xref: /openbmc/linux/fs/proc/proc_sysctl.c (revision 57a8ec387e1441ea5e1232bc0749fb99a8cba7e7)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * /proc/sys support
4  */
5 #include <linux/init.h>
6 #include <linux/sysctl.h>
7 #include <linux/poll.h>
8 #include <linux/proc_fs.h>
9 #include <linux/printk.h>
10 #include <linux/security.h>
11 #include <linux/sched.h>
12 #include <linux/cred.h>
13 #include <linux/namei.h>
14 #include <linux/mm.h>
15 #include <linux/module.h>
16 #include <linux/bpf-cgroup.h>
17 #include "internal.h"
18 
19 static const struct dentry_operations proc_sys_dentry_operations;
20 static const struct file_operations proc_sys_file_operations;
21 static const struct inode_operations proc_sys_inode_operations;
22 static const struct file_operations proc_sys_dir_file_operations;
23 static const struct inode_operations proc_sys_dir_operations;
24 
25 /* Support for permanently empty directories */
26 
27 struct ctl_table sysctl_mount_point[] = {
28 	{ }
29 };
30 
31 static bool is_empty_dir(struct ctl_table_header *head)
32 {
33 	return head->ctl_table[0].child == sysctl_mount_point;
34 }
35 
36 static void set_empty_dir(struct ctl_dir *dir)
37 {
38 	dir->header.ctl_table[0].child = sysctl_mount_point;
39 }
40 
41 static void clear_empty_dir(struct ctl_dir *dir)
42 
43 {
44 	dir->header.ctl_table[0].child = NULL;
45 }
46 
47 void proc_sys_poll_notify(struct ctl_table_poll *poll)
48 {
49 	if (!poll)
50 		return;
51 
52 	atomic_inc(&poll->event);
53 	wake_up_interruptible(&poll->wait);
54 }
55 
56 static struct ctl_table root_table[] = {
57 	{
58 		.procname = "",
59 		.mode = S_IFDIR|S_IRUGO|S_IXUGO,
60 	},
61 	{ }
62 };
63 static struct ctl_table_root sysctl_table_root = {
64 	.default_set.dir.header = {
65 		{{.count = 1,
66 		  .nreg = 1,
67 		  .ctl_table = root_table }},
68 		.ctl_table_arg = root_table,
69 		.root = &sysctl_table_root,
70 		.set = &sysctl_table_root.default_set,
71 	},
72 };
73 
74 static DEFINE_SPINLOCK(sysctl_lock);
75 
76 static void drop_sysctl_table(struct ctl_table_header *header);
77 static int sysctl_follow_link(struct ctl_table_header **phead,
78 	struct ctl_table **pentry);
79 static int insert_links(struct ctl_table_header *head);
80 static void put_links(struct ctl_table_header *header);
81 
82 static void sysctl_print_dir(struct ctl_dir *dir)
83 {
84 	if (dir->header.parent)
85 		sysctl_print_dir(dir->header.parent);
86 	pr_cont("%s/", dir->header.ctl_table[0].procname);
87 }
88 
89 static int namecmp(const char *name1, int len1, const char *name2, int len2)
90 {
91 	int minlen;
92 	int cmp;
93 
94 	minlen = len1;
95 	if (minlen > len2)
96 		minlen = len2;
97 
98 	cmp = memcmp(name1, name2, minlen);
99 	if (cmp == 0)
100 		cmp = len1 - len2;
101 	return cmp;
102 }
103 
104 /* Called under sysctl_lock */
105 static struct ctl_table *find_entry(struct ctl_table_header **phead,
106 	struct ctl_dir *dir, const char *name, int namelen)
107 {
108 	struct ctl_table_header *head;
109 	struct ctl_table *entry;
110 	struct rb_node *node = dir->root.rb_node;
111 
112 	while (node)
113 	{
114 		struct ctl_node *ctl_node;
115 		const char *procname;
116 		int cmp;
117 
118 		ctl_node = rb_entry(node, struct ctl_node, node);
119 		head = ctl_node->header;
120 		entry = &head->ctl_table[ctl_node - head->node];
121 		procname = entry->procname;
122 
123 		cmp = namecmp(name, namelen, procname, strlen(procname));
124 		if (cmp < 0)
125 			node = node->rb_left;
126 		else if (cmp > 0)
127 			node = node->rb_right;
128 		else {
129 			*phead = head;
130 			return entry;
131 		}
132 	}
133 	return NULL;
134 }
135 
136 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
137 {
138 	struct rb_node *node = &head->node[entry - head->ctl_table].node;
139 	struct rb_node **p = &head->parent->root.rb_node;
140 	struct rb_node *parent = NULL;
141 	const char *name = entry->procname;
142 	int namelen = strlen(name);
143 
144 	while (*p) {
145 		struct ctl_table_header *parent_head;
146 		struct ctl_table *parent_entry;
147 		struct ctl_node *parent_node;
148 		const char *parent_name;
149 		int cmp;
150 
151 		parent = *p;
152 		parent_node = rb_entry(parent, struct ctl_node, node);
153 		parent_head = parent_node->header;
154 		parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
155 		parent_name = parent_entry->procname;
156 
157 		cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
158 		if (cmp < 0)
159 			p = &(*p)->rb_left;
160 		else if (cmp > 0)
161 			p = &(*p)->rb_right;
162 		else {
163 			pr_err("sysctl duplicate entry: ");
164 			sysctl_print_dir(head->parent);
165 			pr_cont("/%s\n", entry->procname);
166 			return -EEXIST;
167 		}
168 	}
169 
170 	rb_link_node(node, parent, p);
171 	rb_insert_color(node, &head->parent->root);
172 	return 0;
173 }
174 
175 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
176 {
177 	struct rb_node *node = &head->node[entry - head->ctl_table].node;
178 
179 	rb_erase(node, &head->parent->root);
180 }
181 
182 static void init_header(struct ctl_table_header *head,
183 	struct ctl_table_root *root, struct ctl_table_set *set,
184 	struct ctl_node *node, struct ctl_table *table)
185 {
186 	head->ctl_table = table;
187 	head->ctl_table_arg = table;
188 	head->used = 0;
189 	head->count = 1;
190 	head->nreg = 1;
191 	head->unregistering = NULL;
192 	head->root = root;
193 	head->set = set;
194 	head->parent = NULL;
195 	head->node = node;
196 	INIT_HLIST_HEAD(&head->inodes);
197 	if (node) {
198 		struct ctl_table *entry;
199 		for (entry = table; entry->procname; entry++, node++)
200 			node->header = head;
201 	}
202 }
203 
204 static void erase_header(struct ctl_table_header *head)
205 {
206 	struct ctl_table *entry;
207 	for (entry = head->ctl_table; entry->procname; entry++)
208 		erase_entry(head, entry);
209 }
