xref: /linux/fs/proc/proc_sysctl.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
1 /*
2  * /proc/sys support
3  */
4 #include <linux/init.h>
5 #include <linux/sysctl.h>
6 #include <linux/poll.h>
7 #include <linux/proc_fs.h>
8 #include <linux/printk.h>
9 #include <linux/security.h>
10 #include <linux/sched.h>
11 #include <linux/namei.h>
12 #include <linux/mm.h>
13 #include <linux/module.h>
14 #include "internal.h"
15 
16 static const struct dentry_operations proc_sys_dentry_operations;
17 static const struct file_operations proc_sys_file_operations;
18 static const struct inode_operations proc_sys_inode_operations;
19 static const struct file_operations proc_sys_dir_file_operations;
20 static const struct inode_operations proc_sys_dir_operations;
21 
22 /* Support for permanently empty directories */
23 
24 struct ctl_table sysctl_mount_point[] = {
25 	{ }
26 };
27 
28 static bool is_empty_dir(struct ctl_table_header *head)
29 {
30 	return head->ctl_table[0].child == sysctl_mount_point;
31 }
32 
33 static void set_empty_dir(struct ctl_dir *dir)
34 {
35 	dir->header.ctl_table[0].child = sysctl_mount_point;
36 }
37 
38 static void clear_empty_dir(struct ctl_dir *dir)
39 
40 {
41 	dir->header.ctl_table[0].child = NULL;
42 }
43 
44 void proc_sys_poll_notify(struct ctl_table_poll *poll)
45 {
46 	if (!poll)
47 		return;
48 
49 	atomic_inc(&poll->event);
50 	wake_up_interruptible(&poll->wait);
51 }
52 
53 static struct ctl_table root_table[] = {
54 	{
55 		.procname = "",
56 		.mode = S_IFDIR|S_IRUGO|S_IXUGO,
57 	},
58 	{ }
59 };
60 static struct ctl_table_root sysctl_table_root = {
61 	.default_set.dir.header = {
62 		{{.count = 1,
63 		  .nreg = 1,
64 		  .ctl_table = root_table }},
65 		.ctl_table_arg = root_table,
66 		.root = &sysctl_table_root,
67 		.set = &sysctl_table_root.default_set,
68 	},
69 };
70 
71 static DEFINE_SPINLOCK(sysctl_lock);
72 
73 static void drop_sysctl_table(struct ctl_table_header *header);
74 static int sysctl_follow_link(struct ctl_table_header **phead,
75 	struct ctl_table **pentry);
76 static int insert_links(struct ctl_table_header *head);
77 static void put_links(struct ctl_table_header *header);
78 
79 static void sysctl_print_dir(struct ctl_dir *dir)
80 {
81 	if (dir->header.parent)
82 		sysctl_print_dir(dir->header.parent);
83 	pr_cont("%s/", dir->header.ctl_table[0].procname);
84 }
85 
86 static int namecmp(const char *name1, int len1, const char *name2, int len2)
87 {
88 	int minlen;
89 	int cmp;
90 
91 	minlen = len1;
92 	if (minlen > len2)
93 		minlen = len2;
94 
95 	cmp = memcmp(name1, name2, minlen);
96 	if (cmp == 0)
97 		cmp = len1 - len2;
98 	return cmp;
99 }
100 
101 /* Called under sysctl_lock */
102 static struct ctl_table *find_entry(struct ctl_table_header **phead,
103 	struct ctl_dir *dir, const char *name, int namelen)
104 {
105 	struct ctl_table_header *head;
106 	struct ctl_table *entry;
107 	struct rb_node *node = dir->root.rb_node;
108 
109 	while (node)
110 	{
111 		struct ctl_node *ctl_node;
112 		const char *procname;
113 		int cmp;
114 
115 		ctl_node = rb_entry(node, struct ctl_node, node);
116 		head = ctl_node->header;
117 		entry = &head->ctl_table[ctl_node - head->node];
118 		procname = entry->procname;
119 
120 		cmp = namecmp(name, namelen, procname, strlen(procname));
121 		if (cmp < 0)
122 			node = node->rb_left;
123 		else if (cmp > 0)
124 			node = node->rb_right;
125 		else {
126 			*phead = head;
127 			return entry;
128 		}
129 	}
130 	return NULL;
131 }
132 
133 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
134 {
135 	struct rb_node *node = &head->node[entry - head->ctl_table].node;
136 	struct rb_node **p = &head->parent->root.rb_node;
137 	struct rb_node *parent = NULL;
138 	const char *name = entry->procname;
139 	int namelen = strlen(name);
140 
141 	while (*p) {
142 		struct ctl_table_header *parent_head;
143 		struct ctl_table *parent_entry;
144 		struct ctl_node *parent_node;
145 		const char *parent_name;
146 		int cmp;
147 
148 		parent = *p;
149 		parent_node = rb_entry(parent, struct ctl_node, node);
150 		parent_head = parent_node->header;
151 		parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
152 		parent_name = parent_entry->procname;
153 
154 		cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
155 		if (cmp < 0)
156 			p = &(*p)->rb_left;
157 		else if (cmp > 0)
158 			p = &(*p)->rb_right;
159 		else {
160 			pr_err("sysctl duplicate entry: ");
161 			sysctl_print_dir(head->parent);
162 			pr_cont("/%s\n", entry->procname);
163 			return -EEXIST;
164 		}
165 	}
166 
167 	rb_link_node(node, parent, p);
168 	rb_insert_color(node, &head->parent->root);
169 	return 0;
170 }
171 
172 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
173 {
174 	struct rb_node *node = &head->node[entry - head->ctl_table].node;
175 
176 	rb_erase(node, &head->parent->root);
177 }
178 
179 static void init_header(struct ctl_table_header *head,
180 	struct ctl_table_root *root, struct ctl_table_set *set,
181 	struct ctl_node *node, struct ctl_table *table)
182 {
183 	head->ctl_table = table;
184 	head->ctl_table_arg = table;
185 	head->used = 0;
186 	head->count = 1;
187 	head->nreg = 1;
188 	head->unregistering = NULL;
189 	head->root = root;
190 	head->set = set;
191 	head->parent = NULL;
192 	head->node = node;
193 	if (node) {
194 		struct ctl_table *entry;
195 		for (entry = table; entry->procname; entry++, node++)
196 			node->header = head;
