xref: /linux/fs/namespace.c (revision 14b42963f64b98ab61fa9723c03d71aa5ef4f862)
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/quotaops.h>
17 #include <linux/acct.h>
18 #include <linux/capability.h>
19 #include <linux/module.h>
20 #include <linux/seq_file.h>
21 #include <linux/namespace.h>
22 #include <linux/namei.h>
23 #include <linux/security.h>
24 #include <linux/mount.h>
25 #include <asm/uaccess.h>
26 #include <asm/unistd.h>
27 #include "pnode.h"
28 
29 extern int __init init_rootfs(void);
30 
31 #ifdef CONFIG_SYSFS
32 extern int __init sysfs_init(void);
33 #else
34 static inline int sysfs_init(void)
35 {
36 	return 0;
37 }
38 #endif
39 
40 /* spinlock for vfsmount related operations, inplace of dcache_lock */
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
42 
43 static int event;
44 
45 static struct list_head *mount_hashtable __read_mostly;
46 static int hash_mask __read_mostly, hash_bits __read_mostly;
47 static kmem_cache_t *mnt_cache __read_mostly;
48 static struct rw_semaphore namespace_sem;
49 
50 /* /sys/fs */
51 decl_subsys(fs, NULL, NULL);
52 EXPORT_SYMBOL_GPL(fs_subsys);
53 
54 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
55 {
56 	unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
57 	tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
58 	tmp = tmp + (tmp >> hash_bits);
59 	return tmp & hash_mask;
60 }
61 
62 struct vfsmount *alloc_vfsmnt(const char *name)
63 {
64 	struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
65 	if (mnt) {
66 		memset(mnt, 0, sizeof(struct vfsmount));
67 		atomic_set(&mnt->mnt_count, 1);
68 		INIT_LIST_HEAD(&mnt->mnt_hash);
69 		INIT_LIST_HEAD(&mnt->mnt_child);
70 		INIT_LIST_HEAD(&mnt->mnt_mounts);
71 		INIT_LIST_HEAD(&mnt->mnt_list);
72 		INIT_LIST_HEAD(&mnt->mnt_expire);
73 		INIT_LIST_HEAD(&mnt->mnt_share);
74 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
75 		INIT_LIST_HEAD(&mnt->mnt_slave);
76 		if (name) {
77 			int size = strlen(name) + 1;
78 			char *newname = kmalloc(size, GFP_KERNEL);
79 			if (newname) {
80 				memcpy(newname, name, size);
81 				mnt->mnt_devname = newname;
82 			}
83 		}
84 	}
85 	return mnt;
86 }
87 
88 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
89 {
90 	mnt->mnt_sb = sb;
91 	mnt->mnt_root = dget(sb->s_root);
92 	return 0;
93 }
94 
95 EXPORT_SYMBOL(simple_set_mnt);
96 
97 void free_vfsmnt(struct vfsmount *mnt)
98 {
99 	kfree(mnt->mnt_devname);
100 	kmem_cache_free(mnt_cache, mnt);
101 }
102 
103 /*
104  * find the first or last mount at @dentry on vfsmount @mnt depending on
105  * @dir. If @dir is set return the first mount else return the last mount.
106  */
107 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
108 			      int dir)
109 {
110 	struct list_head *head = mount_hashtable + hash(mnt, dentry);
111 	struct list_head *tmp = head;
112 	struct vfsmount *p, *found = NULL;
113 
114 	for (;;) {
115 		tmp = dir ? tmp->next : tmp->prev;
116 		p = NULL;
117 		if (tmp == head)
118 			break;
119 		p = list_entry(tmp, struct vfsmount, mnt_hash);
120 		if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
121 			found = p;
122 			break;
123 		}
124 	}
125 	return found;
126 }
127 
128 /*
129  * lookup_mnt increments the ref count before returning
130  * the vfsmount struct.
131  */
132 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
133 {
134 	struct vfsmount *child_mnt;
135 	spin_lock(&vfsmount_lock);
136 	if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
137 		mntget(child_mnt);
138 	spin_unlock(&vfsmount_lock);
139 	return child_mnt;
140 }
141 
142 static inline int check_mnt(struct vfsmount *mnt)
143 {
144 	return mnt->mnt_namespace == current->namespace;
145 }
146 
147 static void touch_namespace(struct namespace *ns)
148 {
149 	if (ns) {
150 		ns->event = ++event;
151 		wake_up_interruptible(&ns->poll);
152 	}
153 }
154 
155 static void __touch_namespace(struct namespace *ns)
156 {
157 	if (ns && ns->event != event) {
158 		ns->event = event;
159 		wake_up_interruptible(&ns->poll);
160 	}
161 }
162 
163 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
164 {
165 	old_nd->dentry = mnt->mnt_mountpoint;
166 	old_nd->mnt = mnt->mnt_parent;
167 	mnt->mnt_parent = mnt;
168 	mnt->mnt_mountpoint = mnt->mnt_root;
169 	list_del_init(&mnt->mnt_child);
170 	list_del_init(&mnt->mnt_hash);
171 	old_nd->dentry->d_mounted--;
172 }
173 
174 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
175 			struct vfsmount *child_mnt)
176 {
177 	child_mnt->mnt_parent = mntget(mnt);
178 	child_mnt->mnt_mountpoint = dget(dentry);
179 	dentry->d_mounted++;
180 }
181 
182 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
183 {
184 	mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
185 	list_add_tail(&mnt->mnt_hash, mount_hashtable +
186 			hash(nd->mnt, nd->dentry));
187 	list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
188 }
189 
190 /*
191  * the caller must hold vfsmount_lock
192  */
193 static void commit_tree(struct vfsmount *mnt)
194 {
195 	struct vfsmount *parent = mnt->mnt_parent;
196 	struct vfsmount *m;
197 	LIST_HEAD(head);
198 	struct namespace *n = parent->mnt_namespace;
199 
200 	BUG_ON(parent == mnt);
201 
202 	list_add_tail(&head, &mnt->mnt_list);
203 	list_for_each_entry(m, &head, mnt_list)
204 		m->mnt_namespace = n;
205 	list_splice(&head, n->list.prev);
206 
207 	list_add_tail(&mnt->mnt_hash, mount_hashtable +
208 				hash(parent, mnt->mnt_mountpoint));
209 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
210 	touch_namespace(n);
211 }
212 
213 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
214 {
215 	struct list_head *next = p->mnt_mounts.next;
216 	if (next == &p->mnt_mounts) {
217 		while (1) {
218 			if (p == root)
219 				return NULL;
220 			next = p->mnt_child.next;
221 			if (next != &p->mnt_parent->mnt_mounts)
222 				break;
223 			p = p->mnt_parent;
224 		}
225 	}
226 	return list_entry(next, struct vfsmount, mnt_child);
227 }
228 
229 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
230 {
231 	struct list_head *prev = p->mnt_mounts.prev;
232 	while (prev != &p->mnt_mounts) {
233 		p = list_entry(prev, struct vfsmount, mnt_child);
234 		prev = p->mnt_mounts.prev;
235 	}
236 	return p;
237 }
238 
239 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
240 					int flag)
241 {
242 	struct super_block *sb = old->mnt_sb;
243 	struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
