xref: /linux/fs/proc/proc_sysctl.c (revision 9e9f60108423f18a99c9cc93ef7f23490ecc709b)
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 void proc_sys_poll_notify(struct ctl_table_poll *poll)
23 {
24 	if (!poll)
25 		return;
26 
27 	atomic_inc(&poll->event);
28 	wake_up_interruptible(&poll->wait);
29 }
30 
31 static struct ctl_table root_table[] = {
32 	{
33 		.procname = "",
34 		.mode = S_IFDIR|S_IRUGO|S_IXUGO,
35 	},
36 	{ }
37 };
38 static struct ctl_table_root sysctl_table_root = {
39 	.default_set.dir.header = {
40 		{{.count = 1,
41 		  .nreg = 1,
42 		  .ctl_table = root_table }},
43 		.ctl_table_arg = root_table,
44 		.root = &sysctl_table_root,
45 		.set = &sysctl_table_root.default_set,
46 	},
47 };
48 
49 static DEFINE_SPINLOCK(sysctl_lock);
50 
51 static void drop_sysctl_table(struct ctl_table_header *header);
52 static int sysctl_follow_link(struct ctl_table_header **phead,
53 	struct ctl_table **pentry, struct nsproxy *namespaces);
54 static int insert_links(struct ctl_table_header *head);
55 static void put_links(struct ctl_table_header *header);
56 
57 static void sysctl_print_dir(struct ctl_dir *dir)
58 {
59 	if (dir->header.parent)
60 		sysctl_print_dir(dir->header.parent);
61 	pr_cont("%s/", dir->header.ctl_table[0].procname);
62 }
63 
64 static int namecmp(const char *name1, int len1, const char *name2, int len2)
65 {
66 	int minlen;
67 	int cmp;
68 
69 	minlen = len1;
70 	if (minlen > len2)
71 		minlen = len2;
72 
73 	cmp = memcmp(name1, name2, minlen);
74 	if (cmp == 0)
75 		cmp = len1 - len2;
76 	return cmp;
77 }
78 
79 /* Called under sysctl_lock */
80 static struct ctl_table *find_entry(struct ctl_table_header **phead,
81 	struct ctl_dir *dir, const char *name, int namelen)
82 {
83 	struct ctl_table_header *head;
84 	struct ctl_table *entry;
85 	struct rb_node *node = dir->root.rb_node;
86 
87 	while (node)
88 	{
89 		struct ctl_node *ctl_node;
90 		const char *procname;
91 		int cmp;
92 
93 		ctl_node = rb_entry(node, struct ctl_node, node);
94 		head = ctl_node->header;
95 		entry = &head->ctl_table[ctl_node - head->node];
96 		procname = entry->procname;
97 
98 		cmp = namecmp(name, namelen, procname, strlen(procname));
99 		if (cmp < 0)
100 			node = node->rb_left;
101 		else if (cmp > 0)
102 			node = node->rb_right;
103 		else {
104 			*phead = head;
105 			return entry;
106 		}
107 	}
108 	return NULL;
109 }
110 
111 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
112 {
113 	struct rb_node *node = &head->node[entry - head->ctl_table].node;
114 	struct rb_node **p = &head->parent->root.rb_node;
115 	struct rb_node *parent = NULL;
116 	const char *name = entry->procname;
117 	int namelen = strlen(name);
118 
119 	while (*p) {
120 		struct ctl_table_header *parent_head;
121 		struct ctl_table *parent_entry;
122 		struct ctl_node *parent_node;
123 		const char *parent_name;
124 		int cmp;
125 
126 		parent = *p;
127 		parent_node = rb_entry(parent, struct ctl_node, node);
128 		parent_head = parent_node->header;
129 		parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
130 		parent_name = parent_entry->procname;
131 
132 		cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
133 		if (cmp < 0)
134 			p = &(*p)->rb_left;
135 		else if (cmp > 0)
136 			p = &(*p)->rb_right;
137 		else {
138 			pr_err("sysctl duplicate entry: ");
139 			sysctl_print_dir(head->parent);
140 			pr_cont("/%s\n", entry->procname);
141 			return -EEXIST;
142 		}
143 	}
144 
145 	rb_link_node(node, parent, p);
146 	rb_insert_color(node, &head->parent->root);
147 	return 0;
148 }
149 
150 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
151 {
152 	struct rb_node *node = &head->node[entry - head->ctl_table].node;
153 
154 	rb_erase(node, &head->parent->root);
155 }
156 
157 static void init_header(struct ctl_table_header *head,
158 	struct ctl_table_root *root, struct ctl_table_set *set,
159 	struct ctl_node *node, struct ctl_table *table)
160 {
161 	head->ctl_table = table;
162 	head->ctl_table_arg = table;
163 	head->used = 0;
164 	head->count = 1;
165 	head->nreg = 1;
166 	head->unregistering = NULL;
167 	head->root = root;
168 	head->set = set;
169 	head->parent = NULL;
170 	head->node = node;
171 	if (node) {
172 		struct ctl_table *entry;
173 		for (entry = table; entry->procname; entry++, node++)
174 			node->header = head;
175 	}
176 }
177 
178 static void erase_header(struct ctl_table_header *head)
179 {
180 	struct ctl_table *entry;
181 	for (entry = head->ctl_table; entry->procname; entry++)
182 		erase_entry(head, entry);
183 }
184 
185 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
186 {
187 	struct ctl_table *entry;
188 	int err;
189 
190 	dir->header.nreg++;
191 	header->parent = dir;
192 	err = insert_links(header);
193 	if (err)
194 		goto fail_links;
195 	for (entry = header->ctl_table; entry->procname; entry++) {
196 		err = insert_entry(header, entry);
197 		if (err)
198 			goto fail;
199 	}
200 	return 0;
201 fail:
202 	erase_header(header);
203 	put_links(header);
204 fail_links:
205 	header->parent = NULL;
206 	drop_sysctl_table(&dir->header);
207 	return err;
208 }
209 
210 /* called under sysctl_lock */
211 static int use_table(struct ctl_table_header *p)
212 {
213 	if (unlikely(p->unregistering))
214 		return 0;
215 	p->used++;
216 	return 1;
217 }
218 
219 /* called under sysctl_lock */
220 static void unuse_table(struct ctl_table_header *p)
221 {
222 	if (!--p->used)
223 		if (unlikely(p->unregistering))
224 			complete(p->unregistering);
225 }
226 
227 /* called under sysctl_lock, will reacquire if has to wait */
228 static void start_unregistering(struct ctl_table_header *p)
229 {
230 	/*
231 	 * if p->used is 0, nobody will ever touch that entry again;
232 	 * we'll eliminate all paths to it before dropping sysctl_lock
233 	 */
234 	if (unlikely(p->used)) {
235 		struct completion wait;
236 		init_completion(&wait);
237 		p->unregistering = &wait;
238 		spin_unlock(&sysctl_lock);
239 		wait_for_completion(&wait);
240 		spin_lock(&sysctl_lock);
241 	} else {
242 		/* anything non-NULL; we'll never dereference it */
243 		p->unregistering = ERR_PTR(-EINVAL);
244 	}
245 	/*
246 	 * do not remove from the list until nobody holds it; walking the
247 	 * list in do_sysctl() relies on that.