210 
211 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
212 {
213 	struct ctl_table *entry;
214 	int err;
215 
216 	/* Is this a permanently empty directory? */
217 	if (is_empty_dir(&dir->header))
218 		return -EROFS;
219 
220 	/* Am I creating a permanently empty directory? */
221 	if (header->ctl_table == sysctl_mount_point) {
222 		if (!RB_EMPTY_ROOT(&dir->root))
223 			return -EINVAL;
224 		set_empty_dir(dir);
225 	}
226 
227 	dir->header.nreg++;
228 	header->parent = dir;
229 	err = insert_links(header);
230 	if (err)
231 		goto fail_links;
232 	for (entry = header->ctl_table; entry->procname; entry++) {
233 		err = insert_entry(header, entry);
234 		if (err)
235 			goto fail;
236 	}
237 	return 0;
238 fail:
239 	erase_header(header);
240 	put_links(header);
241 fail_links:
242 	if (header->ctl_table == sysctl_mount_point)
243 		clear_empty_dir(dir);
244 	header->parent = NULL;
245 	drop_sysctl_table(&dir->header);
246 	return err;
247 }
248 
249 /* called under sysctl_lock */
250 static int use_table(struct ctl_table_header *p)
251 {
252 	if (unlikely(p->unregistering))
253 		return 0;
254 	p->used++;
255 	return 1;
256 }
257 
258 /* called under sysctl_lock */
259 static void unuse_table(struct ctl_table_header *p)
260 {
261 	if (!--p->used)
262 		if (unlikely(p->unregistering))
263 			complete(p->unregistering);
264 }
265 
266 static void proc_sys_prune_dcache(struct ctl_table_header *head)
267 {
268 	struct inode *inode;
269 	struct proc_inode *ei;
270 	struct hlist_node *node;
271 	struct super_block *sb;
272 
273 	rcu_read_lock();
274 	for (;;) {
275 		node = hlist_first_rcu(&head->inodes);
276 		if (!node)
277 			break;
278 		ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
279 		spin_lock(&sysctl_lock);
280 		hlist_del_init_rcu(&ei->sysctl_inodes);
281 		spin_unlock(&sysctl_lock);
282 
283 		inode = &ei->vfs_inode;
284 		sb = inode->i_sb;
285 		if (!atomic_inc_not_zero(&sb->s_active))
286 			continue;
287 		inode = igrab(inode);
288 		rcu_read_unlock();
289 		if (unlikely(!inode)) {
290 			deactivate_super(sb);
291 			rcu_read_lock();
292 			continue;
293 		}
294 
295 		d_prune_aliases(inode);
296 		iput(inode);
297 		deactivate_super(sb);
298 
299 		rcu_read_lock();
300 	}
301 	rcu_read_unlock();
302 }
303 
304 /* called under sysctl_lock, will reacquire if has to wait */
305 static void start_unregistering(struct ctl_table_header *p)
306 {
307 	/*
308 	 * if p->used is 0, nobody will ever touch that entry again;
309 	 * we'll eliminate all paths to it before dropping sysctl_lock
310 	 */
311 	if (unlikely(p->used)) {
312 		struct completion wait;
313 		init_completion(&wait);
314 		p->unregistering = &wait;
315 		spin_unlock(&sysctl_lock);
316 		wait_for_completion(&wait);
317 	} else {
318 		/* anything non-NULL; we'll never dereference it */
319 		p->unregistering = ERR_PTR(-EINVAL);
320 		spin_unlock(&sysctl_lock);
321 	}
322 	/*
323 	 * Prune dentries for unregistered sysctls: namespaced sysctls
324 	 * can have duplicate names and contaminate dcache very badly.
325 	 */
326 	proc_sys_prune_dcache(p);
327 	/*
328 	 * do not remove from the list until nobody holds it; walking the
329 	 * list in do_sysctl() relies on that.
330 	 */
331 	spin_lock(&sysctl_lock);
332 	erase_header(p);
333 }
334 
335 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
336 {
337 	BUG_ON(!head);
338 	spin_lock(&sysctl_lock);
339 	if (!use_table(head))
340 		head = ERR_PTR(-ENOENT);
341 	spin_unlock(&sysctl_lock);
342 	return head;
343 }
344 
345 static void sysctl_head_finish(struct ctl_table_header *head)
346 {
347 	if (!head)
348 		return;
349 	spin_lock(&sysctl_lock);
350 	unuse_table(head);
351 	spin_unlock(&sysctl_lock);
352 }
353 
354 static struct ctl_table_set *
355 lookup_header_set(struct ctl_table_root *root)
356 {
357 	struct ctl_table_set *set = &root->default_set;
358 	if (root->lookup)
359 		set = root->lookup(root);
360 	return set;
361 }
362 
363 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
364 				      struct ctl_dir *dir,
365 				      const char *name, int namelen)
366 {
367 	struct ctl_table_header *head;
368 	struct ctl_table *entry;
369 
370 	spin_lock(&sysctl_lock);
371 	entry = find_entry(&head, dir, name, namelen);
372 	if (entry && use_table(head))
373 		*phead = head;
374 	else
375 		entry = NULL;
376 	spin_unlock(&sysctl_lock);
377 	return entry;
378 }
379 
380 static struct ctl_node *first_usable_entry(struct rb_node *node)
381 {
382 	struct ctl_node *ctl_node;
383 
384 	for (;node; node = rb_next(node)) {
385 		ctl_node = rb_entry(node, struct ctl_node, node);
386 		if (use_table(ctl_node->header))
387 			return ctl_node;
388 	}
389 	return NULL;
390 }
391 
392 static void first_entry(struct ctl_dir *dir,
393 	struct ctl_table_header **phead, struct ctl_table **pentry)
394 {
395 	struct ctl_table_header *head = NULL;
396 	struct ctl_table *entry = NULL;
397 	struct ctl_node *ctl_node;
398 
399 	spin_lock(&sysctl_lock);
400 	ctl_node = first_usable_entry(rb_first(&dir->root));
401 	spin_unlock(&sysctl_lock);
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 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
411 {
412 	struct ctl_table_header *head = *phead;
413 	struct ctl_table *entry = *pentry;
414 	struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
415 
416 	spin_lock(&sysctl_lock);
417 	unuse_table(head);
418 
419 	ctl_node = first_usable_entry(rb_next(&ctl_node->node));
420 	spin_unlock(&sysctl_lock);
421 	head = NULL;
422 	if (ctl_node) {
423 		head = ctl_node->header;
424 		entry = &head->ctl_table[ctl_node - head->node];
425 	}
426 	*phead = head;
427 	*pentry = entry;
428 }
429 
430 /*
431  * sysctl_perm does NOT grant the superuser all rights automatically, because
432  * some sysctl variables are readonly even to root.
