xref: /linux/fs/proc/proc_sysctl.c (revision 4e1a33b105ddf201f66dcc44490c6086a25eca0b)
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 	INIT_LIST_HEAD(&head->inodes);
194 	if (node) {
195 		struct ctl_table *entry;
196 		for (entry = table; entry->procname; entry++, node++)
197 			node->header = head;
198 	}
199 }
200 
201 static void erase_header(struct ctl_table_header *head)
202 {
203 	struct ctl_table *entry;
204 	for (entry = head->ctl_table; entry->procname; entry++)
205 		erase_entry(head, entry);
206 }
207 
208 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
209 {
210 	struct ctl_table *entry;
211 	int err;
212 
213 	/* Is this a permanently empty directory? */
214 	if (is_empty_dir(&dir->header))
215 		return -EROFS;
216 
217 	/* Am I creating a permanently empty directory? */
218 	if (header->ctl_table == sysctl_mount_point) {
219 		if (!RB_EMPTY_ROOT(&dir->root))
220 			return -EINVAL;
221 		set_empty_dir(dir);
222 	}
223 
224 	dir->header.nreg++;
225 	header->parent = dir;
226 	err = insert_links(header);
227 	if (err)
228 		goto fail_links;
229 	for (entry = header->ctl_table; entry->procname; entry++) {
230 		err = insert_entry(header, entry);
231 		if (err)
232 			goto fail;
233 	}
234 	return 0;
235 fail:
236 	erase_header(header);
237 	put_links(header);
238 fail_links:
239 	if (header->ctl_table == sysctl_mount_point)
240 		clear_empty_dir(dir);
241 	header->parent = NULL;
242 	drop_sysctl_table(&dir->header);
243 	return err;
244 }
245 
246 /* called under sysctl_lock */
247 static int use_table(struct ctl_table_header *p)
248 {
249 	if (unlikely(p->unregistering))
250 		return 0;
251 	p->used++;
252 	return 1;
253 }
254 
255 /* called under sysctl_lock */
256 static void unuse_table(struct ctl_table_header *p)
257 {
258 	if (!--p->used)
259 		if (unlikely(p->unregistering))
260 			complete(p->unregistering);
261 }
262 
263 /* called under sysctl_lock */
264 static void proc_sys_prune_dcache(struct ctl_table_header *head)
265 {
266 	struct inode *inode, *prev = NULL;
267 	struct proc_inode *ei;
268 
269 	rcu_read_lock();
270 	list_for_each_entry_rcu(ei, &head->inodes, sysctl_inodes) {
271 		inode = igrab(&ei->vfs_inode);
272 		if (inode) {
273 			rcu_read_unlock();
274 			iput(prev);
275 			prev = inode;
276 			d_prune_aliases(inode);
277 			rcu_read_lock();
278 		}
279 	}
280 	rcu_read_unlock();
281 	iput(prev);
282 }
283 
284 /* called under sysctl_lock, will reacquire if has to wait */
285 static void start_unregistering(struct ctl_table_header *p)
286 {
287 	/*
288 	 * if p->used is 0, nobody will ever touch that entry again;
289 	 * we'll eliminate all paths to it before dropping sysctl_lock
290 	 */
291 	if (unlikely(p->used)) {
292 		struct completion wait;
293 		init_completion(&wait);
294 		p->unregistering = &wait;
295 		spin_unlock(&sysctl_lock);
296 		wait_for_completion(&wait);
297 	} else {
298 		/* anything non-NULL; we'll never dereference it */
299 		p->unregistering = ERR_PTR(-EINVAL);
300 		spin_unlock(&sysctl_lock);
301 	}
302 	/*
303 	 * Prune dentries for unregistered sysctls: namespaced sysctls
304 	 * can have duplicate names and contaminate dcache very badly.
305 	 */
306 	proc_sys_prune_dcache(p);
307 	/*
308 	 * do not remove from the list until nobody holds it; walking the
309 	 * list in do_sysctl() relies on that.
310 	 */
311 	spin_lock(&sysctl_lock);
312 	erase_header(p);
313 }
314 
315 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
316 {
317 	BUG_ON(!head);
318 	spin_lock(&sysctl_lock);
319 	if (!use_table(head))
320 		head = ERR_PTR(-ENOENT);
321 	spin_unlock(&sysctl_lock);
322 	return head;
323 }
324 
325 static void sysctl_head_finish(struct ctl_table_header *head)
326 {
327 	if (!head)
328 		return;
329 	spin_lock(&sysctl_lock);
330 	unuse_table(head);
331 	spin_unlock(&sysctl_lock);
332 }
333 
334 static struct ctl_table_set *
335 lookup_header_set(struct ctl_table_root *root)
336 {
337 	struct ctl_table_set *set = &root->default_set;
338 	if (root->lookup)
339 		set = root->lookup(root);
340 	return set;
341 }
342 
343 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
344 				      struct ctl_dir *dir,
345 				      const char *name, int namelen)
346 {
347 	struct ctl_table_header *head;
348 	struct ctl_table *entry;
349 
350 	spin_lock(&sysctl_lock);
351 	entry = find_entry(&head, dir, name, namelen);
352 	if (entry && use_table(head))
353 		*phead = head;
354 	else
355 		entry = NULL;
356 	spin_unlock(&sysctl_lock);
357 	return entry;
358 }
359 
360 static struct ctl_node *first_usable_entry(struct rb_node *node)
361 {
362 	struct ctl_node *ctl_node;
363 
364 	for (;node; node = rb_next(node)) {
365 		ctl_node = rb_entry(node, struct ctl_node, node);
366 		if (use_table(ctl_node->header))
367 			return ctl_node;
368 	}
369 	return NULL;
370 }
371 
372 static void first_entry(struct ctl_dir *dir,
373 	struct ctl_table_header **phead, struct ctl_table **pentry)
374 {
375 	struct ctl_table_header *head = NULL;
376 	struct ctl_table *entry = NULL;
377 	struct ctl_node *ctl_node;
378 
379 	spin_lock(&sysctl_lock);
380 	ctl_node = first_usable_entry(rb_first(&dir->root));
381 	spin_unlock(&sysctl_lock);
382 	if (ctl_node) {
383 		head = ctl_node->header;
384 		entry = &head->ctl_table[ctl_node - head->node];
385 	}
386 	*phead = head;
387 	*pentry = entry;
388 }
389 
390 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
391 {
392 	struct ctl_table_header *head = *phead;
393 	struct ctl_table *entry = *pentry;
394 	struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
395 
396 	spin_lock(&sysctl_lock);
397 	unuse_table(head);
398 
399 	ctl_node = first_usable_entry(rb_next(&ctl_node->node));
400 	spin_unlock(&sysctl_lock);
401 	head = NULL;
402 	if (ctl_node) {
403 		head = ctl_node->header;
404 		entry = &head->ctl_table[ctl_node - head->node];
405 	}
406 	*phead = head;
407 	*pentry = entry;
408 }
409 
410 void register_sysctl_root(struct ctl_table_root *root)
411 {
412 }
413 
414 /*
415  * sysctl_perm does NOT grant the superuser all rights automatically, because
416  * some sysctl variables are readonly even to root.