244 
245 	if (mnt) {
246 		mnt->mnt_flags = old->mnt_flags;
247 		atomic_inc(&sb->s_active);
248 		mnt->mnt_sb = sb;
249 		mnt->mnt_root = dget(root);
250 		mnt->mnt_mountpoint = mnt->mnt_root;
251 		mnt->mnt_parent = mnt;
252 
253 		if (flag & CL_SLAVE) {
254 			list_add(&mnt->mnt_slave, &old->mnt_slave_list);
255 			mnt->mnt_master = old;
256 			CLEAR_MNT_SHARED(mnt);
257 		} else {
258 			if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
259 				list_add(&mnt->mnt_share, &old->mnt_share);
260 			if (IS_MNT_SLAVE(old))
261 				list_add(&mnt->mnt_slave, &old->mnt_slave);
262 			mnt->mnt_master = old->mnt_master;
263 		}
264 		if (flag & CL_MAKE_SHARED)
265 			set_mnt_shared(mnt);
266 
267 		/* stick the duplicate mount on the same expiry list
268 		 * as the original if that was on one */
269 		if (flag & CL_EXPIRE) {
270 			spin_lock(&vfsmount_lock);
271 			if (!list_empty(&old->mnt_expire))
272 				list_add(&mnt->mnt_expire, &old->mnt_expire);
273 			spin_unlock(&vfsmount_lock);
274 		}
275 	}
276 	return mnt;
277 }
278 
279 static inline void __mntput(struct vfsmount *mnt)
280 {
281 	struct super_block *sb = mnt->mnt_sb;
282 	dput(mnt->mnt_root);
283 	free_vfsmnt(mnt);
284 	deactivate_super(sb);
285 }
286 
287 void mntput_no_expire(struct vfsmount *mnt)
288 {
289 repeat:
290 	if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
291 		if (likely(!mnt->mnt_pinned)) {
292 			spin_unlock(&vfsmount_lock);
293 			__mntput(mnt);
294 			return;
295 		}
296 		atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
297 		mnt->mnt_pinned = 0;
298 		spin_unlock(&vfsmount_lock);
299 		acct_auto_close_mnt(mnt);
300 		security_sb_umount_close(mnt);
301 		goto repeat;
302 	}
303 }
304 
305 EXPORT_SYMBOL(mntput_no_expire);
306 
307 void mnt_pin(struct vfsmount *mnt)
308 {
309 	spin_lock(&vfsmount_lock);
310 	mnt->mnt_pinned++;
311 	spin_unlock(&vfsmount_lock);
312 }
313 
314 EXPORT_SYMBOL(mnt_pin);
315 
316 void mnt_unpin(struct vfsmount *mnt)
317 {
318 	spin_lock(&vfsmount_lock);
319 	if (mnt->mnt_pinned) {
320 		atomic_inc(&mnt->mnt_count);
321 		mnt->mnt_pinned--;
322 	}
323 	spin_unlock(&vfsmount_lock);
324 }
325 
326 EXPORT_SYMBOL(mnt_unpin);
327 
328 /* iterator */
329 static void *m_start(struct seq_file *m, loff_t *pos)
330 {
331 	struct namespace *n = m->private;
332 	struct list_head *p;
333 	loff_t l = *pos;
334 
335 	down_read(&namespace_sem);
336 	list_for_each(p, &n->list)
337 		if (!l--)
338 			return list_entry(p, struct vfsmount, mnt_list);
339 	return NULL;
340 }
341 
342 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
343 {
344 	struct namespace *n = m->private;
345 	struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
346 	(*pos)++;
347 	return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
348 }
349 
350 static void m_stop(struct seq_file *m, void *v)
351 {
352 	up_read(&namespace_sem);
353 }
354 
355 static inline void mangle(struct seq_file *m, const char *s)
356 {
357 	seq_escape(m, s, " \t\n\\");
358 }
359 
360 static int show_vfsmnt(struct seq_file *m, void *v)
361 {
362 	struct vfsmount *mnt = v;
363 	int err = 0;
364 	static struct proc_fs_info {
365 		int flag;
366 		char *str;
367 	} fs_info[] = {
368 		{ MS_SYNCHRONOUS, ",sync" },
369 		{ MS_DIRSYNC, ",dirsync" },
370 		{ MS_MANDLOCK, ",mand" },
371 		{ 0, NULL }
372 	};
373 	static struct proc_fs_info mnt_info[] = {
374 		{ MNT_NOSUID, ",nosuid" },
375 		{ MNT_NODEV, ",nodev" },
376 		{ MNT_NOEXEC, ",noexec" },
377 		{ MNT_NOATIME, ",noatime" },
378 		{ MNT_NODIRATIME, ",nodiratime" },
379 		{ 0, NULL }
380 	};
381 	struct proc_fs_info *fs_infop;
382 
383 	mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
384 	seq_putc(m, ' ');
385 	seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
386 	seq_putc(m, ' ');
387 	mangle(m, mnt->mnt_sb->s_type->name);
388 	seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
389 	for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
390 		if (mnt->mnt_sb->s_flags & fs_infop->flag)
391 			seq_puts(m, fs_infop->str);
392 	}
393 	for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
394 		if (mnt->mnt_flags & fs_infop->flag)
395 			seq_puts(m, fs_infop->str);
396 	}
397 	if (mnt->mnt_sb->s_op->show_options)
398 		err = mnt->mnt_sb->s_op->show_options(m, mnt);
399 	seq_puts(m, " 0 0\n");
400 	return err;
401 }
402 
403 struct seq_operations mounts_op = {
404 	.start	= m_start,
405 	.next	= m_next,
406 	.stop	= m_stop,
407 	.show	= show_vfsmnt
408 };
409 
410 static int show_vfsstat(struct seq_file *m, void *v)
411 {
412 	struct vfsmount *mnt = v;
413 	int err = 0;
414 
415 	/* device */
416 	if (mnt->mnt_devname) {
417 		seq_puts(m, "device ");
418 		mangle(m, mnt->mnt_devname);
419 	} else
420 		seq_puts(m, "no device");
421 
422 	/* mount point */
423 	seq_puts(m, " mounted on ");
424 	seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
425 	seq_putc(m, ' ');
426 
427 	/* file system type */
428 	seq_puts(m, "with fstype ");
429 	mangle(m, mnt->mnt_sb->s_type->name);
430 
431 	/* optional statistics */
432 	if (mnt->mnt_sb->s_op->show_stats) {
433 		seq_putc(m, ' ');
434 		err = mnt->mnt_sb->s_op->show_stats(m, mnt);
435 	}
436 
437 	seq_putc(m, '\n');
438 	return err;
439 }
440 
441 struct seq_operations mountstats_op = {
442 	.start	= m_start,
443 	.next	= m_next,
444 	.stop	= m_stop,
445 	.show	= show_vfsstat,
446 };
447 
448 /**
449  * may_umount_tree - check if a mount tree is busy
450  * @mnt: root of mount tree
451  *
452  * This is called to check if a tree of mounts has any
453  * open files, pwds, chroots or sub mounts that are
454  * busy.
455  */
456 int may_umount_tree(struct vfsmount *mnt)
457 {
458 	int actual_refs = 0;
459 	int minimum_refs = 0;
460 	struct vfsmount *p;
461 
462 	spin_lock(&vfsmount_lock);
463 	for (p = mnt; p; p = next_mnt(p, mnt)) {
464 		actual_refs += atomic_read(&p->mnt_count);
465 		minimum_refs += 2;
466 	}
467 	spin_unlock(&vfsmount_lock);
468 
469 	if (actual_refs > minimum_refs)
470 		return 0;
471 
472 	return 1;
473 }
474 
475 EXPORT_SYMBOL(may_umount_tree);
476 
477 /**
478  * may_umount - check if a mount point is busy
479  * @mnt: root of mount
480  *
481  * This is called to check if a mount point has any
482  * open files, pwds, chroots or sub mounts. If the
483  * mount has sub mounts this will return busy
484  * regardless of whether the sub mounts are busy.