433  */
434 
435 static int test_perm(int mode, int op)
436 {
437 	if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
438 		mode >>= 6;
439 	else if (in_egroup_p(GLOBAL_ROOT_GID))
440 		mode >>= 3;
441 	if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
442 		return 0;
443 	return -EACCES;
444 }
445 
446 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
447 {
448 	struct ctl_table_root *root = head->root;
449 	int mode;
450 
451 	if (root->permissions)
452 		mode = root->permissions(head, table);
453 	else
454 		mode = table->mode;
455 
456 	return test_perm(mode, op);
457 }
458 
459 static struct inode *proc_sys_make_inode(struct super_block *sb,
460 		struct ctl_table_header *head, struct ctl_table *table)
461 {
462 	struct ctl_table_root *root = head->root;
463 	struct inode *inode;
464 	struct proc_inode *ei;
465 
466 	inode = new_inode(sb);
467 	if (!inode)
468 		return ERR_PTR(-ENOMEM);
469 
470 	inode->i_ino = get_next_ino();
471 
472 	ei = PROC_I(inode);
473 
474 	spin_lock(&sysctl_lock);
475 	if (unlikely(head->unregistering)) {
476 		spin_unlock(&sysctl_lock);
477 		iput(inode);
478 		return ERR_PTR(-ENOENT);
479 	}
480 	ei->sysctl = head;
481 	ei->sysctl_entry = table;
482 	hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
483 	head->count++;
484 	spin_unlock(&sysctl_lock);
485 
486 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
487 	inode->i_mode = table->mode;
488 	if (!S_ISDIR(table->mode)) {
489 		inode->i_mode |= S_IFREG;
490 		inode->i_op = &proc_sys_inode_operations;
491 		inode->i_fop = &proc_sys_file_operations;
492 	} else {
493 		inode->i_mode |= S_IFDIR;
494 		inode->i_op = &proc_sys_dir_operations;
495 		inode->i_fop = &proc_sys_dir_file_operations;
496 		if (is_empty_dir(head))
497 			make_empty_dir_inode(inode);
498 	}
499 
500 	if (root->set_ownership)
501 		root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
502 	else {
503 		inode->i_uid = GLOBAL_ROOT_UID;
504 		inode->i_gid = GLOBAL_ROOT_GID;
505 	}
506 
507 	return inode;
508 }
509 
510 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
511 {
512 	spin_lock(&sysctl_lock);
513 	hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
514 	if (!--head->count)
515 		kfree_rcu(head, rcu);
516 	spin_unlock(&sysctl_lock);
517 }
518 
519 static struct ctl_table_header *grab_header(struct inode *inode)
520 {
521 	struct ctl_table_header *head = PROC_I(inode)->sysctl;
522 	if (!head)
523 		head = &sysctl_table_root.default_set.dir.header;
524 	return sysctl_head_grab(head);
525 }
526 
527 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
528 					unsigned int flags)
529 {
530 	struct ctl_table_header *head = grab_header(dir);
531 	struct ctl_table_header *h = NULL;
532 	const struct qstr *name = &dentry->d_name;
533 	struct ctl_table *p;
534 	struct inode *inode;
535 	struct dentry *err = ERR_PTR(-ENOENT);
536 	struct ctl_dir *ctl_dir;
537 	int ret;
538 
539 	if (IS_ERR(head))
540 		return ERR_CAST(head);
541 
542 	ctl_dir = container_of(head, struct ctl_dir, header);
543 
544 	p = lookup_entry(&h, ctl_dir, name->name, name->len);
545 	if (!p)
546 		goto out;
547 
548 	if (S_ISLNK(p->mode)) {
549 		ret = sysctl_follow_link(&h, &p);
550 		err = ERR_PTR(ret);
551 		if (ret)
552 			goto out;
553 	}
554 
555 	inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
556 	if (IS_ERR(inode)) {
557 		err = ERR_CAST(inode);
558 		goto out;
559 	}
560 
561 	d_set_d_op(dentry, &proc_sys_dentry_operations);
562 	err = d_splice_alias(inode, dentry);
563 
564 out:
565 	if (h)
566 		sysctl_head_finish(h);
567 	sysctl_head_finish(head);
568 	return err;
569 }
570 
571 static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
572 		size_t count, loff_t *ppos, int write)
573 {
574 	struct inode *inode = file_inode(filp);
575 	struct ctl_table_header *head = grab_header(inode);
576 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
577 	void *new_buf = NULL;
578 	ssize_t error;
579 
580 	if (IS_ERR(head))
581 		return PTR_ERR(head);
582 
583 	/*
584 	 * At this point we know that the sysctl was not unregistered
585 	 * and won't be until we finish.
586 	 */
587 	error = -EPERM;
588 	if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
589 		goto out;
590 
591 	/* if that can happen at all, it should be -EINVAL, not -EISDIR */
592 	error = -EINVAL;
593 	if (!table->proc_handler)
594 		goto out;
595 
596 	error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, &count,
597 					   ppos, &new_buf);
598 	if (error)
599 		goto out;
600 
601 	/* careful: calling conventions are nasty here */
602 	if (new_buf) {
603 		mm_segment_t old_fs;
604 
605 		old_fs = get_fs();
606 		set_fs(KERNEL_DS);
607 		error = table->proc_handler(table, write, (void __user *)new_buf,
608 					    &count, ppos);
609 		set_fs(old_fs);
610 		kfree(new_buf);
611 	} else {
612 		error = table->proc_handler(table, write, buf, &count, ppos);
613 	}
614 
615 	if (!error)
616 		error = count;
617 out:
618 	sysctl_head_finish(head);
619 
620 	return error;
621 }
622 
623 static ssize_t proc_sys_read(struct file *filp, char __user *buf,
624 				size_t count, loff_t *ppos)
625 {
626 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
627 }
628 
629 static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
630 				size_t count, loff_t *ppos)
631 {
632 	return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
633 }
634 
635 static int proc_sys_open(struct inode *inode, struct file *filp)
636 {
637 	struct ctl_table_header *head = grab_header(inode);
638 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
639 
640 	/* sysctl was unregistered */
641 	if (IS_ERR(head))
642 		return PTR_ERR(head);
643 
644 	if (table->poll)
645 		filp->private_data = proc_sys_poll_event(table->poll);
646 
647 	sysctl_head_finish(head);
648 
649 	return 0;
650 }
651 
652 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
653 {
654 	struct inode *inode = file_inode(filp);
655 	struct ctl_table_header *head = grab_header(inode);
656 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
657 	__poll_t ret = DEFAULT_POLLMASK;
658 	unsigned long event;
659 
660 	/* sysctl was unregistered */
661 	if (IS_ERR(head))
662 		return EPOLLERR | EPOLLHUP;
663 
664 	if (!table->proc_handler)
665 		goto out;
666 
667 	if (!table->poll)
668 		goto out;
669 