197 	}
198 }
199 
200 static void erase_header(struct ctl_table_header *head)
201 {
202 	struct ctl_table *entry;
203 	for (entry = head->ctl_table; entry->procname; entry++)
204 		erase_entry(head, entry);
205 }
206 
207 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
208 {
209 	struct ctl_table *entry;
210 	int err;
211 
212 	/* Is this a permanently empty directory? */
213 	if (is_empty_dir(&dir->header))
214 		return -EROFS;
215 
216 	/* Am I creating a permanently empty directory? */
217 	if (header->ctl_table == sysctl_mount_point) {
218 		if (!RB_EMPTY_ROOT(&dir->root))
219 			return -EINVAL;
220 		set_empty_dir(dir);
221 	}
222 
223 	dir->header.nreg++;
224 	header->parent = dir;
225 	err = insert_links(header);
226 	if (err)
227 		goto fail_links;
228 	for (entry = header->ctl_table; entry->procname; entry++) {
229 		err = insert_entry(header, entry);
230 		if (err)
231 			goto fail;
232 	}
233 	return 0;
234 fail:
235 	erase_header(header);
236 	put_links(header);
237 fail_links:
238 	if (header->ctl_table == sysctl_mount_point)
239 		clear_empty_dir(dir);
240 	header->parent = NULL;
241 	drop_sysctl_table(&dir->header);
242 	return err;
243 }
244 
245 /* called under sysctl_lock */
246 static int use_table(struct ctl_table_header *p)
247 {
248 	if (unlikely(p->unregistering))
249 		return 0;
250 	p->used++;
251 	return 1;
252 }
253 
254 /* called under sysctl_lock */
255 static void unuse_table(struct ctl_table_header *p)
256 {
257 	if (!--p->used)
258 		if (unlikely(p->unregistering))
259 			complete(p->unregistering);
260 }
261 
262 /* called under sysctl_lock, will reacquire if has to wait */
263 static void start_unregistering(struct ctl_table_header *p)
264 {
265 	/*
266 	 * if p->used is 0, nobody will ever touch that entry again;
267 	 * we'll eliminate all paths to it before dropping sysctl_lock
268 	 */
269 	if (unlikely(p->used)) {
270 		struct completion wait;
271 		init_completion(&wait);
272 		p->unregistering = &wait;
273 		spin_unlock(&sysctl_lock);
274 		wait_for_completion(&wait);
275 		spin_lock(&sysctl_lock);
276 	} else {
277 		/* anything non-NULL; we'll never dereference it */
278 		p->unregistering = ERR_PTR(-EINVAL);
279 	}
280 	/*
281 	 * do not remove from the list until nobody holds it; walking the
282 	 * list in do_sysctl() relies on that.
283 	 */
284 	erase_header(p);
285 }
286 
287 static void sysctl_head_get(struct ctl_table_header *head)
288 {
289 	spin_lock(&sysctl_lock);
290 	head->count++;
291 	spin_unlock(&sysctl_lock);
292 }
293 
294 void sysctl_head_put(struct ctl_table_header *head)
295 {
296 	spin_lock(&sysctl_lock);
297 	if (!--head->count)
298 		kfree_rcu(head, rcu);
299 	spin_unlock(&sysctl_lock);
300 }
301 
302 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
303 {
304 	BUG_ON(!head);
305 	spin_lock(&sysctl_lock);
306 	if (!use_table(head))
307 		head = ERR_PTR(-ENOENT);
308 	spin_unlock(&sysctl_lock);
309 	return head;
310 }
311 
312 static void sysctl_head_finish(struct ctl_table_header *head)
313 {
314 	if (!head)
315 		return;
316 	spin_lock(&sysctl_lock);
317 	unuse_table(head);
318 	spin_unlock(&sysctl_lock);
319 }
320 
321 static struct ctl_table_set *
322 lookup_header_set(struct ctl_table_root *root)
323 {
324 	struct ctl_table_set *set = &root->default_set;
325 	if (root->lookup)
326 		set = root->lookup(root);
327 	return set;
328 }
329 
330 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
331 				      struct ctl_dir *dir,
332 				      const char *name, int namelen)
333 {
334 	struct ctl_table_header *head;
335 	struct ctl_table *entry;
336 
337 	spin_lock(&sysctl_lock);
338 	entry = find_entry(&head, dir, name, namelen);
339 	if (entry && use_table(head))
340 		*phead = head;
341 	else
342 		entry = NULL;
343 	spin_unlock(&sysctl_lock);
344 	return entry;
345 }
346 
347 static struct ctl_node *first_usable_entry(struct rb_node *node)
348 {
349 	struct ctl_node *ctl_node;
350 
351 	for (;node; node = rb_next(node)) {
352 		ctl_node = rb_entry(node, struct ctl_node, node);
353 		if (use_table(ctl_node->header))
354 			return ctl_node;
355 	}
356 	return NULL;
357 }
358 
359 static void first_entry(struct ctl_dir *dir,
360 	struct ctl_table_header **phead, struct ctl_table **pentry)
361 {
362 	struct ctl_table_header *head = NULL;
363 	struct ctl_table *entry = NULL;
364 	struct ctl_node *ctl_node;
365 
366 	spin_lock(&sysctl_lock);
367 	ctl_node = first_usable_entry(rb_first(&dir->root));
368 	spin_unlock(&sysctl_lock);
369 	if (ctl_node) {
370 		head = ctl_node->header;
371 		entry = &head->ctl_table[ctl_node - head->node];
372 	}
373 	*phead = head;
374 	*pentry = entry;
375 }
376 
377 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
378 {
379 	struct ctl_table_header *head = *phead;
380 	struct ctl_table *entry = *pentry;
381 	struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
382 
383 	spin_lock(&sysctl_lock);
384 	unuse_table(head);
385 
386 	ctl_node = first_usable_entry(rb_next(&ctl_node->node));
387 	spin_unlock(&sysctl_lock);
388 	head = NULL;
389 	if (ctl_node) {
390 		head = ctl_node->header;
391 		entry = &head->ctl_table[ctl_node - head->node];
392 	}
393 	*phead = head;
394 	*pentry = entry;
395 }
396 
397 void register_sysctl_root(struct ctl_table_root *root)
398 {
399 }
400 
401 /*
402  * sysctl_perm does NOT grant the superuser all rights automatically, because
403  * some sysctl variables are readonly even to root.
