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