670 	event = (unsigned long)filp->private_data;
671 	poll_wait(filp, &table->poll->wait, wait);
672 
673 	if (event != atomic_read(&table->poll->event)) {
674 		filp->private_data = proc_sys_poll_event(table->poll);
675 		ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
676 	}
677 
678 out:
679 	sysctl_head_finish(head);
680 
681 	return ret;
682 }
683 
684 static bool proc_sys_fill_cache(struct file *file,
685 				struct dir_context *ctx,
686 				struct ctl_table_header *head,
687 				struct ctl_table *table)
688 {
689 	struct dentry *child, *dir = file->f_path.dentry;
690 	struct inode *inode;
691 	struct qstr qname;
692 	ino_t ino = 0;
693 	unsigned type = DT_UNKNOWN;
694 
695 	qname.name = table->procname;
696 	qname.len  = strlen(table->procname);
697 	qname.hash = full_name_hash(dir, qname.name, qname.len);
698 
699 	child = d_lookup(dir, &qname);
700 	if (!child) {
701 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
702 		child = d_alloc_parallel(dir, &qname, &wq);
703 		if (IS_ERR(child))
704 			return false;
705 		if (d_in_lookup(child)) {
706 			struct dentry *res;
707 			inode = proc_sys_make_inode(dir->d_sb, head, table);
708 			if (IS_ERR(inode)) {
709 				d_lookup_done(child);
710 				dput(child);
711 				return false;
712 			}
713 			d_set_d_op(child, &proc_sys_dentry_operations);
714 			res = d_splice_alias(inode, child);
715 			d_lookup_done(child);
716 			if (unlikely(res)) {
717 				if (IS_ERR(res)) {
718 					dput(child);
719 					return false;
720 				}
721 				dput(child);
722 				child = res;
723 			}
724 		}
725 	}
726 	inode = d_inode(child);
727 	ino  = inode->i_ino;
728 	type = inode->i_mode >> 12;
729 	dput(child);
730 	return dir_emit(ctx, qname.name, qname.len, ino, type);
731 }
732 
733 static bool proc_sys_link_fill_cache(struct file *file,
734 				    struct dir_context *ctx,
735 				    struct ctl_table_header *head,
736 				    struct ctl_table *table)
737 {
738 	bool ret = true;
739 
740 	head = sysctl_head_grab(head);
741 	if (IS_ERR(head))
742 		return false;
743 
744 	/* It is not an error if we can not follow the link ignore it */
745 	if (sysctl_follow_link(&head, &table))
746 		goto out;
747 
748 	ret = proc_sys_fill_cache(file, ctx, head, table);
749 out:
750 	sysctl_head_finish(head);
751 	return ret;
752 }
753 
754 static int scan(struct ctl_table_header *head, struct ctl_table *table,
755 		unsigned long *pos, struct file *file,
756 		struct dir_context *ctx)
757 {
758 	bool res;
759 
760 	if ((*pos)++ < ctx->pos)
761 		return true;
762 
763 	if (unlikely(S_ISLNK(table->mode)))
764 		res = proc_sys_link_fill_cache(file, ctx, head, table);
765 	else
766 		res = proc_sys_fill_cache(file, ctx, head, table);
767 
768 	if (res)
769 		ctx->pos = *pos;
770 
771 	return res;
772 }
773 
774 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
775 {
776 	struct ctl_table_header *head = grab_header(file_inode(file));
777 	struct ctl_table_header *h = NULL;
778 	struct ctl_table *entry;
779 	struct ctl_dir *ctl_dir;
780 	unsigned long pos;
781 
782 	if (IS_ERR(head))
783 		return PTR_ERR(head);
784 
785 	ctl_dir = container_of(head, struct ctl_dir, header);
786 
787 	if (!dir_emit_dots(file, ctx))
788 		goto out;
789 
790 	pos = 2;
791 
792 	for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
793 		if (!scan(h, entry, &pos, file, ctx)) {
794 			sysctl_head_finish(h);
795 			break;
796 		}
797 	}
798 out:
799 	sysctl_head_finish(head);
800 	return 0;
801 }
802 
803 static int proc_sys_permission(struct inode *inode, int mask)
804 {
805 	/*
806 	 * sysctl entries that are not writeable,
807 	 * are _NOT_ writeable, capabilities or not.
808 	 */
809 	struct ctl_table_header *head;
810 	struct ctl_table *table;
811 	int error;
812 
813 	/* Executable files are not allowed under /proc/sys/ */
814 	if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
815 		return -EACCES;
816 
817 	head = grab_header(inode);
818 	if (IS_ERR(head))
819 		return PTR_ERR(head);
820 
821 	table = PROC_I(inode)->sysctl_entry;
822 	if (!table) /* global root - r-xr-xr-x */
823 		error = mask & MAY_WRITE ? -EACCES : 0;
824 	else /* Use the permissions on the sysctl table entry */
825 		error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
826 
827 	sysctl_head_finish(head);
828 	return error;
829 }
830 
831 static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
832 {
833 	struct inode *inode = d_inode(dentry);
834 	int error;
835 
836 	if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
837 		return -EPERM;
838 
839 	error = setattr_prepare(dentry, attr);
840 	if (error)
841 		return error;
842 
843 	setattr_copy(inode, attr);
844 	mark_inode_dirty(inode);
845 	return 0;
846 }
847 
848 static int proc_sys_getattr(const struct path *path, struct kstat *stat,
849 			    u32 request_mask, unsigned int query_flags)
850 {
851 	struct inode *inode = d_inode(path->dentry);
852 	struct ctl_table_header *head = grab_header(inode);
853 	struct ctl_table *table = PROC_I(inode)->sysctl_entry;
854 
855 	if (IS_ERR(head))
856 		return PTR_ERR(head);
857 
858 	generic_fillattr(inode, stat);
859 	if (table)
860 		stat->mode = (stat->mode & S_IFMT) | table->mode;
861 
862 	sysctl_head_finish(head);
863 	return 0;
864 }
865 
866 static const struct file_operations proc_sys_file_operations = {
867 	.open		= proc_sys_open,
868 	.poll		= proc_sys_poll,
869 	.read		= proc_sys_read,
870 	.write		= proc_sys_write,
871 	.llseek		= default_llseek,
872 };
873 
874 static const struct file_operations proc_sys_dir_file_operations = {
875 	.read		= generic_read_dir,
876 	.iterate_shared	= proc_sys_readdir,
877 	.llseek		= generic_file_llseek,
878 };
879 
880 static const struct inode_operations proc_sys_inode_operations = {
881 	.permission	= proc_sys_permission,
882 	.setattr	= proc_sys_setattr,
883 	.getattr	= proc_sys_getattr,
884 };
885 
886 static const struct inode_operations proc_sys_dir_operations = {
887 	.lookup		= proc_sys_lookup,
888 	.permission	= proc_sys_permission,
889 	.setattr	= proc_sys_setattr,
890 	.getattr	= proc_sys_getattr,
891 };
892 
893 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
894 {
895 	if (flags & LOOKUP_RCU)
896 		return -ECHILD;
897 	return !PROC_I(d_inode(dentry))->sysctl->unregistering;
898 }
899 
900 static int proc_sys_delete(const struct dentry *dentry)
901 {
