir. If @dir is set return the first mount else return the last mount.
98  */
99 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
100 			      int dir)
101 {
102 	struct list_head *head = mount_hashtable + hash(mnt, dentry);
103 	struct list_head *tmp = head;
104 	struct vfsmount *p, *found = NULL;
105 
106 	for (;;) {
107 		tmp = dir ? tmp->next : tmp->prev;
108 		p = NULL;
109 		if (tmp == head)
110 			break;
111 		p = list_entry(tmp, struct vfsmount, mnt_hash);
112 		if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
113 			found = p;
114 			break;
115 		}
116 	}
117 	return found;
118 }
119 
120 /*
121  * lookup_mnt increments the ref count before returning
122  * the vfsmount struct.
123  */
124 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
125 {
126 	struct vfsmount *child_mnt;
127 	spin_lock(&vfsmount_lock);
128 	if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
129 		mntget(child_mnt);
130 	spin_unlock(&vfsmount_lock);
131 	return child_mnt;
132 }
133 
134 static inline int check_mnt(struct vfsmount *mnt)
135 {
136 	return mnt->mnt_ns == current->nsproxy->mnt_ns;
137 }
138 
139 static void touch_mnt_namespace(struct mnt_namespace *ns)
140 {
141 	if (ns) {
142 		ns->event = ++event;
143 		wake_up_interruptible(&ns->poll);
144 	}
145 }
146 
147 static void __touch_mnt_namespace(struct mnt_namespace *ns)
148 {
149 	if (ns && ns->event != event) {
150 		ns->event = event;
151 		wake_up_interruptible(&ns->poll);
152 	}
153 }
154 
155 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
156 {
157 	old_nd->dentry = mnt->mnt_mountpoint;
158 	old_nd->mnt = mnt->mnt_parent;
159 	mnt->mnt_parent = mnt;
160 	mnt->mnt_mountpoint = mnt->mnt_root;
161 	list_del_init(&mnt->mnt_child);
162 	list_del_init(&mnt->mnt_hash);
163 	old_nd->dentry->d_mounted--;
164 }
165 
166 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
167 			struct vfsmount *child_mnt)
168 {
169 	child_mnt->mnt_parent = mntget(mnt);
170 	child_mnt->mnt_mountpoint = dget(dentry);
171 	dentry->d_mounted++;
172 }
173 
174 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
175 {
176 	mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
177 	list_add_tail(&mnt->mnt_hash, mount_hashtable +
178 			hash(nd->mnt, nd->dentry));
179 	list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
180 }
181 
182 /*
183  * the caller must hold vfsmount_lock
184  */
185 static void commit_tree(struct vfsmount *mnt)
186 {
187 	struct vfsmount *parent = mnt->mnt_parent;
188 	struct vfsmount *m;
189 	LIST_HEAD(head);
190 	struct mnt_namespace *n = parent->mnt_ns;
191 
192 	BUG_ON(parent == mnt);
193 
194 	list_add_tail(&head, &mnt->mnt_list);
195 	list_for_each_entry(m, &head, mnt_list)
196 		m->mnt_ns = n;
197 	list_splice(&head, n->list.prev);
198 
199 	list_add_tail(&mnt->mnt_hash, mount_hashtable +
200 				hash(parent, mnt->mnt_mountpoint));
201 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
202 	touch_mnt_namespace(n);
203 }
204 
205 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
206 {
207 	struct list_head *next = p->mnt_mounts.next;
208 	if (next == &p->mnt_mounts) {
209 		while (1) {
210 			if (p == root)
211 				return NULL;
212 			next = p->mnt_child.next;
213 			if (next != &p->mnt_parent->mnt_mounts)
214 				break;
215 			p = p->mnt_parent;
216 		}
217 	}
218 	return list_entry(next, struct vfsmount, mnt_child);
219 }
220 
221 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
222 {
223 	struct list_head *prev = p->mnt_mounts.prev;
224 	while (prev != &p->mnt_mounts) {
225 		p = list_entry(prev, struct vfsmount, mnt_child);
226 		prev = p->mnt_mounts.prev;
227 	}
228 	return p;
229 }
230 
231 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
232 					int flag)
233 {
234 	struct super_block *sb = old->mnt_sb;
235 	struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
236 
237 	if (mnt) {
238 		mnt->mnt_flags = old->mnt_flags;
239 		atomic_inc(&sb->s_active);
240 		mnt->mnt_sb = sb;
241 		mnt->mnt_root = dget(root);
242 		mnt->mnt_mountpoint = mnt->mnt_root;
243 		mnt->mnt_parent = mnt;
244 
245 		if (flag & CL_SLAVE) {
246 			list_add(&mnt->mnt_slave, &old->mnt_slave_list);
247 			mnt->mnt_master = old;
248 			CLEAR_MNT_SHARED(mnt);
249 		} else {
250 			if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
251 				list_add(&mnt->mnt_share, &old->mnt_share);
252 			if (IS_MNT_SLAVE(old))
253 				list_add(&mnt->mnt_slave, &old->mnt_slave);
254 			mnt->mnt_master = old->mnt_master;
255 		}
256 		if (flag & CL_MAKE_SHARED)
257 			set_mnt_shared(mnt);
258 
259 		/* stick the duplicate mount on the same expiry list
260 		 * as the original if that was on one */
261 		if (flag & CL_EXPIRE) {
262 			spin_lock(&vfsmount_lock);
263 			if (!list_empty(&old->mnt_expire))
264 				list_add(&mnt->mnt_expire, &old->mnt_expire);
265 			spin_unlock(&vfsmount_lock);
266 		}
267 	}
268 	return mnt;
269 }
270 
271 static inline void __mntput(struct vfsmount *mnt)
272 {
273 	struct super_block *sb = mnt->mnt_sb;
274 	dput(mnt->mnt_root);
275 	free_vfsmnt(mnt);
276 	deactivate_super(sb);
277 }
278 
279 void mntput_no_expire(struct vfsmount *mnt)
280 {
281 repeat:
282 	if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
283 		if (likely(!mnt->mnt_pinned)) {
284 			spin_unlock(&vfsmount_lock);
285 			__mntput(mnt);
286 			return;
287 		}
288 		atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
289 		mnt->mnt_pinned = 0;
290 		spin_unlock(&vfsmount_lock);
291 		acct_auto_close_mnt(mnt);
292 		security_sb_umount_close(mnt);
293 		goto repeat;
294 	}
295 }
296 
297 EXPORT_SYMBOL(mntput_no_expire);
298 
299 void mnt_pin(struct vfsmount *mnt)
300 {
301 	spin_lock(&vfsmount_lock);
302 	mnt->mnt_pinned++;
303 	spin_unlock(&vfsmount_lock);
304 }
305 
306 EXPORT_SYMBOL(mnt_pin);
307 
308 void mnt_unpin(struct vfsmount *mnt)
309 {
310 	spin_lock(&vfsmount_lock);
311 	if (mnt->mnt_pinned) {
312 		atomic_inc(&mnt->mnt_count);
313 		mnt->mnt_pinned--;
314 	}
315 	spin_unlock(&vfsmount_lock);
316 }
317 
318 EXPORT_SYMBOL(mnt_unpin);
319 
320 /* iterator */
321 static void *m_start(struct seq_file *m, loff_t *pos)
322 {
323 	struct mnt_namespace *n = m->private;
324 
325 	down_read(&namespace_sem);
326 	return seq_list_start(&n->list, *pos);
327 }
328 
329 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
330 {
331 	struct mnt_namespace *n = m->private;
332 
333 	return seq_list_next(v, &n->list, pos);
334 }
335 
336 static void m_stop(struct seq_file *m, void *v)
337 {
338 	up_read(&namespace_sem);
339 }
340 
341 static inline void mangle(struct seq_file *m, const char *s)
342 {