485  *
486  * Doesn't take quota and stuff into account. IOW, in some cases it will
487  * give false negatives. The main reason why it's here is that we need
488  * a non-destructive way to look for easily umountable filesystems.
489  */
490 int may_umount(struct vfsmount *mnt)
491 {
492 	int ret = 1;
493 	spin_lock(&vfsmount_lock);
494 	if (propagate_mount_busy(mnt, 2))
495 		ret = 0;
496 	spin_unlock(&vfsmount_lock);
497 	return ret;
498 }
499 
500 EXPORT_SYMBOL(may_umount);
501 
502 void release_mounts(struct list_head *head)
503 {
504 	struct vfsmount *mnt;
505 	while (!list_empty(head)) {
506 		mnt = list_entry(head->next, struct vfsmount, mnt_hash);
507 		list_del_init(&mnt->mnt_hash);
508 		if (mnt->mnt_parent != mnt) {
509 			struct dentry *dentry;
510 			struct vfsmount *m;
511 			spin_lock(&vfsmount_lock);
512 			dentry = mnt->mnt_mountpoint;
513 			m = mnt->mnt_parent;
514 			mnt->mnt_mountpoint = mnt->mnt_root;
515 			mnt->mnt_parent = mnt;
516 			spin_unlock(&vfsmount_lock);
517 			dput(dentry);
518 			mntput(m);
519 		}
520 		mntput(mnt);
521 	}
522 }
523 
524 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
525 {
526 	struct vfsmount *p;
527 
528 	for (p = mnt; p; p = next_mnt(p, mnt))
529 		list_move(&p->mnt_hash, kill);
530 
531 	if (propagate)
532 		propagate_umount(kill);
533 
534 	list_for_each_entry(p, kill, mnt_hash) {
535 		list_del_init(&p->mnt_expire);
536 		list_del_init(&p->mnt_list);
537 		__touch_namespace(p->mnt_namespace);
538 		p->mnt_namespace = NULL;
539 		list_del_init(&p->mnt_child);
540 		if (p->mnt_parent != p)
541 			p->mnt_mountpoint->d_mounted--;
542 		change_mnt_propagation(p, MS_PRIVATE);
543 	}
544 }
545 
546 static int do_umount(struct vfsmount *mnt, int flags)
547 {
548 	struct super_block *sb = mnt->mnt_sb;
549 	int retval;
550 	LIST_HEAD(umount_list);
551 
552 	retval = security_sb_umount(mnt, flags);
553 	if (retval)
554 		return retval;
555 
556 	/*
557 	 * Allow userspace to request a mountpoint be expired rather than
558 	 * unmounting unconditionally. Unmount only happens if:
559 	 *  (1) the mark is already set (the mark is cleared by mntput())
560 	 *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
561 	 */
562 	if (flags & MNT_EXPIRE) {
563 		if (mnt == current->fs->rootmnt ||
564 		    flags & (MNT_FORCE | MNT_DETACH))
565 			return -EINVAL;
566 
567 		if (atomic_read(&mnt->mnt_count) != 2)
568 			return -EBUSY;
569 
570 		if (!xchg(&mnt->mnt_expiry_mark, 1))
571 			return -EAGAIN;
572 	}
573 
574 	/*
575 	 * If we may have to abort operations to get out of this
576 	 * mount, and they will themselves hold resources we must
577 	 * allow the fs to do things. In the Unix tradition of
578 	 * 'Gee thats tricky lets do it in userspace' the umount_begin
579 	 * might fail to complete on the first run through as other tasks
580 	 * must return, and the like. Thats for the mount program to worry
581 	 * about for the moment.
582 	 */
583 
584 	lock_kernel();
585 	if (sb->s_op->umount_begin)
586 		sb->s_op->umount_begin(mnt, flags);
587 	unlock_kernel();
588 
589 	/*
590 	 * No sense to grab the lock for this test, but test itself looks
591 	 * somewhat bogus. Suggestions for better replacement?
592 	 * Ho-hum... In principle, we might treat that as umount + switch
593 	 * to rootfs. GC would eventually take care of the old vfsmount.
594 	 * Actually it makes sense, especially if rootfs would contain a
595 	 * /reboot - static binary that would close all descriptors and
596 	 * call reboot(9). Then init(8) could umount root and exec /reboot.
597 	 */
598 	if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
599 		/*
600 		 * Special case for "unmounting" root ...
601 		 * we just try to remount it readonly.
602 		 */
603 		down_write(&sb->s_umount);
604 		if (!(sb->s_flags & MS_RDONLY)) {
605 			lock_kernel();
606 			DQUOT_OFF(sb);
607 			retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
608 			unlock_kernel();
609 		}
610 		up_write(&sb->s_umount);
611 		return retval;
612 	}
613 
614 	down_write(&namespace_sem);
615 	spin_lock(&vfsmount_lock);
616 	event++;
617 
618 	retval = -EBUSY;
619 	if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
620 		if (!list_empty(&mnt->mnt_list))
621 			umount_tree(mnt, 1, &umount_list);
622 		retval = 0;
623 	}
624 	spin_unlock(&vfsmount_lock);
625 	if (retval)
626 		security_sb_umount_busy(mnt);
627 	up_write(&namespace_sem);
628 	release_mounts(&umount_list);
629 	return retval;
630 }
631 
632 /*
633  * Now umount can handle mount points as well as block devices.
634  * This is important for filesystems which use unnamed block devices.
635  *
636  * We now support a flag for forced unmount like the other 'big iron'
637  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
638  */
639 
640 asmlinkage long sys_umount(char __user * name, int flags)
641 {
642 	struct nameidata nd;
643 	int retval;
644 
645 	retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
646 	if (retval)
647 		goto out;
648 	retval = -EINVAL;
649 	if (nd.dentry != nd.mnt->mnt_root)
650 		goto dput_and_out;
651 	if (!check_mnt(nd.mnt))
652 		goto dput_and_out;
653 
654 	retval = -EPERM;
655 	if (!capable(CAP_SYS_ADMIN))
656 		goto dput_and_out;
657 
658 	retval = do_umount(nd.mnt, flags);
659 dput_and_out:
660 	path_release_on_umount(&nd);
661 out:
662 	return retval;
663 }
664 
665 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
666 
667 /*
668  *	The 2.0 compatible umount. No flags.
669  */
670 asmlinkage long sys_oldumount(char __user * name)
671 {
672 	return sys_umount(name, 0);
673 }
674 
675 #endif
676 
677 static int mount_is_safe(struct nameidata *nd)
678 {
679 	if (capable(CAP_SYS_ADMIN))
680 		return 0;
681 	return -EPERM;
682 #ifdef notyet
683 	if (S_ISLNK(nd->dentry->d_inode->i_mode))
684 		return -EPERM;
685 	if (nd->dentry->d_inode->i_mode & S_ISVTX) {
686 		if (current->uid != nd->dentry->d_inode->i_uid)
687 			return -EPERM;
688 	}
689 	if (vfs_permission(nd, MAY_WRITE))
690 		return -EPERM;
691 	return 0;
692 #endif
693 }
694 
695 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
696 {
697 	while (1) {
698 		if (d == dentry)
699 			return 1;
700 		if (d == NULL || d == d->d_parent)
701 			return 0;
702 		d = d->d_parent;
703 	}
704 }
705 
706 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
707 					int flag)
708 {
709 	struct vfsmount *res, *p, *q, *r, *s;
710 	struct nameidata nd;
711 
712 	if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
713 		return NULL;
714 
715 	res = q = clone_mnt(mnt, dentry, flag);
716 	if (!q)
717 		goto Enomem;
718 	q->mnt_mountpoint = mnt->mnt_mountpoint;
719 
720 	p = mnt;
721 	list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
722 		if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
723 			continue;
724 
725 		for (s = r; s; s = next_mnt(s, r)) {
726 			if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
727 				s = skip_mnt_tree(s);
728 				continue;
729 			}
730 			while (p != s->mnt_parent) {
731 				p = p->mnt_parent;
732 				q = q->mnt_parent;
733 			}
734 			p = s;
735 			nd.mnt = q;
736 			nd.dentry = p->mnt_mountpoint;
737 			q = clone_mnt(p, p->mnt_root, flag);
738 			if (!q)
739 				goto Enomem;
740 			spin_lock(&vfsmount_lock);
741 			list_add_tail(&q->mnt_list, &res->mnt_list);
742 			attach_mnt(q, &nd);
743 			spin_unlock(&vfsmount_lock);
744 		}
745 	}
746 	return res;
747 Enomem:
748 	if (res) {
749 		LIST_HEAD(umount_list);
750 		spin_lock(&vfsmount_lock);
751 		umount_tree(res, 0, &umount_list);
752 		spin_unlock(&vfsmount_lock);
753 		release_mounts(&umount_list);
754 	}
755 	return NULL;
756 }
757 
758 /*
759  *  @source_mnt : mount tree to be attached
760  *  @nd         : place the mount tree @source_mnt is attached
761  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
762  *  		   store the parent mount and mountpoint dentry.
