xref: /openbmc/linux/net/netlink/genetlink.c (revision 232b0b08)
1 /*
2  * NETLINK      Generic Netlink Family
3  *
4  * 		Authors:	Jamal Hadi Salim
5  * 				Thomas Graf <tgraf@suug.ch>
6  *				Johannes Berg <johannes@sipsolutions.net>
7  */
8 
9 #include <linux/module.h>
10 #include <linux/kernel.h>
11 #include <linux/slab.h>
12 #include <linux/errno.h>
13 #include <linux/types.h>
14 #include <linux/socket.h>
15 #include <linux/string.h>
16 #include <linux/skbuff.h>
17 #include <linux/mutex.h>
18 #include <linux/bitmap.h>
19 #include <linux/rwsem.h>
20 #include <linux/idr.h>
21 #include <net/sock.h>
22 #include <net/genetlink.h>
23 
24 static DEFINE_MUTEX(genl_mutex); /* serialization of message processing */
25 static DECLARE_RWSEM(cb_lock);
26 
27 atomic_t genl_sk_destructing_cnt = ATOMIC_INIT(0);
28 DECLARE_WAIT_QUEUE_HEAD(genl_sk_destructing_waitq);
29 
30 void genl_lock(void)
31 {
32 	mutex_lock(&genl_mutex);
33 }
34 EXPORT_SYMBOL(genl_lock);
35 
36 void genl_unlock(void)
37 {
38 	mutex_unlock(&genl_mutex);
39 }
40 EXPORT_SYMBOL(genl_unlock);
41 
42 #ifdef CONFIG_LOCKDEP
43 bool lockdep_genl_is_held(void)
44 {
45 	return lockdep_is_held(&genl_mutex);
46 }
47 EXPORT_SYMBOL(lockdep_genl_is_held);
48 #endif
49 
50 static void genl_lock_all(void)
51 {
52 	down_write(&cb_lock);
53 	genl_lock();
54 }
55 
56 static void genl_unlock_all(void)
57 {
58 	genl_unlock();
59 	up_write(&cb_lock);
60 }
61 
62 static DEFINE_IDR(genl_fam_idr);
63 
64 /*
65  * Bitmap of multicast groups that are currently in use.
66  *
67  * To avoid an allocation at boot of just one unsigned long,
68  * declare it global instead.
69  * Bit 0 is marked as already used since group 0 is invalid.
70  * Bit 1 is marked as already used since the drop-monitor code
71  * abuses the API and thinks it can statically use group 1.
72  * That group will typically conflict with other groups that
73  * any proper users use.
74  * Bit 16 is marked as used since it's used for generic netlink
75  * and the code no longer marks pre-reserved IDs as used.
76  * Bit 17 is marked as already used since the VFS quota code
77  * also abused this API and relied on family == group ID, we
78  * cater to that by giving it a static family and group ID.
79  * Bit 18 is marked as already used since the PMCRAID driver
80  * did the same thing as the VFS quota code (maybe copied?)