343 	seq_escape(m, s, " \t\n\\");
344 }
345 
346 static int show_vfsmnt(struct seq_file *m, void *v)
347 {
348 	struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
349 	int err = 0;
350 	static struct proc_fs_info {
351 		int flag;
352 		char *str;
353 	} fs_info[] = {
354 		{ MS_SYNCHRONOUS, ",sync" },
355 		{ MS_DIRSYNC, ",dirsync" },
356 		{ MS_MANDLOCK, ",mand" },
357 		{ 0, NULL }
358 	};
359 	static struct proc_fs_info mnt_info[] = {
360 		{ MNT_NOSUID, ",nosuid" },
361 		{ MNT_NODEV, ",nodev" },
362 		{ MNT_NOEXEC, ",noexec" },
363 		{ MNT_NOATIME, ",noatime" },
364 		{ MNT_NODIRATIME, ",nodiratime" },
365 		{ MNT_RELATIME, ",relatime" },
366 		{ 0, NULL }
367 	};
368 	struct proc_fs_info *fs_infop;
369 
370 	mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
371 	seq_putc(m, ' ');
372 	seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
373 	seq_putc(m, ' ');
374 	mangle(m, mnt->mnt_sb->s_type->name);
375 	if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
376 		seq_putc(m, '.');
377 		mangle(m, mnt->mnt_sb->s_subtype);
378 	}
379 	seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
380 	for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
381 		if (mnt->mnt_sb->s_flags & fs_infop->flag)
382 			seq_puts(m, fs_infop->str);
383 	}
384 	for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
385 		if (mnt->mnt_flags & fs_infop->flag)
386 			seq_puts(m, fs_infop->str);
387 	}
388 	if (mnt->mnt_sb->s_op->show_options)
389 		err = mnt->mnt_sb->s_op->show_options(m, mnt);
390 	seq_puts(m, " 0 0\n");
391 	return err;
392 }
393 
394 struct seq_operations mounts_op = {
395 	.start	= m_start,
396 	.next	= m_next,
397 	.stop	= m_stop,
398 	.show	= show_vfsmnt
399 };
400 
401 static int show_vfsstat(struct seq_file *m, void *v)
402 {
403 	struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
404 	int err = 0;
405 
406 	/* device */
407 	if (mnt->mnt_devname) {
408 		seq_puts(m, "device ");
409 		mangle(m, mnt->mnt_devname);
410 	} else
411 		seq_puts(m, "no device");
412 
413 	/* mount point */
414 	seq_puts(m, " mounted on ");
415 	seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
416 	seq_putc(m, ' ');
417 
418 	/* file system type */
419 	seq_puts(m, "with fstype ");
420 	mangle(m, mnt->mnt_sb->s_type->name);
421 
422 	/* optional statistics */
423 	if (mnt->mnt_sb->s_op->show_stats) {
424 		seq_putc(m, ' ');
425 		err = mnt->mnt_sb->s_op->show_stats(m, mnt);
426 	}
427 
428 	seq_putc(m, '\n');
429 	return err;
430 }
431 
432 struct seq_operations mountstats_op = {
433 	.start	= m_start,
434 	.next	= m_next,
435 	.stop	= m_stop,
436 	.show	= show_vfsstat,
437 };
438 
439 /**
440  * may_umount_tree - check if a mount tree is busy
441  * @mnt: root of mount tree
442  *
443  * This is called to check if a tree of mounts has any
444  * open files, pwds, chroots or sub mounts that are
445  * busy.
446  */
447 int may_umount_tree(struct vfsmount *mnt)
448 {
449 	int actual_refs = 0;
450 	int minimum_refs = 0;
451 	struct vfsmount *p;
452 
453 	spin_lock(&vfsmount_lock);
454 	for (p = mnt; p; p = next_mnt(p, mnt)) {
455 		actual_refs += atomic_read(&p->mnt_count);
456 		minimum_refs += 2;
457 	}
458 	spin_unlock(&vfsmount_lock);
459 
460 	if (actual_refs > minimum_refs)
461 		return 0;
462 
463 	return 1;
464 }
465 
466 EXPORT_SYMBOL(may_umount_tree);
467 
468 /**
469  * may_umount - check if a mount point is busy
470  * @mnt: root of mount
471  *
472  * This is called to check if a mount point has any
473  * open files, pwds, chroots or sub mounts. If the
474  * mount has sub mounts this will return busy
475  * regardless of whether the sub mounts are busy.
476  *
477  * Doesn't take quota and stuff into account. IOW, in some cases it will
478  * give false negatives. The main reason why it's here is that we need
479  * a non-destructive way to look for easily umountable filesystems.
480  */
481 int may_umount(struct vfsmount *mnt)
482 {
483 	int ret = 1;
484 	spin_lock(&vfsmount_lock);
485 	if (propagate_mount_busy(mnt, 2))
486 		ret = 0;
487 	spin_unlock(&vfsmount_lock);
488 	return ret;
489 }
490 
491 EXPORT_SYMBOL(may_umount);
492 
493 void release_mounts(struct list_head *head)
494 {
495 	struct vfsmount *mnt;
496 	while (!list_empty(head)) {
497 		mnt = list_first_entry(head, struct vfsmount, mnt_hash);
498 		list_del_init(&mnt->mnt_hash);
499 		if (mnt->mnt_parent != mnt) {
500 			struct dentry *dentry;
501 			struct vfsmount *m;
502 			spin_lock(&vfsmount_lock);
503 			dentry = mnt->mnt_mountpoint;
504 			m = mnt->mnt_parent;
505 			mnt->mnt_mountpoint = mnt->mnt_root;
506 			mnt->mnt_parent = mnt;
507 			spin_unlock(&vfsmount_lock);
508 			dput(dentry);
509 			mntput(m);
510 		}
511 		mntput(mnt);
512 	}
513 }
514 
515 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
516 {
517 	struct vfsmount *p;
518 
519 	for (p = mnt; p; p = next_mnt(p, mnt))
520 		list_move(&p->mnt_hash, kill);
521 
522 	if (propagate)
523 		propagate_umount(kill);
524 
525 	list_for_each_entry(p, kill, mnt_hash) {
526 		list_del_init(&p->mnt_expire);
527 		list_del_init(&p->mnt_list);
528 		__touch_mnt_namespace(p->mnt_ns);
529 		p->mnt_ns = NULL;
530 		list_del_init(&p->mnt_child);
531 		if (p->mnt_parent != p)
532 			p->mnt_mountpoint->d_mounted--;
533 		change_mnt_propagation(p, MS_PRIVATE);
534 	}
535 }
536 
537 static int do_umount(struct vfsmount *mnt, int flags)
538 {
539 	struct super_block *sb = mnt->mnt_sb;
540 	int retval;
541 	LIST_HEAD(umount_list);
542 
543 	retval = security_sb_umount(mnt, flags);
544 	if (retval)
545 		return retval;
546 
547 	/*
548 	 * Allow userspace to request a mountpoint be expired rather than
549 	 * unmounting unconditionally. Unmount only happens if:
550 	 *  (1) the mark is already set (the mark is cleared by mntput())
551 	 *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
552 	 */
553 	if (flags & MNT_EXPIRE) {
554 		if (mnt == current->fs->rootmnt ||
555 		    flags & (MNT_FORCE | MNT_DETACH))
556 			return -EINVAL;
557 
558 		if (atomic_read(&mnt->mnt_count) != 2)
559 			return -EBUSY;
560 
561 		if (!xchg(&mnt->mnt_expiry_mark, 1))
562 			return -EAGAIN;
563 	}
564 
565 	/*
566 	 * If we may have to abort operations to get out of this
567 	 * mount, and they will themselves hold resources we must
568 	 * allow the fs to do things. In the Unix tradition of
569 	 * 'Gee thats tricky lets do it in userspace' the umount_begin
570 	 * might fail to complete on the first run through as other tasks
571 	 * must return, and the like. Thats for the mount program to worry
572 	 * about for the moment.