248 	 */
249 	erase_header(p);
250 }
251 
252 static void sysctl_head_get(struct ctl_table_header *head)
253 {
254 	spin_lock(&sysctl_lock);
255 	head->count++;
256 	spin_unlock(&sysctl_lock);
257 }
258 
259 void sysctl_head_put(struct ctl_table_header *head)
260 {
261 	spin_lock(&sysctl_lock);
262 	if (!--head->count)
263 		kfree_rcu(head, rcu);
264 	spin_unlock(&sysctl_lock);
265 }
266 
267 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
268 {
269 	BUG_ON(!head);
270 	spin_lock(&sysctl_lock);
271 	if (!use_table(head))
272 		head = ERR_PTR(-ENOENT);
273 	spin_unlock(&sysctl_lock);
274 	return head;
275 }
276 
277 static void sysctl_head_finish(struct ctl_table_header *head)
278 {
279 	if (!head)
280 		return;
281 	spin_lock(&sysctl_lock);
282 	unuse_table(head);
283 	spin_unlock(&sysctl_lock);
284 }
285 
286 static struct ctl_table_set *
287 lookup_header_set(struct ctl_table_root *root, struct nsproxy *namespaces)
288 {
289 	struct ctl_table_set *set = &root->default_set;
290 	if (root->lookup)
291 		set = root->lookup(root, namespaces);
292 	return set;
293 }
294 
295 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
296 				      struct ctl_dir *dir,
297 				      const char *name, int namelen)
298 {
299 	struct ctl_table_header *head;
300 	struct ctl_table *entry;
301 
302 	spin_lock(&sysctl_lock);
303 	entry = find_entry(&head, dir, name, namelen);
304 	if (entry && use_table(head))
305 		*phead = head;
306 	else
307 		entry = NULL;
308 	spin_unlock(&sysctl_lock);
309 	return entry;
310 }
311 
312 static struct ctl_node *first_usable_entry(struct rb_node *node)
313 {
314 	struct ctl_node *ctl_node;
315 
316 	for (;node; node = rb_next(node)) {
317 		ctl_node = rb_entry(node, struct ctl_node, node);
318 		if (use_table(ctl_node->header))
319 			return ctl_node;
320 	}
321 	return NULL;
322 }
323 
324 static void first_entry(struct ctl_dir *dir,
325 	struct ctl_table_header **phead, struct ctl_table **pentry)
326 {
327 	struct ctl_table_header *head = NULL;
328 	struct ctl_table *entry = NULL;
329 	struct ctl_node *ctl_node;
330 
331 	spin_lock(&sysctl_lock);
332 	ctl_node = first_usable_entry(rb_first(&dir->root));
333 	spin_unlock(&sysctl_lock);
334 	if (ctl_node) {
335 		head = ctl_node->header;
336 		entry = &head->ctl_table[ctl_node - head->node];
337 	}
338 	*phead = head;
339 	*pentry = entry;
340 }
341 
342 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
343 {
344 	struct ctl_table_header *head = *phead;
345 	struct ctl_table *entry = *pentry;
346 	struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
347 
348 	spin_lock(&sysctl_lock);
349 	unuse_table(head);
350 
351 	ctl_node = first_usable_entry(rb_next(&ctl_node->node));
352 	spin_unlock(&sysctl_lock);
353 	head = NULL;
354 	if (ctl_node) {
355 		head = ctl_node->header;
356 		entry = &head->ctl_table[ctl_node - head->node];
357 	}
358 	*phead = head;
359 	*pentry = entry;
360 }
361 
362 void register_sysctl_root(struct ctl_table_root *root)
363 {
364 }
365 
366 /*
367  * sysctl_perm does NOT grant the superuser all rights automatically, because
368  * some sysctl variables are readonly even to root.