417  */
418 
419 static int test_perm(int mode, int op)
420 {
421 	if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
422 		mode >>= 6;
423 	else if (in_egroup_p(GLOBAL_ROOT_GID))
424 		mode >>= 3;
425 	if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
426 		return 0;
427 	return -EACCES;
428 }
429 
430 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
431 {
432 	struct ctl_table_root *root = head->root;
433 	int mode;
434 
435 	if (root->permissions)
436 		mode = root->permissions(head, table);
437 	else
438 		mode = table->mode;
439 
440 	return test_perm(mode, op);
441 }
442 
443 static struct inode *proc_sys_make_inode(struct super_block *sb,
444 		struct ctl_table_header *head, struct ctl_table *table)
445 {
446 	struct ctl_table_root *root = head->root;
447 	struct inode *inode;
448 	struct proc_inode *ei;
449 
450 	inode = new_inode(sb);
451 	if (!inode)
452 		goto out;
453 
454 	inode->i_ino = get_next_ino();
455 
456 	ei = PROC_I(inode);
457 
458 	spin_lock(&sysctl_lock);
459 	if (unlikely(head->unregistering)) {
460 		spin_unlock(&sysctl_lock);
461 		iput(inode);
462 		inode = NULL;
463 		goto out;
464 	}
465 	ei->sysctl = head;
466 	ei->sysctl_entry = table;
467 	list_add_rcu(&ei->sysctl_inodes, &head->inodes);
468 	head->count++;
469 	spin_unlock(&sysctl_lock);
470 
471 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
472 	inode->i_mode = table->mode;
473 	if (!S_ISDIR(table->mode)) {
474 		inode->i_mode |= S_IFREG;
475 		inode->i_op = &proc_sys_inode_operations;
476 		inode->i_fop = &proc_sys_file_operations;
477 	} else {
478 		inode->i_mode |= S_IFDIR;
479 		inode->i_op = &proc_sys_dir_operations;
480 		inode->i_fop = &proc_sys_dir_file_operations;
481 		if (is_empty_dir(head))
482 			make_empty_dir_inode(inode);
483 	}
484 
485 	if (root->set_ownership)
486 		root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
487 
488 out:
489 	return inode;
490 }
491 
492 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
493 {
494 	spin_lock(&sysctl_lock);
495 	list_del_rcu(&PROC_I(inode)->sysctl_inodes);
496 	if (!--head->count)
497 		kfree_rcu(head, rcu);
498 	spin_unlock(&sysctl_lock);
499 }
500 
501 static struct ctl_table_header *grab_header(struct inode *inode)
502 {
503 	struct ctl_table_header *head = PROC_I(inode)->sysctl;
504 	if (!head)
505 		head = &sysctl_table_root.default_set.dir.header;
506 	return sysctl_head_grab(head);
507 }
508 
509 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
510 					unsigned int flags)
511 {
512 	struct ctl_table_header *head = grab_header(dir);
513 	struct ctl_table_header *h = NULL;
514 	const struct qstr *name = &dentry->d_name;
515 	struct ctl_table *p;
516 	struct inode *inode;
517 	struct dentry *err = ERR_PTR(-ENOENT);
518 	struct ctl_dir *ctl_dir;
519 	int ret;
520 
521 	if (IS_ERR(head))
522 		return ERR_CAST(head);
523 
524 	ctl_dir = container_of(head, struct ctl_dir, header);
525 
526 	p = lookup_entry(&h, ctl_dir, name->name, name->len);
527 	if (!p)
528 		goto out;
529 
530 	if (S_ISLNK(p->mode)) {
531 		ret = sysctl_follow_link(&h, &p);
532 		err = ERR_PTR(ret);
533 		if (ret)
534 			goto out;
535 	}
536 
537 	err = ERR_PTR(-ENOMEM);
538 	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
539 	if (!inode)
540 		goto out;
541 
542 	err = NULL;
543 	d_set_d_op(dentry, &proc_sys_dentry_operations);
544 	d_add(dentry, inode);
545 
546 out:
547 	if (h)
548 		sysctl_head_finish(h);
549 	sysctl_head_finish(head);
550 	return err;
551 }
552 
553 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
554 		size_t count, loff_t *ppos, int write)
555 {
556 	struct inode *inode = file_inode(filp);
557 	struct ctl_table_header *head = grab_header(inode);
558 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
559 	ssize_t error;
560 	size_t res;
561 
562 	if (IS_ERR(head))
563 		return PTR_ERR(head);
564 
565 	/*
566 	 * At this point we know that the sysctl was not unregistered
567 	 * and won't be until we finish.