573 	 */
574 
575 	lock_kernel();
576 	if (sb->s_op->umount_begin)
577 		sb->s_op->umount_begin(mnt, flags);
578 	unlock_kernel();
579 
580 	/*
581 	 * No sense to grab the lock for this test, but test itself looks
582 	 * somewhat bogus. Suggestions for better replacement?
583 	 * Ho-hum... In principle, we might treat that as umount + switch
584 	 * to rootfs. GC would eventually take care of the old vfsmount.
585 	 * Actually it makes sense, especially if rootfs would contain a
586 	 * /reboot - static binary that would close all descriptors and
587 	 * call reboot(9). Then init(8) could umount root and exec /reboot.
588 	 */
589 	if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
590 		/*
591 		 * Special case for "unmounting" root ...
592 		 * we just try to remount it readonly.
593 		 */
594 		down_write(&sb->s_umount);
595 		if (!(sb->s_flags & MS_RDONLY)) {
596 			lock_kernel();
597 			DQUOT_OFF(sb);
598 			retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
599 			unlock_kernel();
600 		}
601 		up_write(&sb->s_umount);
602 		return retval;
603 	}
604 
605 	down_write(&namespace_sem);
606 	spin_lock(&vfsmount_lock);
607 	event++;
608 
609 	retval = -EBUSY;
610 	if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
611 		if (!list_empty(&mnt->mnt_list))
612 			umount_tree(mnt, 1, &umount_list);
613 		retval = 0;
614 	}
615 	spin_unlock(&vfsmount_lock);
616 	if (retval)
617 		security_sb_umount_busy(mnt);
618 	up_write(&namespace_sem);
619 	release_mounts(&umount_list);
620 	return retval;
621 }
622 
623 /*
624  * Now umount can handle mount points as well as block devices.
625  * This is important for filesystems which use unnamed block devices.
626  *
627  * We now support a flag for forced unmount like the other 'big iron'
628  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
629  */
630 
631 asmlinkage long sys_umount(char __user * name, int flags)
632 {
633 	struct nameidata nd;
634 	int retval;
635 
636 	retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
637 	if (retval)
638 		goto out;
639 	retval = -EINVAL;
640 	if (nd.dentry != nd.mnt->mnt_root)
641 		goto dput_and_out;
642 	if (!check_mnt(nd.mnt))
643 		goto dput_and_out;
644 
645 	retval = -EPERM;
646 	if (!capable(CAP_SYS_ADMIN))
647 		goto dput_and_out;
648 
649 	retval = do_umount(nd.mnt, flags);
650 dput_and_out:
651 	path_release_on_umount(&nd);
652 out:
653 	return retval;
654 }
655 
656 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
657 
658 /*
659  *	The 2.0 compatible umount. No flags.
660  */
661 asmlinkage long sys_oldumount(char __user * name)
662 {
663 	return sys_umount(name, 0);
664 }
665 
666 #endif
667 
668 static int mount_is_safe(struct nameidata *nd)
669 {
670 	if (capable(CAP_SYS_ADMIN))
671 		return 0;
672 	return -EPERM;
673 #ifdef notyet
674 	if (S_ISLNK(nd->dentry->d_inode->i_mode))
675 		return -EPERM;
676 	if (nd->dentry->d_inode->i_mode & S_ISVTX) {
677 		if (current->uid != nd->dentry->d_inode->i_uid)
678 			return -EPERM;
679 	}
680 	if (vfs_permission(nd, MAY_WRITE))
681 		return -EPERM;
682 	return 0;
683 #endif
684 }
685 
686 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
687 {
688 	while (1) {
689 		if (d == dentry)
690 			return 1;
691 		if (d == NULL || d == d->d_parent)
692 			return 0;
693 		d = d->d_parent;
694 	}
695 }
696 
697 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
698 					int flag)
699 {
700 	struct vfsmount *res, *p, *q, *r, *s;
701 	struct nameidata nd;
702 
703 	if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
704 		return NULL;
705 
706 	res = q = clone_mnt(mnt, dentry, flag);
707 	if (!q)
708 		goto Enomem;
709 	q->mnt_mountpoint = mnt->mnt_mountpoint;
710 
711 	p = mnt;
712 	list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
713 		if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
714 			continue;
715 
716 		for (s = r; s; s = next_mnt(s, r)) {
717 			if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
718 				s = skip_mnt_tree(s);
719 				continue;
720 			}
721 			while (p != s->mnt_parent) {
722 				p = p->mnt_parent;
723 				q = q->mnt_parent;
724 			}
725 			p = s;
726 			nd.mnt = q;
727 			nd.dentry = p->mnt_mountpoint;
728 			q = clone_mnt(p, p->mnt_root, flag);
729 			if (!q)
730 				goto Enomem;
731 			spin_lock(&vfsmount_lock);
732 			list_add_tail(&q->mnt_list, &res->mnt_list);
733 			attach_mnt(q, &nd);
734 			spin_unlock(&vfsmount_lock);
735 		}
736 	}
737 	return res;
738 Enomem:
739 	if (res) {
740 		LIST_HEAD(umount_list);
741 		spin_lock(&vfsmount_lock);
742 		umount_tree(res, 0, &umount_list);
743 		spin_unlock(&vfsmount_lock);
744 		release_mounts(&umount_list);
745 	}
746 	return NULL;
747 }
748 
749 /*
750  *  @source_mnt : mount tree to be attached
751  *  @nd         : place the mount tree @source_mnt is attached
752  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
753  *  		   store the parent mount and mountpoint dentry.
754  *  		   (done when source_mnt is moved)
755  *
756  *  NOTE: in the table below explains the semantics when a source mount
757  *  of a given type is attached to a destination mount of a given type.
758  * ---------------------------------------------------------------------------
759  * |         BIND MOUNT OPERATION                                            |
760  * |**************************************************************************
761  * | source-->| shared        |       private  |       slave    | unbindable |
762  * | dest     |               |                |                |            |
763  * |   |      |               |                |                |            |
764  * |   v      |               |                |                |            |
765  * |**************************************************************************
766  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
767  * |          |               |                |                |            |
768  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
769  * ***************************************************************************
770  * A bind operation clones the source mount and mounts the clone on the
771  * destination mount.