404  */
405 
406 static int test_perm(int mode, int op)
407 {
408 	if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
409 		mode >>= 6;
410 	else if (in_egroup_p(GLOBAL_ROOT_GID))
411 		mode >>= 3;
412 	if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
413 		return 0;
414 	return -EACCES;
415 }
416 
417 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
418 {
419 	struct ctl_table_root *root = head->root;
420 	int mode;
421 
422 	if (root->permissions)
423 		mode = root->permissions(head, table);
424 	else
425 		mode = table->mode;
426 
427 	return test_perm(mode, op);
428 }
429 
430 static struct inode *proc_sys_make_inode(struct super_block *sb,
431 		struct ctl_table_header *head, struct ctl_table *table)
432 {
433 	struct ctl_table_root *root = head->root;
434 	struct inode *inode;
435 	struct proc_inode *ei;
436 
437 	inode = new_inode(sb);
438 	if (!inode)
439 		goto out;
440 
441 	inode->i_ino = get_next_ino();
442 
443 	sysctl_head_get(head);
444 	ei = PROC_I(inode);
445 	ei->sysctl = head;
446 	ei->sysctl_entry = table;
447 
448 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
449 	inode->i_mode = table->mode;
450 	if (!S_ISDIR(table->mode)) {
451 		inode->i_mode |= S_IFREG;
452 		inode->i_op = &proc_sys_inode_operations;
453 		inode->i_fop = &proc_sys_file_operations;
454 	} else {
455 		inode->i_mode |= S_IFDIR;
456 		inode->i_op = &proc_sys_dir_operations;
457 		inode->i_fop = &proc_sys_dir_file_operations;
458 		if (is_empty_dir(head))
459 			make_empty_dir_inode(inode);
460 	}
461 
462 	if (root->set_ownership)
463 		root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
464 
465 out:
466 	return inode;
467 }
468 
469 static struct ctl_table_header *grab_header(struct inode *inode)
470 {
471 	struct ctl_table_header *head = PROC_I(inode)->sysctl;
472 	if (!head)
473 		head = &sysctl_table_root.default_set.dir.header;
474 	return sysctl_head_grab(head);
475 }
476 
477 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
478 					unsigned int flags)
479 {
480 	struct ctl_table_header *head = grab_header(dir);
481 	struct ctl_table_header *h = NULL;
482 	const struct qstr *name = &dentry->d_name;
483 	struct ctl_table *p;
484 	struct inode *inode;
485 	struct dentry *err = ERR_PTR(-ENOENT);
486 	struct ctl_dir *ctl_dir;
487 	int ret;
488 
489 	if (IS_ERR(head))
490 		return ERR_CAST(head);
491 
492 	ctl_dir = container_of(head, struct ctl_dir, header);
493 
494 	p = lookup_entry(&h, ctl_dir, name->name, name->len);
495 	if (!p)
496 		goto out;
497 
498 	if (S_ISLNK(p->mode)) {
499 		ret = sysctl_follow_link(&h, &p);
500 		err = ERR_PTR(ret);
501 		if (ret)
502 			goto out;
503 	}
504 
505 	err = ERR_PTR(-ENOMEM);
506 	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
507 	if (!inode)
508 		goto out;
509 
510 	err = NULL;
511 	d_set_d_op(dentry, &proc_sys_dentry_operations);
512 	d_add(dentry, inode);
513 
514 out:
515 	if (h)
516 		sysctl_head_finish(h);
517 	sysctl_head_finish(head);
518 	return err;
519 }
520 
521 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
522 		size_t count, loff_t *ppos, int write)
523 {
524 	struct inode *inode = file_inode(filp);
525 	struct ctl_table_header *head = grab_header(inode);
526 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
527 	ssize_t error;
528 	size_t res;
529 
530 	if (IS_ERR(head))
531 		return PTR_ERR(head);
532 
533 	/*
534 	 * At this point we know that the sysctl was not unregistered
535 	 * and won't be until we finish.
536 	 */
537 	error = -EPERM;
538 	if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
539 		goto out;
540 
541 	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
542 	error = -EINVAL;
543 	if (!table->proc_handler)
544 		goto out;
545 
546 	/* careful: calling conventions are nasty here */
547 	res = count;
548 	error = table->proc_handler(table, write, buf, &res, ppos);
549 	if (!error)
550 		error = res;
551 out:
552 	sysctl_head_finish(head);
553 
554 	return error;
555 }
556 
557 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
558 				size_t count, loff_t *ppos)
559 {
560 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
561 }
562 
563 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
564 				size_t count, loff_t *ppos)
565 {
566 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
567 }
568 
569 static int proc_sys_open(struct inode *inode, struct file *filp)
570 {
571 	struct ctl_table_header *head = grab_header(inode);
572 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
573 
574 	/* sysctl was unregistered */
575 	if (IS_ERR(head))
576 		return PTR_ERR(head);
577 
578 	if (table->poll)
579 		filp->private_data = proc_sys_poll_event(table->poll);
580 
581 	sysctl_head_finish(head);
582 
583 	return 0;
584 }
585 
586 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
587 {
588 	struct inode *inode = file_inode(filp);
589 	struct ctl_table_header *head = grab_header(inode);
590 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
591 	unsigned int ret = DEFAULT_POLLMASK;
592 	unsigned long event;
593 
594 	/* sysctl was unregistered */
595 	if (IS_ERR(head))
596 		return POLLERR | POLLHUP;
597 
598 	if (!table->proc_handler)
599 		goto out;
600 
601 	if (!table->poll)
602 		goto out;
603 
604 	event = (unsigned long)filp->private_data;
605 	poll_wait(filp, &table->poll->wait, wait);
606 
607 	if (event != atomic_read(&table->poll->event)) {
608 		filp->private_data = proc_sys_poll_event(table->poll);
609 		ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
610 	}
611 
612 out:
613 	sysctl_head_finish(head);
614 
615 	return ret;
616 }
617 
618 static bool proc_sys_fill_cache(struct file *file,
619 				struct dir_context *ctx,
620 				struct ctl_table_header *head,
621 				struct ctl_table *table)
622 {
623 	struct dentry *child, *dir = file->f_path.dentry;
624 	struct inode *inode;
625 	struct qstr qname;
626 	ino_t ino = 0;
627 	unsigned type = DT_UNKNOWN;
628 
629 	qname.name = table->procname;
630 	qname.len  = strlen(table->procname);
631 	qname.hash = full_name_hash(dir, qname.name, qname.len);
632 
633 	child = d_lookup(dir, &qname);
634 	if (!child) {
635 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
636 		child = d_alloc_parallel(dir, &qname, &wq);
637 		if (IS_ERR(child))
638 			return false;
639 		if (d_in_lookup(child)) {
640 			inode = proc_sys_make_inode(dir->d_sb, head, table);
641 			if (!inode) {
642 				d_lookup_done(child);
643 				dput(child);
644 				return false;
645 			}
646 			d_set_d_op(child, &proc_sys_dentry_operations);
647 			d_add(child, inode);
648 		}
649 	}
650 	inode = d_inode(child);
651 	ino  = inode->i_ino;
652 	type = inode->i_mode >> 12;
653 	dput(child);
654 	return dir_emit(ctx, qname.name, qname.len, ino, type);
655 }
656 
657 static bool proc_sys_link_fill_cache(struct file *file,
658 				    struct dir_context *ctx,
659 				    struct ctl_table_header *head,