568 	 */
569 	error = -EPERM;
570 	if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
571 		goto out;
572 
573 	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
574 	error = -EINVAL;
575 	if (!table->proc_handler)
576 		goto out;
577 
578 	/* careful: calling conventions are nasty here */
579 	res = count;
580 	error = table->proc_handler(table, write, buf, &res, ppos);
581 	if (!error)
582 		error = res;
583 out:
584 	sysctl_head_finish(head);
585 
586 	return error;
587 }
588 
589 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
590 				size_t count, loff_t *ppos)
591 {
592 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
593 }
594 
595 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
596 				size_t count, loff_t *ppos)
597 {
598 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
599 }
600 
601 static int proc_sys_open(struct inode *inode, struct file *filp)
602 {
603 	struct ctl_table_header *head = grab_header(inode);
604 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
605 
606 	/* sysctl was unregistered */
607 	if (IS_ERR(head))
608 		return PTR_ERR(head);
609 
610 	if (table->poll)
611 		filp->private_data = proc_sys_poll_event(table->poll);
612 
613 	sysctl_head_finish(head);
614 
615 	return 0;
616 }
617 
618 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
619 {
620 	struct inode *inode = file_inode(filp);
621 	struct ctl_table_header *head = grab_header(inode);
622 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
623 	unsigned int ret = DEFAULT_POLLMASK;
624 	unsigned long event;
625 
626 	/* sysctl was unregistered */
627 	if (IS_ERR(head))
628 		return POLLERR | POLLHUP;
629 
630 	if (!table->proc_handler)
631 		goto out;
632 
633 	if (!table->poll)
634 		goto out;
635 
636 	event = (unsigned long)filp->private_data;
637 	poll_wait(filp, &table->poll->wait, wait);
638 
639 	if (event != atomic_read(&table->poll->event)) {
640 		filp->private_data = proc_sys_poll_event(table->poll);
641 		ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
642 	}
643 
644 out:
645 	sysctl_head_finish(head);
646 
647 	return ret;
648 }
649 
650 static bool proc_sys_fill_cache(struct file *file,
651 				struct dir_context *ctx,
652 				struct ctl_table_header *head,
653 				struct ctl_table *table)
654 {
655 	struct dentry *child, *dir = file->f_path.dentry;
656 	struct inode *inode;
657 	struct qstr qname;
658 	ino_t ino = 0;
659 	unsigned type = DT_UNKNOWN;
660 
661 	qname.name = table->procname;
662 	qname.len  = strlen(table->procname);
663 	qname.hash = full_name_hash(dir, qname.name, qname.len);
664 
665 	child = d_lookup(dir, &qname);
666 	if (!child) {
667 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
668 		child = d_alloc_parallel(dir, &qname, &wq);
669 		if (IS_ERR(child))
670 			return false;
671 		if (d_in_lookup(child)) {
672 			inode = proc_sys_make_inode(dir->d_sb, head, table);
673 			if (!inode) {
674 				d_lookup_done(child);
675 				dput(child);
676 				return false;
677 			}
678 			d_set_d_op(child, &proc_sys_dentry_operations);
679 			d_add(child, inode);
680 		}
681 	}
682 	inode = d_inode(child);
683 	ino  = inode->i_ino;
684 	type = inode->i_mode >> 12;
685 	dput(child);
686 	return dir_emit(ctx, qname.name, qname.len, ino, type);
687 }
688 
689 static bool proc_sys_link_fill_cache(struct file *file,
690 				    struct dir_context *ctx,
691 				    struct ctl_table_header *head,
692 				    struct ctl_table *table)
693 {
694 	bool ret = true;
695 	head = sysctl_head_grab(head);
696 
697 	if (S_ISLNK(table->mode)) {
698 		/* It is not an error if we can not follow the link ignore it */
699 		int err = sysctl_follow_link(&head, &table);
700 		if (err)
701 			goto out;
702 	}
703 
704 	ret = proc_sys_fill_cache(file, ctx, head, table);
705 out:
706 	sysctl_head_finish(head);
707 	return ret;
708 }
709 
710 static int scan(struct ctl_table_header *head, struct ctl_table *table,
711 		unsigned long *pos, struct file *file,
712 		struct dir_context *ctx)
713 {
714 	bool res;
715 
716 	if ((*pos)++ < ctx->pos)
717 		return true;
718 
719 	if (unlikely(S_ISLNK(table->mode)))
720 		res = proc_sys_link_fill_cache(file, ctx, head, table);
721 	else
722 		res = proc_sys_fill_cache(file, ctx, head, table);
723 
724 	if (res)
725 		ctx->pos = *pos;
726 
727 	return res;
728 }
729 
730 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
731 {
732 	struct ctl_table_header *head = grab_header(file_inode(file));
733 	struct ctl_table_header *h = NULL;
734 	struct ctl_table *entry;
735 	struct ctl_dir *ctl_dir;
736 	unsigned long pos;
737 
738 	if (IS_ERR(head))
739 		return PTR_ERR(head);
740 
741 	ctl_dir = container_of(head, struct ctl_dir, header);
742 
743 	if (!dir_emit_dots(file, ctx))
744 		goto out;
745 
746 	pos = 2;
747 
748 	for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
749 		if (!scan(h, entry, &pos, file, ctx)) {
750 			sysctl_head_finish(h);
751 			break;
752 		}
753 	}
754 out:
755 	sysctl_head_finish(head);
756 	return 0;
757 }
758 
759 static int proc_sys_permission(struct inode *inode, int mask)
760 {
761 	/*
762 	 * sysctl entries that are not writeable,
763 	 * are _NOT_ writeable, capabilities or not.