369  */
370 
371 static int test_perm(int mode, int op)
372 {
373 	if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
374 		mode >>= 6;
375 	else if (in_egroup_p(GLOBAL_ROOT_GID))
376 		mode >>= 3;
377 	if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
378 		return 0;
379 	return -EACCES;
380 }
381 
382 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
383 {
384 	struct ctl_table_root *root = head->root;
385 	int mode;
386 
387 	if (root->permissions)
388 		mode = root->permissions(head, table);
389 	else
390 		mode = table->mode;
391 
392 	return test_perm(mode, op);
393 }
394 
395 static struct inode *proc_sys_make_inode(struct super_block *sb,
396 		struct ctl_table_header *head, struct ctl_table *table)
397 {
398 	struct inode *inode;
399 	struct proc_inode *ei;
400 
401 	inode = new_inode(sb);
402 	if (!inode)
403 		goto out;
404 
405 	inode->i_ino = get_next_ino();
406 
407 	sysctl_head_get(head);
408 	ei = PROC_I(inode);
409 	ei->sysctl = head;
410 	ei->sysctl_entry = table;
411 
412 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
413 	inode->i_mode = table->mode;
414 	if (!S_ISDIR(table->mode)) {
415 		inode->i_mode |= S_IFREG;
416 		inode->i_op = &proc_sys_inode_operations;
417 		inode->i_fop = &proc_sys_file_operations;
418 	} else {
419 		inode->i_mode |= S_IFDIR;
420 		inode->i_op = &proc_sys_dir_operations;
421 		inode->i_fop = &proc_sys_dir_file_operations;
422 	}
423 out:
424 	return inode;
425 }
426 
427 static struct ctl_table_header *grab_header(struct inode *inode)
428 {
429 	struct ctl_table_header *head = PROC_I(inode)->sysctl;
430 	if (!head)
431 		head = &sysctl_table_root.default_set.dir.header;
432 	return sysctl_head_grab(head);
433 }
434 
435 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
436 					unsigned int flags)
437 {
438 	struct ctl_table_header *head = grab_header(dir);
439 	struct ctl_table_header *h = NULL;
440 	struct qstr *name = &dentry->d_name;
441 	struct ctl_table *p;
442 	struct inode *inode;
443 	struct dentry *err = ERR_PTR(-ENOENT);
444 	struct ctl_dir *ctl_dir;
445 	int ret;
446 
447 	if (IS_ERR(head))
448 		return ERR_CAST(head);
449 
450 	ctl_dir = container_of(head, struct ctl_dir, header);
451 
452 	p = lookup_entry(&h, ctl_dir, name->name, name->len);
453 	if (!p)
454 		goto out;
455 
456 	if (S_ISLNK(p->mode)) {
457 		ret = sysctl_follow_link(&h, &p, current->nsproxy);
458 		err = ERR_PTR(ret);
459 		if (ret)
460 			goto out;
461 	}
462 
463 	err = ERR_PTR(-ENOMEM);
464 	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
465 	if (!inode)
466 		goto out;
467 
468 	err = NULL;
469 	d_set_d_op(dentry, &proc_sys_dentry_operations);
470 	d_add(dentry, inode);
471 
472 out:
473 	if (h)
474 		sysctl_head_finish(h);
475 	sysctl_head_finish(head);
476 	return err;
477 }
478 
479 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
480 		size_t count, loff_t *ppos, int write)
481 {
482 	struct inode *inode = file_inode(filp);
483 	struct ctl_table_header *head = grab_header(inode);
484 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
485 	ssize_t error;
486 	size_t res;
487 
488 	if (IS_ERR(head))
489 		return PTR_ERR(head);
490 
491 	/*
492 	 * At this point we know that the sysctl was not unregistered
493 	 * and won't be until we finish.
494 	 */
495 	error = -EPERM;
496 	if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
497 		goto out;
498 
499 	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
500 	error = -EINVAL;
501 	if (!table->proc_handler)
502 		goto out;
503 
504 	/* careful: calling conventions are nasty here */
505 	res = count;
506 	error = table->proc_handler(table, write, buf, &res, ppos);
507 	if (!error)
508 		error = res;
509 out:
510 	sysctl_head_finish(head);
511 
512 	return error;
513 }
514 
515 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
516 				size_t count, loff_t *ppos)
517 {
518 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
519 }
520 
521 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
522 				size_t count, loff_t *ppos)
523 {
524 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
525 }
526 
527 static int proc_sys_open(struct inode *inode, struct file *filp)
528 {
529 	struct ctl_table_header *head = grab_header(inode);
530 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
531 
532 	/* sysctl was unregistered */
533 	if (IS_ERR(head))
534 		return PTR_ERR(head);
535 
536 	if (table->poll)
537 		filp->private_data = proc_sys_poll_event(table->poll);
538 
539 	sysctl_head_finish(head);
540 
541 	return 0;
542 }
543 
544 static unsigned int proc_sys_poll(struct file *filp, poll_table *wait)
545 {
546 	struct inode *inode = file_inode(filp);
547 	struct ctl_table_header *head = grab_header(inode);
548 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
549 	unsigned int ret = DEFAULT_POLLMASK;
550 	unsigned long event;
551 
552 	/* sysctl was unregistered */
553 	if (IS_ERR(head))
554 		return POLLERR | POLLHUP;
555 
556 	if (!table->proc_handler)
557 		goto out;
558 
559 	if (!table->poll)
560 		goto out;
561 
562 	event = (unsigned long)filp->private_data;
563 	poll_wait(filp, &table->poll->wait, wait);
564 
565 	if (event != atomic_read(&table->poll->event)) {
566 		filp->private_data = proc_sys_poll_event(table->poll);
567 		ret = POLLIN | POLLRDNORM | POLLERR | POLLPRI;
568 	}
569 
570 out:
571 	sysctl_head_finish(head);
572 
573 	return ret;
574 }
575 
576 static bool proc_sys_fill_cache(struct file *file,
577 				struct dir_context *ctx,
578 				struct ctl_table_header *head,
579 				struct ctl_table *table)
580 {
581 	struct dentry *child, *dir = file->f_path.dentry;
582 	struct inode *inode;
583 	struct qstr qname;
584 	ino_t ino = 0;
585 	unsigned type = DT_UNKNOWN;
586 
587 	qname.name = table->procname;
588 	qname.len  = strlen(table->procname);
589 	qname.hash = full_name_hash(qname.name, qname.len);
590 
591 	child = d_lookup(dir, &qname);
592 	if (!child) {
593 		child = d_alloc(dir, &qname);
594 		if (child) {
595 			inode = proc_sys_make_inode(dir->d_sb, head, table);
596 			if (!inode) {
597 				dput(child);
598 				return false;
599 			} else {
600 				d_set_d_op(child, &proc_sys_dentry_operations);
601 				d_add(child, inode);
602 			}
603 		} else {
604 			return false;
605 		}
606 	}
607 	inode = child->d_inode;
608 	ino  = inode->i_ino;
609 	type = inode->i_mode >> 12;
610 	dput(child);
611 	return dir_emit(ctx, qname.name, qname.len, ino, type);
612 }
613 
614 static bool proc_sys_link_fill_cache(struct file *file,
615 				    struct dir_context *ctx,
616 				    struct ctl_table_header *head,
617 				    struct ctl_table *table)
618 {
619 	bool ret = true;
620 	head = sysctl_head_grab(head);
621 
622 	if (S_ISLNK(table->mode)) {
623 		/* It is not an error if we can not follow the link ignore it */
624 		int err = sysctl_follow_link(&head, &table, current->nsproxy);
625 		if (err)
626 			goto out;
627 	}
628 
629 	ret = proc_sys_fill_cache(file, ctx, head, table);
630 out:
631 	sysctl_head_finish(head);
632 	return ret;
633 }
634 
635 static int scan(struct ctl_table_header *head, struct ctl_table *table,
636 		unsigned long *pos, struct file *file,
637 		struct dir_context *ctx)
638 {
639 	bool res;
640 
641 	if ((*pos)++ < ctx->pos)
642 		return true;
643 
644 	if (unlikely(S_ISLNK(table->mode)))
645 		res = proc_sys_link_fill_cache(file, ctx, head, table);
646 	else
647 		res = proc_sys_fill_cache(file, ctx, head, table);
648 
649 	if (res)
650 		ctx->pos = *pos;
651 
652 	return res;
653 }
654 
655 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
656 {
657 	struct ctl_table_header *head = grab_header(file_inode(file));
658 	struct ctl_table_header *h = NULL;
659 	struct ctl_table *entry;
660 	struct ctl_dir *ctl_dir;
661 	unsigned long pos;
662 
663 	if (IS_ERR(head))
664 		return PTR_ERR(head);
665 
666 	ctl_dir = container_of(head, struct ctl_dir, header);
667 
668 	if (!dir_emit_dots(file, ctx))
669 		return 0;
670 
671 	pos = 2;
672 
673 	for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
674 		if (!scan(h, entry, &pos, file, ctx)) {
675 			sysctl_head_finish(h);
676 			break;
677 		}
678 	}
679 	sysctl_head_finish(head);
680 	return 0;
681 }
682 
683 static int proc_sys_permission(struct inode *inode, int mask)
684 {
685 	/*
686 	 * sysctl entries that are not writeable,
687 	 * are _NOT_ writeable, capabilities or not.