772  *
773  * (++)  the cloned mount is propagated to all the mounts in the propagation
774  * 	 tree of the destination mount and the cloned mount is added to
775  * 	 the peer group of the source mount.
776  * (+)   the cloned mount is created under the destination mount and is marked
777  *       as shared. The cloned mount is added to the peer group of the source
778  *       mount.
779  * (+++) the mount is propagated to all the mounts in the propagation tree
780  *       of the destination mount and the cloned mount is made slave
781  *       of the same master as that of the source mount. The cloned mount
782  *       is marked as 'shared and slave'.
783  * (*)   the cloned mount is made a slave of the same master as that of the
784  * 	 source mount.
785  *
786  * ---------------------------------------------------------------------------
787  * |         		MOVE MOUNT OPERATION                                 |
788  * |**************************************************************************
789  * | source-->| shared        |       private  |       slave    | unbindable |
790  * | dest     |               |                |                |            |
791  * |   |      |               |                |                |            |
792  * |   v      |               |                |                |            |
793  * |**************************************************************************
794  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
795  * |          |               |                |                |            |
796  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
797  * ***************************************************************************
798  *
799  * (+)  the mount is moved to the destination. And is then propagated to
800  * 	all the mounts in the propagation tree of the destination mount.
801  * (+*)  the mount is moved to the destination.
802  * (+++)  the mount is moved to the destination and is then propagated to
803  * 	all the mounts belonging to the destination mount's propagation tree.
804  * 	the mount is marked as 'shared and slave'.
805  * (*)	the mount continues to be a slave at the new location.
806  *
807  * if the source mount is a tree, the operations explained above is
808  * applied to each mount in the tree.
809  * Must be called without spinlocks held, since this function can sleep
810  * in allocations.
811  */
812 static int attach_recursive_mnt(struct vfsmount *source_mnt,
813 			struct nameidata *nd, struct nameidata *parent_nd)
814 {
815 	LIST_HEAD(tree_list);
816 	struct vfsmount *dest_mnt = nd->mnt;
817 	struct dentry *dest_dentry = nd->dentry;
818 	struct vfsmount *child, *p;
819 
820 	if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
821 		return -EINVAL;
822 
823 	if (IS_MNT_SHARED(dest_mnt)) {
824 		for (p = source_mnt; p; p = next_mnt(p, source_mnt))
825 			set_mnt_shared(p);
826 	}
827 
828 	spin_lock(&vfsmount_lock);
829 	if (parent_nd) {
830 		detach_mnt(source_mnt, parent_nd);
831 		attach_mnt(source_mnt, nd);
832 		touch_mnt_namespace(current->nsproxy->mnt_ns);
833 	} else {
834 		mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
835 		commit_tree(source_mnt);
836 	}
837 
838 	list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
839 		list_del_init(&child->mnt_hash);
840 		commit_tree(child);
841 	}
842 	spin_unlock(&vfsmount_lock);
843 	return 0;
844 }
845 
846 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
847 {
848 	int err;
849 	if (mnt->mnt_sb->s_flags & MS_NOUSER)
850 		return -EINVAL;
851 
852 	if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
853 	      S_ISDIR(mnt->mnt_root->d_inode->i_mode))
854 		return -ENOTDIR;
855 
856 	err = -ENOENT;
857 	mutex_lock(&nd->dentry->d_inode->i_mutex);
858 	if (IS_DEADDIR(nd->dentry->d_inode))
859 		goto out_unlock;
860 
861 	err = security_sb_check_sb(mnt, nd);
862 	if (err)
863 		goto out_unlock;
864 
865 	err = -ENOENT;
866 	if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
867 		err = attach_recursive_mnt(mnt, nd, NULL);
868 out_unlock:
869 	mutex_unlock(&nd->dentry->d_inode->i_mutex);
870 	if (!err)
871 		security_sb_post_addmount(mnt, nd);
872 	return err;
873 }
874 
875 /*
876  * recursively change the type of the mountpoint.
877  */
878 static int do_change_type(struct nameidata *nd, int flag)
879 {
880 	struct vfsmount *m, *mnt = nd->mnt;
881 	int recurse = flag & MS_REC;
882 	int type = flag & ~MS_REC;
883 
884 	if (!capable(CAP_SYS_ADMIN))
885 		return -EPERM;
886 
887 	if (nd->dentry != nd->mnt->mnt_root)
888 		return -EINVAL;
889 
890 	down_write(&namespace_sem);
891 	spin_lock(&vfsmount_lock);
892 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
893 		change_mnt_propagation(m, type);
894 	spin_unlock(&vfsmount_lock);
895 	up_write(&namespace_sem);
896 	return 0;
897 }
898 
899 /*
900  * do loopback mount.
901  */
902 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
903 {
904 	struct nameidata old_nd;
905 	struct vfsmount *mnt = NULL;
906 	int err = mount_is_safe(nd);
907 	if (err)
908 		return err;
909 	if (!old_name || !*old_name)
910 		return -EINVAL;
911 	err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
912 	if (err)
913 		return err;
914 
915 	down_write(&namespace_sem);
916 	err = -EINVAL;
917 	if (IS_MNT_UNBINDABLE(old_nd.mnt))
918  		goto out;
919 
920 	if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
921 		goto out;
922 
923 	err = -ENOMEM;
924 	if (recurse)
925 		mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
926 	else
927 		mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
928 
929 	if (!mnt)
930 		goto out;
931 
932 	err = graft_tree(mnt, nd);
933 	if (err) {
934 		LIST_HEAD(umount_list);
935 		spin_lock(&vfsmount_lock);
936 		umount_tree(mnt, 0, &umount_list);
937 		spin_unlock(&vfsmount_lock);
938 		release_mounts(&umount_list);
939 	}
940 
941 out:
942 	up_write(&namespace_sem);
943 	path_release(&old_nd);
944 	return err;
945 }
946 
947 /*
948  * change filesystem flags. dir should be a physical root of filesystem.
949  * If you've mounted a non-root directory somewhere and want to do remount
950  * on it - tough luck.