81  */
82 static unsigned long mc_group_start = 0x3 | BIT(GENL_ID_CTRL) |
83 				      BIT(GENL_ID_VFS_DQUOT) |
84 				      BIT(GENL_ID_PMCRAID);
85 static unsigned long *mc_groups = &mc_group_start;
86 static unsigned long mc_groups_longs = 1;
87 
88 static int genl_ctrl_event(int event, const struct genl_family *family,
89 			   const struct genl_multicast_group *grp,
90 			   int grp_id);
91 
92 static const struct genl_family *genl_family_find_byid(unsigned int id)
93 {
94 	return idr_find(&genl_fam_idr, id);
95 }
96 
97 static const struct genl_family *genl_family_find_byname(char *name)
98 {
99 	const struct genl_family *family;
100 	unsigned int id;
101 
102 	idr_for_each_entry(&genl_fam_idr, family, id)
103 		if (strcmp(family->name, name) == 0)
104 			return family;
105 
106 	return NULL;
107 }
108 
109 static const struct genl_ops *genl_get_cmd(u8 cmd,
110 					   const struct genl_family *family)
111 {
112 	int i;
113 
114 	for (i = 0; i < family->n_ops; i++)
115 		if (family->ops[i].cmd == cmd)
116 			return &family->ops[i];
117 
118 	return NULL;
119 }
120 
121 static int genl_allocate_reserve_groups(int n_groups, int *first_id)
122 {
123 	unsigned long *new_groups;
124 	int start = 0;
125 	int i;
126 	int id;
127 	bool fits;
128 
129 	do {
130 		if (start == 0)
131 			id = find_first_zero_bit(mc_groups,
132 						 mc_groups_longs *
133 						 BITS_PER_LONG);
134 		else
135 			id = find_next_zero_bit(mc_groups,
136 						mc_groups_longs * BITS_PER_LONG,
137 						start);
138 
139 		fits = true;
140 		for (i = id;
141 		     i < min_t(int, id + n_groups,
142 			       mc_groups_longs * BITS_PER_LONG);
143 		     i++) {
144 			if (test_bit(i, mc_groups)) {
145 				start = i;
146 				fits = false;
147 				break;
148 			}
149 		}
150 
151 		if (id + n_groups > mc_groups_longs * BITS_PER_LONG) {
152 			unsigned long new_longs = mc_groups_longs +
153 						  BITS_TO_LONGS(n_groups);
154 			size_t nlen = new_longs * sizeof(unsigned long);
155 
156 			if (mc_groups == &mc_group_start) {
157 				new_groups = kzalloc(nlen, GFP_KERNEL);
158 				if (!new_groups)
159 					return -ENOMEM;
160 				mc_groups = new_groups;
161 				*mc_groups = mc_group_start;
162 			} else {
163 				new_groups = krealloc(mc_groups, nlen,
164 						      GFP_KERNEL);
165 				if (!new_groups)
166 					return -ENOMEM;
167 				mc_groups = new_groups;
168 				for (i = 0; i < BITS_TO_LONGS(n_groups); i++)
169 					mc_groups[mc_groups_longs + i] = 0;
170 			}
171 			mc_groups_longs = new_longs;
172 		}
173 	} while (!fits);
174 
175 	for (i = id; i < id + n_groups; i++)
176 		set_bit(i, mc_groups);
177 	*first_id = id;
178 	return 0;
179 }
180 
181 static struct genl_family genl_ctrl;
182 
183 static int genl_validate_assign_mc_groups(struct genl_family *family)
184 {
185 	int first_id;
186 	int n_groups = family->n_mcgrps;
187 	int err = 0, i;
188 	bool groups_allocated = false;
189 
190 	if (!n_groups)
191 		return 0;
192 
193 	for (i = 0; i < n_groups; i++) {
194 		const struct genl_multicast_group *grp = &family->mcgrps[i];
195 
196 		if (WARN_ON(grp->name[0] == '\0'))
197 			return -EINVAL;
198 		if (WARN_ON(memchr(grp->name, '\0', GENL_NAMSIZ) == NULL))
199 			return -EINVAL;
200 	}
201 
202 	/* special-case our own group and hacks */
203 	if (family == &genl_ctrl) {
204 		first_id = GENL_ID_CTRL;
205 		BUG_ON(n_groups != 1);
206 	} else if (strcmp(family->name, "NET_DM") == 0) {
207 		first_id = 1;
208 		BUG_ON(n_groups != 1);
209 	} else if (family->id == GENL_ID_VFS_DQUOT) {
210 		first_id = GENL_ID_VFS_DQUOT;
211 		BUG_ON(n_groups != 1);
212 	} else if (family->id == GENL_ID_PMCRAID) {
213 		first_id = GENL_ID_PMCRAID;
214 		BUG_ON(n_groups != 1);
215 	} else {
216 		groups_allocated = true;
217 		err = genl_allocate_reserve_groups(n_groups, &first_id);
218 		if (err)
219 			return err;
220 	}
221 
222 	family->mcgrp_offset = first_id;
223 
224 	/* if still initializing, can't and don't need to to realloc bitmaps */
225 	if (!init_net.genl_sock)
226 		return 0;
227 
228 	if (family->netnsok) {
229 		struct net *net;
230 
231 		netlink_table_grab();
232 		rcu_read_lock();
233 		for_each_net_rcu(net) {
234 			err = __netlink_change_ngroups(net->genl_sock,
235 					mc_groups_longs * BITS_PER_LONG);
236 			if (err) {
237 				/*
238 				 * No need to roll back, can only fail if
239 				 * memory allocation fails and then the
240 				 * number of _possible_ groups has been
241 				 * increased on some sockets which is ok.