763  *  		   (done when source_mnt is moved)
764  *
765  *  NOTE: in the table below explains the semantics when a source mount
766  *  of a given type is attached to a destination mount of a given type.
767  * ---------------------------------------------------------------------------
768  * |         BIND MOUNT OPERATION                                            |
769  * |**************************************************************************
770  * | source-->| shared        |       private  |       slave    | unbindable |
771  * | dest     |               |                |                |            |
772  * |   |      |               |                |                |            |
773  * |   v      |               |                |                |            |
774  * |**************************************************************************
775  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
776  * |          |               |                |                |            |
777  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
778  * ***************************************************************************
779  * A bind operation clones the source mount and mounts the clone on the
780  * destination mount.
781  *
782  * (++)  the cloned mount is propagated to all the mounts in the propagation
783  * 	 tree of the destination mount and the cloned mount is added to
784  * 	 the peer group of the source mount.
785  * (+)   the cloned mount is created under the destination mount and is marked
786  *       as shared. The cloned mount is added to the peer group of the source
787  *       mount.
788  * (+++) the mount is propagated to all the mounts in the propagation tree
789  *       of the destination mount and the cloned mount is made slave
790  *       of the same master as that of the source mount. The cloned mount
791  *       is marked as 'shared and slave'.
792  * (*)   the cloned mount is made a slave of the same master as that of the
793  * 	 source mount.
794  *
795  * ---------------------------------------------------------------------------
796  * |         		MOVE MOUNT OPERATION                                 |
797  * |**************************************************************************
798  * | source-->| shared        |       private  |       slave    | unbindable |
799  * | dest     |               |                |                |            |
800  * |   |      |               |                |                |            |
801  * |   v      |               |                |                |            |
802  * |**************************************************************************
803  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
804  * |          |               |                |                |            |
805  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
806  * ***************************************************************************
807  *
808  * (+)  the mount is moved to the destination. And is then propagated to
809  * 	all the mounts in the propagation tree of the destination mount.
810  * (+*)  the mount is moved to the destination.
811  * (+++)  the mount is moved to the destination and is then propagated to
812  * 	all the mounts belonging to the destination mount's propagation tree.
813  * 	the mount is marked as 'shared and slave'.
814  * (*)	the mount continues to be a slave at the new location.
815  *
816  * if the source mount is a tree, the operations explained above is
817  * applied to each mount in the tree.
818  * Must be called without spinlocks held, since this function can sleep
819  * in allocations.
820  */
821 static int attach_recursive_mnt(struct vfsmount *source_mnt,
822 			struct nameidata *nd, struct nameidata *parent_nd)
823 {
824 	LIST_HEAD(tree_list);
825 	struct vfsmount *dest_mnt = nd->mnt;
826 	struct dentry *dest_dentry = nd->dentry;
827 	struct vfsmount *child, *p;
828 
829 	if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
830 		return -EINVAL;
831 
832 	if (IS_MNT_SHARED(dest_mnt)) {
833 		for (p = source_mnt; p; p = next_mnt(p, source_mnt))
834 			set_mnt_shared(p);
835 	}
836 
837 	spin_lock(&vfsmount_lock);
838 	if (parent_nd) {
839 		detach_mnt(source_mnt, parent_nd);
840 		attach_mnt(source_mnt, nd);
841 		touch_namespace(current->namespace);
842 	} else {
843 		mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
844 		commit_tree(source_mnt);
845 	}
846 
847 	list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
848 		list_del_init(&child->mnt_hash);
849 		commit_tree(child);
850 	}
851 	spin_unlock(&vfsmount_lock);
852 	return 0;
853 }
854 
855 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
856 {
857 	int err;
858 	if (mnt->mnt_sb->s_flags & MS_NOUSER)
859 		return -EINVAL;
860 
861 	if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
862 	      S_ISDIR(mnt->mnt_root->d_inode->i_mode))
863 		return -ENOTDIR;
864 
865 	err = -ENOENT;
866 	mutex_lock(&nd->dentry->d_inode->i_mutex);
867 	if (IS_DEADDIR(nd->dentry->d_inode))
868 		goto out_unlock;
869 
870 	err = security_sb_check_sb(mnt, nd);
871 	if (err)
872 		goto out_unlock;
873 
874 	err = -ENOENT;
875 	if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
876 		err = attach_recursive_mnt(mnt, nd, NULL);
877 out_unlock:
878 	mutex_unlock(&nd->dentry->d_inode->i_mutex);
879 	if (!err)
880 		security_sb_post_addmount(mnt, nd);
881 	return err;
882 }
883 
884 /*
885  * recursively change the type of the mountpoint.
886  */
887 static int do_change_type(struct nameidata *nd, int flag)
888 {
889 	struct vfsmount *m, *mnt = nd->mnt;
890 	int recurse = flag & MS_REC;
891 	int type = flag & ~MS_REC;
892 
893 	if (nd->dentry != nd->mnt->mnt_root)
894 		return -EINVAL;
895 
896 	down_write(&namespace_sem);
897 	spin_lock(&vfsmount_lock);
898 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
899 		change_mnt_propagation(m, type);
900 	spin_unlock(&vfsmount_lock);
901 	up_write(&namespace_sem);
902 	return 0;
903 }
904 
905 /*
906  * do loopback mount.
907  */
908 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
909 {
910 	struct nameidata old_nd;
911 	struct vfsmount *mnt = NULL;
912 	int err = mount_is_safe(nd);
913 	if (err)
914 		return err;
915 	if (!old_name || !*old_name)
916 		return -EINVAL;
917 	err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
918 	if (err)
919 		return err;
920 
921 	down_write(&namespace_sem);
922 	err = -EINVAL;
923 	if (IS_MNT_UNBINDABLE(old_nd.mnt))
924  		goto out;
925 
926 	if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
927 		goto out;
928 
929 	err = -ENOMEM;
930 	if (recurse)
931 		mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
932 	else
933 		mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
934 
935 	if (!mnt)
936 		goto out;
937 
938 	err = graft_tree(mnt, nd);
939 	if (err) {
940 		LIST_HEAD(umount_list);
941 		spin_lock(&vfsmount_lock);
942 		umount_tree(mnt, 0, &umount_list);
943 		spin_unlock(&vfsmount_lock);
944 		release_mounts(&umount_list);
945 	}
946 
947 out:
948 	up_write(&namespace_sem);
949 	path_release(&old_nd);
950 	return err;
951 }
952 
953 /*
954  * change filesystem flags. dir should be a physical root of filesystem.