951  */
952 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
953 		      void *data)
954 {
955 	int err;
956 	struct super_block *sb = nd->mnt->mnt_sb;
957 
958 	if (!capable(CAP_SYS_ADMIN))
959 		return -EPERM;
960 
961 	if (!check_mnt(nd->mnt))
962 		return -EINVAL;
963 
964 	if (nd->dentry != nd->mnt->mnt_root)
965 		return -EINVAL;
966 
967 	down_write(&sb->s_umount);
968 	err = do_remount_sb(sb, flags, data, 0);
969 	if (!err)
970 		nd->mnt->mnt_flags = mnt_flags;
971 	up_write(&sb->s_umount);
972 	if (!err)
973 		security_sb_post_remount(nd->mnt, flags, data);
974 	return err;
975 }
976 
977 static inline int tree_contains_unbindable(struct vfsmount *mnt)
978 {
979 	struct vfsmount *p;
980 	for (p = mnt; p; p = next_mnt(p, mnt)) {
981 		if (IS_MNT_UNBINDABLE(p))
982 			return 1;
983 	}
984 	return 0;
985 }
986 
987 static int do_move_mount(struct nameidata *nd, char *old_name)
988 {
989 	struct nameidata old_nd, parent_nd;
990 	struct vfsmount *p;
991 	int err = 0;
992 	if (!capable(CAP_SYS_ADMIN))
993 		return -EPERM;
994 	if (!old_name || !*old_name)
995 		return -EINVAL;
996 	err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
997 	if (err)
998 		return err;
999 
1000 	down_write(&namespace_sem);
1001 	while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1002 		;
1003 	err = -EINVAL;
1004 	if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1005 		goto out;
1006 
1007 	err = -ENOENT;
1008 	mutex_lock(&nd->dentry->d_inode->i_mutex);
1009 	if (IS_DEADDIR(nd->dentry->d_inode))
1010 		goto out1;
1011 
1012 	if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1013 		goto out1;
1014 
1015 	err = -EINVAL;
1016 	if (old_nd.dentry != old_nd.mnt->mnt_root)
1017 		goto out1;
1018 
1019 	if (old_nd.mnt == old_nd.mnt->mnt_parent)
1020 		goto out1;
1021 
1022 	if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1023 	      S_ISDIR(old_nd.dentry->d_inode->i_mode))
1024 		goto out1;
1025 	/*
1026 	 * Don't move a mount residing in a shared parent.
1027 	 */
1028 	if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1029 		goto out1;
1030 	/*
1031 	 * Don't move a mount tree containing unbindable mounts to a destination
1032 	 * mount which is shared.
1033 	 */
1034 	if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1035 		goto out1;
1036 	err = -ELOOP;
1037 	for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1038 		if (p == old_nd.mnt)
1039 			goto out1;
1040 
1041 	if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1042 		goto out1;
1043 
1044 	spin_lock(&vfsmount_lock);
1045 	/* if the mount is moved, it should no longer be expire
1046 	 * automatically */
1047 	list_del_init(&old_nd.mnt->mnt_expire);
1048 	spin_unlock(&vfsmount_lock);
1049 out1:
1050 	mutex_unlock(&nd->dentry->d_inode->i_mutex);
1051 out:
1052 	up_write(&namespace_sem);
1053 	if (!err)
1054 		path_release(&parent_nd);
1055 	path_release(&old_nd);
1056 	return err;
1057 }
1058 
1059 /*
1060  * create a new mount for userspace and request it to be added into the
1061  * namespace's tree
1062  */
1063 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1064 			int mnt_flags, char *name, void *data)
1065 {
1066 	struct vfsmount *mnt;
1067 
1068 	if (!type || !memchr(type, 0, PAGE_SIZE))
1069 		return -EINVAL;
1070 
1071 	/* we need capabilities... */
1072 	if (!capable(CAP_SYS_ADMIN))
1073 		return -EPERM;
1074 
1075 	mnt = do_kern_mount(type, flags, name, data);
1076 	if (IS_ERR(mnt))
1077 		return PTR_ERR(mnt);
1078 
1079 	return do_add_mount(mnt, nd, mnt_flags, NULL);
1080 }
1081 
1082 /*
1083  * add a mount into a namespace's mount tree
1084  * - provide the option of adding the new mount to an expiration list
1085  */
1086 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1087 		 int mnt_flags, struct list_head *fslist)
1088 {
1089 	int err;
1090 
1091 	down_write(&namespace_sem);
1092 	/* Something was mounted here while we slept */
1093 	while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1094 		;
1095 	err = -EINVAL;
1096 	if (!check_mnt(nd->mnt))
1097 		goto unlock;
1098 
1099 	/* Refuse the same filesystem on the same mount point */
1100 	err = -EBUSY;
1101 	if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1102 	    nd->mnt->mnt_root == nd->dentry)
1103 		goto unlock;
1104 
1105 	err = -EINVAL;
1106 	if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1107 		goto unlock;
1108 
1109 	newmnt->mnt_flags = mnt_flags;
1110 	if ((err = graft_tree(newmnt, nd)))
1111 		goto unlock;
1112 
1113 	if (fslist) {
1114 		/* add to the specified expiration list */
1115 		spin_lock(&vfsmount_lock);
1116 		list_add_tail(&newmnt->mnt_expire, fslist);
1117 		spin_unlock(&vfsmount_lock);
1118 	}
1119 	up_write(&namespace_sem);
1120 	return 0;
1121 
1122 unlock:
1123 	up_write(&namespace_sem);
1124 	mntput(newmnt);
1125 	return err;
1126 }
1127 
1128 EXPORT_SYMBOL_GPL(do_add_mount);
1129 
1130 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1131 				struct list_head *umounts)
1132 {
1133 	spin_lock(&vfsmount_lock);
1134 
1135 	/*
1136 	 * Check if mount is still attached, if not, let whoever holds it deal
1137 	 * with the sucker
1138 	 */
1139 	if (mnt->mnt_parent == mnt) {
1140 		spin_unlock(&vfsmount_lock);
1141 		return;
1142 	}
1143 
1144 	/*
1145 	 * Check that it is still dead: the count should now be 2 - as
1146 	 * contributed by the vfsmount parent and the mntget above
1147 	 */
1148 	if (!propagate_mount_busy(mnt, 2)) {
1149 		/* delete from the namespace */
1150 		touch_mnt_namespace(mnt->mnt_ns);
1151 		list_del_init(&mnt->mnt_list);
1152 		mnt->mnt_ns = NULL;
1153 		umount_tree(mnt, 1, umounts);
1154 		spin_unlock(&vfsmount_lock);
1155 	} else {
1156 		/*
1157 		 * Someone brought it back to life whilst we didn't have any
1158 		 * locks held so return it to the expiration list
1159 		 */
1160 		list_add_tail(&mnt->mnt_expire, mounts);
1161 		spin_unlock(&vfsmount_lock);
1162 	}
1163 }
1164 
1165 /*
1166  * go through the vfsmounts we've just consigned to the graveyard to
1167  * - check that they're still dead
1168  * - delete the vfsmount from the appropriate namespace under lock
1169  * - dispose of the corpse
1170  */
1171 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1172 {
1173 	struct mnt_namespace *ns;
1174 	struct vfsmount *mnt;
1175 
1176 	while (!list_empty(graveyard)) {
1177 		LIST_HEAD(umounts);
1178 		mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire);
1179 		list_del_init(&mnt->mnt_expire);
1180 
1181 		/* don't do anything if the namespace is dead - all the
1182 		 * vfsmounts from it are going away anyway */
1183 		ns = mnt->mnt_ns;
1184 		if (!ns || !ns->root)
1185 			continue;
1186 		get_mnt_ns(ns);
1187 
1188 		spin_unlock(&vfsmount_lock);
1189 		down_write(&namespace_sem);
1190 		expire_mount(mnt, mounts, &umounts);
1191 		up_write(&namespace_sem);
1192 		release_mounts(&umounts);
1193 		mntput(mnt);
1194 		put_mnt_ns(ns);
1195 		spin_lock(&vfsmount_lock);
1196 	}
1197 }
1198 
1199 /*
1200  * process a list of expirable mountpoints with the intent of discarding any
1201  * mountpoints that aren't in use and haven't been touched since last we came
1202  * here
1203  */
1204 void mark_mounts_for_expiry(struct list_head *mounts)
1205 {
1206 	struct vfsmount *mnt, *next;
1207 	LIST_HEAD(graveyard);
1208 
1209 	if (list_empty(mounts))
1210 		return;
1211 
1212 	spin_lock(&vfsmount_lock);
1213 
1214 	/* extract from the expiration list every vfsmount that matches the
1215 	 * following criteria:
1216 	 * - only referenced by its parent vfsmount
1217 	 * - still marked for expiry (marked on the last call here; marks are
1218 	 *   cleared by mntput())
1219 	 */
1220 	list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1221 		if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1222 		    atomic_read(&mnt->mnt_count) != 1)
1223 			continue;
1224 
1225 		mntget(mnt);
1226 		list_move(&mnt->mnt_expire, &graveyard);
1227 	}
1228 
1229 	expire_mount_list(&graveyard, mounts);
1230 
1231 	spin_unlock(&vfsmount_lock);
1232 }
1233 
1234 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1235 
1236 /*
1237  * Ripoff of 'select_parent()'
1238  *
1239  * search the list of submounts for a given mountpoint, and move any
1240  * shrinkable submounts to the 'graveyard' list.