902 	return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
903 }
904 
905 static int sysctl_is_seen(struct ctl_table_header *p)
906 {
907 	struct ctl_table_set *set = p->set;
908 	int res;
909 	spin_lock(&sysctl_lock);
910 	if (p->unregistering)
911 		res = 0;
912 	else if (!set->is_seen)
913 		res = 1;
914 	else
915 		res = set->is_seen(set);
916 	spin_unlock(&sysctl_lock);
917 	return res;
918 }
919 
920 static int proc_sys_compare(const struct dentry *dentry,
921 		unsigned int len, const char *str, const struct qstr *name)
922 {
923 	struct ctl_table_header *head;
924 	struct inode *inode;
925 
926 	/* Although proc doesn't have negative dentries, rcu-walk means
927 	 * that inode here can be NULL */
928 	/* AV: can it, indeed? */
929 	inode = d_inode_rcu(dentry);
930 	if (!inode)
931 		return 1;
932 	if (name->len != len)
933 		return 1;
934 	if (memcmp(name->name, str, len))
935 		return 1;
936 	head = rcu_dereference(PROC_I(inode)->sysctl);
937 	return !head || !sysctl_is_seen(head);
938 }
939 
940 static const struct dentry_operations proc_sys_dentry_operations = {
941 	.d_revalidate	= proc_sys_revalidate,
942 	.d_delete	= proc_sys_delete,
943 	.d_compare	= proc_sys_compare,
944 };
945 
946 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
947 				   const char *name, int namelen)
948 {
949 	struct ctl_table_header *head;
950 	struct ctl_table *entry;
951 
952 	entry = find_entry(&head, dir, name, namelen);
953 	if (!entry)
954 		return ERR_PTR(-ENOENT);
955 	if (!S_ISDIR(entry->mode))
956 		return ERR_PTR(-ENOTDIR);
957 	return container_of(head, struct ctl_dir, header);
958 }
959 
960 static struct ctl_dir *new_dir(struct ctl_table_set *set,
961 			       const char *name, int namelen)
962 {
963 	struct ctl_table *table;
964 	struct ctl_dir *new;
965 	struct ctl_node *node;
966 	char *new_name;
967 
968 	new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
969 		      sizeof(struct ctl_table)*2 +  namelen + 1,
970 		      GFP_KERNEL);
971 	if (!new)
972 		return NULL;
973 
974 	node = (struct ctl_node *)(new + 1);
975 	table = (struct ctl_table *)(node + 1);
976 	new_name = (char *)(table + 2);
977 	memcpy(new_name, name, namelen);
978 	new_name[namelen] = '\0';
979 	table[0].procname = new_name;
980 	table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
981 	init_header(&new->header, set->dir.header.root, set, node, table);
982 
983 	return new;
984 }
985 
986 /**
987  * get_subdir - find or create a subdir with the specified name.
988  * @dir:  Directory to create the subdirectory in
989  * @name: The name of the subdirectory to find or create
990  * @namelen: The length of name
991  *
992  * Takes a directory with an elevated reference count so we know that
993  * if we drop the lock the directory will not go away.  Upon success
994  * the reference is moved from @dir to the returned subdirectory.
995  * Upon error an error code is returned and the reference on @dir is
996  * simply dropped.
997  */
998 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
999 				  const char *name, int namelen)
1000 {
1001 	struct ctl_table_set *set = dir->header.set;
1002 	struct ctl_dir *subdir, *new = NULL;
1003 	int err;
1004 
1005 	spin_lock(&sysctl_lock);
1006 	subdir = find_subdir(dir, name, namelen);
1007 	if (!IS_ERR(subdir))
1008 		goto found;
1009 	if (PTR_ERR(subdir) != -ENOENT)
1010 		goto failed;
1011 
1012 	spin_unlock(&sysctl_lock);
1013 	new = new_dir(set, name, namelen);
1014 	spin_lock(&sysctl_lock);
1015 	subdir = ERR_PTR(-ENOMEM);
1016 	if (!new)
1017 		goto failed;
1018 
1019 	/* Was the subdir added while we dropped the lock? */
1020 	subdir = find_subdir(dir, name, namelen);
1021 	if (!IS_ERR(subdir))
1022 		goto found;
1023 	if (PTR_ERR(subdir) != -ENOENT)
1024 		goto failed;
1025 
1026 	/* Nope.  Use the our freshly made directory entry. */
1027 	err = insert_header(dir, &new->header);
1028 	subdir = ERR_PTR(err);
1029 	if (err)
1030 		goto failed;
1031 	subdir = new;
1032 found:
1033 	subdir->header.nreg++;
1034 failed:
1035 	if (IS_ERR(subdir)) {
1036 		pr_err("sysctl could not get directory: ");
1037 		sysctl_print_dir(dir);
1038 		pr_cont("/%*.*s %ld\n",
1039 			namelen, namelen, name, PTR_ERR(subdir));
1040 	}
1041 	drop_sysctl_table(&dir->header);
1042 	if (new)
1043 		drop_sysctl_table(&new->header);
1044 	spin_unlock(&sysctl_lock);
1045 	return subdir;
1046 }
1047 
1048 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1049 {
1050 	struct ctl_dir *parent;
1051 	const char *procname;
1052 	if (!dir->header.parent)
1053 		return &set->dir;
1054 	parent = xlate_dir(set, dir->header.parent);
1055 	if (IS_ERR(parent))
1056 		return parent;
1057 	procname = dir->header.ctl_table[0].procname;
1058 	return find_subdir(parent, procname, strlen(procname));
1059 }
1060 
1061 static int sysctl_follow_link(struct ctl_table_header **phead,
1062 	struct ctl_table **pentry)
1063 {
1064 	struct ctl_table_header *head;
1065 	struct ctl_table_root *root;
1066 	struct ctl_table_set *set;
1067 	struct ctl_table *entry;
1068 	struct ctl_dir *dir;
1069 	int ret;
1070 
1071 	ret = 0;
1072 	spin_lock(&sysctl_lock);
1073 	root = (*pentry)->data;
1074 	set = lookup_header_set(root);
1075 	dir = xlate_dir(set, (*phead)->parent);
1076 	if (IS_ERR(dir))
1077 		ret = PTR_ERR(dir);
1078 	else {
1079 		const char *procname = (*pentry)->procname;
1080 		head = NULL;
1081 		entry = find_entry(&head, dir, procname, strlen(procname));
1082 		ret = -ENOENT;
1083 		if (entry && use_table(head)) {
1084 			unuse_table(*phead);
1085 			*phead = head;
1086 			*pentry = entry;
1087 			ret = 0;
1088 		}
1089 	}
1090 
1091 	spin_unlock(&sysctl_lock);
1092 	return ret;
1093 }
1094 
1095 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1096 {
1097 	struct va_format vaf;
1098 	va_list args;
1099 
1100 	va_start(args, fmt);
1101 	vaf.fmt = fmt;
1102 	vaf.va = &args;
1103 
1104 	pr_err("sysctl table check failed: %s/%s %pV\n",
1105 	       path, table->procname, &vaf);
1106 
1107 	va_end(args);
1108 	return -EINVAL;
1109 }
1110 
1111 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1112 {
1113 	int err = 0;
1114 
1115 	if ((table->proc_handler == proc_douintvec) ||
1116 	    (table->proc_handler == proc_douintvec_minmax)) {
1117 		if (table->maxlen != sizeof(unsigned int))
1118 			err |= sysctl_err(path, table, "array not allowed");