660 				    struct ctl_table *table)
661 {
662 	bool ret = true;
663 	head = sysctl_head_grab(head);
664 
665 	if (S_ISLNK(table->mode)) {
666 		/* It is not an error if we can not follow the link ignore it */
667 		int err = sysctl_follow_link(&head, &table);
668 		if (err)
669 			goto out;
670 	}
671 
672 	ret = proc_sys_fill_cache(file, ctx, head, table);
673 out:
674 	sysctl_head_finish(head);
675 	return ret;
676 }
677 
678 static int scan(struct ctl_table_header *head, struct ctl_table *table,
679 		unsigned long *pos, struct file *file,
680 		struct dir_context *ctx)
681 {
682 	bool res;
683 
684 	if ((*pos)++ < ctx->pos)
685 		return true;
686 
687 	if (unlikely(S_ISLNK(table->mode)))
688 		res = proc_sys_link_fill_cache(file, ctx, head, table);
689 	else
690 		res = proc_sys_fill_cache(file, ctx, head, table);
691 
692 	if (res)
693 		ctx->pos = *pos;
694 
695 	return res;
696 }
697 
698 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
699 {
700 	struct ctl_table_header *head = grab_header(file_inode(file));
701 	struct ctl_table_header *h = NULL;
702 	struct ctl_table *entry;
703 	struct ctl_dir *ctl_dir;
704 	unsigned long pos;
705 
706 	if (IS_ERR(head))
707 		return PTR_ERR(head);
708 
709 	ctl_dir = container_of(head, struct ctl_dir, header);
710 
711 	if (!dir_emit_dots(file, ctx))
712 		return 0;
713 
714 	pos = 2;
715 
716 	for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
717 		if (!scan(h, entry, &pos, file, ctx)) {
718 			sysctl_head_finish(h);
719 			break;
720 		}
721 	}
722 	sysctl_head_finish(head);
723 	return 0;
724 }
725 
726 static int proc_sys_permission(struct inode *inode, int mask)
727 {
728 	/*
729 	 * sysctl entries that are not writeable,
730 	 * are _NOT_ writeable, capabilities or not.
731 	 */
732 	struct ctl_table_header *head;
733 	struct ctl_table *table;
734 	int error;
735 
736 	/* Executable files are not allowed under /proc/sys/ */
737 	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
738 		return -EACCES;
739 
740 	head = grab_header(inode);
741 	if (IS_ERR(head))
742 		return PTR_ERR(head);
743 
744 	table = PROC_I(inode)->sysctl_entry;
745 	if (!table) /* global root - r-xr-xr-x */
746 		error = mask & MAY_WRITE ? -EACCES : 0;
747 	else /* Use the permissions on the sysctl table entry */
748 		error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
749 
750 	sysctl_head_finish(head);
751 	return error;
752 }
753 
754 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
755 {
756 	struct inode *inode = d_inode(dentry);
757 	int error;
758 
759 	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
760 		return -EPERM;
761 
762 	error = setattr_prepare(dentry, attr);
763 	if (error)
764 		return error;
765 
766 	setattr_copy(inode, attr);
767 	mark_inode_dirty(inode);
768 	return 0;
769 }
770 
771 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
772 {
773 	struct inode *inode = d_inode(dentry);
774 	struct ctl_table_header *head = grab_header(inode);
775 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
776 
777 	if (IS_ERR(head))
778 		return PTR_ERR(head);
779 
780 	generic_fillattr(inode, stat);
781 	if (table)
782 		stat->mode = (stat->mode & S_IFMT) | table->mode;
783 
784 	sysctl_head_finish(head);
785 	return 0;
786 }
787 
788 static const struct file_operations proc_sys_file_operations = {
789 	.open		= proc_sys_open,
790 	.poll		= proc_sys_poll,
791 	.read		= proc_sys_read,
792 	.write		= proc_sys_write,
793 	.llseek		= default_llseek,
794 };
795 
796 static const struct file_operations proc_sys_dir_file_operations = {
797 	.read		= generic_read_dir,
798 	.iterate_shared	= proc_sys_readdir,
799 	.llseek		= generic_file_llseek,
800 };
801 
802 static const struct inode_operations proc_sys_inode_operations = {
803 	.permission	= proc_sys_permission,
804 	.setattr	= proc_sys_setattr,
805 	.getattr	= proc_sys_getattr,
806 };
807 
808 static const struct inode_operations proc_sys_dir_operations = {
809 	.lookup		= proc_sys_lookup,
810 	.permission	= proc_sys_permission,
811 	.setattr	= proc_sys_setattr,
812 	.getattr	= proc_sys_getattr,
813 };
814 
815 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
816 {
817 	if (flags & LOOKUP_RCU)
818 		return -ECHILD;
819 	return !PROC_I(d_inode(dentry))->sysctl->unregistering;
820 }
821 
822 static int proc_sys_delete(const struct dentry *dentry)
823 {
824 	return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
825 }
826 
827 static int sysctl_is_seen(struct ctl_table_header *p)
828 {
829 	struct ctl_table_set *set = p->set;
830 	int res;
831 	spin_lock(&sysctl_lock);
832 	if (p->unregistering)
833 		res = 0;
834 	else if (!set->is_seen)
835 		res = 1;
836 	else
837 		res = set->is_seen(set);
838 	spin_unlock(&sysctl_lock);
839 	return res;
840 }
841 
842 static int proc_sys_compare(const struct dentry *dentry,
843 		unsigned int len, const char *str, const struct qstr *name)
844 {
845 	struct ctl_table_header *head;
846 	struct inode *inode;
847 
848 	/* Although proc doesn't have negative dentries, rcu-walk means
849 	 * that inode here can be NULL */
850 	/* AV: can it, indeed? */
851 	inode = d_inode_rcu(dentry);
852 	if (!inode)
853 		return 1;
854 	if (name->len != len)
855 		return 1;
856 	if (memcmp(name->name, str, len))
857 		return 1;
858 	head = rcu_dereference(PROC_I(inode)->sysctl);
859 	return !head || !sysctl_is_seen(head);
860 }
861 
862 static const struct dentry_operations proc_sys_dentry_operations = {
863 	.d_revalidate	= proc_sys_revalidate,
864 	.d_delete	= proc_sys_delete,
865 	.d_compare	= proc_sys_compare,
866 };
867 
868 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
869 				   const char *name, int namelen)
870 {
871 	struct ctl_table_header *head;
872 	struct ctl_table *entry;
873 
874 	entry = find_entry(&head, dir, name, namelen);
875 	if (!entry)
876 		return ERR_PTR(-ENOENT);
877 	if (!S_ISDIR(entry->mode))
878 		return ERR_PTR(-ENOTDIR);
879 	return container_of(head, struct ctl_dir, header);
880 }
881 
882 static struct ctl_dir *new_dir(struct ctl_table_set *set,
883 			       const char *name, int namelen)
884 {
885 	struct ctl_table *table;
886 	struct ctl_dir *new;
887 	struct ctl_node *node;
888 	char *new_name;
889 
890 	new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
891 		      sizeof(struct ctl_table)*2 +  namelen + 1,
892 		      GFP_KERNEL);
893 	if (!new)
894 		return NULL;
895 
896 	node = (struct ctl_node *)(new + 1);
897 	table = (struct ctl_table *)(node + 1);
898 	new_name = (char *)(table + 2);
899 	memcpy(new_name, name, namelen);
900 	new_name[namelen] = '\0';
901 	table[0].procname = new_name;
902 	table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
903 	init_header(&new->header, set->dir.header.root, set, node, table);
904 
905 	return new;
906 }
907 
908 /**
909  * get_subdir - find or create a subdir with the specified name.