764 	 */
765 	struct ctl_table_header *head;
766 	struct ctl_table *table;
767 	int error;
768 
769 	/* Executable files are not allowed under /proc/sys/ */
770 	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
771 		return -EACCES;
772 
773 	head = grab_header(inode);
774 	if (IS_ERR(head))
775 		return PTR_ERR(head);
776 
777 	table = PROC_I(inode)->sysctl_entry;
778 	if (!table) /* global root - r-xr-xr-x */
779 		error = mask & MAY_WRITE ? -EACCES : 0;
780 	else /* Use the permissions on the sysctl table entry */
781 		error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
782 
783 	sysctl_head_finish(head);
784 	return error;
785 }
786 
787 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
788 {
789 	struct inode *inode = d_inode(dentry);
790 	int error;
791 
792 	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
793 		return -EPERM;
794 
795 	error = setattr_prepare(dentry, attr);
796 	if (error)
797 		return error;
798 
799 	setattr_copy(inode, attr);
800 	mark_inode_dirty(inode);
801 	return 0;
802 }
803 
804 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
805 {
806 	struct inode *inode = d_inode(dentry);
807 	struct ctl_table_header *head = grab_header(inode);
808 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
809 
810 	if (IS_ERR(head))
811 		return PTR_ERR(head);
812 
813 	generic_fillattr(inode, stat);
814 	if (table)
815 		stat->mode = (stat->mode & S_IFMT) | table->mode;
816 
817 	sysctl_head_finish(head);
818 	return 0;
819 }
820 
821 static const struct file_operations proc_sys_file_operations = {
822 	.open		= proc_sys_open,
823 	.poll		= proc_sys_poll,
824 	.read		= proc_sys_read,
825 	.write		= proc_sys_write,
826 	.llseek		= default_llseek,
827 };
828 
829 static const struct file_operations proc_sys_dir_file_operations = {
830 	.read		= generic_read_dir,
831 	.iterate_shared	= proc_sys_readdir,
832 	.llseek		= generic_file_llseek,
833 };
834 
835 static const struct inode_operations proc_sys_inode_operations = {
836 	.permission	= proc_sys_permission,
837 	.setattr	= proc_sys_setattr,
838 	.getattr	= proc_sys_getattr,
839 };
840 
841 static const struct inode_operations proc_sys_dir_operations = {
842 	.lookup		= proc_sys_lookup,
843 	.permission	= proc_sys_permission,
844 	.setattr	= proc_sys_setattr,
845 	.getattr	= proc_sys_getattr,
846 };
847 
848 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
849 {
850 	if (flags & LOOKUP_RCU)
851 		return -ECHILD;
852 	return !PROC_I(d_inode(dentry))->sysctl->unregistering;
853 }
854 
855 static int proc_sys_delete(const struct dentry *dentry)
856 {
857 	return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
858 }
859 
860 static int sysctl_is_seen(struct ctl_table_header *p)
861 {
862 	struct ctl_table_set *set = p->set;
863 	int res;
864 	spin_lock(&sysctl_lock);
865 	if (p->unregistering)
866 		res = 0;
867 	else if (!set->is_seen)
868 		res = 1;
869 	else
870 		res = set->is_seen(set);
871 	spin_unlock(&sysctl_lock);
872 	return res;
873 }
874 
875 static int proc_sys_compare(const struct dentry *dentry,
876 		unsigned int len, const char *str, const struct qstr *name)
877 {
878 	struct ctl_table_header *head;
879 	struct inode *inode;
880 
881 	/* Although proc doesn't have negative dentries, rcu-walk means
882 	 * that inode here can be NULL */
883 	/* AV: can it, indeed? */
884 	inode = d_inode_rcu(dentry);
885 	if (!inode)
886 		return 1;
887 	if (name->len != len)
888 		return 1;
889 	if (memcmp(name->name, str, len))
890 		return 1;
891 	head = rcu_dereference(PROC_I(inode)->sysctl);
892 	return !head || !sysctl_is_seen(head);
893 }
894 
895 static const struct dentry_operations proc_sys_dentry_operations = {
896 	.d_revalidate	= proc_sys_revalidate,
897 	.d_delete	= proc_sys_delete,
898 	.d_compare	= proc_sys_compare,
899 };
900 
901 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
902 				   const char *name, int namelen)
903 {
904 	struct ctl_table_header *head;
905 	struct ctl_table *entry;
906 
907 	entry = find_entry(&head, dir, name, namelen);
908 	if (!entry)
909 		return ERR_PTR(-ENOENT);
910 	if (!S_ISDIR(entry->mode))
911 		return ERR_PTR(-ENOTDIR);
912 	return container_of(head, struct ctl_dir, header);
913 }
914 
915 static struct ctl_dir *new_dir(struct ctl_table_set *set,
916 			       const char *name, int namelen)
917 {
918 	struct ctl_table *table;
919 	struct ctl_dir *new;
920 	struct ctl_node *node;
921 	char *new_name;
922 
923 	new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
924 		      sizeof(struct ctl_table)*2 +  namelen + 1,
925 		      GFP_KERNEL);
926 	if (!new)
927 		return NULL;
928 
929 	node = (struct ctl_node *)(new + 1);
930 	table = (struct ctl_table *)(node + 1);
931 	new_name = (char *)(table + 2);
932 	memcpy(new_name, name, namelen);
933 	new_name[namelen] = '\0';
934 	table[0].procname = new_name;
935 	table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
936 	init_header(&new->header, set->dir.header.root, set, node, table);
937 
938 	return new;
939 }
940 
941 /**
942  * get_subdir - find or create a subdir with the specified name.
943  * @dir:  Directory to create the subdirectory in
944  * @name: The name of the subdirectory to find or create
945  * @namelen: The length of name
946  *
947  * Takes a directory with an elevated reference count so we know that
948  * if we drop the lock the directory will not go away.  Upon success
949  * the reference is moved from @dir to the returned subdirectory.
950  * Upon error an error code is returned and the reference on @dir is
951  * simply dropped.