242 				 */
243 				break;
244 			}
245 		}
246 		rcu_read_unlock();
247 		netlink_table_ungrab();
248 	} else {
249 		err = netlink_change_ngroups(init_net.genl_sock,
250 					     mc_groups_longs * BITS_PER_LONG);
251 	}
252 
253 	if (groups_allocated && err) {
254 		for (i = 0; i < family->n_mcgrps; i++)
255 			clear_bit(family->mcgrp_offset + i, mc_groups);
256 	}
257 
258 	return err;
259 }
260 
261 static void genl_unregister_mc_groups(const struct genl_family *family)
262 {
263 	struct net *net;
264 	int i;
265 
266 	netlink_table_grab();
267 	rcu_read_lock();
268 	for_each_net_rcu(net) {
269 		for (i = 0; i < family->n_mcgrps; i++)
270 			__netlink_clear_multicast_users(
271 				net->genl_sock, family->mcgrp_offset + i);
272 	}
273 	rcu_read_unlock();
274 	netlink_table_ungrab();
275 
276 	for (i = 0; i < family->n_mcgrps; i++) {
277 		int grp_id = family->mcgrp_offset + i;
278 
279 		if (grp_id != 1)
280 			clear_bit(grp_id, mc_groups);
281 		genl_ctrl_event(CTRL_CMD_DELMCAST_GRP, family,
282 				&family->mcgrps[i], grp_id);
283 	}
284 }
285 
286 static int genl_validate_ops(const struct genl_family *family)
287 {
288 	const struct genl_ops *ops = family->ops;
289 	unsigned int n_ops = family->n_ops;
290 	int i, j;
291 
292 	if (WARN_ON(n_ops && !ops))
293 		return -EINVAL;
294 
295 	if (!n_ops)
296 		return 0;
297 
298 	for (i = 0; i < n_ops; i++) {
299 		if (ops[i].dumpit == NULL && ops[i].doit == NULL)
300 			return -EINVAL;
301 		for (j = i + 1; j < n_ops; j++)
302 			if (ops[i].cmd == ops[j].cmd)
303 				return -EINVAL;
304 	}
305 
306 	return 0;
307 }
308 
309 /**
310  * genl_register_family - register a generic netlink family
311  * @family: generic netlink family
312  *
313  * Registers the specified family after validating it first. Only one
314  * family may be registered with the same family name or identifier.
315  *
316  * The family's ops, multicast groups and module pointer must already
317  * be assigned.
318  *
319  * Return 0 on success or a negative error code.
320  */
321 int genl_register_family(struct genl_family *family)
322 {
323 	int err, i;
324 	int start = GENL_START_ALLOC, end = GENL_MAX_ID;
325 
326 	err = genl_validate_ops(family);
327 	if (err)
328 		return err;
329 
330 	genl_lock_all();
331 
332 	if (genl_family_find_byname(family->name)) {
333 		err = -EEXIST;
334 		goto errout_locked;
335 	}
336 
337 	/*
338 	 * Sadly, a few cases need to be special-cased
339 	 * due to them having previously abused the API
340 	 * and having used their family ID also as their
341 	 * multicast group ID, so we use reserved IDs
342 	 * for both to be sure we can do that mapping.