910  * @dir:  Directory to create the subdirectory in
911  * @name: The name of the subdirectory to find or create
912  * @namelen: The length of name
913  *
914  * Takes a directory with an elevated reference count so we know that
915  * if we drop the lock the directory will not go away.  Upon success
916  * the reference is moved from @dir to the returned subdirectory.
917  * Upon error an error code is returned and the reference on @dir is
918  * simply dropped.
919  */
920 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
921 				  const char *name, int namelen)
922 {
923 	struct ctl_table_set *set = dir->header.set;
924 	struct ctl_dir *subdir, *new = NULL;
925 	int err;
926 
927 	spin_lock(&sysctl_lock);
928 	subdir = find_subdir(dir, name, namelen);
929 	if (!IS_ERR(subdir))
930 		goto found;
931 	if (PTR_ERR(subdir) != -ENOENT)
932 		goto failed;
933 
934 	spin_unlock(&sysctl_lock);
935 	new = new_dir(set, name, namelen);
936 	spin_lock(&sysctl_lock);
937 	subdir = ERR_PTR(-ENOMEM);
938 	if (!new)
939 		goto failed;
940 
941 	/* Was the subdir added while we dropped the lock? */
942 	subdir = find_subdir(dir, name, namelen);
943 	if (!IS_ERR(subdir))
944 		goto found;
945 	if (PTR_ERR(subdir) != -ENOENT)
946 		goto failed;
947 
948 	/* Nope.  Use the our freshly made directory entry. */
949 	err = insert_header(dir, &new->header);
950 	subdir = ERR_PTR(err);
951 	if (err)
952 		goto failed;
953 	subdir = new;
954 found:
955 	subdir->header.nreg++;
956 failed:
957 	if (IS_ERR(subdir)) {
958 		pr_err("sysctl could not get directory: ");
959 		sysctl_print_dir(dir);
960 		pr_cont("/%*.*s %ld\n",
961 			namelen, namelen, name, PTR_ERR(subdir));
962 	}
963 	drop_sysctl_table(&dir->header);
964 	if (new)
965 		drop_sysctl_table(&new->header);
966 	spin_unlock(&sysctl_lock);
967 	return subdir;
968 }
969 
970 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
971 {
972 	struct ctl_dir *parent;
973 	const char *procname;
974 	if (!dir->header.parent)
975 		return &set->dir;
976 	parent = xlate_dir(set, dir->header.parent);
977 	if (IS_ERR(parent))
978 		return parent;
979 	procname = dir->header.ctl_table[0].procname;
980 	return find_subdir(parent, procname, strlen(procname));
981 }
982 
983 static int sysctl_follow_link(struct ctl_table_header **phead,
984 	struct ctl_table **pentry)
985 {
986 	struct ctl_table_header *head;
987 	struct ctl_table_root *root;
988 	struct ctl_table_set *set;
989 	struct ctl_table *entry;
990 	struct ctl_dir *dir;
991 	int ret;
992 
993 	ret = 0;
994 	spin_lock(&sysctl_lock);
995 	root = (*pentry)->data;
996 	set = lookup_header_set(root);
997 	dir = xlate_dir(set, (*phead)->parent);
998 	if (IS_ERR(dir))
999 		ret = PTR_ERR(dir);
1000 	else {
1001 		const char *procname = (*pentry)->procname;
1002 		head = NULL;
1003 		entry = find_entry(&head, dir, procname, strlen(procname));
1004 		ret = -ENOENT;
1005 		if (entry && use_table(head)) {
1006 			unuse_table(*phead);
1007 			*phead = head;
1008 			*pentry = entry;
1009 			ret = 0;
1010 		}
1011 	}
1012 
1013 	spin_unlock(&sysctl_lock);
1014 	return ret;
1015 }
1016 
1017 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1018 {
1019 	struct va_format vaf;
1020 	va_list args;
1021 
1022 	va_start(args, fmt);
1023 	vaf.fmt = fmt;
1024 	vaf.va = &args;
1025 
1026 	pr_err("sysctl table check failed: %s/%s %pV\n",
1027 	       path, table->procname, &vaf);
1028 
1029 	va_end(args);
1030 	return -EINVAL;
1031 }
1032 
1033 static int sysctl_check_table(const char *path, struct ctl_table *table)
1034 {
1035 	int err = 0;
1036 	for (; table->procname; table++) {
1037 		if (table->child)
1038 			err = sysctl_err(path, table, "Not a file");
1039 
1040 		if ((table->proc_handler == proc_dostring) ||
1041 		    (table->proc_handler == proc_dointvec) ||
1042 		    (table->proc_handler == proc_dointvec_minmax) ||
1043 		    (table->proc_handler == proc_dointvec_jiffies) ||
1044 		    (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1045 		    (table->proc_handler == proc_dointvec_ms_jiffies) ||
1046 		    (table->proc_handler == proc_doulongvec_minmax) ||
1047 		    (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1048 			if (!table->data)
1049 				err = sysctl_err(path, table, "No data");
1050 			if (!table->maxlen)
1051 				err = sysctl_err(path, table, "No maxlen");
1052 		}
1053 		if (!table->proc_handler)
1054 			err = sysctl_err(path, table, "No proc_handler");
1055 
1056 		if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1057 			err = sysctl_err(path, table, "bogus .mode 0%o",
1058 				table->mode);
1059 	}
1060 	return err;
1061 }
1062 
1063 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1064 	struct ctl_table_root *link_root)
1065 {
1066 	struct ctl_table *link_table, *entry, *link;
1067 	struct ctl_table_header *links;