1119 	}
1120 
1121 	return err;
1122 }
1123 
1124 static int sysctl_check_table(const char *path, struct ctl_table *table)
1125 {
1126 	int err = 0;
1127 	for (; table->procname; table++) {
1128 		if (table->child)
1129 			err |= sysctl_err(path, table, "Not a file");
1130 
1131 		if ((table->proc_handler == proc_dostring) ||
1132 		    (table->proc_handler == proc_dointvec) ||
1133 		    (table->proc_handler == proc_douintvec) ||
1134 		    (table->proc_handler == proc_douintvec_minmax) ||
1135 		    (table->proc_handler == proc_dointvec_minmax) ||
1136 		    (table->proc_handler == proc_dointvec_jiffies) ||
1137 		    (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1138 		    (table->proc_handler == proc_dointvec_ms_jiffies) ||
1139 		    (table->proc_handler == proc_doulongvec_minmax) ||
1140 		    (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1141 			if (!table->data)
1142 				err |= sysctl_err(path, table, "No data");
1143 			if (!table->maxlen)
1144 				err |= sysctl_err(path, table, "No maxlen");
1145 			else
1146 				err |= sysctl_check_table_array(path, table);
1147 		}
1148 		if (!table->proc_handler)
1149 			err |= sysctl_err(path, table, "No proc_handler");
1150 
1151 		if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1152 			err |= sysctl_err(path, table, "bogus .mode 0%o",
1153 				table->mode);
1154 	}
1155 	return err;
1156 }
1157 
1158 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1159 	struct ctl_table_root *link_root)
1160 {
1161 	struct ctl_table *link_table, *entry, *link;
1162 	struct ctl_table_header *links;
1163 	struct ctl_node *node;
1164 	char *link_name;
1165 	int nr_entries, name_bytes;
1166 
1167 	name_bytes = 0;
1168 	nr_entries = 0;
1169 	for (entry = table; entry->procname; entry++) {
1170 		nr_entries++;
1171 		name_bytes += strlen(entry->procname) + 1;
1172 	}
1173 
1174 	links = kzalloc(sizeof(struct ctl_table_header) +
1175 			sizeof(struct ctl_node)*nr_entries +
1176 			sizeof(struct ctl_table)*(nr_entries + 1) +
1177 			name_bytes,
1178 			GFP_KERNEL);
1179 
1180 	if (!links)
1181 		return NULL;
1182 
1183 	node = (struct ctl_node *)(links + 1);
1184 	link_table = (struct ctl_table *)(node + nr_entries);
1185 	link_name = (char *)&link_table[nr_entries + 1];
1186 
1187 	for (link = link_table, entry = table; entry->procname; link++, entry++) {
1188 		int len = strlen(entry->procname) + 1;
1189 		memcpy(link_name, entry->procname, len);
1190 		link->procname = link_name;
1191 		link->mode = S_IFLNK|S_IRWXUGO;
1192 		link->data = link_root;
1193 		link_name += len;
1194 	}
1195 	init_header(links, dir->header.root, dir->header.set, node, link_table);
1196 	links->nreg = nr_entries;
1197 
1198 	return links;
1199 }
1200 
1201 static bool get_links(struct ctl_dir *dir,
1202 	struct ctl_table *table, struct ctl_table_root *link_root)
1203 {
1204 	struct ctl_table_header *head;
1205 	struct ctl_table *entry, *link;
1206 
1207 	/* Are there links available for every entry in table? */
1208 	for (entry = table; entry->procname; entry++) {
1209 		const char *procname = entry->procname;
1210 		link = find_entry(&head, dir, procname, strlen(procname));
1211 		if (!link)
1212 			return false;
1213 		if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1214 			continue;
1215 		if (S_ISLNK(link->mode) && (link->data == link_root))
1216 			continue;
1217 		return false;
1218 	}
1219 
1220 	/* The checks passed.  Increase the registration count on the links */
1221 	for (entry = table; entry->procname; entry++) {
1222 		const char *procname = entry->procname;
1223 		link = find_entry(&head, dir, procname, strlen(procname));
1224 		head->nreg++;
1225 	}
1226 	return true;
1227 }
1228 
1229 static int insert_links(struct ctl_table_header *head)
1230 {
1231 	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1232 	struct ctl_dir *core_parent = NULL;
1233 	struct ctl_table_header *links;
1234 	int err;
1235 
1236 	if (head->set == root_set)
1237 		return 0;
1238 
1239 	core_parent = xlate_dir(root_set, head->parent);
1240 	if (IS_ERR(core_parent))
1241 		return 0;
1242 
1243 	if (get_links(core_parent, head->ctl_table, head->root))
1244 		return 0;
1245 
1246 	core_parent->header.nreg++;
1247 	spin_unlock(&sysctl_lock);
1248 
1249 	links = new_links(core_parent, head->ctl_table, head->root);
1250 
1251 	spin_lock(&sysctl_lock);
1252 	err = -ENOMEM;
1253 	if (!links)
1254 		goto out;
1255 
1256 	err = 0;
1257 	if (get_links(core_parent, head->ctl_table, head->root)) {
1258 		kfree(links);
1259 		goto out;
1260 	}
1261 
1262 	err = insert_header(core_parent, links);
1263 	if (err)
1264 		kfree(links);
1265 out:
1266 	drop_sysctl_table(&core_parent->header);
1267 	return err;
1268 }
1269 
1270 /**
1271  * __register_sysctl_table - register a leaf sysctl table
1272  * @set: Sysctl tree to register on
1273  * @path: The path to the directory the sysctl table is in.
1274  * @table: the top-level table structure
1275  *
1276  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1277  * array. A completely 0 filled entry terminates the table.
1278  *
1279  * The members of the &struct ctl_table structure are used as follows:
1280  *
1281  * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1282  *            enter a sysctl file
1283  *
1284  * data - a pointer to data for use by proc_handler
1285  *
1286  * maxlen - the maximum size in bytes of the data
1287  *
1288  * mode - the file permissions for the /proc/sys file
1289  *
1290  * child - must be %NULL.
1291  *
1292  * proc_handler - the text handler routine (described below)
1293  *
1294  * extra1, extra2 - extra pointers usable by the proc handler routines
1295  *
1296  * Leaf nodes in the sysctl tree will be represented by a single file
1297  * under /proc; non-leaf nodes will be represented by directories.
1298  *
1299  * There must be a proc_handler routine for any terminal nodes.
1300  * Several default handlers are available to cover common cases -
1301  *
1302  * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1303  * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1304  * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1305  *
1306  * It is the handler's job to read the input buffer from user memory
1307  * and process it. The handler should return 0 on success.