343 	 */
344 	if (family == &genl_ctrl) {
345 		/* and this needs to be special for initial family lookups */
346 		start = end = GENL_ID_CTRL;
347 	} else if (strcmp(family->name, "pmcraid") == 0) {
348 		start = end = GENL_ID_PMCRAID;
349 	} else if (strcmp(family->name, "VFS_DQUOT") == 0) {
350 		start = end = GENL_ID_VFS_DQUOT;
351 	}
352 
353 	if (family->maxattr && !family->parallel_ops) {
354 		family->attrbuf = kmalloc((family->maxattr+1) *
355 					sizeof(struct nlattr *), GFP_KERNEL);
356 		if (family->attrbuf == NULL) {
357 			err = -ENOMEM;
358 			goto errout_locked;
359 		}
360 	} else
361 		family->attrbuf = NULL;
362 
363 	family->id = idr_alloc(&genl_fam_idr, family,
364 			       start, end + 1, GFP_KERNEL);
365 	if (family->id < 0) {
366 		err = family->id;
367 		goto errout_locked;
368 	}
369 
370 	err = genl_validate_assign_mc_groups(family);
371 	if (err)
372 		goto errout_remove;
373 
374 	genl_unlock_all();
375 
376 	/* send all events */
377 	genl_ctrl_event(CTRL_CMD_NEWFAMILY, family, NULL, 0);
378 	for (i = 0; i < family->n_mcgrps; i++)
379 		genl_ctrl_event(CTRL_CMD_NEWMCAST_GRP, family,
380 				&family->mcgrps[i], family->mcgrp_offset + i);
381 
382 	return 0;
383 
384 errout_remove:
385 	idr_remove(&genl_fam_idr, family->id);
386 	kfree(family->attrbuf);
387 errout_locked:
388 	genl_unlock_all();
389 	return err;
390 }
391 EXPORT_SYMBOL(genl_register_family);
392 
393 /**
394  * genl_unregister_family - unregister generic netlink family
395  * @family: generic netlink family
396  *
397  * Unregisters the specified family.
398  *
399  * Returns 0 on success or a negative error code.
400  */
401 int genl_unregister_family(const struct genl_family *family)
402 {
403 	genl_lock_all();
404 
405 	if (!genl_family_find_byid(family->id)) {
406 		genl_unlock_all();
407 		return -ENOENT;
408 	}
409 
410 	genl_unregister_mc_groups(family);
411 
412 	idr_remove(&genl_fam_idr, family->id);
413 
414 	up_write(&cb_lock);
415 	wait_event(genl_sk_destructing_waitq,
416 		   atomic_read(&genl_sk_destructing_cnt) == 0);
417 	genl_unlock();
418 
419 	kfree(family->attrbuf);
420 
421 	genl_ctrl_event(CTRL_CMD_DELFAMILY, family, NULL, 0);
422 
423 	return 0;
424 }
425 EXPORT_SYMBOL(genl_unregister_family);
426 
427 /**
428  * genlmsg_put - Add generic netlink header to netlink message
429  * @skb: socket buffer holding the message
430  * @portid: netlink portid the message is addressed to
431  * @seq: sequence number (usually the one of the sender)
432  * @family: generic netlink family
433  * @flags: netlink message flags
434  * @cmd: generic netlink command
435  *
436  * Returns pointer to user specific header
437  */
438 void *genlmsg_put(struct sk_buff *skb, u32 portid, u32 seq,
439 		  const struct genl_family *family, int flags, u8 cmd)
440 {
441 	struct nlmsghdr *nlh;
442 	struct genlmsghdr *hdr;
443 
444 	nlh = nlmsg_put(skb, portid, seq, family->id, GENL_HDRLEN +
445 			family->hdrsize, flags);
446 	if (nlh == NULL)
447 		return NULL;
448 
449 	hdr = nlmsg_data(nlh);