1241  */
1242 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1243 {
1244 	struct vfsmount *this_parent = parent;
1245 	struct list_head *next;
1246 	int found = 0;
1247 
1248 repeat:
1249 	next = this_parent->mnt_mounts.next;
1250 resume:
1251 	while (next != &this_parent->mnt_mounts) {
1252 		struct list_head *tmp = next;
1253 		struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1254 
1255 		next = tmp->next;
1256 		if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1257 			continue;
1258 		/*
1259 		 * Descend a level if the d_mounts list is non-empty.
1260 		 */
1261 		if (!list_empty(&mnt->mnt_mounts)) {
1262 			this_parent = mnt;
1263 			goto repeat;
1264 		}
1265 
1266 		if (!propagate_mount_busy(mnt, 1)) {
1267 			mntget(mnt);
1268 			list_move_tail(&mnt->mnt_expire, graveyard);
1269 			found++;
1270 		}
1271 	}
1272 	/*
1273 	 * All done at this level ... ascend and resume the search
1274 	 */
1275 	if (this_parent != parent) {
1276 		next = this_parent->mnt_child.next;
1277 		this_parent = this_parent->mnt_parent;
1278 		goto resume;
1279 	}
1280 	return found;
1281 }
1282 
1283 /*
1284  * process a list of expirable mountpoints with the intent of discarding any
1285  * submounts of a specific parent mountpoint
1286  */
1287 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1288 {
1289 	LIST_HEAD(graveyard);
1290 	int found;
1291 
1292 	spin_lock(&vfsmount_lock);
1293 
1294 	/* extract submounts of 'mountpoint' from the expiration list */
1295 	while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1296 		expire_mount_list(&graveyard, mounts);
1297 
1298 	spin_unlock(&vfsmount_lock);
1299 }
1300 
1301 EXPORT_SYMBOL_GPL(shrink_submounts);
1302 
1303 /*
1304  * Some copy_from_user() implementations do not return the exact number of
1305  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
1306  * Note that this function differs from copy_from_user() in that it will oops
1307  * on bad values of `to', rather than returning a short copy.
1308  */
1309 static long exact_copy_from_user(void *to, const void __user * from,
1310 				 unsigned long n)
1311 {
1312 	char *t = to;
1313 	const char __user *f = from;
1314 	char c;
1315 
1316 	if (!access_ok(VERIFY_READ, from, n))
1317 		return n;
1318 
1319 	while (n) {
1320 		if (__get_user(c, f)) {
1321 			memset(t, 0, n);
1322 			break;
1323 		}
1324 		*t++ = c;
1325 		f++;
1326 		n--;
1327 	}
1328 	return n;
1329 }
1330 
1331 int copy_mount_options(const void __user * data, unsigned long *where)
1332 {
1333 	int i;
1334 	unsigned long page;
1335 	unsigned long size;
1336 
1337 	*where = 0;
1338 	if (!data)
1339 		return 0;
1340 
1341 	if (!(page = __get_free_page(GFP_KERNEL)))
1342 		return -ENOMEM;
1343 
1344 	/* We only care that *some* data at the address the user
1345 	 * gave us is valid.  Just in case, we'll zero
1346 	 * the remainder of the page.
1347 	 */
1348 	/* copy_from_user cannot cross TASK_SIZE ! */
1349 	size = TASK_SIZE - (unsigned long)data;
1350 	if (size > PAGE_SIZE)
1351 		size = PAGE_SIZE;
1352 
1353 	i = size - exact_copy_from_user((void *)page, data, size);
1354 	if (!i) {
1355 		free_page(page);
1356 		return -EFAULT;
1357 	}
1358 	if (i != PAGE_SIZE)
1359 		memset((char *)page + i, 0, PAGE_SIZE - i);
1360 	*where = page;
1361 	return 0;
1362 }
1363 
1364 /*
1365  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1366  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1367  *
1368  * data is a (void *) that can point to any structure up to
1369  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1370  * information (or be NULL).
1371  *
1372  * Pre-0.97 versions of mount() didn't have a flags word.
1373  * When the flags word was introduced its top half was required
1374  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1375  * Therefore, if this magic number is present, it carries no information
1376  * and must be discarded.
1377  */
1378 long do_mount(char *dev_name, char *dir_name, char *type_page,
1379 		  unsigned long flags, void *data_page)
1380 {
1381 	struct nameidata nd;
1382 	int retval = 0;
1383 	int mnt_flags = 0;
1384 
1385 	/* Discard magic */
1386 	if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1387 		flags &= ~MS_MGC_MSK;
1388 
1389 	/* Basic sanity checks */
1390 
1391 	if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1392 		return -EINVAL;
1393 	if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1394 		return -EINVAL;
1395 
1396 	if (data_page)
1397 		((char *)data_page)[PAGE_SIZE - 1] = 0;
1398 
1399 	/* Separate the per-mountpoint flags */
1400 	if (flags & MS_NOSUID)
1401 		mnt_flags |= MNT_NOSUID;
1402 	if (flags & MS_NODEV)
1403 		mnt_flags |= MNT_NODEV;
1404 	if (flags & MS_NOEXEC)
1405 		mnt_flags |= MNT_NOEXEC;
1406 	if (flags & MS_NOATIME)
1407 		mnt_flags |= MNT_NOATIME;
1408 	if (flags & MS_NODIRATIME)
1409 		mnt_flags |= MNT_NODIRATIME;
1410 	if (flags & MS_RELATIME)
1411 		mnt_flags |= MNT_RELATIME;
1412 
1413 	flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1414 		   MS_NOATIME | MS_NODIRATIME | MS_RELATIME);
1415 
1416 	/* ... and get the mountpoint */
1417 	retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1418 	if (retval)
1419 		return retval;
1420 
1421 	retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1422 	if (retval)
1423 		goto dput_out;
1424 
1425 	if (flags & MS_REMOUNT)
1426 		retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1427 				    data_page);
1428 	else if (flags & MS_BIND)
1429 		retval = do_loopback(&nd, dev_name, flags & MS_REC);
1430 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1431 		retval = do_change_type(&nd, flags);
1432 	else if (flags & MS_MOVE)
1433 		retval = do_move_mount(&nd, dev_name);
1434 	else
1435 		retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1436 				      dev_name, data_page);
1437 dput_out:
1438 	path_release(&nd);
1439 	return retval;
1440 }
1441 
1442 /*
1443  * Allocate a new namespace structure and populate it with contents
1444  * copied from the namespace of the passed in task structure.