688 	 */
689 	struct ctl_table_header *head;
690 	struct ctl_table *table;
691 	int error;
692 
693 	/* Executable files are not allowed under /proc/sys/ */
694 	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
695 		return -EACCES;
696 
697 	head = grab_header(inode);
698 	if (IS_ERR(head))
699 		return PTR_ERR(head);
700 
701 	table = PROC_I(inode)->sysctl_entry;
702 	if (!table) /* global root - r-xr-xr-x */
703 		error = mask & MAY_WRITE ? -EACCES : 0;
704 	else /* Use the permissions on the sysctl table entry */
705 		error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
706 
707 	sysctl_head_finish(head);
708 	return error;
709 }
710 
711 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
712 {
713 	struct inode *inode = dentry->d_inode;
714 	int error;
715 
716 	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
717 		return -EPERM;
718 
719 	error = inode_change_ok(inode, attr);
720 	if (error)
721 		return error;
722 
723 	setattr_copy(inode, attr);
724 	mark_inode_dirty(inode);
725 	return 0;
726 }
727 
728 static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
729 {
730 	struct inode *inode = dentry->d_inode;
731 	struct ctl_table_header *head = grab_header(inode);
732 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
733 
734 	if (IS_ERR(head))
735 		return PTR_ERR(head);
736 
737 	generic_fillattr(inode, stat);
738 	if (table)
739 		stat->mode = (stat->mode & S_IFMT) | table->mode;
740 
741 	sysctl_head_finish(head);
742 	return 0;
743 }
744 
745 static const struct file_operations proc_sys_file_operations = {
746 	.open		= proc_sys_open,
747 	.poll		= proc_sys_poll,
748 	.read		= proc_sys_read,
749 	.write		= proc_sys_write,
750 	.llseek		= default_llseek,
751 };
752 
753 static const struct file_operations proc_sys_dir_file_operations = {
754 	.read		= generic_read_dir,
755 	.iterate	= proc_sys_readdir,
756 	.llseek		= generic_file_llseek,
757 };
758 
759 static const struct inode_operations proc_sys_inode_operations = {
760 	.permission	= proc_sys_permission,
761 	.setattr	= proc_sys_setattr,
762 	.getattr	= proc_sys_getattr,
763 };
764 
765 static const struct inode_operations proc_sys_dir_operations = {
766 	.lookup		= proc_sys_lookup,
767 	.permission	= proc_sys_permission,
768 	.setattr	= proc_sys_setattr,
769 	.getattr	= proc_sys_getattr,
770 };
771 
772 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
773 {
774 	if (flags & LOOKUP_RCU)
775 		return -ECHILD;
776 	return !PROC_I(dentry->d_inode)->sysctl->unregistering;
777 }
778 
779 static int proc_sys_delete(const struct dentry *dentry)
780 {
781 	return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
782 }
783 
784 static int sysctl_is_seen(struct ctl_table_header *p)
785 {
786 	struct ctl_table_set *set = p->set;
787 	int res;
788 	spin_lock(&sysctl_lock);
789 	if (p->unregistering)
790 		res = 0;
791 	else if (!set->is_seen)
792 		res = 1;
793 	else
794 		res = set->is_seen(set);
795 	spin_unlock(&sysctl_lock);
796 	return res;
797 }
798 
799 static int proc_sys_compare(const struct dentry *parent, const struct dentry *dentry,
800 		unsigned int len, const char *str, const struct qstr *name)
801 {
802 	struct ctl_table_header *head;
803 	struct inode *inode;
804 
805 	/* Although proc doesn't have negative dentries, rcu-walk means
806 	 * that inode here can be NULL */
807 	/* AV: can it, indeed? */
808 	inode = ACCESS_ONCE(dentry->d_inode);
809 	if (!inode)
810 		return 1;
811 	if (name->len != len)
812 		return 1;
813 	if (memcmp(name->name, str, len))
814 		return 1;
815 	head = rcu_dereference(PROC_I(inode)->sysctl);
816 	return !head || !sysctl_is_seen(head);
817 }
818 
819 static const struct dentry_operations proc_sys_dentry_operations = {
820 	.d_revalidate	= proc_sys_revalidate,
821 	.d_delete	= proc_sys_delete,
822 	.d_compare	= proc_sys_compare,
823 };
824 
825 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
826 				   const char *name, int namelen)
827 {
828 	struct ctl_table_header *head;
829 	struct ctl_table *entry;
830 
831 	entry = find_entry(&head, dir, name, namelen);
832 	if (!entry)
833 		return ERR_PTR(-ENOENT);
834 	if (!S_ISDIR(entry->mode))
835 		return ERR_PTR(-ENOTDIR);
836 	return container_of(head, struct ctl_dir, header);
837 }
838 
839 static struct ctl_dir *new_dir(struct ctl_table_set *set,
840 			       const char *name, int namelen)
841 {
842 	struct ctl_table *table;
843 	struct ctl_dir *new;
844 	struct ctl_node *node;
845 	char *new_name;
846 
847 	new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
848 		      sizeof(struct ctl_table)*2 +  namelen + 1,
849 		      GFP_KERNEL);
850 	if (!new)
851 		return NULL;
852 
853 	node = (struct ctl_node *)(new + 1);
854 	table = (struct ctl_table *)(node + 1);
855 	new_name = (char *)(table + 2);
856 	memcpy(new_name, name, namelen);
857 	new_name[namelen] = '\0';
858 	table[0].procname = new_name;
859 	table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
860 	init_header(&new->header, set->dir.header.root, set, node, table);
861 
862 	return new;
863 }
864 
865 /**
866  * get_subdir - find or create a subdir with the specified name.