952  */
953 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
954 				  const char *name, int namelen)
955 {
956 	struct ctl_table_set *set = dir->header.set;
957 	struct ctl_dir *subdir, *new = NULL;
958 	int err;
959 
960 	spin_lock(&sysctl_lock);
961 	subdir = find_subdir(dir, name, namelen);
962 	if (!IS_ERR(subdir))
963 		goto found;
964 	if (PTR_ERR(subdir) != -ENOENT)
965 		goto failed;
966 
967 	spin_unlock(&sysctl_lock);
968 	new = new_dir(set, name, namelen);
969 	spin_lock(&sysctl_lock);
970 	subdir = ERR_PTR(-ENOMEM);
971 	if (!new)
972 		goto failed;
973 
974 	/* Was the subdir added while we dropped the lock? */
975 	subdir = find_subdir(dir, name, namelen);
976 	if (!IS_ERR(subdir))
977 		goto found;
978 	if (PTR_ERR(subdir) != -ENOENT)
979 		goto failed;
980 
981 	/* Nope.  Use the our freshly made directory entry. */
982 	err = insert_header(dir, &new->header);
983 	subdir = ERR_PTR(err);
984 	if (err)
985 		goto failed;
986 	subdir = new;
987 found:
988 	subdir->header.nreg++;
989 failed:
990 	if (IS_ERR(subdir)) {
991 		pr_err("sysctl could not get directory: ");
992 		sysctl_print_dir(dir);
993 		pr_cont("/%*.*s %ld\n",
994 			namelen, namelen, name, PTR_ERR(subdir));
995 	}
996 	drop_sysctl_table(&dir->header);
997 	if (new)
998 		drop_sysctl_table(&new->header);
999 	spin_unlock(&sysctl_lock);
1000 	return subdir;
1001 }
1002 
1003 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1004 {
1005 	struct ctl_dir *parent;
1006 	const char *procname;
1007 	if (!dir->header.parent)
1008 		return &set->dir;
1009 	parent = xlate_dir(set, dir->header.parent);
1010 	if (IS_ERR(parent))
1011 		return parent;
1012 	procname = dir->header.ctl_table[0].procname;
1013 	return find_subdir(parent, procname, strlen(procname));
1014 }
1015 
1016 static int sysctl_follow_link(struct ctl_table_header **phead,
1017 	struct ctl_table **pentry)
1018 {
1019 	struct ctl_table_header *head;
1020 	struct ctl_table_root *root;
1021 	struct ctl_table_set *set;
1022 	struct ctl_table *entry;
1023 	struct ctl_dir *dir;
1024 	int ret;
1025 
1026 	ret = 0;
1027 	spin_lock(&sysctl_lock);
1028 	root = (*pentry)->data;
1029 	set = lookup_header_set(root);
1030 	dir = xlate_dir(set, (*phead)->parent);
1031 	if (IS_ERR(dir))
1032 		ret = PTR_ERR(dir);
1033 	else {
1034 		const char *procname = (*pentry)->procname;
1035 		head = NULL;
1036 		entry = find_entry(&head, dir, procname, strlen(procname));
1037 		ret = -ENOENT;
1038 		if (entry && use_table(head)) {
1039 			unuse_table(*phead);
1040 			*phead = head;
1041 			*pentry = entry;
1042 			ret = 0;
1043 		}
1044 	}
1045 
1046 	spin_unlock(&sysctl_lock);
1047 	return ret;
1048 }
1049 
1050 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1051 {
1052 	struct va_format vaf;
1053 	va_list args;
1054 
1055 	va_start(args, fmt);
1056 	vaf.fmt = fmt;
1057 	vaf.va = &args;
1058 
1059 	pr_err("sysctl table check failed: %s/%s %pV\n",
1060 	       path, table->procname, &vaf);
1061 
1062 	va_end(args);
1063 	return -EINVAL;
1064 }
1065 
1066 static int sysctl_check_table(const char *path, struct ctl_table *table)
1067 {
1068 	int err = 0;
1069 	for (; table->procname; table++) {
1070 		if (table->child)
1071 			err = sysctl_err(path, table, "Not a file");
1072 
1073 		if ((table->proc_handler == proc_dostring) ||
1074 		    (table->proc_handler == proc_dointvec) ||
1075 		    (table->proc_handler == proc_dointvec_minmax) ||
1076 		    (table->proc_handler == proc_dointvec_jiffies) ||
1077 		    (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1078 		    (table->proc_handler == proc_dointvec_ms_jiffies) ||
1079 		    (table->proc_handler == proc_doulongvec_minmax) ||
1080 		    (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1081 			if (!table->data)
1082 				err = sysctl_err(path, table, "No data");
1083 			if (!table->maxlen)
1084 				err = sysctl_err(path, table, "No maxlen");
1085 		}
1086 		if (!table->proc_handler)
1087 			err = sysctl_err(path, table, "No proc_handler");
1088 
1089 		if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1090 			err = sysctl_err(path, table, "bogus .mode 0%o",
1091 				table->mode);
1092 	}
1093 	return err;
1094 }
1095 
1096 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1097 	struct ctl_table_root *link_root)
1098 {
1099 	struct ctl_table *link_table, *entry, *link;
1100 	struct ctl_table_header *links;
1101 	struct ctl_node *node;
1102 	char *link_name;
1103 	int nr_entries, name_bytes;
1104 
1105 	name_bytes = 0;
1106 	nr_entries = 0;
1107 	for (entry = table; entry->procname; entry++) {
1108 		nr_entries++;
1109 		name_bytes += strlen(entry->procname) + 1;
1110 	}
1111 
1112 	links = kzalloc(sizeof(struct ctl_table_header) +
1113 			sizeof(struct ctl_node)*nr_entries +
1114 			sizeof(struct ctl_table)*(nr_entries + 1) +
1115 			name_bytes,
1116 			GFP_KERNEL);
1117 
1118 	if (!links)
1119 		return NULL;
1120 
1121 	node = (struct ctl_node *)(links + 1);
1122 	link_table = (struct ctl_table *)(node + nr_entries);
1123 	link_name = (char *)&link_table[nr_entries + 1];
1124 
1125 	for (link = link_table, entry = table; entry->procname; link++, entry++) {
1126 		int len = strlen(entry->procname) + 1;
1127 		memcpy(link_name, entry->procname, len);
1128 		link->procname = link_name;
1129 		link->mode = S_IFLNK|S_IRWXUGO;
1130 		link->data = link_root;
1131 		link_name += len;
1132 	}
1133 	init_header(links, dir->header.root, dir->header.set, node, link_table);
1134 	links->nreg = nr_entries;
1135 
1136 	return links;
1137 }
1138 
1139 static bool get_links(struct ctl_dir *dir,