1308  *
1309  * This routine returns %NULL on a failure to register, and a pointer
1310  * to the table header on success.
1311  */
1312 struct ctl_table_header *__register_sysctl_table(
1313 	struct ctl_table_set *set,
1314 	const char *path, struct ctl_table *table)
1315 {
1316 	struct ctl_table_root *root = set->dir.header.root;
1317 	struct ctl_table_header *header;
1318 	const char *name, *nextname;
1319 	struct ctl_dir *dir;
1320 	struct ctl_table *entry;
1321 	struct ctl_node *node;
1322 	int nr_entries = 0;
1323 
1324 	for (entry = table; entry->procname; entry++)
1325 		nr_entries++;
1326 
1327 	header = kzalloc(sizeof(struct ctl_table_header) +
1328 			 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1329 	if (!header)
1330 		return NULL;
1331 
1332 	node = (struct ctl_node *)(header + 1);
1333 	init_header(header, root, set, node, table);
1334 	if (sysctl_check_table(path, table))
1335 		goto fail;
1336 
1337 	spin_lock(&sysctl_lock);
1338 	dir = &set->dir;
1339 	/* Reference moved down the diretory tree get_subdir */
1340 	dir->header.nreg++;
1341 	spin_unlock(&sysctl_lock);
1342 
1343 	/* Find the directory for the ctl_table */
1344 	for (name = path; name; name = nextname) {
1345 		int namelen;
1346 		nextname = strchr(name, '/');
1347 		if (nextname) {
1348 			namelen = nextname - name;
1349 			nextname++;
1350 		} else {
1351 			namelen = strlen(name);
1352 		}
1353 		if (namelen == 0)
1354 			continue;
1355 
1356 		dir = get_subdir(dir, name, namelen);
1357 		if (IS_ERR(dir))
1358 			goto fail;
1359 	}
1360 
1361 	spin_lock(&sysctl_lock);
1362 	if (insert_header(dir, header))
1363 		goto fail_put_dir_locked;
1364 
1365 	drop_sysctl_table(&dir->header);
1366 	spin_unlock(&sysctl_lock);
1367 
1368 	return header;
1369 
1370 fail_put_dir_locked:
1371 	drop_sysctl_table(&dir->header);
1372 	spin_unlock(&sysctl_lock);
1373 fail:
1374 	kfree(header);
1375 	dump_stack();
1376 	return NULL;
1377 }
1378 
1379 /**
1380  * register_sysctl - register a sysctl table
1381  * @path: The path to the directory the sysctl table is in.
1382  * @table: the table structure
1383  *
1384  * Register a sysctl table. @table should be a filled in ctl_table
1385  * array. A completely 0 filled entry terminates the table.
1386  *
1387  * See __register_sysctl_table for more details.
1388  */
1389 struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1390 {
1391 	return __register_sysctl_table(&sysctl_table_root.default_set,
1392 					path, table);
1393 }
1394 EXPORT_SYMBOL(register_sysctl);
1395 
1396 static char *append_path(const char *path, char *pos, const char *name)
1397 {
1398 	int namelen;
1399 	namelen = strlen(name);
1400 	if (((pos - path) + namelen + 2) >= PATH_MAX)
1401 		return NULL;
1402 	memcpy(pos, name, namelen);
1403 	pos[namelen] = '/';
1404 	pos[namelen + 1] = '\0';
1405 	pos += namelen + 1;
1406 	return pos;
1407 }
1408 
1409 static int count_subheaders(struct ctl_table *table)
1410 {
1411 	int has_files = 0;
1412 	int nr_subheaders = 0;
1413 	struct ctl_table *entry;
1414 
1415 	/* special case: no directory and empty directory */
1416 	if (!table || !table->procname)
1417 		return 1;
1418 
1419 	for (entry = table; entry->procname; entry++) {
1420 		if (entry->child)
1421 			nr_subheaders += count_subheaders(entry->child);
1422 		else
1423 			has_files = 1;
1424 	}
1425 	return nr_subheaders + has_files;
1426 }
1427 
1428 static int register_leaf_sysctl_tables(const char *path, char *pos,
1429 	struct ctl_table_header ***subheader, struct ctl_table_set *set,
1430 	struct ctl_table *table)
1431 {
1432 	struct ctl_table *ctl_table_arg = NULL;
1433 	struct ctl_table *entry, *files;
1434 	int nr_files = 0;
1435 	int nr_dirs = 0;
1436 	int err = -ENOMEM;
1437 
1438 	for (entry = table; entry->procname; entry++) {
1439 		if (entry->child)
1440 			nr_dirs++;
1441 		else
1442 			nr_files++;
1443 	}
1444 
1445 	files = table;
1446 	/* If there are mixed files and directories we need a new table */
1447 	if (nr_dirs && nr_files) {
1448 		struct ctl_table *new;
1449 		files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1450 				GFP_KERNEL);
1451 		if (!files)
1452 			goto out;
1453 
1454 		ctl_table_arg = files;
1455 		for (new = files, entry = table; entry->procname; entry++) {
1456 			if (entry->child)
1457 				continue;
1458 			*new = *entry;
1459 			new++;
1460 		}
1461 	}
1462 
1463 	/* Register everything except a directory full of subdirectories */
1464 	if (nr_files || !nr_dirs) {
1465 		struct ctl_table_header *header;
1466 		header = __register_sysctl_table(set, path, files);
1467 		if (!header) {
1468 			kfree(ctl_table_arg);
1469 			goto out;
1470 		}
1471 
1472 		/* Remember if we need to free the file table */
1473 		header->ctl_table_arg = ctl_table_arg;
1474 		**subheader = header;
1475 		(*subheader)++;
1476 	}
1477 
1478 	/* Recurse into the subdirectories. */
1479 	for (entry = table; entry->procname; entry++) {
1480 		char *child_pos;
1481 
1482 		if (!entry->child)
1483 			continue;
1484 
1485 		err = -ENAMETOOLONG;
1486 		child_pos = append_path(path, pos, entry->procname);
1487 		if (!child_pos)
1488 			goto out;
1489 
1490 		err = register_leaf_sysctl_tables(path, child_pos, subheader,
1491 						  set, entry->child);
1492 		pos[0] = '\0';
1493 		if (err)
1494 			goto out;
1495 	}
1496 	err = 0;
1497 out:
1498 	/* On failure our caller will unregister all registered subheaders */
1499 	return err;
1500 }
1501 
1502 /**
1503  * __register_sysctl_paths - register a sysctl table hierarchy
1504  * @set: Sysctl tree to register on
1505  * @path: The path to the directory the sysctl table is in.
1506  * @table: the top-level table structure
1507  *
1508  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1509  * array. A completely 0 filled entry terminates the table.
1510  *
1511  * See __register_sysctl_table for more details.