955  * If you've mounted a non-root directory somewhere and want to do remount
956  * on it - tough luck.
957  */
958 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
959 		      void *data)
960 {
961 	int err;
962 	struct super_block *sb = nd->mnt->mnt_sb;
963 
964 	if (!capable(CAP_SYS_ADMIN))
965 		return -EPERM;
966 
967 	if (!check_mnt(nd->mnt))
968 		return -EINVAL;
969 
970 	if (nd->dentry != nd->mnt->mnt_root)
971 		return -EINVAL;
972 
973 	down_write(&sb->s_umount);
974 	err = do_remount_sb(sb, flags, data, 0);
975 	if (!err)
976 		nd->mnt->mnt_flags = mnt_flags;
977 	up_write(&sb->s_umount);
978 	if (!err)
979 		security_sb_post_remount(nd->mnt, flags, data);
980 	return err;
981 }
982 
983 static inline int tree_contains_unbindable(struct vfsmount *mnt)
984 {
985 	struct vfsmount *p;
986 	for (p = mnt; p; p = next_mnt(p, mnt)) {
987 		if (IS_MNT_UNBINDABLE(p))
988 			return 1;
989 	}
990 	return 0;
991 }
992 
993 static int do_move_mount(struct nameidata *nd, char *old_name)
994 {
995 	struct nameidata old_nd, parent_nd;
996 	struct vfsmount *p;
997 	int err = 0;
998 	if (!capable(CAP_SYS_ADMIN))
999 		return -EPERM;
1000 	if (!old_name || !*old_name)
1001 		return -EINVAL;
1002 	err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1003 	if (err)
1004 		return err;
1005 
1006 	down_write(&namespace_sem);
1007 	while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1008 		;
1009 	err = -EINVAL;
1010 	if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1011 		goto out;
1012 
1013 	err = -ENOENT;
1014 	mutex_lock(&nd->dentry->d_inode->i_mutex);
1015 	if (IS_DEADDIR(nd->dentry->d_inode))
1016 		goto out1;
1017 
1018 	if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1019 		goto out1;
1020 
1021 	err = -EINVAL;
1022 	if (old_nd.dentry != old_nd.mnt->mnt_root)
1023 		goto out1;
1024 
1025 	if (old_nd.mnt == old_nd.mnt->mnt_parent)
1026 		goto out1;
1027 
1028 	if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1029 	      S_ISDIR(old_nd.dentry->d_inode->i_mode))
1030 		goto out1;
1031 	/*
1032 	 * Don't move a mount residing in a shared parent.
1033 	 */
1034 	if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1035 		goto out1;
1036 	/*
1037 	 * Don't move a mount tree containing unbindable mounts to a destination
1038 	 * mount which is shared.
1039 	 */
1040 	if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1041 		goto out1;
1042 	err = -ELOOP;
1043 	for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1044 		if (p == old_nd.mnt)
1045 			goto out1;
1046 
1047 	if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1048 		goto out1;
1049 
1050 	spin_lock(&vfsmount_lock);
1051 	/* if the mount is moved, it should no longer be expire
1052 	 * automatically */
1053 	list_del_init(&old_nd.mnt->mnt_expire);
1054 	spin_unlock(&vfsmount_lock);
1055 out1:
1056 	mutex_unlock(&nd->dentry->d_inode->i_mutex);
1057 out:
1058 	up_write(&namespace_sem);
1059 	if (!err)
1060 		path_release(&parent_nd);
1061 	path_release(&old_nd);
1062 	return err;
1063 }
1064 
1065 /*
1066  * create a new mount for userspace and request it to be added into the
1067  * namespace's tree
1068  */
1069 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1070 			int mnt_flags, char *name, void *data)
1071 {
1072 	struct vfsmount *mnt;
1073 
1074 	if (!type || !memchr(type, 0, PAGE_SIZE))
1075 		return -EINVAL;
1076 
1077 	/* we need capabilities... */
1078 	if (!capable(CAP_SYS_ADMIN))
1079 		return -EPERM;
1080 
1081 	mnt = do_kern_mount(type, flags, name, data);
1082 	if (IS_ERR(mnt))
1083 		return PTR_ERR(mnt);
1084 
1085 	return do_add_mount(mnt, nd, mnt_flags, NULL);
1086 }
1087 
1088 /*
1089  * add a mount into a namespace's mount tree
1090  * - provide the option of adding the new mount to an expiration list
1091  */
1092 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1093 		 int mnt_flags, struct list_head *fslist)
1094 {
1095 	int err;
1096 
1097 	down_write(&namespace_sem);
1098 	/* Something was mounted here while we slept */
1099 	while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1100 		;
1101 	err = -EINVAL;
1102 	if (!check_mnt(nd->mnt))
1103 		goto unlock;
1104 
1105 	/* Refuse the same filesystem on the same mount point */
1106 	err = -EBUSY;
1107 	if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1108 	    nd->mnt->mnt_root == nd->dentry)
1109 		goto unlock;
1110 
1111 	err = -EINVAL;
1112 	if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1113 		goto unlock;
1114 
1115 	newmnt->mnt_flags = mnt_flags;
1116 	if ((err = graft_tree(newmnt, nd)))
1117 		goto unlock;
1118 
1119 	if (fslist) {
1120 		/* add to the specified expiration list */
1121 		spin_lock(&vfsmount_lock);
1122 		list_add_tail(&newmnt->mnt_expire, fslist);
1123 		spin_unlock(&vfsmount_lock);
1124 	}
1125 	up_write(&namespace_sem);
1126 	return 0;
1127 
1128 unlock:
1129 	up_write(&namespace_sem);
1130 	mntput(newmnt);
1131 	return err;
1132 }
1133 
1134 EXPORT_SYMBOL_GPL(do_add_mount);
1135 
1136 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1137 				struct list_head *umounts)
1138 {
1139 	spin_lock(&vfsmount_lock);
1140 
1141 	/*
1142 	 * Check if mount is still attached, if not, let whoever holds it deal
1143 	 * with the sucker
1144 	 */
1145 	if (mnt->mnt_parent == mnt) {
1146 		spin_unlock(&vfsmount_lock);
1147 		return;
1148 	}
1149 
1150 	/*
1151 	 * Check that it is still dead: the count should now be 2 - as
1152 	 * contributed by the vfsmount parent and the mntget above
1153 	 */
1154 	if (!propagate_mount_busy(mnt, 2)) {
1155 		/* delete from the namespace */
1156 		touch_namespace(mnt->mnt_namespace);
1157 		list_del_init(&mnt->mnt_list);
1158 		mnt->mnt_namespace = NULL;
1159 		umount_tree(mnt, 1, umounts);
1160 		spin_unlock(&vfsmount_lock);
1161 	} else {
1162 		/*
1163 		 * Someone brought it back to life whilst we didn't have any
1164 		 * locks held so return it to the expiration list
1165 		 */
1166 		list_add_tail(&mnt->mnt_expire, mounts);
1167 		spin_unlock(&vfsmount_lock);
1168 	}
1169 }
1170 
1171 /*
1172  * go through the vfsmounts we've just consigned to the graveyard to
1173  * - check that they're still dead
1174  * - delete the vfsmount from the appropriate namespace under lock
1175  * - dispose of the corpse
1176  */
1177 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1178 {
1179 	struct namespace *namespace;
1180 	struct vfsmount *mnt;
1181 
1182 	while (!list_empty(graveyard)) {
1183 		LIST_HEAD(umounts);
1184 		mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire);
1185 		list_del_init(&mnt->mnt_expire);
1186 
1187 		/* don't do anything if the namespace is dead - all the
1188 		 * vfsmounts from it are going away anyway */
1189 		namespace = mnt->mnt_namespace;
1190 		if (!namespace || !namespace->root)
1191 			continue;
1192 		get_namespace(namespace);
1193 
1194 		spin_unlock(&vfsmount_lock);
1195 		down_write(&namespace_sem);
1196 		expire_mount(mnt, mounts, &umounts);
1197 		up_write(&namespace_sem);
1198 		release_mounts(&umounts);
1199 		mntput(mnt);
1200 		put_namespace(namespace);
1201 		spin_lock(&vfsmount_lock);
1202 	}
1203 }
1204 
1205 /*
1206  * process a list of expirable mountpoints with the intent of discarding any
1207  * mountpoints that aren't in use and haven't been touched since last we came
1208  * here
1209  */
1210 void mark_mounts_for_expiry(struct list_head *mounts)
1211 {
1212 	struct vfsmount *mnt, *next;
1213 	LIST_HEAD(graveyard);
1214 
1215 	if (list_empty(mounts))
1216 		return;
1217 
1218 	spin_lock(&vfsmount_lock);
1219 
1220 	/* extract from the expiration list every vfsmount that matches the
1221 	 * following criteria:
1222 	 * - only referenced by its parent vfsmount
1223 	 * - still marked for expiry (marked on the last call here; marks are
1224 	 *   cleared by mntput())
1225 	 */
1226 	list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1227 		if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1228 		    atomic_read(&mnt->mnt_count) != 1)
1229 			continue;
1230 
1231 		mntget(mnt);
1232 		list_move(&mnt->mnt_expire, &graveyard);
1233 	}
1234 
1235 	expire_mount_list(&graveyard, mounts);
1236 
1237 	spin_unlock(&vfsmount_lock);
1238 }
1239 
1240 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1241 
1242 /*
1243  * Ripoff of 'select_parent()'
1244  *
1245  * search the list of submounts for a given mountpoint, and move any
1246  * shrinkable submounts to the 'graveyard' list.