1068 	struct ctl_node *node;
1069 	char *link_name;
1070 	int nr_entries, name_bytes;
1071 
1072 	name_bytes = 0;
1073 	nr_entries = 0;
1074 	for (entry = table; entry->procname; entry++) {
1075 		nr_entries++;
1076 		name_bytes += strlen(entry->procname) + 1;
1077 	}
1078 
1079 	links = kzalloc(sizeof(struct ctl_table_header) +
1080 			sizeof(struct ctl_node)*nr_entries +
1081 			sizeof(struct ctl_table)*(nr_entries + 1) +
1082 			name_bytes,
1083 			GFP_KERNEL);
1084 
1085 	if (!links)
1086 		return NULL;
1087 
1088 	node = (struct ctl_node *)(links + 1);
1089 	link_table = (struct ctl_table *)(node + nr_entries);
1090 	link_name = (char *)&link_table[nr_entries + 1];
1091 
1092 	for (link = link_table, entry = table; entry->procname; link++, entry++) {
1093 		int len = strlen(entry->procname) + 1;
1094 		memcpy(link_name, entry->procname, len);
1095 		link->procname = link_name;
1096 		link->mode = S_IFLNK|S_IRWXUGO;
1097 		link->data = link_root;
1098 		link_name += len;
1099 	}
1100 	init_header(links, dir->header.root, dir->header.set, node, link_table);
1101 	links->nreg = nr_entries;
1102 
1103 	return links;
1104 }
1105 
1106 static bool get_links(struct ctl_dir *dir,
1107 	struct ctl_table *table, struct ctl_table_root *link_root)
1108 {
1109 	struct ctl_table_header *head;
1110 	struct ctl_table *entry, *link;
1111 
1112 	/* Are there links available for every entry in table? */
1113 	for (entry = table; entry->procname; entry++) {
1114 		const char *procname = entry->procname;
1115 		link = find_entry(&head, dir, procname, strlen(procname));
1116 		if (!link)
1117 			return false;
1118 		if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1119 			continue;
1120 		if (S_ISLNK(link->mode) && (link->data == link_root))
1121 			continue;
1122 		return false;
1123 	}
1124 
1125 	/* The checks passed.  Increase the registration count on the links */
1126 	for (entry = table; entry->procname; entry++) {
1127 		const char *procname = entry->procname;
1128 		link = find_entry(&head, dir, procname, strlen(procname));
1129 		head->nreg++;
1130 	}
1131 	return true;
1132 }
1133 
1134 static int insert_links(struct ctl_table_header *head)
1135 {
1136 	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1137 	struct ctl_dir *core_parent = NULL;
1138 	struct ctl_table_header *links;
1139 	int err;
1140 
1141 	if (head->set == root_set)
1142 		return 0;
1143 
1144 	core_parent = xlate_dir(root_set, head->parent);
1145 	if (IS_ERR(core_parent))
1146 		return 0;
1147 
1148 	if (get_links(core_parent, head->ctl_table, head->root))
1149 		return 0;
1150 
1151 	core_parent->header.nreg++;
1152 	spin_unlock(&sysctl_lock);
1153 
1154 	links = new_links(core_parent, head->ctl_table, head->root);
1155 
1156 	spin_lock(&sysctl_lock);
1157 	err = -ENOMEM;
1158 	if (!links)
1159 		goto out;
1160 
1161 	err = 0;
1162 	if (get_links(core_parent, head->ctl_table, head->root)) {
1163 		kfree(links);
1164 		goto out;
1165 	}
1166 
1167 	err = insert_header(core_parent, links);
1168 	if (err)
1169 		kfree(links);
1170 out:
1171 	drop_sysctl_table(&core_parent->header);
1172 	return err;
1173 }
1174 
1175 /**
1176  * __register_sysctl_table - register a leaf sysctl table
1177  * @set: Sysctl tree to register on
1178  * @path: The path to the directory the sysctl table is in.
1179  * @table: the top-level table structure
1180  *
1181  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1182  * array. A completely 0 filled entry terminates the table.
1183  *
1184  * The members of the &struct ctl_table structure are used as follows:
1185  *
1186  * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1187  *            enter a sysctl file
1188  *
1189  * data - a pointer to data for use by proc_handler
1190  *
1191  * maxlen - the maximum size in bytes of the data
1192  *
1193  * mode - the file permissions for the /proc/sys file
1194  *
1195  * child - must be %NULL.
1196  *
1197  * proc_handler - the text handler routine (described below)
1198  *
1199  * extra1, extra2 - extra pointers usable by the proc handler routines
1200  *
1201  * Leaf nodes in the sysctl tree will be represented by a single file
1202  * under /proc; non-leaf nodes will be represented by directories.
1203  *
1204  * There must be a proc_handler routine for any terminal nodes.
1205  * Several default handlers are available to cover common cases -
1206  *
1207  * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1208  * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1209  * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1210  *
1211  * It is the handler's job to read the input buffer from user memory
1212  * and process it. The handler should return 0 on success.
1213  *
1214  * This routine returns %NULL on a failure to register, and a pointer
1215  * to the table header on success.