1140 	struct ctl_table *table, struct ctl_table_root *link_root)
1141 {
1142 	struct ctl_table_header *head;
1143 	struct ctl_table *entry, *link;
1144 
1145 	/* Are there links available for every entry in table? */
1146 	for (entry = table; entry->procname; entry++) {
1147 		const char *procname = entry->procname;
1148 		link = find_entry(&head, dir, procname, strlen(procname));
1149 		if (!link)
1150 			return false;
1151 		if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1152 			continue;
1153 		if (S_ISLNK(link->mode) && (link->data == link_root))
1154 			continue;
1155 		return false;
1156 	}
1157 
1158 	/* The checks passed.  Increase the registration count on the links */
1159 	for (entry = table; entry->procname; entry++) {
1160 		const char *procname = entry->procname;
1161 		link = find_entry(&head, dir, procname, strlen(procname));
1162 		head->nreg++;
1163 	}
1164 	return true;
1165 }
1166 
1167 static int insert_links(struct ctl_table_header *head)
1168 {
1169 	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1170 	struct ctl_dir *core_parent = NULL;
1171 	struct ctl_table_header *links;
1172 	int err;
1173 
1174 	if (head->set == root_set)
1175 		return 0;
1176 
1177 	core_parent = xlate_dir(root_set, head->parent);
1178 	if (IS_ERR(core_parent))
1179 		return 0;
1180 
1181 	if (get_links(core_parent, head->ctl_table, head->root))
1182 		return 0;
1183 
1184 	core_parent->header.nreg++;
1185 	spin_unlock(&sysctl_lock);
1186 
1187 	links = new_links(core_parent, head->ctl_table, head->root);
1188 
1189 	spin_lock(&sysctl_lock);
1190 	err = -ENOMEM;
1191 	if (!links)
1192 		goto out;
1193 
1194 	err = 0;
1195 	if (get_links(core_parent, head->ctl_table, head->root)) {
1196 		kfree(links);
1197 		goto out;
1198 	}
1199 
1200 	err = insert_header(core_parent, links);
1201 	if (err)
1202 		kfree(links);
1203 out:
1204 	drop_sysctl_table(&core_parent->header);
1205 	return err;
1206 }
1207 
1208 /**
1209  * __register_sysctl_table - register a leaf sysctl table
1210  * @set: Sysctl tree to register on
1211  * @path: The path to the directory the sysctl table is in.
1212  * @table: the top-level table structure
1213  *
1214  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1215  * array. A completely 0 filled entry terminates the table.
1216  *
1217  * The members of the &struct ctl_table structure are used as follows:
1218  *
1219  * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1220  *            enter a sysctl file
1221  *
1222  * data - a pointer to data for use by proc_handler
1223  *
1224  * maxlen - the maximum size in bytes of the data
1225  *
1226  * mode - the file permissions for the /proc/sys file
1227  *
1228  * child - must be %NULL.
1229  *
1230  * proc_handler - the text handler routine (described below)
1231  *
1232  * extra1, extra2 - extra pointers usable by the proc handler routines
1233  *
1234  * Leaf nodes in the sysctl tree will be represented by a single file
1235  * under /proc; non-leaf nodes will be represented by directories.
1236  *
1237  * There must be a proc_handler routine for any terminal nodes.
1238  * Several default handlers are available to cover common cases -
1239  *
1240  * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1241  * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1242  * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1243  *
1244  * It is the handler's job to read the input buffer from user memory
1245  * and process it. The handler should return 0 on success.
1246  *
1247  * This routine returns %NULL on a failure to register, and a pointer
1248  * to the table header on success.
1249  */
1250 struct ctl_table_header *__register_sysctl_table(
1251 	struct ctl_table_set *set,
1252 	const char *path, struct ctl_table *table)
1253 {
1254 	struct ctl_table_root *root = set->dir.header.root;
1255 	struct ctl_table_header *header;
1256 	const char *name, *nextname;
1257 	struct ctl_dir *dir;
1258 	struct ctl_table *entry;
1259 	struct ctl_node *node;
1260 	int nr_entries = 0;
1261 
1262 	for (entry = table; entry->procname; entry++)
1263 		nr_entries++;
1264 
1265 	header = kzalloc(sizeof(struct ctl_table_header) +
1266 			 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1267 	if (!header)
1268 		return NULL;
1269 
1270 	node = (struct ctl_node *)(header + 1);
1271 	init_header(header, root, set, node, table);
1272 	if (sysctl_check_table(path, table))
1273 		goto fail;
1274 
1275 	spin_lock(&sysctl_lock);
1276 	dir = &set->dir;
1277 	/* Reference moved down the diretory tree get_subdir */
1278 	dir->header.nreg++;
1279 	spin_unlock(&sysctl_lock);
1280 
1281 	/* Find the directory for the ctl_table */
1282 	for (name = path; name; name = nextname) {
1283 		int namelen;
1284 		nextname = strchr(name, '/');
1285 		if (nextname) {
1286 			namelen = nextname - name;
1287 			nextname++;
1288 		} else {
1289 			namelen = strlen(name);
1290 		}
1291 		if (namelen == 0)
1292 			continue;
1293 
1294 		dir = get_subdir(dir, name, namelen);
1295 		if (IS_ERR(dir))
1296 			goto fail;
1297 	}
1298 
1299 	spin_lock(&sysctl_lock);
1300 	if (insert_header(dir, header))
1301 		goto fail_put_dir_locked;
1302 
1303 	drop_sysctl_table(&dir->header);
1304 	spin_unlock(&sysctl_lock);
1305 
1306 	return header;
1307 
1308 fail_put_dir_locked:
1309 	drop_sysctl_table(&dir->header);
1310 	spin_unlock(&sysctl_lock);
1311 fail:
1312 	kfree(header);
1313 	dump_stack();
1314 	return NULL;
1315 }
1316 
1317 /**
1318  * register_sysctl - register a sysctl table
1319  * @path: The path to the directory the sysctl table is in.