1445  */
1446 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1447 		struct fs_struct *fs)
1448 {
1449 	struct mnt_namespace *new_ns;
1450 	struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1451 	struct vfsmount *p, *q;
1452 
1453 	new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1454 	if (!new_ns)
1455 		return ERR_PTR(-ENOMEM);
1456 
1457 	atomic_set(&new_ns->count, 1);
1458 	INIT_LIST_HEAD(&new_ns->list);
1459 	init_waitqueue_head(&new_ns->poll);
1460 	new_ns->event = 0;
1461 
1462 	down_write(&namespace_sem);
1463 	/* First pass: copy the tree topology */
1464 	new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1465 					CL_COPY_ALL | CL_EXPIRE);
1466 	if (!new_ns->root) {
1467 		up_write(&namespace_sem);
1468 		kfree(new_ns);
1469 		return ERR_PTR(-ENOMEM);;
1470 	}
1471 	spin_lock(&vfsmount_lock);
1472 	list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1473 	spin_unlock(&vfsmount_lock);
1474 
1475 	/*
1476 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1477 	 * as belonging to new namespace.  We have already acquired a private
1478 	 * fs_struct, so tsk->fs->lock is not needed.
1479 	 */
1480 	p = mnt_ns->root;
1481 	q = new_ns->root;
1482 	while (p) {
1483 		q->mnt_ns = new_ns;
1484 		if (fs) {
1485 			if (p == fs->rootmnt) {
1486 				rootmnt = p;
1487 				fs->rootmnt = mntget(q);
1488 			}
1489 			if (p == fs->pwdmnt) {
1490 				pwdmnt = p;
1491 				fs->pwdmnt = mntget(q);
1492 			}
1493 			if (p == fs->altrootmnt) {
1494 				altrootmnt = p;
1495 				fs->altrootmnt = mntget(q);
1496 			}
1497 		}
1498 		p = next_mnt(p, mnt_ns->root);
1499 		q = next_mnt(q, new_ns->root);
1500 	}
1501 	up_write(&namespace_sem);
1502 
1503 	if (rootmnt)
1504 		mntput(rootmnt);
1505 	if (pwdmnt)
1506 		mntput(pwdmnt);
1507 	if (altrootmnt)
1508 		mntput(altrootmnt);
1509 
1510 	return new_ns;
1511 }
1512 
1513 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
1514 		struct fs_struct *new_fs)
1515 {
1516 	struct mnt_namespace *new_ns;
1517 
1518 	BUG_ON(!ns);
1519 	get_mnt_ns(ns);
1520 
1521 	if (!(flags & CLONE_NEWNS))
1522 		return ns;
1523 
1524 	new_ns = dup_mnt_ns(ns, new_fs);
1525 
1526 	put_mnt_ns(ns);
1527 	return new_ns;
1528 }
1529 
1530 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1531 			  char __user * type, unsigned long flags,
1532 			  void __user * data)
1533 {
1534 	int retval;
1535 	unsigned long data_page;
1536 	unsigned long type_page;
1537 	unsigned long dev_page;
1538 	char *dir_page;
1539 
1540 	retval = copy_mount_options(type, &type_page);
1541 	if (retval < 0)
1542 		return retval;
1543 
1544 	dir_page = getname(dir_name);
1545 	retval = PTR_ERR(dir_page);
1546 	if (IS_ERR(dir_page))
1547 		goto out1;
1548 
1549 	retval = copy_mount_options(dev_name, &dev_page);
1550 	if (retval < 0)
1551 		goto out2;
1552 
1553 	retval = copy_mount_options(data, &data_page);
1554 	if (retval < 0)
1555 		goto out3;
1556 
1557 	lock_kernel();
1558 	retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1559 			  flags, (void *)data_page);
1560 	unlock_kernel();
1561 	free_page(data_page);
1562 
1563 out3:
1564 	free_page(dev_page);
1565 out2:
1566 	putname(dir_page);
1567 out1:
1568 	free_page(type_page);
1569 	return retval;
1570 }
1571 
1572 /*
1573  * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1574  * It can block. Requires the big lock held.
1575  */
1576 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1577 		 struct dentry *dentry)
1578 {
1579 	struct dentry *old_root;
1580 	struct vfsmount *old_rootmnt;
1581 	write_lock(&fs->lock);
1582 	old_root = fs->root;
1583 	old_rootmnt = fs->rootmnt;
1584 	fs->rootmnt = mntget(mnt);
1585 	fs->root = dget(dentry);
1586 	write_unlock(&fs->lock);
1587 	if (old_root) {
1588 		dput(old_root);
1589 		mntput(old_rootmnt);
1590 	}
1591 }
1592 
1593 /*
1594  * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1595  * It can block. Requires the big lock held.
1596  */
1597 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1598 		struct dentry *dentry)
1599 {
1600 	struct dentry *old_pwd;
1601 	struct vfsmount *old_pwdmnt;
1602 
1603 	write_lock(&fs->lock);
1604 	old_pwd = fs->pwd;
1605 	old_pwdmnt = fs->pwdmnt;
1606 	fs->pwdmnt = mntget(mnt);
1607 	fs->pwd = dget(dentry);
1608 	write_unlock(&fs->lock);
1609 
1610 	if (old_pwd) {
1611 		dput(old_pwd);
1612 		mntput(old_pwdmnt);
1613 	}
1614 }
1615 
1616 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1617 {
1618 	struct task_struct *g, *p;
1619 	struct fs_struct *fs;
1620 
1621 	read_lock(&tasklist_lock);
1622 	do_each_thread(g, p) {
1623 		task_lock(p);
1624 		fs = p->fs;
1625 		if (fs) {
1626 			atomic_inc(&fs->count);
1627 			task_unlock(p);
1628 			if (fs->root == old_nd->dentry
1629 			    && fs->rootmnt == old_nd->mnt)
1630 				set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1631 			if (fs->pwd == old_nd->dentry
1632 			    && fs->pwdmnt == old_nd->mnt)
1633 				set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1634 			put_fs_struct(fs);
1635 		} else
1636 			task_unlock(p);
1637 	} while_each_thread(g, p);
1638 	read_unlock(&tasklist_lock);
1639 }
1640 
1641 /*
1642  * pivot_root Semantics:
1643  * Moves the root file system of the current process to the directory put_old,
1644  * makes new_root as the new root file system of the current process, and sets
1645  * root/cwd of all processes which had them on the current root to new_root.
1646  *
1647  * Restrictions:
1648  * The new_root and put_old must be directories, and  must not be on the
1649  * same file  system as the current process root. The put_old  must  be
1650  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
1651  * pointed to by put_old must yield the same directory as new_root. No other
1652  * file system may be mounted on put_old. After all, new_root is a mountpoint.
1653  *
1654  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1655  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1656  * in this situation.
1657  *
1658  * Notes:
1659  *  - we don't move root/cwd if they are not at the root (reason: if something
1660  *    cared enough to change them, it's probably wrong to force them elsewhere)
1661  *  - it's okay to pick a root that isn't the root of a file system, e.g.