450 	hdr->cmd = cmd;
451 	hdr->version = family->version;
452 	hdr->reserved = 0;
453 
454 	return (char *) hdr + GENL_HDRLEN;
455 }
456 EXPORT_SYMBOL(genlmsg_put);
457 
458 static int genl_lock_start(struct netlink_callback *cb)
459 {
460 	/* our ops are always const - netlink API doesn't propagate that */
461 	const struct genl_ops *ops = cb->data;
462 	int rc = 0;
463 
464 	if (ops->start) {
465 		genl_lock();
466 		rc = ops->start(cb);
467 		genl_unlock();
468 	}
469 	return rc;
470 }
471 
472 static int genl_lock_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
473 {
474 	/* our ops are always const - netlink API doesn't propagate that */
475 	const struct genl_ops *ops = cb->data;
476 	int rc;
477 
478 	genl_lock();
479 	rc = ops->dumpit(skb, cb);
480 	genl_unlock();
481 	return rc;
482 }
483 
484 static int genl_lock_done(struct netlink_callback *cb)
485 {
486 	/* our ops are always const - netlink API doesn't propagate that */
487 	const struct genl_ops *ops = cb->data;
488 	int rc = 0;
489 
490 	if (ops->done) {
491 		genl_lock();
492 		rc = ops->done(cb);
493 		genl_unlock();
494 	}
495 	return rc;
496 }
497 
498 static int genl_family_rcv_msg(const struct genl_family *family,
499 			       struct sk_buff *skb,
500 			       struct nlmsghdr *nlh)
501 {
502 	const struct genl_ops *ops;
503 	struct net *net = sock_net(skb->sk);
504 	struct genl_info info;
505 	struct genlmsghdr *hdr = nlmsg_data(nlh);
506 	struct nlattr **attrbuf;
507 	int hdrlen, err;
508 
509 	/* this family doesn't exist in this netns */
510 	if (!family->netnsok && !net_eq(net, &init_net))
511 		return -ENOENT;
512 
513 	hdrlen = GENL_HDRLEN + family->hdrsize;
514 	if (nlh->nlmsg_len < nlmsg_msg_size(hdrlen))
515 		return -EINVAL;
516 
517 	ops = genl_get_cmd(hdr->cmd, family);
518 	if (ops == NULL)
519 		return -EOPNOTSUPP;
520 
521 	if ((ops->flags & GENL_ADMIN_PERM) &&
522 	    !netlink_capable(skb, CAP_NET_ADMIN))
523 		return -EPERM;
524 
525 	if ((ops->flags & GENL_UNS_ADMIN_PERM) &&
526 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
527 		return -EPERM;
528 
529 	if ((nlh->nlmsg_flags & NLM_F_DUMP) == NLM_F_DUMP) {
530 		int rc;
531 
532 		if (ops->dumpit == NULL)
533 			return -EOPNOTSUPP;
534 
535 		if (!family->parallel_ops) {
536 			struct netlink_dump_control c = {
537 				.module = family->module,
538 				/* we have const, but the netlink API doesn't */
539 				.data = (void *)ops,
540 				.start = genl_lock_start,
541 				.dump = genl_lock_dumpit,
542 				.done = genl_lock_done,
543 			};
544 
545 			genl_unlock();
546 			rc = __netlink_dump_start(net->genl_sock, skb, nlh, &c);
547 			genl_lock();
548 
549 		} else {
550 			struct netlink_dump_control c = {
551 				.module = family->module,
552 				.start = ops->start,
553 				.dump = ops->dumpit,
554 				.done = ops->done,
555 			};
556 
557 			rc = __netlink_dump_start(net->genl_sock, skb, nlh, &c);
558 		}
559 
560 		return rc;
561 	}
562 
563 	if (ops->doit == NULL)
564 		return -EOPNOTSUPP;
565 