867  * @dir:  Directory to create the subdirectory in
868  * @name: The name of the subdirectory to find or create
869  * @namelen: The length of name
870  *
871  * Takes a directory with an elevated reference count so we know that
872  * if we drop the lock the directory will not go away.  Upon success
873  * the reference is moved from @dir to the returned subdirectory.
874  * Upon error an error code is returned and the reference on @dir is
875  * simply dropped.
876  */
877 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
878 				  const char *name, int namelen)
879 {
880 	struct ctl_table_set *set = dir->header.set;
881 	struct ctl_dir *subdir, *new = NULL;
882 	int err;
883 
884 	spin_lock(&sysctl_lock);
885 	subdir = find_subdir(dir, name, namelen);
886 	if (!IS_ERR(subdir))
887 		goto found;
888 	if (PTR_ERR(subdir) != -ENOENT)
889 		goto failed;
890 
891 	spin_unlock(&sysctl_lock);
892 	new = new_dir(set, name, namelen);
893 	spin_lock(&sysctl_lock);
894 	subdir = ERR_PTR(-ENOMEM);
895 	if (!new)
896 		goto failed;
897 
898 	/* Was the subdir added while we dropped the lock? */
899 	subdir = find_subdir(dir, name, namelen);
900 	if (!IS_ERR(subdir))
901 		goto found;
902 	if (PTR_ERR(subdir) != -ENOENT)
903 		goto failed;
904 
905 	/* Nope.  Use the our freshly made directory entry. */
906 	err = insert_header(dir, &new->header);
907 	subdir = ERR_PTR(err);
908 	if (err)
909 		goto failed;
910 	subdir = new;
911 found:
912 	subdir->header.nreg++;
913 failed:
914 	if (unlikely(IS_ERR(subdir))) {
915 		pr_err("sysctl could not get directory: ");
916 		sysctl_print_dir(dir);
917 		pr_cont("/%*.*s %ld\n",
918 			namelen, namelen, name, PTR_ERR(subdir));
919 	}
920 	drop_sysctl_table(&dir->header);
921 	if (new)
922 		drop_sysctl_table(&new->header);
923 	spin_unlock(&sysctl_lock);
924 	return subdir;
925 }
926 
927 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
928 {
929 	struct ctl_dir *parent;
930 	const char *procname;
931 	if (!dir->header.parent)
932 		return &set->dir;
933 	parent = xlate_dir(set, dir->header.parent);
934 	if (IS_ERR(parent))
935 		return parent;
936 	procname = dir->header.ctl_table[0].procname;
937 	return find_subdir(parent, procname, strlen(procname));
938 }
939 
940 static int sysctl_follow_link(struct ctl_table_header **phead,
941 	struct ctl_table **pentry, struct nsproxy *namespaces)
942 {
943 	struct ctl_table_header *head;
944 	struct ctl_table_root *root;
945 	struct ctl_table_set *set;
946 	struct ctl_table *entry;
947 	struct ctl_dir *dir;
948 	int ret;
949 
950 	ret = 0;
951 	spin_lock(&sysctl_lock);
952 	root = (*pentry)->data;
953 	set = lookup_header_set(root, namespaces);
954 	dir = xlate_dir(set, (*phead)->parent);
955 	if (IS_ERR(dir))
956 		ret = PTR_ERR(dir);
957 	else {
958 		const char *procname = (*pentry)->procname;
959 		head = NULL;
960 		entry = find_entry(&head, dir, procname, strlen(procname));
961 		ret = -ENOENT;
962 		if (entry && use_table(head)) {
963 			unuse_table(*phead);
964 			*phead = head;
965 			*pentry = entry;
966 			ret = 0;
967 		}
968 	}
969 
970 	spin_unlock(&sysctl_lock);
971 	return ret;
972 }
973 
974 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
975 {
976 	struct va_format vaf;
977 	va_list args;
978 
979 	va_start(args, fmt);
980 	vaf.fmt = fmt;
981 	vaf.va = &args;
982 
983 	pr_err("sysctl table check failed: %s/%s %pV\n",
984 	       path, table->procname, &vaf);
985 
986 	va_end(args);
987 	return -EINVAL;
988 }
989 
990 static int sysctl_check_table(const char *path, struct ctl_table *table)
991 {
992 	int err = 0;
993 	for (; table->procname; table++) {
994 		if (table->child)
995 			err = sysctl_err(path, table, "Not a file");
996 
997 		if ((table->proc_handler == proc_dostring) ||
998 		    (table->proc_handler == proc_dointvec) ||
999 		    (table->proc_handler == proc_dointvec_minmax) ||
1000 		    (table->proc_handler == proc_dointvec_jiffies) ||
1001 		    (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1002 		    (table->proc_handler == proc_dointvec_ms_jiffies) ||
1003 		    (table->proc_handler == proc_doulongvec_minmax) ||
1004 		    (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1005 			if (!table->data)
1006 				err = sysctl_err(path, table, "No data");
1007 			if (!table->maxlen)
1008 				err = sysctl_err(path, table, "No maxlen");
1009 		}
1010 		if (!table->proc_handler)
1011 			err = sysctl_err(path, table, "No proc_handler");
1012 
1013 		if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1014 			err = sysctl_err(path, table, "bogus .mode 0%o",
1015 				table->mode);
1016 	}
1017 	return err;
1018 }
1019 
1020 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1021 	struct ctl_table_root *link_root)
1022 {
1023 	struct ctl_table *link_table, *entry, *link;
1024 	struct ctl_table_header *links;