1662  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1663  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1664  *    first.
1665  */
1666 asmlinkage long sys_pivot_root(const char __user * new_root,
1667 			       const char __user * put_old)
1668 {
1669 	struct vfsmount *tmp;
1670 	struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1671 	int error;
1672 
1673 	if (!capable(CAP_SYS_ADMIN))
1674 		return -EPERM;
1675 
1676 	lock_kernel();
1677 
1678 	error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1679 			    &new_nd);
1680 	if (error)
1681 		goto out0;
1682 	error = -EINVAL;
1683 	if (!check_mnt(new_nd.mnt))
1684 		goto out1;
1685 
1686 	error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1687 	if (error)
1688 		goto out1;
1689 
1690 	error = security_sb_pivotroot(&old_nd, &new_nd);
1691 	if (error) {
1692 		path_release(&old_nd);
1693 		goto out1;
1694 	}
1695 
1696 	read_lock(&current->fs->lock);
1697 	user_nd.mnt = mntget(current->fs->rootmnt);
1698 	user_nd.dentry = dget(current->fs->root);
1699 	read_unlock(&current->fs->lock);
1700 	down_write(&namespace_sem);
1701 	mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1702 	error = -EINVAL;
1703 	if (IS_MNT_SHARED(old_nd.mnt) ||
1704 		IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1705 		IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1706 		goto out2;
1707 	if (!check_mnt(user_nd.mnt))
1708 		goto out2;
1709 	error = -ENOENT;
1710 	if (IS_DEADDIR(new_nd.dentry->d_inode))
1711 		goto out2;
1712 	if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1713 		goto out2;
1714 	if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1715 		goto out2;
1716 	error = -EBUSY;
1717 	if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1718 		goto out2; /* loop, on the same file system  */
1719 	error = -EINVAL;
1720 	if (user_nd.mnt->mnt_root != user_nd.dentry)
1721 		goto out2; /* not a mountpoint */
1722 	if (user_nd.mnt->mnt_parent == user_nd.mnt)
1723 		goto out2; /* not attached */
1724 	if (new_nd.mnt->mnt_root != new_nd.dentry)
1725 		goto out2; /* not a mountpoint */
1726 	if (new_nd.mnt->mnt_parent == new_nd.mnt)
1727 		goto out2; /* not attached */
1728 	tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1729 	spin_lock(&vfsmount_lock);
1730 	if (tmp != new_nd.mnt) {
1731 		for (;;) {
1732 			if (tmp->mnt_parent == tmp)
1733 				goto out3; /* already mounted on put_old */
1734 			if (tmp->mnt_parent == new_nd.mnt)
1735 				break;
1736 			tmp = tmp->mnt_parent;
1737 		}
1738 		if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1739 			goto out3;
1740 	} else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1741 		goto out3;
1742 	detach_mnt(new_nd.mnt, &parent_nd);
1743 	detach_mnt(user_nd.mnt, &root_parent);
1744 	attach_mnt(user_nd.mnt, &old_nd);     /* mount old root on put_old */
1745 	attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1746 	touch_mnt_namespace(current->nsproxy->mnt_ns);
1747 	spin_unlock(&vfsmount_lock);
1748 	chroot_fs_refs(&user_nd, &new_nd);
1749 	security_sb_post_pivotroot(&user_nd, &new_nd);
1750 	error = 0;
1751 	path_release(&root_parent);
1752 	path_release(&parent_nd);
1753 out2:
1754 	mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1755 	up_write(&namespace_sem);
1756 	path_release(&user_nd);
1757 	path_release(&old_nd);
1758 out1:
1759 	path_release(&new_nd);
1760 out0:
1761 	unlock_kernel();
1762 	return error;
1763 out3:
1764 	spin_unlock(&vfsmount_lock);
1765 	goto out2;
1766 }
1767 
1768 static void __init init_mount_tree(void)
1769 {
1770 	struct vfsmount *mnt;
1771 	struct mnt_namespace *ns;
1772 
1773 	mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1774 	if (IS_ERR(mnt))
1775 		panic("Can't create rootfs");
1776 	ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1777 	if (!ns)
1778 		panic("Can't allocate initial namespace");
1779 	atomic_set(&ns->count, 1);
1780 	INIT_LIST_HEAD(&ns->list);
1781 	init_waitqueue_head(&ns->poll);
1782 	ns->event = 0;
1783 	list_add(&mnt->mnt_list, &ns->list);
1784 	ns->root = mnt;
1785 	mnt->mnt_ns = ns;
1786 
1787 	init_task.nsproxy->mnt_ns = ns;
1788 	get_mnt_ns(ns);
1789 
1790 	set_fs_pwd(current->fs, ns->root, ns->root->mnt_root);
1791 	set_fs_root(current->fs, ns->root, ns->root->mnt_root);
1792 }
1793 
1794 void __init mnt_init(unsigned long mempages)
1795 {
1796 	struct list_head *d;
1797 	unsigned int nr_hash;
1798 	int i;
1799 	int err;
1800 
1801 	init_rwsem(&namespace_sem);
1802 
1803 	mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1804 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1805 
1806 	mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1807 
1808 	if (!mount_hashtable)
1809 		panic("Failed to allocate mount hash table\n");
1810 
1811 	/*
1812 	 * Find the power-of-two list-heads that can fit into the allocation..
1813 	 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1814 	 * a power-of-two.
1815 	 */
1816 	nr_hash = PAGE_SIZE / sizeof(struct list_head);
1817 	hash_bits = 0;
1818 	do {
1819 		hash_bits++;
1820 	} while ((nr_hash >> hash_bits) != 0);
1821 	hash_bits--;
1822 
1823 	/*
1824 	 * Re-calculate the actual number of entries and the mask
1825 	 * from the number of bits we can fit.
1826 	 */
1827 	nr_hash = 1UL << hash_bits;
1828 	hash_mask = nr_hash - 1;
1829 
1830 	printk("Mount-cache hash table entries: %d\n", nr_hash);
1831 
1832 	/* And initialize the newly allocated array */
1833 	d = mount_hashtable;
1834 	i = nr_hash;
1835 	do {
1836 		INIT_LIST_HEAD(d);
1837 		d++;
1838 		i--;
1839 	} while (i);
1840 	err = sysfs_init();
1841 	if (err)
1842 		printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1843 			__FUNCTION__, err);
1844 	err = subsystem_register(&fs_subsys);
1845 	if (err)
1846 		printk(KERN_WARNING "%s: subsystem_register error: %d\n",
1847 			__FUNCTION__, err);
1848 	init_rootfs();
1849 	init_mount_tree();
1850 }
1851 
1852 void __put_mnt_ns(struct mnt_namespace *ns)
1853 {
1854 	struct vfsmount *root = ns->root;
1855 	LIST_HEAD(umount_list);
1856 	ns->root = NULL;
1857 	spin_unlock(&vfsmount_lock);
1858 	down_write(&namespace_sem);
1859 	spin_lock(&vfsmount_lock);
1860 	umount_tree(root, 0, &umount_list);
1861 	spin_unlock(&vfsmount_lock);
1862 	up_write(&namespace_sem);
1863 	release_mounts(&umount_list);
1864 	kfree(ns);
1865 }
1866