566 	if (family->maxattr && family->parallel_ops) {
567 		attrbuf = kmalloc((family->maxattr+1) *
568 					sizeof(struct nlattr *), GFP_KERNEL);
569 		if (attrbuf == NULL)
570 			return -ENOMEM;
571 	} else
572 		attrbuf = family->attrbuf;
573 
574 	if (attrbuf) {
575 		err = nlmsg_parse(nlh, hdrlen, attrbuf, family->maxattr,
576 				  ops->policy);
577 		if (err < 0)
578 			goto out;
579 	}
580 
581 	info.snd_seq = nlh->nlmsg_seq;
582 	info.snd_portid = NETLINK_CB(skb).portid;
583 	info.nlhdr = nlh;
584 	info.genlhdr = nlmsg_data(nlh);
585 	info.userhdr = nlmsg_data(nlh) + GENL_HDRLEN;
586 	info.attrs = attrbuf;
587 	genl_info_net_set(&info, net);
588 	memset(&info.user_ptr, 0, sizeof(info.user_ptr));
589 
590 	if (family->pre_doit) {
591 		err = family->pre_doit(ops, skb, &info);
592 		if (err)
593 			goto out;
594 	}
595 
596 	err = ops->doit(skb, &info);
597 
598 	if (family->post_doit)
599 		family->post_doit(ops, skb, &info);
600 
601 out:
602 	if (family->parallel_ops)
603 		kfree(attrbuf);
604 
605 	return err;
606 }
607 
608 static int genl_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
609 {
610 	const struct genl_family *family;
611 	int err;
612 
613 	family = genl_family_find_byid(nlh->nlmsg_type);
614 	if (family == NULL)
615 		return -ENOENT;
616 
617 	if (!family->parallel_ops)
618 		genl_lock();
619 
620 	err = genl_family_rcv_msg(family, skb, nlh);
621 
622 	if (!family->parallel_ops)
623 		genl_unlock();
624 
625 	return err;
626 }
627 
628 static void genl_rcv(struct sk_buff *skb)
629 {
630 	down_read(&cb_lock);
631 	netlink_rcv_skb(skb, &genl_rcv_msg);
632 	up_read(&cb_lock);
633 }
634 
635 /**************************************************************************
636  * Controller
637  **************************************************************************/
638 
639 static struct genl_family genl_ctrl;
640 
641 static int ctrl_fill_info(const struct genl_family *family, u32 portid, u32 seq,
642 			  u32 flags, struct sk_buff *skb, u8 cmd)
643 {
644 	void *hdr;
645 
646 	hdr = genlmsg_put(skb, portid, seq, &genl_ctrl, flags, cmd);
647 	if (hdr == NULL)
648 		return -1;
649 
650 	if (nla_put_string(skb, CTRL_ATTR_FAMILY_NAME, family->name) ||
651 	    nla_put_u16(skb, CTRL_ATTR_FAMILY_ID, family->id) ||
652 	    nla_put_u32(skb, CTRL_ATTR_VERSION, family->version) ||
653 	    nla_put_u32(skb, CTRL_ATTR_HDRSIZE, family->hdrsize) ||
654 	    nla_put_u32(skb, CTRL_ATTR_MAXATTR, family->maxattr))
655 		goto nla_put_failure;
656 
657 	if (family->n_ops) {
658 		struct nlattr *nla_ops;
659 		int i;
660 
661 		nla_ops = nla_nest_start(skb, CTRL_ATTR_OPS);
662 		if (nla_ops == NULL)
663 			goto nla_put_failure;
664 
665 		for (i = 0; i < family->n_ops; i++) {
666 			struct nlattr *nest;
667 			const struct genl_ops *ops = &family->ops[i];
668 			u32 op_flags = ops->flags;
669 
670 			if (ops->dumpit)
671 				op_flags |= GENL_CMD_CAP_DUMP;
672 			if (ops->doit)
673 				op_flags |= GENL_CMD_CAP_DO;
674 			if (ops->policy)
675 				op_flags |= GENL_CMD_CAP_HASPOL;
676 