1247  */
1248 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1249 {
1250 	struct vfsmount *this_parent = parent;
1251 	struct list_head *next;
1252 	int found = 0;
1253 
1254 repeat:
1255 	next = this_parent->mnt_mounts.next;
1256 resume:
1257 	while (next != &this_parent->mnt_mounts) {
1258 		struct list_head *tmp = next;
1259 		struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1260 
1261 		next = tmp->next;
1262 		if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1263 			continue;
1264 		/*
1265 		 * Descend a level if the d_mounts list is non-empty.
1266 		 */
1267 		if (!list_empty(&mnt->mnt_mounts)) {
1268 			this_parent = mnt;
1269 			goto repeat;
1270 		}
1271 
1272 		if (!propagate_mount_busy(mnt, 1)) {
1273 			mntget(mnt);
1274 			list_move_tail(&mnt->mnt_expire, graveyard);
1275 			found++;
1276 		}
1277 	}
1278 	/*
1279 	 * All done at this level ... ascend and resume the search
1280 	 */
1281 	if (this_parent != parent) {
1282 		next = this_parent->mnt_child.next;
1283 		this_parent = this_parent->mnt_parent;
1284 		goto resume;
1285 	}
1286 	return found;
1287 }
1288 
1289 /*
1290  * process a list of expirable mountpoints with the intent of discarding any
1291  * submounts of a specific parent mountpoint
1292  */
1293 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1294 {
1295 	LIST_HEAD(graveyard);
1296 	int found;
1297 
1298 	spin_lock(&vfsmount_lock);
1299 
1300 	/* extract submounts of 'mountpoint' from the expiration list */
1301 	while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1302 		expire_mount_list(&graveyard, mounts);
1303 
1304 	spin_unlock(&vfsmount_lock);
1305 }
1306 
1307 EXPORT_SYMBOL_GPL(shrink_submounts);
1308 
1309 /*
1310  * Some copy_from_user() implementations do not return the exact number of
1311  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
1312  * Note that this function differs from copy_from_user() in that it will oops
1313  * on bad values of `to', rather than returning a short copy.
1314  */
1315 static long exact_copy_from_user(void *to, const void __user * from,
1316 				 unsigned long n)
1317 {
1318 	char *t = to;
1319 	const char __user *f = from;
1320 	char c;
1321 
1322 	if (!access_ok(VERIFY_READ, from, n))
1323 		return n;
1324 
1325 	while (n) {
1326 		if (__get_user(c, f)) {
1327 			memset(t, 0, n);
1328 			break;
1329 		}
1330 		*t++ = c;
1331 		f++;
1332 		n--;
1333 	}
1334 	return n;
1335 }
1336 
1337 int copy_mount_options(const void __user * data, unsigned long *where)
1338 {
1339 	int i;
1340 	unsigned long page;
1341 	unsigned long size;
1342 
1343 	*where = 0;
1344 	if (!data)
1345 		return 0;
1346 
1347 	if (!(page = __get_free_page(GFP_KERNEL)))
1348 		return -ENOMEM;
1349 
1350 	/* We only care that *some* data at the address the user
1351 	 * gave us is valid.  Just in case, we'll zero
1352 	 * the remainder of the page.
1353 	 */
1354 	/* copy_from_user cannot cross TASK_SIZE ! */
1355 	size = TASK_SIZE - (unsigned long)data;
1356 	if (size > PAGE_SIZE)
1357 		size = PAGE_SIZE;
1358 
1359 	i = size - exact_copy_from_user((void *)page, data, size);
1360 	if (!i) {
1361 		free_page(page);
1362 		return -EFAULT;
1363 	}
1364 	if (i != PAGE_SIZE)
1365 		memset((char *)page + i, 0, PAGE_SIZE - i);
1366 	*where = page;
1367 	return 0;
1368 }
1369 
1370 /*
1371  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1372  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1373  *
1374  * data is a (void *) that can point to any structure up to
1375  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1376  * information (or be NULL).
1377  *
1378  * Pre-0.97 versions of mount() didn't have a flags word.
1379  * When the flags word was introduced its top half was required
1380  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1381  * Therefore, if this magic number is present, it carries no information
1382  * and must be discarded.
1383  */
1384 long do_mount(char *dev_name, char *dir_name, char *type_page,
1385 		  unsigned long flags, void *data_page)
1386 {
1387 	struct nameidata nd;
1388 	int retval = 0;
1389 	int mnt_flags = 0;
1390 
1391 	/* Discard magic */
1392 	if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1393 		flags &= ~MS_MGC_MSK;
1394 
1395 	/* Basic sanity checks */
1396 
1397 	if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1398 		return -EINVAL;
1399 	if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1400 		return -EINVAL;
1401 
1402 	if (data_page)
1403 		((char *)data_page)[PAGE_SIZE - 1] = 0;
1404 
1405 	/* Separate the per-mountpoint flags */
1406 	if (flags & MS_NOSUID)
1407 		mnt_flags |= MNT_NOSUID;
1408 	if (flags & MS_NODEV)
1409 		mnt_flags |= MNT_NODEV;
1410 	if (flags & MS_NOEXEC)
1411 		mnt_flags |= MNT_NOEXEC;
1412 	if (flags & MS_NOATIME)
1413 		mnt_flags |= MNT_NOATIME;
1414 	if (flags & MS_NODIRATIME)
1415 		mnt_flags |= MNT_NODIRATIME;
1416 
1417 	flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1418 		   MS_NOATIME | MS_NODIRATIME);
1419 
1420 	/* ... and get the mountpoint */
1421 	retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1422 	if (retval)
1423 		return retval;
1424 
1425 	retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1426 	if (retval)
1427 		goto dput_out;
1428 
1429 	if (flags & MS_REMOUNT)
1430 		retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1431 				    data_page);
1432 	else if (flags & MS_BIND)
1433 		retval = do_loopback(&nd, dev_name, flags & MS_REC);
1434 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1435 		retval = do_change_type(&nd, flags);
1436 	else if (flags & MS_MOVE)
1437 		retval = do_move_mount(&nd, dev_name);
1438 	else
1439 		retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1440 				      dev_name, data_page);
1441 dput_out:
1442 	path_release(&nd);
1443 	return retval;
1444 }
1445 
1446 /*
1447  * Allocate a new namespace structure and populate it with contents
1448  * copied from the namespace of the passed in task structure.