1025 	struct ctl_node *node;
1026 	char *link_name;
1027 	int nr_entries, name_bytes;
1028 
1029 	name_bytes = 0;
1030 	nr_entries = 0;
1031 	for (entry = table; entry->procname; entry++) {
1032 		nr_entries++;
1033 		name_bytes += strlen(entry->procname) + 1;
1034 	}
1035 
1036 	links = kzalloc(sizeof(struct ctl_table_header) +
1037 			sizeof(struct ctl_node)*nr_entries +
1038 			sizeof(struct ctl_table)*(nr_entries + 1) +
1039 			name_bytes,
1040 			GFP_KERNEL);
1041 
1042 	if (!links)
1043 		return NULL;
1044 
1045 	node = (struct ctl_node *)(links + 1);
1046 	link_table = (struct ctl_table *)(node + nr_entries);
1047 	link_name = (char *)&link_table[nr_entries + 1];
1048 
1049 	for (link = link_table, entry = table; entry->procname; link++, entry++) {
1050 		int len = strlen(entry->procname) + 1;
1051 		memcpy(link_name, entry->procname, len);
1052 		link->procname = link_name;
1053 		link->mode = S_IFLNK|S_IRWXUGO;
1054 		link->data = link_root;
1055 		link_name += len;
1056 	}
1057 	init_header(links, dir->header.root, dir->header.set, node, link_table);
1058 	links->nreg = nr_entries;
1059 
1060 	return links;
1061 }
1062 
1063 static bool get_links(struct ctl_dir *dir,
1064 	struct ctl_table *table, struct ctl_table_root *link_root)
1065 {
1066 	struct ctl_table_header *head;
1067 	struct ctl_table *entry, *link;
1068 
1069 	/* Are there links available for every entry in table? */
1070 	for (entry = table; entry->procname; entry++) {
1071 		const char *procname = entry->procname;
1072 		link = find_entry(&head, dir, procname, strlen(procname));
1073 		if (!link)
1074 			return false;
1075 		if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1076 			continue;
1077 		if (S_ISLNK(link->mode) && (link->data == link_root))
1078 			continue;
1079 		return false;
1080 	}
1081 
1082 	/* The checks passed.  Increase the registration count on the links */
1083 	for (entry = table; entry->procname; entry++) {
1084 		const char *procname = entry->procname;
1085 		link = find_entry(&head, dir, procname, strlen(procname));
1086 		head->nreg++;
1087 	}
1088 	return true;
1089 }
1090 
1091 static int insert_links(struct ctl_table_header *head)
1092 {
1093 	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1094 	struct ctl_dir *core_parent = NULL;
1095 	struct ctl_table_header *links;
1096 	int err;
1097 
1098 	if (head->set == root_set)
1099 		return 0;
1100 
1101 	core_parent = xlate_dir(root_set, head->parent);
1102 	if (IS_ERR(core_parent))
1103 		return 0;
1104 
1105 	if (get_links(core_parent, head->ctl_table, head->root))
1106 		return 0;
1107 
1108 	core_parent->header.nreg++;
1109 	spin_unlock(&sysctl_lock);
1110 
1111 	links = new_links(core_parent, head->ctl_table, head->root);
1112 
1113 	spin_lock(&sysctl_lock);
1114 	err = -ENOMEM;
1115 	if (!links)
1116 		goto out;
1117 
1118 	err = 0;
1119 	if (get_links(core_parent, head->ctl_table, head->root)) {
1120 		kfree(links);
1121 		goto out;
1122 	}
1123 
1124 	err = insert_header(core_parent, links);
1125 	if (err)
1126 		kfree(links);
1127 out:
1128 	drop_sysctl_table(&core_parent->header);
1129 	return err;
1130 }
1131 
1132 /**
1133  * __register_sysctl_table - register a leaf sysctl table
1134  * @set: Sysctl tree to register on
1135  * @path: The path to the directory the sysctl table is in.
1136  * @table: the top-level table structure
1137  *
1138  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1139  * array. A completely 0 filled entry terminates the table.
1140  *
1141  * The members of the &struct ctl_table structure are used as follows:
1142  *
1143  * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1144  *            enter a sysctl file
1145  *
1146  * data - a pointer to data for use by proc_handler
1147  *
1148  * maxlen - the maximum size in bytes of the data
1149  *
1150  * mode - the file permissions for the /proc/sys file
1151  *
1152  * child - must be %NULL.
1153  *
1154  * proc_handler - the text handler routine (described below)
1155  *
1156  * extra1, extra2 - extra pointers usable by the proc handler routines
1157  *
1158  * Leaf nodes in the sysctl tree will be represented by a single file
1159  * under /proc; non-leaf nodes will be represented by directories.
1160  *
1161  * There must be a proc_handler routine for any terminal nodes.
1162  * Several default handlers are available to cover common cases -
1163  *
1164  * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1165  * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1166  * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1167  *
1168  * It is the handler's job to read the input buffer from user memory
1169  * and process it. The handler should return 0 on success.
1170  *
1171  * This routine returns %NULL on a failure to register, and a pointer
1172  * to the table header on success.