677 			nest = nla_nest_start(skb, i + 1);
678 			if (nest == NULL)
679 				goto nla_put_failure;
680 
681 			if (nla_put_u32(skb, CTRL_ATTR_OP_ID, ops->cmd) ||
682 			    nla_put_u32(skb, CTRL_ATTR_OP_FLAGS, op_flags))
683 				goto nla_put_failure;
684 
685 			nla_nest_end(skb, nest);
686 		}
687 
688 		nla_nest_end(skb, nla_ops);
689 	}
690 
691 	if (family->n_mcgrps) {
692 		struct nlattr *nla_grps;
693 		int i;
694 
695 		nla_grps = nla_nest_start(skb, CTRL_ATTR_MCAST_GROUPS);
696 		if (nla_grps == NULL)
697 			goto nla_put_failure;
698 
699 		for (i = 0; i < family->n_mcgrps; i++) {
700 			struct nlattr *nest;
701 			const struct genl_multicast_group *grp;
702 
703 			grp = &family->mcgrps[i];
704 
705 			nest = nla_nest_start(skb, i + 1);
706 			if (nest == NULL)
707 				goto nla_put_failure;
708 
709 			if (nla_put_u32(skb, CTRL_ATTR_MCAST_GRP_ID,
710 					family->mcgrp_offset + i) ||
711 			    nla_put_string(skb, CTRL_ATTR_MCAST_GRP_NAME,
712 					   grp->name))
713 				goto nla_put_failure;
714 
715 			nla_nest_end(skb, nest);
716 		}
717 		nla_nest_end(skb, nla_grps);
718 	}
719 
720 	genlmsg_end(skb, hdr);
721 	return 0;
722 
723 nla_put_failure:
724 	genlmsg_cancel(skb, hdr);
725 	return -EMSGSIZE;
726 }
727 
728 static int ctrl_fill_mcgrp_info(const struct genl_family *family,
729 				const struct genl_multicast_group *grp,
730 				int grp_id, u32 portid, u32 seq, u32 flags,
731 				struct sk_buff *skb, u8 cmd)
732 {
733 	void *hdr;
734 	struct nlattr *nla_grps;
735 	struct nlattr *nest;
736 
737 	hdr = genlmsg_put(skb, portid, seq, &genl_ctrl, flags, cmd);
738 	if (hdr == NULL)
739 		return -1;
740 
741 	if (nla_put_string(skb, CTRL_ATTR_FAMILY_NAME, family->name) ||
742 	    nla_put_u16(skb, CTRL_ATTR_FAMILY_ID, family->id))
743 		goto nla_put_failure;
744 
745 	nla_grps = nla_nest_start(skb, CTRL_ATTR_MCAST_GROUPS);
746 	if (nla_grps == NULL)
747 		goto nla_put_failure;
748 
749 	nest = nla_nest_start(skb, 1);
750 	if (nest == NULL)
751 		goto nla_put_failure;
752 
753 	if (nla_put_u32(skb, CTRL_ATTR_MCAST_GRP_ID, grp_id) ||
754 	    nla_put_string(skb, CTRL_ATTR_MCAST_GRP_NAME,
755 			   grp->name))
756 		goto nla_put_failure;
757 
758 	nla_nest_end(skb, nest);
759 	nla_nest_end(skb, nla_grps);
760 
761 	genlmsg_end(skb, hdr);
762 	return 0;
763 
764 nla_put_failure:
765 	genlmsg_cancel(skb, hdr);
766 	return -EMSGSIZE;
767 }
768 
769 static int ctrl_dumpfamily(struct sk_buff *skb, struct netlink_callback *cb)
770 {
771 	int n = 0;
772 	struct genl_family *rt;
773 	struct net *net = sock_net(skb->sk);
774 	int fams_to_skip = cb->args[0];
775 	unsigned int id;
776 
777 	idr_for_each_entry(&genl_fam_idr, rt, id) {
778 		if (!rt->netnsok && !net_eq(net, &init_net))
779 			continue;
780 
781 		if (n++ < fams_to_skip)
782 			continue;
783 
784 		if (ctrl_fill_info(rt, NETLINK_CB(cb->skb).portid,
785 				   cb->nlh->nlmsg_seq, NLM_F_MULTI,
786 				   skb, CTRL_CMD_NEWFAMILY) < 0) {
787 			n--;
788 			break;
789 		}
790 	}
791 
792 	cb->args[0] = n;
793 	return skb->len;
794 }
795 