1449  */
1450 struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
1451 {
1452 	struct namespace *namespace = tsk->namespace;
1453 	struct namespace *new_ns;
1454 	struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1455 	struct vfsmount *p, *q;
1456 
1457 	new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1458 	if (!new_ns)
1459 		return NULL;
1460 
1461 	atomic_set(&new_ns->count, 1);
1462 	INIT_LIST_HEAD(&new_ns->list);
1463 	init_waitqueue_head(&new_ns->poll);
1464 	new_ns->event = 0;
1465 
1466 	down_write(&namespace_sem);
1467 	/* First pass: copy the tree topology */
1468 	new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1469 					CL_COPY_ALL | CL_EXPIRE);
1470 	if (!new_ns->root) {
1471 		up_write(&namespace_sem);
1472 		kfree(new_ns);
1473 		return NULL;
1474 	}
1475 	spin_lock(&vfsmount_lock);
1476 	list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1477 	spin_unlock(&vfsmount_lock);
1478 
1479 	/*
1480 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1481 	 * as belonging to new namespace.  We have already acquired a private
1482 	 * fs_struct, so tsk->fs->lock is not needed.
1483 	 */
1484 	p = namespace->root;
1485 	q = new_ns->root;
1486 	while (p) {
1487 		q->mnt_namespace = new_ns;
1488 		if (fs) {
1489 			if (p == fs->rootmnt) {
1490 				rootmnt = p;
1491 				fs->rootmnt = mntget(q);
1492 			}
1493 			if (p == fs->pwdmnt) {
1494 				pwdmnt = p;
1495 				fs->pwdmnt = mntget(q);
1496 			}
1497 			if (p == fs->altrootmnt) {
1498 				altrootmnt = p;
1499 				fs->altrootmnt = mntget(q);
1500 			}
1501 		}
1502 		p = next_mnt(p, namespace->root);
1503 		q = next_mnt(q, new_ns->root);
1504 	}
1505 	up_write(&namespace_sem);
1506 
1507 	if (rootmnt)
1508 		mntput(rootmnt);
1509 	if (pwdmnt)
1510 		mntput(pwdmnt);
1511 	if (altrootmnt)
1512 		mntput(altrootmnt);
1513 
1514 	return new_ns;
1515 }
1516 
1517 int copy_namespace(int flags, struct task_struct *tsk)
1518 {
1519 	struct namespace *namespace = tsk->namespace;
1520 	struct namespace *new_ns;
1521 	int err = 0;
1522 
1523 	if (!namespace)
1524 		return 0;
1525 
1526 	get_namespace(namespace);
1527 
1528 	if (!(flags & CLONE_NEWNS))
1529 		return 0;
1530 
1531 	if (!capable(CAP_SYS_ADMIN)) {
1532 		err = -EPERM;
1533 		goto out;
1534 	}
1535 
1536 	new_ns = dup_namespace(tsk, tsk->fs);
1537 	if (!new_ns) {
1538 		err = -ENOMEM;
1539 		goto out;
1540 	}
1541 
1542 	tsk->namespace = new_ns;
1543 
1544 out:
1545 	put_namespace(namespace);
1546 	return err;
1547 }
1548 
1549 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1550 			  char __user * type, unsigned long flags,
1551 			  void __user * data)
1552 {
1553 	int retval;
1554 	unsigned long data_page;
1555 	unsigned long type_page;
1556 	unsigned long dev_page;
1557 	char *dir_page;
1558 
1559 	retval = copy_mount_options(type, &type_page);
1560 	if (retval < 0)
1561 		return retval;
1562 
1563 	dir_page = getname(dir_name);
1564 	retval = PTR_ERR(dir_page);
1565 	if (IS_ERR(dir_page))
1566 		goto out1;
1567 
1568 	retval = copy_mount_options(dev_name, &dev_page);
1569 	if (retval < 0)
1570 		goto out2;
1571 
1572 	retval = copy_mount_options(data, &data_page);
1573 	if (retval < 0)
1574 		goto out3;
1575 
1576 	lock_kernel();
1577 	retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1578 			  flags, (void *)data_page);
1579 	unlock_kernel();
1580 	free_page(data_page);
1581 
1582 out3:
1583 	free_page(dev_page);
1584 out2:
1585 	putname(dir_page);
1586 out1:
1587 	free_page(type_page);
1588 	return retval;
1589 }
1590 
1591 /*
1592  * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1593  * It can block. Requires the big lock held.
1594  */
1595 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1596 		 struct dentry *dentry)
1597 {
1598 	struct dentry *old_root;
1599 	struct vfsmount *old_rootmnt;
1600 	write_lock(&fs->lock);
1601 	old_root = fs->root;
1602 	old_rootmnt = fs->rootmnt;
1603 	fs->rootmnt = mntget(mnt);
1604 	fs->root = dget(dentry);
1605 	write_unlock(&fs->lock);
1606 	if (old_root) {
1607 		dput(old_root);
1608 		mntput(old_rootmnt);
1609 	}
1610 }
1611 
1612 /*
1613  * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1614  * It can block. Requires the big lock held.
1615  */
1616 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1617 		struct dentry *dentry)
1618 {
1619 	struct dentry *old_pwd;
1620 	struct vfsmount *old_pwdmnt;
1621 
1622 	write_lock(&fs->lock);
1623 	old_pwd = fs->pwd;
1624 	old_pwdmnt = fs->pwdmnt;
1625 	fs->pwdmnt = mntget(mnt);
1626 	fs->pwd = dget(dentry);
1627 	write_unlock(&fs->lock);
1628 
1629 	if (old_pwd) {
1630 		dput(old_pwd);
1631 		mntput(old_pwdmnt);
1632 	}
1633 }
1634 
1635 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1636 {
1637 	struct task_struct *g, *p;
1638 	struct fs_struct *fs;
1639 
1640 	read_lock(&tasklist_lock);
1641 	do_each_thread(g, p) {
1642 		task_lock(p);
1643 		fs = p->fs;
1644 		if (fs) {
1645 			atomic_inc(&fs->count);
1646 			task_unlock(p);
1647 			if (fs->root == old_nd->dentry
1648 			    && fs->rootmnt == old_nd->mnt)
1649 				set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1650 			if (fs->pwd == old_nd->dentry
1651 			    && fs->pwdmnt == old_nd->mnt)
1652 				set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1653 			put_fs_struct(fs);
1654 		} else
1655 			task_unlock(p);
1656 	} while_each_thread(g, p);
1657 	read_unlock(&tasklist_lock);
1658 }
1659 
1660 /*
1661  * pivot_root Semantics:
1662  * Moves the root file system of the current process to the directory put_old,
1663  * makes new_root as the new root file system of the current process, and sets
1664  * root/cwd of all processes which had them on the current root to new_root.