1173  */
1174 struct ctl_table_header *__register_sysctl_table(
1175 	struct ctl_table_set *set,
1176 	const char *path, struct ctl_table *table)
1177 {
1178 	struct ctl_table_root *root = set->dir.header.root;
1179 	struct ctl_table_header *header;
1180 	const char *name, *nextname;
1181 	struct ctl_dir *dir;
1182 	struct ctl_table *entry;
1183 	struct ctl_node *node;
1184 	int nr_entries = 0;
1185 
1186 	for (entry = table; entry->procname; entry++)
1187 		nr_entries++;
1188 
1189 	header = kzalloc(sizeof(struct ctl_table_header) +
1190 			 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1191 	if (!header)
1192 		return NULL;
1193 
1194 	node = (struct ctl_node *)(header + 1);
1195 	init_header(header, root, set, node, table);
1196 	if (sysctl_check_table(path, table))
1197 		goto fail;
1198 
1199 	spin_lock(&sysctl_lock);
1200 	dir = &set->dir;
1201 	/* Reference moved down the diretory tree get_subdir */
1202 	dir->header.nreg++;
1203 	spin_unlock(&sysctl_lock);
1204 
1205 	/* Find the directory for the ctl_table */
1206 	for (name = path; name; name = nextname) {
1207 		int namelen;
1208 		nextname = strchr(name, '/');
1209 		if (nextname) {
1210 			namelen = nextname - name;
1211 			nextname++;
1212 		} else {
1213 			namelen = strlen(name);
1214 		}
1215 		if (namelen == 0)
1216 			continue;
1217 
1218 		dir = get_subdir(dir, name, namelen);
1219 		if (IS_ERR(dir))
1220 			goto fail;
1221 	}
1222 
1223 	spin_lock(&sysctl_lock);
1224 	if (insert_header(dir, header))
1225 		goto fail_put_dir_locked;
1226 
1227 	drop_sysctl_table(&dir->header);
1228 	spin_unlock(&sysctl_lock);
1229 
1230 	return header;
1231 
1232 fail_put_dir_locked:
1233 	drop_sysctl_table(&dir->header);
1234 	spin_unlock(&sysctl_lock);
1235 fail:
1236 	kfree(header);
1237 	dump_stack();
1238 	return NULL;
1239 }
1240 
1241 /**
1242  * register_sysctl - register a sysctl table
1243  * @path: The path to the directory the sysctl table is in.
1244  * @table: the table structure
1245  *
1246  * Register a sysctl table. @table should be a filled in ctl_table
1247  * array. A completely 0 filled entry terminates the table.
1248  *
1249  * See __register_sysctl_table for more details.
1250  */
1251 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1252 {
1253 	return __register_sysctl_table(&sysctl_table_root.default_set,
1254 					path, table);
1255 }
1256 EXPORT_SYMBOL(register_sysctl);
1257 
1258 static char *append_path(const char *path, char *pos, const char *name)
1259 {
1260 	int namelen;
1261 	namelen = strlen(name);
1262 	if (((pos - path) + namelen + 2) >= PATH_MAX)
1263 		return NULL;
1264 	memcpy(pos, name, namelen);
1265 	pos[namelen] = '/';
1266 	pos[namelen + 1] = '\0';
1267 	pos += namelen + 1;
1268 	return pos;
1269 }
1270 
1271 static int count_subheaders(struct ctl_table *table)
1272 {
1273 	int has_files = 0;
1274 	int nr_subheaders = 0;
1275 	struct ctl_table *entry;
1276 
1277 	/* special case: no directory and empty directory */
1278 	if (!table || !table->procname)
1279 		return 1;
1280 
1281 	for (entry = table; entry->procname; entry++) {
1282 		if (entry->child)
1283 			nr_subheaders += count_subheaders(entry->child);
1284 		else
1285 			has_files = 1;
1286 	}
1287 	return nr_subheaders + has_files;
1288 }
1289 
1290 static int register_leaf_sysctl_tables(const char *path, char *pos,
1291 	struct ctl_table_header ***subheader, struct ctl_table_set *set,
1292 	struct ctl_table *table)
1293 {
1294 	struct ctl_table *ctl_table_arg = NULL;
1295 	struct ctl_table *entry, *files;
1296 	int nr_files = 0;
1297 	int nr_dirs = 0;
1298 	int err = -ENOMEM;
1299 
1300 	for (entry = table; entry->procname; entry++) {
1301 		if (entry->child)
1302 			nr_dirs++;
1303 		else
1304 			nr_files++;
1305 	}
1306 
1307 	files = table;
1308 	/* If there are mixed files and directories we need a new table */
1309 	if (nr_dirs && nr_files) {
1310 		struct ctl_table *new;
1311 		files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1312 				GFP_KERNEL);
1313 		if (!files)
1314 			goto out;
1315 
1316 		ctl_table_arg = files;
1317 		for (new = files, entry = table; entry->procname; entry++) {
1318 			if (entry->child)
1319 				continue;
1320 			*new = *entry;
1321 			new++;
1322 		}
1323 	}
1324 
1325 	/* Register everything except a directory full of subdirectories */
1326 	if (nr_files || !nr_dirs) {
1327 		struct ctl_table_header *header;
1328 		header = __register_sysctl_table(set, path, files);
1329 		if (!header) {
1330 			kfree(ctl_table_arg);
1331 			goto out;
1332 		}
1333 
1334 		/* Remember if we need to free the file table */
1335 		header->ctl_table_arg = ctl_table_arg;
1336 		**subheader = header;
1337 		(*subheader)++;
1338 	}
1339 
1340 	/* Recurse into the subdirectories. */
1341 	for (entry = table; entry->procname; entry++) {
1342 		char *child_pos;
1343 
1344 		if (!entry->child)
1345 			continue;
1346 
1347 		err = -ENAMETOOLONG;
1348 		child_pos = append_path(path, pos, entry->procname);
1349 		if (!child_pos)
1350 			goto out;
1351 
1352 		err = register_leaf_sysctl_tables(path, child_pos, subheader,
1353 						  set, entry->child);
1354 		pos[0] = '\0';
1355 		if (err)
1356 			goto out;
1357 	}
1358 	err = 0;
1359 out:
1360 	/* On failure our caller will unregister all registered subheaders */
1361 	return err;
1362 }
1363 
1364 /**
1365  * __register_sysctl_paths - register a sysctl table hierarchy
1366  * @set: Sysctl tree to register on
1367  * @path: The path to the directory the sysctl table is in.
1368  * @table: the top-level table structure
1369  *
1370  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1371  * array. A completely 0 filled entry terminates the table.
1372  *
1373  * See __register_sysctl_table for more details.