1216  */
1217 struct ctl_table_header *__register_sysctl_table(
1218 	struct ctl_table_set *set,
1219 	const char *path, struct ctl_table *table)
1220 {
1221 	struct ctl_table_root *root = set->dir.header.root;
1222 	struct ctl_table_header *header;
1223 	const char *name, *nextname;
1224 	struct ctl_dir *dir;
1225 	struct ctl_table *entry;
1226 	struct ctl_node *node;
1227 	int nr_entries = 0;
1228 
1229 	for (entry = table; entry->procname; entry++)
1230 		nr_entries++;
1231 
1232 	header = kzalloc(sizeof(struct ctl_table_header) +
1233 			 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1234 	if (!header)
1235 		return NULL;
1236 
1237 	node = (struct ctl_node *)(header + 1);
1238 	init_header(header, root, set, node, table);
1239 	if (sysctl_check_table(path, table))
1240 		goto fail;
1241 
1242 	spin_lock(&sysctl_lock);
1243 	dir = &set->dir;
1244 	/* Reference moved down the diretory tree get_subdir */
1245 	dir->header.nreg++;
1246 	spin_unlock(&sysctl_lock);
1247 
1248 	/* Find the directory for the ctl_table */
1249 	for (name = path; name; name = nextname) {
1250 		int namelen;
1251 		nextname = strchr(name, '/');
1252 		if (nextname) {
1253 			namelen = nextname - name;
1254 			nextname++;
1255 		} else {
1256 			namelen = strlen(name);
1257 		}
1258 		if (namelen == 0)
1259 			continue;
1260 
1261 		dir = get_subdir(dir, name, namelen);
1262 		if (IS_ERR(dir))
1263 			goto fail;
1264 	}
1265 
1266 	spin_lock(&sysctl_lock);
1267 	if (insert_header(dir, header))
1268 		goto fail_put_dir_locked;
1269 
1270 	drop_sysctl_table(&dir->header);
1271 	spin_unlock(&sysctl_lock);
1272 
1273 	return header;
1274 
1275 fail_put_dir_locked:
1276 	drop_sysctl_table(&dir->header);
1277 	spin_unlock(&sysctl_lock);
1278 fail:
1279 	kfree(header);
1280 	dump_stack();
1281 	return NULL;
1282 }
1283 
1284 /**
1285  * register_sysctl - register a sysctl table
1286  * @path: The path to the directory the sysctl table is in.
1287  * @table: the table structure
1288  *
1289  * Register a sysctl table. @table should be a filled in ctl_table
1290  * array. A completely 0 filled entry terminates the table.
1291  *
1292  * See __register_sysctl_table for more details.
1293  */
1294 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1295 {
1296 	return __register_sysctl_table(&sysctl_table_root.default_set,
1297 					path, table);
1298 }
1299 EXPORT_SYMBOL(register_sysctl);
1300 
1301 static char *append_path(const char *path, char *pos, const char *name)
1302 {
1303 	int namelen;
1304 	namelen = strlen(name);
1305 	if (((pos - path) + namelen + 2) >= PATH_MAX)
1306 		return NULL;
1307 	memcpy(pos, name, namelen);
1308 	pos[namelen] = '/';
1309 	pos[namelen + 1] = '\0';
1310 	pos += namelen + 1;
1311 	return pos;
1312 }
1313 
1314 static int count_subheaders(struct ctl_table *table)
1315 {
1316 	int has_files = 0;
1317 	int nr_subheaders = 0;
1318 	struct ctl_table *entry;
1319 
1320 	/* special case: no directory and empty directory */
1321 	if (!table || !table->procname)
1322 		return 1;
1323 
1324 	for (entry = table; entry->procname; entry++) {
1325 		if (entry->child)
1326 			nr_subheaders += count_subheaders(entry->child);
1327 		else
1328 			has_files = 1;
1329 	}
1330 	return nr_subheaders + has_files;
1331 }
1332 
1333 static int register_leaf_sysctl_tables(const char *path, char *pos,
1334 	struct ctl_table_header ***subheader, struct ctl_table_set *set,
1335 	struct ctl_table *table)
1336 {
1337 	struct ctl_table *ctl_table_arg = NULL;
1338 	struct ctl_table *entry, *files;
1339 	int nr_files = 0;
1340 	int nr_dirs = 0;
1341 	int err = -ENOMEM;
1342 
1343 	for (entry = table; entry->procname; entry++) {
1344 		if (entry->child)
1345 			nr_dirs++;
1346 		else
1347 			nr_files++;
1348 	}
1349 
1350 	files = table;
1351 	/* If there are mixed files and directories we need a new table */
1352 	if (nr_dirs && nr_files) {
1353 		struct ctl_table *new;
1354 		files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1355 				GFP_KERNEL);
1356 		if (!files)
1357 			goto out;
1358 
1359 		ctl_table_arg = files;
1360 		for (new = files, entry = table; entry->procname; entry++) {
1361 			if (entry->child)
1362 				continue;
1363 			*new = *entry;
1364 			new++;
1365 		}
1366 	}
1367 
1368 	/* Register everything except a directory full of subdirectories */
1369 	if (nr_files || !nr_dirs) {
1370 		struct ctl_table_header *header;
1371 		header = __register_sysctl_table(set, path, files);
1372 		if (!header) {
1373 			kfree(ctl_table_arg);
1374 			goto out;
1375 		}
1376 
1377 		/* Remember if we need to free the file table */
1378 		header->ctl_table_arg = ctl_table_arg;
1379 		**subheader = header;
1380 		(*subheader)++;
1381 	}
1382 
1383 	/* Recurse into the subdirectories. */
1384 	for (entry = table; entry->procname; entry++) {
1385 		char *child_pos;
1386 
1387 		if (!entry->child)
1388 			continue;
1389 
1390 		err = -ENAMETOOLONG;
1391 		child_pos = append_path(path, pos, entry->procname);
1392 		if (!child_pos)
1393 			goto out;
1394 
1395 		err = register_leaf_sysctl_tables(path, child_pos, subheader,
1396 						  set, entry->child);
1397 		pos[0] = '\0';
1398 		if (err)
1399 			goto out;
1400 	}
1401 	err = 0;
1402 out:
1403 	/* On failure our caller will unregister all registered subheaders */
1404 	return err;
1405 }
1406 
1407 /**
1408  * __register_sysctl_paths - register a sysctl table hierarchy
1409  * @set: Sysctl tree to register on
1410  * @path: The path to the directory the sysctl table is in.
1411  * @table: the top-level table structure
1412  *
1413  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1414  * array. A completely 0 filled entry terminates the table.
1415  *
1416  * See __register_sysctl_table for more details.