1665  *
1666  * Restrictions:
1667  * The new_root and put_old must be directories, and  must not be on the
1668  * same file  system as the current process root. The put_old  must  be
1669  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
1670  * pointed to by put_old must yield the same directory as new_root. No other
1671  * file system may be mounted on put_old. After all, new_root is a mountpoint.
1672  *
1673  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1674  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1675  * in this situation.
1676  *
1677  * Notes:
1678  *  - we don't move root/cwd if they are not at the root (reason: if something
1679  *    cared enough to change them, it's probably wrong to force them elsewhere)
1680  *  - it's okay to pick a root that isn't the root of a file system, e.g.
1681  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1682  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1683  *    first.
1684  */
1685 asmlinkage long sys_pivot_root(const char __user * new_root,
1686 			       const char __user * put_old)
1687 {
1688 	struct vfsmount *tmp;
1689 	struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1690 	int error;
1691 
1692 	if (!capable(CAP_SYS_ADMIN))
1693 		return -EPERM;
1694 
1695 	lock_kernel();
1696 
1697 	error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1698 			    &new_nd);
1699 	if (error)
1700 		goto out0;
1701 	error = -EINVAL;
1702 	if (!check_mnt(new_nd.mnt))
1703 		goto out1;
1704 
1705 	error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1706 	if (error)
1707 		goto out1;
1708 
1709 	error = security_sb_pivotroot(&old_nd, &new_nd);
1710 	if (error) {
1711 		path_release(&old_nd);
1712 		goto out1;
1713 	}
1714 
1715 	read_lock(&current->fs->lock);
1716 	user_nd.mnt = mntget(current->fs->rootmnt);
1717 	user_nd.dentry = dget(current->fs->root);
1718 	read_unlock(&current->fs->lock);
1719 	down_write(&namespace_sem);
1720 	mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1721 	error = -EINVAL;
1722 	if (IS_MNT_SHARED(old_nd.mnt) ||
1723 		IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1724 		IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1725 		goto out2;
1726 	if (!check_mnt(user_nd.mnt))
1727 		goto out2;
1728 	error = -ENOENT;
1729 	if (IS_DEADDIR(new_nd.dentry->d_inode))
1730 		goto out2;
1731 	if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1732 		goto out2;
1733 	if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1734 		goto out2;
1735 	error = -EBUSY;
1736 	if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1737 		goto out2; /* loop, on the same file system  */
1738 	error = -EINVAL;
1739 	if (user_nd.mnt->mnt_root != user_nd.dentry)
1740 		goto out2; /* not a mountpoint */
1741 	if (user_nd.mnt->mnt_parent == user_nd.mnt)
1742 		goto out2; /* not attached */
1743 	if (new_nd.mnt->mnt_root != new_nd.dentry)
1744 		goto out2; /* not a mountpoint */
1745 	if (new_nd.mnt->mnt_parent == new_nd.mnt)
1746 		goto out2; /* not attached */
1747 	tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1748 	spin_lock(&vfsmount_lock);
1749 	if (tmp != new_nd.mnt) {
1750 		for (;;) {
1751 			if (tmp->mnt_parent == tmp)
1752 				goto out3; /* already mounted on put_old */
1753 			if (tmp->mnt_parent == new_nd.mnt)
1754 				break;
1755 			tmp = tmp->mnt_parent;
1756 		}
1757 		if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1758 			goto out3;
1759 	} else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1760 		goto out3;
1761 	detach_mnt(new_nd.mnt, &parent_nd);
1762 	detach_mnt(user_nd.mnt, &root_parent);
1763 	attach_mnt(user_nd.mnt, &old_nd);     /* mount old root on put_old */
1764 	attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1765 	touch_namespace(current->namespace);
1766 	spin_unlock(&vfsmount_lock);
1767 	chroot_fs_refs(&user_nd, &new_nd);
1768 	security_sb_post_pivotroot(&user_nd, &new_nd);
1769 	error = 0;
1770 	path_release(&root_parent);
1771 	path_release(&parent_nd);
1772 out2:
1773 	mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1774 	up_write(&namespace_sem);
1775 	path_release(&user_nd);
1776 	path_release(&old_nd);
1777 out1:
1778 	path_release(&new_nd);
1779 out0:
1780 	unlock_kernel();
1781 	return error;
1782 out3:
1783 	spin_unlock(&vfsmount_lock);
1784 	goto out2;
1785 }
1786 
1787 static void __init init_mount_tree(void)
1788 {
1789 	struct vfsmount *mnt;
1790 	struct namespace *namespace;
1791 	struct task_struct *g, *p;
1792 
1793 	mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1794 	if (IS_ERR(mnt))
1795 		panic("Can't create rootfs");
1796 	namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1797 	if (!namespace)
1798 		panic("Can't allocate initial namespace");
1799 	atomic_set(&namespace->count, 1);
1800 	INIT_LIST_HEAD(&namespace->list);
1801 	init_waitqueue_head(&namespace->poll);
1802 	namespace->event = 0;
1803 	list_add(&mnt->mnt_list, &namespace->list);
1804 	namespace->root = mnt;
1805 	mnt->mnt_namespace = namespace;
1806 
1807 	init_task.namespace = namespace;
1808 	read_lock(&tasklist_lock);
1809 	do_each_thread(g, p) {
1810 		get_namespace(namespace);
1811 		p->namespace = namespace;
1812 	} while_each_thread(g, p);
1813 	read_unlock(&tasklist_lock);
1814 
1815 	set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1816 	set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1817 }
1818 
1819 void __init mnt_init(unsigned long mempages)
1820 {
1821 	struct list_head *d;
1822 	unsigned int nr_hash;
1823 	int i;
1824 
1825 	init_rwsem(&namespace_sem);
1826 
1827 	mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1828 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1829 
1830 	mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1831 
1832 	if (!mount_hashtable)
1833 		panic("Failed to allocate mount hash table\n");
1834 
1835 	/*
1836 	 * Find the power-of-two list-heads that can fit into the allocation..
1837 	 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1838 	 * a power-of-two.
1839 	 */
1840 	nr_hash = PAGE_SIZE / sizeof(struct list_head);
1841 	hash_bits = 0;
1842 	do {
1843 		hash_bits++;
1844 	} while ((nr_hash >> hash_bits) != 0);
1845 	hash_bits--;
1846 
1847 	/*
1848 	 * Re-calculate the actual number of entries and the mask
1849 	 * from the number of bits we can fit.
1850 	 */
1851 	nr_hash = 1UL << hash_bits;
1852 	hash_mask = nr_hash - 1;
1853 
1854 	printk("Mount-cache hash table entries: %d\n", nr_hash);
1855 
1856 	/* And initialize the newly allocated array */
1857 	d = mount_hashtable;
1858 	i = nr_hash;
1859 	do {
1860 		INIT_LIST_HEAD(d);
1861 		d++;
1862 		i--;
1863 	} while (i);
1864 	sysfs_init();
1865 	subsystem_register(&fs_subsys);
1866 	init_rootfs();
1867 	init_mount_tree();
1868 }
1869 
1870 void __put_namespace(struct namespace *namespace)
1871 {
1872 	struct vfsmount *root = namespace->root;
1873 	LIST_HEAD(umount_list);
1874 	namespace->root = NULL;
1875 	spin_unlock(&vfsmount_lock);
1876 	down_write(&namespace_sem);
1877 	spin_lock(&vfsmount_lock);
1878 	umount_tree(root, 0, &umount_list);
1879 	spin_unlock(&vfsmount_lock);
1880 	up_write(&namespace_sem);
1881 	release_mounts(&umount_list);
1882 	kfree(namespace);
1883 }
1884