1374  */
1375 struct ctl_table_header *__register_sysctl_paths(
1376 	struct ctl_table_set *set,
1377 	const struct ctl_path *path, struct ctl_table *table)
1378 {
1379 	struct ctl_table *ctl_table_arg = table;
1380 	int nr_subheaders = count_subheaders(table);
1381 	struct ctl_table_header *header = NULL, **subheaders, **subheader;
1382 	const struct ctl_path *component;
1383 	char *new_path, *pos;
1384 
1385 	pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1386 	if (!new_path)
1387 		return NULL;
1388 
1389 	pos[0] = '\0';
1390 	for (component = path; component->procname; component++) {
1391 		pos = append_path(new_path, pos, component->procname);
1392 		if (!pos)
1393 			goto out;
1394 	}
1395 	while (table->procname && table->child && !table[1].procname) {
1396 		pos = append_path(new_path, pos, table->procname);
1397 		if (!pos)
1398 			goto out;
1399 		table = table->child;
1400 	}
1401 	if (nr_subheaders == 1) {
1402 		header = __register_sysctl_table(set, new_path, table);
1403 		if (header)
1404 			header->ctl_table_arg = ctl_table_arg;
1405 	} else {
1406 		header = kzalloc(sizeof(*header) +
1407 				 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1408 		if (!header)
1409 			goto out;
1410 
1411 		subheaders = (struct ctl_table_header **) (header + 1);
1412 		subheader = subheaders;
1413 		header->ctl_table_arg = ctl_table_arg;
1414 
1415 		if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1416 						set, table))
1417 			goto err_register_leaves;
1418 	}
1419 
1420 out:
1421 	kfree(new_path);
1422 	return header;
1423 
1424 err_register_leaves:
1425 	while (subheader > subheaders) {
1426 		struct ctl_table_header *subh = *(--subheader);
1427 		struct ctl_table *table = subh->ctl_table_arg;
1428 		unregister_sysctl_table(subh);
1429 		kfree(table);
1430 	}
1431 	kfree(header);
1432 	header = NULL;
1433 	goto out;
1434 }
1435 
1436 /**
1437  * register_sysctl_table_path - register a sysctl table hierarchy
1438  * @path: The path to the directory the sysctl table is in.
1439  * @table: the top-level table structure
1440  *
1441  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1442  * array. A completely 0 filled entry terminates the table.
1443  *
1444  * See __register_sysctl_paths for more details.
1445  */
1446 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1447 						struct ctl_table *table)
1448 {
1449 	return __register_sysctl_paths(&sysctl_table_root.default_set,
1450 					path, table);
1451 }
1452 EXPORT_SYMBOL(register_sysctl_paths);
1453 
1454 /**
1455  * register_sysctl_table - register a sysctl table hierarchy
1456  * @table: the top-level table structure
1457  *
1458  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1459  * array. A completely 0 filled entry terminates the table.
1460  *
1461  * See register_sysctl_paths for more details.
1462  */
1463 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1464 {
1465 	static const struct ctl_path null_path[] = { {} };
1466 
1467 	return register_sysctl_paths(null_path, table);
1468 }
1469 EXPORT_SYMBOL(register_sysctl_table);
1470 
1471 static void put_links(struct ctl_table_header *header)
1472 {
1473 	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1474 	struct ctl_table_root *root = header->root;
1475 	struct ctl_dir *parent = header->parent;
1476 	struct ctl_dir *core_parent;
1477 	struct ctl_table *entry;
1478 
1479 	if (header->set == root_set)
1480 		return;
1481 
1482 	core_parent = xlate_dir(root_set, parent);
1483 	if (IS_ERR(core_parent))
1484 		return;
1485 
1486 	for (entry = header->ctl_table; entry->procname; entry++) {
1487 		struct ctl_table_header *link_head;
1488 		struct ctl_table *link;
1489 		const char *name = entry->procname;
1490 
1491 		link = find_entry(&link_head, core_parent, name, strlen(name));
1492 		if (link &&
1493 		    ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1494 		     (S_ISLNK(link->mode) && (link->data == root)))) {
1495 			drop_sysctl_table(link_head);
1496 		}
1497 		else {
1498 			pr_err("sysctl link missing during unregister: ");
1499 			sysctl_print_dir(parent);
1500 			pr_cont("/%s\n", name);
1501 		}
1502 	}
1503 }
1504 
1505 static void drop_sysctl_table(struct ctl_table_header *header)
1506 {
1507 	struct ctl_dir *parent = header->parent;
1508 
1509 	if (--header->nreg)
1510 		return;
1511 
1512 	put_links(header);
1513 	start_unregistering(header);
1514 	if (!--header->count)
1515 		kfree_rcu(header, rcu);
1516 
1517 	if (parent)
1518 		drop_sysctl_table(&parent->header);
1519 }
1520 
1521 /**
1522  * unregister_sysctl_table - unregister a sysctl table hierarchy
1523  * @header: the header returned from register_sysctl_table
1524  *
1525  * Unregisters the sysctl table and all children. proc entries may not
1526  * actually be removed until they are no longer used by anyone.
1527  */
1528 void unregister_sysctl_table(struct ctl_table_header * header)
1529 {
1530 	int nr_subheaders;
1531 	might_sleep();
1532 
1533 	if (header == NULL)
1534 		return;
1535 
1536 	nr_subheaders = count_subheaders(header->ctl_table_arg);
1537 	if (unlikely(nr_subheaders > 1)) {
1538 		struct ctl_table_header **subheaders;
1539 		int i;
1540 
1541 		subheaders = (struct ctl_table_header **)(header + 1);
1542 		for (i = nr_subheaders -1; i >= 0; i--) {
1543 			struct ctl_table_header *subh = subheaders[i];
1544 			struct ctl_table *table = subh->ctl_table_arg;
1545 			unregister_sysctl_table(subh);
1546 			kfree(table);
1547 		}
1548 		kfree(header);
1549 		return;
1550 	}
1551 
1552 	spin_lock(&sysctl_lock);
1553 	drop_sysctl_table(header);
1554 	spin_unlock(&sysctl_lock);
1555 }
1556 EXPORT_SYMBOL(unregister_sysctl_table);
1557 
1558 void setup_sysctl_set(struct ctl_table_set *set,
1559 	struct ctl_table_root *root,
1560 	int (*is_seen)(struct ctl_table_set *))
1561 {
1562 	memset(set, 0, sizeof(*set));
1563 	set->is_seen = is_seen;
1564 	init_header(&set->dir.header, root, set, NULL, root_table);
1565 }
1566 
1567 void retire_sysctl_set(struct ctl_table_set *set)
1568 {
1569 	WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1570 }
1571 
1572 int __init proc_sys_init(void)
1573 {
1574 	struct proc_dir_entry *proc_sys_root;
1575 
1576 	proc_sys_root = proc_mkdir("sys", NULL);
1577 	proc_sys_root->proc_iops = &proc_sys_dir_operations;
1578 	proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1579 	proc_sys_root->nlink = 0;
1580 
1581 	return sysctl_init();
1582 }
1583