1320  * @table: the table structure
1321  *
1322  * Register a sysctl table. @table should be a filled in ctl_table
1323  * array. A completely 0 filled entry terminates the table.
1324  *
1325  * See __register_sysctl_table for more details.
1326  */
1327 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1328 {
1329 	return __register_sysctl_table(&sysctl_table_root.default_set,
1330 					path, table);
1331 }
1332 EXPORT_SYMBOL(register_sysctl);
1333 
1334 static char *append_path(const char *path, char *pos, const char *name)
1335 {
1336 	int namelen;
1337 	namelen = strlen(name);
1338 	if (((pos - path) + namelen + 2) >= PATH_MAX)
1339 		return NULL;
1340 	memcpy(pos, name, namelen);
1341 	pos[namelen] = '/';
1342 	pos[namelen + 1] = '\0';
1343 	pos += namelen + 1;
1344 	return pos;
1345 }
1346 
1347 static int count_subheaders(struct ctl_table *table)
1348 {
1349 	int has_files = 0;
1350 	int nr_subheaders = 0;
1351 	struct ctl_table *entry;
1352 
1353 	/* special case: no directory and empty directory */
1354 	if (!table || !table->procname)
1355 		return 1;
1356 
1357 	for (entry = table; entry->procname; entry++) {
1358 		if (entry->child)
1359 			nr_subheaders += count_subheaders(entry->child);
1360 		else
1361 			has_files = 1;
1362 	}
1363 	return nr_subheaders + has_files;
1364 }
1365 
1366 static int register_leaf_sysctl_tables(const char *path, char *pos,
1367 	struct ctl_table_header ***subheader, struct ctl_table_set *set,
1368 	struct ctl_table *table)
1369 {
1370 	struct ctl_table *ctl_table_arg = NULL;
1371 	struct ctl_table *entry, *files;
1372 	int nr_files = 0;
1373 	int nr_dirs = 0;
1374 	int err = -ENOMEM;
1375 
1376 	for (entry = table; entry->procname; entry++) {
1377 		if (entry->child)
1378 			nr_dirs++;
1379 		else
1380 			nr_files++;
1381 	}
1382 
1383 	files = table;
1384 	/* If there are mixed files and directories we need a new table */
1385 	if (nr_dirs && nr_files) {
1386 		struct ctl_table *new;
1387 		files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1388 				GFP_KERNEL);
1389 		if (!files)
1390 			goto out;
1391 
1392 		ctl_table_arg = files;
1393 		for (new = files, entry = table; entry->procname; entry++) {
1394 			if (entry->child)
1395 				continue;
1396 			*new = *entry;
1397 			new++;
1398 		}
1399 	}
1400 
1401 	/* Register everything except a directory full of subdirectories */
1402 	if (nr_files || !nr_dirs) {
1403 		struct ctl_table_header *header;
1404 		header = __register_sysctl_table(set, path, files);
1405 		if (!header) {
1406 			kfree(ctl_table_arg);
1407 			goto out;
1408 		}
1409 
1410 		/* Remember if we need to free the file table */
1411 		header->ctl_table_arg = ctl_table_arg;
1412 		**subheader = header;
1413 		(*subheader)++;
1414 	}
1415 
1416 	/* Recurse into the subdirectories. */
1417 	for (entry = table; entry->procname; entry++) {
1418 		char *child_pos;
1419 
1420 		if (!entry->child)
1421 			continue;
1422 
1423 		err = -ENAMETOOLONG;
1424 		child_pos = append_path(path, pos, entry->procname);
1425 		if (!child_pos)
1426 			goto out;
1427 
1428 		err = register_leaf_sysctl_tables(path, child_pos, subheader,
1429 						  set, entry->child);
1430 		pos[0] = '\0';
1431 		if (err)
1432 			goto out;
1433 	}
1434 	err = 0;
1435 out:
1436 	/* On failure our caller will unregister all registered subheaders */
1437 	return err;
1438 }
1439 
1440 /**
1441  * __register_sysctl_paths - register a sysctl table hierarchy
1442  * @set: Sysctl tree to register on
1443  * @path: The path to the directory the sysctl table is in.
1444  * @table: the top-level table structure
1445  *
1446  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1447  * array. A completely 0 filled entry terminates the table.
1448  *
1449  * See __register_sysctl_table for more details.
1450  */
1451 struct ctl_table_header *__register_sysctl_paths(
1452 	struct ctl_table_set *set,
1453 	const struct ctl_path *path, struct ctl_table *table)
1454 {
1455 	struct ctl_table *ctl_table_arg = table;
1456 	int nr_subheaders = count_subheaders(table);
1457 	struct ctl_table_header *header = NULL, **subheaders, **subheader;
1458 	const struct ctl_path *component;
1459 	char *new_path, *pos;
1460 
1461 	pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1462 	if (!new_path)
1463 		return NULL;
1464 
1465 	pos[0] = '\0';
1466 	for (component = path; component->procname; component++) {
1467 		pos = append_path(new_path, pos, component->procname);
1468 		if (!pos)
1469 			goto out;
1470 	}
1471 	while (table->procname && table->child && !table[1].procname) {
1472 		pos = append_path(new_path, pos, table->procname);
1473 		if (!pos)
1474 			goto out;
1475 		table = table->child;
1476 	}
1477 	if (nr_subheaders == 1) {
1478 		header = __register_sysctl_table(set, new_path, table);
1479 		if (header)
1480 			header->ctl_table_arg = ctl_table_arg;
1481 	} else {
1482 		header = kzalloc(sizeof(*header) +
1483 				 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1484 		if (!header)
1485 			goto out;
1486 
1487 		subheaders = (struct ctl_table_header **) (header + 1);
1488 		subheader = subheaders;
1489 		header->ctl_table_arg = ctl_table_arg;
1490 
1491 		if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1492 						set, table))
1493 			goto err_register_leaves;
1494 	}
1495 
1496 out:
1497 	kfree(new_path);
1498 	return header;
1499 
1500 err_register_leaves:
1501 	while (subheader > subheaders) {
1502 		struct ctl_table_header *subh = *(--subheader);
1503 		struct ctl_table *table = subh->ctl_table_arg;
1504 		unregister_sysctl_table(subh);
1505 		kfree(table);
1506 	}
1507 	kfree(header);
1508 	header = NULL;
1509 	goto out;
1510 }
1511 
1512 /**
1513  * register_sysctl_table_path - register a sysctl table hierarchy
1514  * @path: The path to the directory the sysctl table is in.