1417  */
1418 struct ctl_table_header *__register_sysctl_paths(
1419 	struct ctl_table_set *set,
1420 	const struct ctl_path *path, struct ctl_table *table)
1421 {
1422 	struct ctl_table *ctl_table_arg = table;
1423 	int nr_subheaders = count_subheaders(table);
1424 	struct ctl_table_header *header = NULL, **subheaders, **subheader;
1425 	const struct ctl_path *component;
1426 	char *new_path, *pos;
1427 
1428 	pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1429 	if (!new_path)
1430 		return NULL;
1431 
1432 	pos[0] = '\0';
1433 	for (component = path; component->procname; component++) {
1434 		pos = append_path(new_path, pos, component->procname);
1435 		if (!pos)
1436 			goto out;
1437 	}
1438 	while (table->procname && table->child && !table[1].procname) {
1439 		pos = append_path(new_path, pos, table->procname);
1440 		if (!pos)
1441 			goto out;
1442 		table = table->child;
1443 	}
1444 	if (nr_subheaders == 1) {
1445 		header = __register_sysctl_table(set, new_path, table);
1446 		if (header)
1447 			header->ctl_table_arg = ctl_table_arg;
1448 	} else {
1449 		header = kzalloc(sizeof(*header) +
1450 				 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1451 		if (!header)
1452 			goto out;
1453 
1454 		subheaders = (struct ctl_table_header **) (header + 1);
1455 		subheader = subheaders;
1456 		header->ctl_table_arg = ctl_table_arg;
1457 
1458 		if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1459 						set, table))
1460 			goto err_register_leaves;
1461 	}
1462 
1463 out:
1464 	kfree(new_path);
1465 	return header;
1466 
1467 err_register_leaves:
1468 	while (subheader > subheaders) {
1469 		struct ctl_table_header *subh = *(--subheader);
1470 		struct ctl_table *table = subh->ctl_table_arg;
1471 		unregister_sysctl_table(subh);
1472 		kfree(table);
1473 	}
1474 	kfree(header);
1475 	header = NULL;
1476 	goto out;
1477 }
1478 
1479 /**
1480  * register_sysctl_table_path - register a sysctl table hierarchy
1481  * @path: The path to the directory the sysctl table is in.
1482  * @table: the top-level table structure
1483  *
1484  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1485  * array. A completely 0 filled entry terminates the table.
1486  *
1487  * See __register_sysctl_paths for more details.
1488  */
1489 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1490 						struct ctl_table *table)
1491 {
1492 	return __register_sysctl_paths(&sysctl_table_root.default_set,
1493 					path, table);
1494 }
1495 EXPORT_SYMBOL(register_sysctl_paths);
1496 
1497 /**
1498  * register_sysctl_table - register a sysctl table hierarchy
1499  * @table: the top-level table structure
1500  *
1501  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1502  * array. A completely 0 filled entry terminates the table.
1503  *
1504  * See register_sysctl_paths for more details.
1505  */
1506 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1507 {
1508 	static const struct ctl_path null_path[] = { {} };
1509 
1510 	return register_sysctl_paths(null_path, table);
1511 }
1512 EXPORT_SYMBOL(register_sysctl_table);
1513 
1514 static void put_links(struct ctl_table_header *header)
1515 {
1516 	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1517 	struct ctl_table_root *root = header->root;
1518 	struct ctl_dir *parent = header->parent;
1519 	struct ctl_dir *core_parent;
1520 	struct ctl_table *entry;
1521 
1522 	if (header->set == root_set)
1523 		return;
1524 
1525 	core_parent = xlate_dir(root_set, parent);
1526 	if (IS_ERR(core_parent))
1527 		return;
1528 
1529 	for (entry = header->ctl_table; entry->procname; entry++) {
1530 		struct ctl_table_header *link_head;
1531 		struct ctl_table *link;
1532 		const char *name = entry->procname;
1533 
1534 		link = find_entry(&link_head, core_parent, name, strlen(name));
1535 		if (link &&
1536 		    ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1537 		     (S_ISLNK(link->mode) && (link->data == root)))) {
1538 			drop_sysctl_table(link_head);
1539 		}
1540 		else {
1541 			pr_err("sysctl link missing during unregister: ");
1542 			sysctl_print_dir(parent);
1543 			pr_cont("/%s\n", name);
1544 		}
1545 	}
1546 }
1547 
1548 static void drop_sysctl_table(struct ctl_table_header *header)
1549 {
1550 	struct ctl_dir *parent = header->parent;
1551 
1552 	if (--header->nreg)
1553 		return;
1554 
1555 	put_links(header);
1556 	start_unregistering(header);
1557 	if (!--header->count)
1558 		kfree_rcu(header, rcu);
1559 
1560 	if (parent)
1561 		drop_sysctl_table(&parent->header);
1562 }
1563 
1564 /**
1565  * unregister_sysctl_table - unregister a sysctl table hierarchy
1566  * @header: the header returned from register_sysctl_table
1567  *
1568  * Unregisters the sysctl table and all children. proc entries may not
1569  * actually be removed until they are no longer used by anyone.
1570  */
1571 void unregister_sysctl_table(struct ctl_table_header * header)
1572 {
1573 	int nr_subheaders;
1574 	might_sleep();
1575 
1576 	if (header == NULL)
1577 		return;
1578 
1579 	nr_subheaders = count_subheaders(header->ctl_table_arg);
1580 	if (unlikely(nr_subheaders > 1)) {
1581 		struct ctl_table_header **subheaders;
1582 		int i;
1583 
1584 		subheaders = (struct ctl_table_header **)(header + 1);
1585 		for (i = nr_subheaders -1; i >= 0; i--) {
1586 			struct ctl_table_header *subh = subheaders[i];
1587 			struct ctl_table *table = subh->ctl_table_arg;
1588 			unregister_sysctl_table(subh);
1589 			kfree(table);
1590 		}
1591 		kfree(header);
1592 		return;
1593 	}
1594 
1595 	spin_lock(&sysctl_lock);
1596 	drop_sysctl_table(header);
1597 	spin_unlock(&sysctl_lock);
1598 }
1599 EXPORT_SYMBOL(unregister_sysctl_table);
1600 
1601 void setup_sysctl_set(struct ctl_table_set *set,
1602 	struct ctl_table_root *root,
1603 	int (*is_seen)(struct ctl_table_set *))
1604 {
1605 	memset(set, 0, sizeof(*set));
1606 	set->is_seen = is_seen;
1607 	init_header(&set->dir.header, root, set, NULL, root_table);
1608 }
1609 
1610 void retire_sysctl_set(struct ctl_table_set *set)
1611 {
1612 	WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1613 }
1614 
1615 int __init proc_sys_init(void)
1616 {
1617 	struct proc_dir_entry *proc_sys_root;
1618 
1619 	proc_sys_root = proc_mkdir("sys", NULL);
1620 	proc_sys_root->proc_iops = &proc_sys_dir_operations;
1621 	proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1622 	proc_sys_root->nlink = 0;
1623 
1624 	return sysctl_init();
1625 }
1626