1512  */
1513 struct ctl_table_header *__register_sysctl_paths(
1514 	struct ctl_table_set *set,
1515 	const struct ctl_path *path, struct ctl_table *table)
1516 {
1517 	struct ctl_table *ctl_table_arg = table;
1518 	int nr_subheaders = count_subheaders(table);
1519 	struct ctl_table_header *header = NULL, **subheaders, **subheader;
1520 	const struct ctl_path *component;
1521 	char *new_path, *pos;
1522 
1523 	pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1524 	if (!new_path)
1525 		return NULL;
1526 
1527 	pos[0] = '\0';
1528 	for (component = path; component->procname; component++) {
1529 		pos = append_path(new_path, pos, component->procname);
1530 		if (!pos)
1531 			goto out;
1532 	}
1533 	while (table->procname && table->child && !table[1].procname) {
1534 		pos = append_path(new_path, pos, table->procname);
1535 		if (!pos)
1536 			goto out;
1537 		table = table->child;
1538 	}
1539 	if (nr_subheaders == 1) {
1540 		header = __register_sysctl_table(set, new_path, table);
1541 		if (header)
1542 			header->ctl_table_arg = ctl_table_arg;
1543 	} else {
1544 		header = kzalloc(sizeof(*header) +
1545 				 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1546 		if (!header)
1547 			goto out;
1548 
1549 		subheaders = (struct ctl_table_header **) (header + 1);
1550 		subheader = subheaders;
1551 		header->ctl_table_arg = ctl_table_arg;
1552 
1553 		if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1554 						set, table))
1555 			goto err_register_leaves;
1556 	}
1557 
1558 out:
1559 	kfree(new_path);
1560 	return header;
1561 
1562 err_register_leaves:
1563 	while (subheader > subheaders) {
1564 		struct ctl_table_header *subh = *(--subheader);
1565 		struct ctl_table *table = subh->ctl_table_arg;
1566 		unregister_sysctl_table(subh);
1567 		kfree(table);
1568 	}
1569 	kfree(header);
1570 	header = NULL;
1571 	goto out;
1572 }
1573 
1574 /**
1575  * register_sysctl_table_path - register a sysctl table hierarchy
1576  * @path: The path to the directory the sysctl table is in.
1577  * @table: the top-level table structure
1578  *
1579  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1580  * array. A completely 0 filled entry terminates the table.
1581  *
1582  * See __register_sysctl_paths for more details.
1583  */
1584 struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1585 						struct ctl_table *table)
1586 {
1587 	return __register_sysctl_paths(&sysctl_table_root.default_set,
1588 					path, table);
1589 }
1590 EXPORT_SYMBOL(register_sysctl_paths);
1591 
1592 /**
1593  * register_sysctl_table - register a sysctl table hierarchy
1594  * @table: the top-level table structure
1595  *
1596  * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1597  * array. A completely 0 filled entry terminates the table.
1598  *
1599  * See register_sysctl_paths for more details.
1600  */
1601 struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1602 {
1603 	static const struct ctl_path null_path[] = { {} };
1604 
1605 	return register_sysctl_paths(null_path, table);
1606 }
1607 EXPORT_SYMBOL(register_sysctl_table);
1608 
1609 static void put_links(struct ctl_table_header *header)
1610 {
1611 	struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1612 	struct ctl_table_root *root = header->root;
1613 	struct ctl_dir *parent = header->parent;
1614 	struct ctl_dir *core_parent;
1615 	struct ctl_table *entry;
1616 
1617 	if (header->set == root_set)
1618 		return;
1619 
1620 	core_parent = xlate_dir(root_set, parent);
1621 	if (IS_ERR(core_parent))
1622 		return;
1623 
1624 	for (entry = header->ctl_table; entry->procname; entry++) {
1625 		struct ctl_table_header *link_head;
1626 		struct ctl_table *link;
1627 		const char *name = entry->procname;
1628 
1629 		link = find_entry(&link_head, core_parent, name, strlen(name));
1630 		if (link &&
1631 		    ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1632 		     (S_ISLNK(link->mode) && (link->data == root)))) {
1633 			drop_sysctl_table(link_head);
1634 		}
1635 		else {
1636 			pr_err("sysctl link missing during unregister: ");
1637 			sysctl_print_dir(parent);
1638 			pr_cont("/%s\n", name);
1639 		}
1640 	}
1641 }
1642 
1643 static void drop_sysctl_table(struct ctl_table_header *header)
1644 {
1645 	struct ctl_dir *parent = header->parent;
1646 
1647 	if (--header->nreg)
1648 		return;
1649 
1650 	if (parent) {
1651 		put_links(header);
1652 		start_unregistering(header);
1653 	}
1654 
1655 	if (!--header->count)
1656 		kfree_rcu(header, rcu);
1657 
1658 	if (parent)
1659 		drop_sysctl_table(&parent->header);
1660 }
1661 
1662 /**
1663  * unregister_sysctl_table - unregister a sysctl table hierarchy
1664  * @header: the header returned from register_sysctl_table
1665  *
1666  * Unregisters the sysctl table and all children. proc entries may not
1667  * actually be removed until they are no longer used by anyone.
1668  */
1669 void unregister_sysctl_table(struct ctl_table_header * header)
1670 {
1671 	int nr_subheaders;
1672 	might_sleep();
1673 
1674 	if (header == NULL)
1675 		return;
1676 
1677 	nr_subheaders = count_subheaders(header->ctl_table_arg);
1678 	if (unlikely(nr_subheaders > 1)) {
1679 		struct ctl_table_header **subheaders;
1680 		int i;
1681 
1682 		subheaders = (struct ctl_table_header **)(header + 1);
1683 		for (i = nr_subheaders -1; i >= 0; i--) {
1684 			struct ctl_table_header *subh = subheaders[i];
1685 			struct ctl_table *table = subh->ctl_table_arg;
1686 			unregister_sysctl_table(subh);
1687 			kfree(table);
1688 		}
1689 		kfree(header);
1690 		return;
1691 	}
1692 
1693 	spin_lock(&sysctl_lock);
1694 	drop_sysctl_table(header);
1695 	spin_unlock(&sysctl_lock);
1696 }
1697 EXPORT_SYMBOL(unregister_sysctl_table);
1698 
1699 void setup_sysctl_set(struct ctl_table_set *set,
1700 	struct ctl_table_root *root,
1701 	int (*is_seen)(struct ctl_table_set *))
1702 {
1703 	memset(set, 0, sizeof(*set));
1704 	set->is_seen = is_seen;
1705 	init_header(&set->dir.header, root, set, NULL, root_table);
1706 }
1707 
1708 void retire_sysctl_set(struct ctl_table_set *set)
1709 {
1710 	WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1711 }
1712 
1713 int __init proc_sys_init(void)
1714 {
1715 	struct proc_dir_entry *proc_sys_root;
1716 
1717 	proc_sys_root = proc_mkdir("sys", NULL);
1718 	proc_sys_root->proc_iops = &proc_sys_dir_operations;
1719 	proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1720 	proc_sys_root->nlink = 0;
1721 
1722 	return sysctl_init();
1723 }
1724