1515  * @table: the top-level table structure
1516  *
1517  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1518  * array. A completely 0 filled entry terminates the table.
1519  *
1520  * See __register_sysctl_paths for more details.
1521  */
1522 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1523 						struct ctl_table *table)
1524 {
1525 	return __register_sysctl_paths(&sysctl_table_root.default_set,
1526 					path, table);
1527 }
1528 EXPORT_SYMBOL(register_sysctl_paths);
1529 
1530 /**
1531  * register_sysctl_table - register a sysctl table hierarchy
1532  * @table: the top-level table structure
1533  *
1534  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1535  * array. A completely 0 filled entry terminates the table.
1536  *
1537  * See register_sysctl_paths for more details.
1538  */
1539 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1540 {
1541 	static const struct ctl_path null_path[] = { {} };
1542 
1543 	return register_sysctl_paths(null_path, table);
1544 }
1545 EXPORT_SYMBOL(register_sysctl_table);
1546 
1547 static void put_links(struct ctl_table_header *header)
1548 {
1549 	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1550 	struct ctl_table_root *root = header->root;
1551 	struct ctl_dir *parent = header->parent;
1552 	struct ctl_dir *core_parent;
1553 	struct ctl_table *entry;
1554 
1555 	if (header->set == root_set)
1556 		return;
1557 
1558 	core_parent = xlate_dir(root_set, parent);
1559 	if (IS_ERR(core_parent))
1560 		return;
1561 
1562 	for (entry = header->ctl_table; entry->procname; entry++) {
1563 		struct ctl_table_header *link_head;
1564 		struct ctl_table *link;
1565 		const char *name = entry->procname;
1566 
1567 		link = find_entry(&link_head, core_parent, name, strlen(name));
1568 		if (link &&
1569 		    ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1570 		     (S_ISLNK(link->mode) && (link->data == root)))) {
1571 			drop_sysctl_table(link_head);
1572 		}
1573 		else {
1574 			pr_err("sysctl link missing during unregister: ");
1575 			sysctl_print_dir(parent);
1576 			pr_cont("/%s\n", name);
1577 		}
1578 	}
1579 }
1580 
1581 static void drop_sysctl_table(struct ctl_table_header *header)
1582 {
1583 	struct ctl_dir *parent = header->parent;
1584 
1585 	if (--header->nreg)
1586 		return;
1587 
1588 	put_links(header);
1589 	start_unregistering(header);
1590 	if (!--header->count)
1591 		kfree_rcu(header, rcu);
1592 
1593 	if (parent)
1594 		drop_sysctl_table(&parent->header);
1595 }
1596 
1597 /**
1598  * unregister_sysctl_table - unregister a sysctl table hierarchy
1599  * @header: the header returned from register_sysctl_table
1600  *
1601  * Unregisters the sysctl table and all children. proc entries may not
1602  * actually be removed until they are no longer used by anyone.
1603  */
1604 void unregister_sysctl_table(struct ctl_table_header * header)
1605 {
1606 	int nr_subheaders;
1607 	might_sleep();
1608 
1609 	if (header == NULL)
1610 		return;
1611 
1612 	nr_subheaders = count_subheaders(header->ctl_table_arg);
1613 	if (unlikely(nr_subheaders > 1)) {
1614 		struct ctl_table_header **subheaders;
1615 		int i;
1616 
1617 		subheaders = (struct ctl_table_header **)(header + 1);
1618 		for (i = nr_subheaders -1; i >= 0; i--) {
1619 			struct ctl_table_header *subh = subheaders[i];
1620 			struct ctl_table *table = subh->ctl_table_arg;
1621 			unregister_sysctl_table(subh);
1622 			kfree(table);
1623 		}
1624 		kfree(header);
1625 		return;
1626 	}
1627 
1628 	spin_lock(&sysctl_lock);
1629 	drop_sysctl_table(header);
1630 	spin_unlock(&sysctl_lock);
1631 }
1632 EXPORT_SYMBOL(unregister_sysctl_table);
1633 
1634 void setup_sysctl_set(struct ctl_table_set *set,
1635 	struct ctl_table_root *root,
1636 	int (*is_seen)(struct ctl_table_set *))
1637 {
1638 	memset(set, 0, sizeof(*set));
1639 	set->is_seen = is_seen;
1640 	init_header(&set->dir.header, root, set, NULL, root_table);
1641 }
1642 
1643 void retire_sysctl_set(struct ctl_table_set *set)
1644 {
1645 	WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1646 }
1647 
1648 int __init proc_sys_init(void)
1649 {
1650 	struct proc_dir_entry *proc_sys_root;
1651 
1652 	proc_sys_root = proc_mkdir("sys", NULL);
1653 	proc_sys_root->proc_iops = &proc_sys_dir_operations;
1654 	proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1655 	proc_sys_root->nlink = 0;
1656 
1657 	return sysctl_init();
1658 }
1659