796 static struct sk_buff *ctrl_build_family_msg(const struct genl_family *family,
797 					     u32 portid, int seq, u8 cmd)
798 {
799 	struct sk_buff *skb;
800 	int err;
801 
802 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
803 	if (skb == NULL)
804 		return ERR_PTR(-ENOBUFS);
805 
806 	err = ctrl_fill_info(family, portid, seq, 0, skb, cmd);
807 	if (err < 0) {
808 		nlmsg_free(skb);
809 		return ERR_PTR(err);
810 	}
811 
812 	return skb;
813 }
814 
815 static struct sk_buff *
816 ctrl_build_mcgrp_msg(const struct genl_family *family,
817 		     const struct genl_multicast_group *grp,
818 		     int grp_id, u32 portid, int seq, u8 cmd)
819 {
820 	struct sk_buff *skb;
821 	int err;
822 
823 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
824 	if (skb == NULL)
825 		return ERR_PTR(-ENOBUFS);
826 
827 	err = ctrl_fill_mcgrp_info(family, grp, grp_id, portid,
828 				   seq, 0, skb, cmd);
829 	if (err < 0) {
830 		nlmsg_free(skb);
831 		return ERR_PTR(err);
832 	}
833 
834 	return skb;
835 }
836 
837 static const struct nla_policy ctrl_policy[CTRL_ATTR_MAX+1] = {
838 	[CTRL_ATTR_FAMILY_ID]	= { .type = NLA_U16 },
839 	[CTRL_ATTR_FAMILY_NAME]	= { .type = NLA_NUL_STRING,
840 				    .len = GENL_NAMSIZ - 1 },
841 };
842 
843 static int ctrl_getfamily(struct sk_buff *skb, struct genl_info *info)
844 {
845 	struct sk_buff *msg;
846 	const struct genl_family *res = NULL;
847 	int err = -EINVAL;
848 
849 	if (info->attrs[CTRL_ATTR_FAMILY_ID]) {
850 		u16 id = nla_get_u16(info->attrs[CTRL_ATTR_FAMILY_ID]);
851 		res = genl_family_find_byid(id);
852 		err = -ENOENT;
853 	}
854 
855 	if (info->attrs[CTRL_ATTR_FAMILY_NAME]) {
856 		char *name;
857 
858 		name = nla_data(info->attrs[CTRL_ATTR_FAMILY_NAME]);
859 		res = genl_family_find_byname(name);
860 #ifdef CONFIG_MODULES
861 		if (res == NULL) {
862 			genl_unlock();
863 			up_read(&cb_lock);
864 			request_module("net-pf-%d-proto-%d-family-%s",
865 				       PF_NETLINK, NETLINK_GENERIC, name);
866 			down_read(&cb_lock);
867 			genl_lock();
868 			res = genl_family_find_byname(name);
869 		}
870 #endif
871 		err = -ENOENT;
872 	}
873 
874 	if (res == NULL)
875 		return err;
876 
877 	if (!res->netnsok && !net_eq(genl_info_net(info), &init_net)) {
878 		/* family doesn't exist here */
879 		return -ENOENT;
880 	}
881 
882 	msg = ctrl_build_family_msg(res, info->snd_portid, info->snd_seq,
883 				    CTRL_CMD_NEWFAMILY);
884 	if (IS_ERR(msg))
885 		return PTR_ERR(msg);
886 
887 	return genlmsg_reply(msg, info);
888 }
889 
890 static int genl_ctrl_event(int event, const struct genl_family *family,
891 			   const struct genl_multicast_group *grp,
892 			   int grp_id)
893 {
894 	struct sk_buff *msg;
895 
896 	/* genl is still initialising */
897 	if (!init_net.genl_sock)
898 		return 0;
899 
900 	switch (event) {
901 	case CTRL_CMD_NEWFAMILY:
902 	case CTRL_CMD_DELFAMILY:
903 		WARN_ON(grp);
904 		msg = ctrl_build_family_msg(family, 0, 0, event);
905 		break;
906 	case CTRL_CMD_NEWMCAST_GRP:
907 	case CTRL_CMD_DELMCAST_GRP:
908 		BUG_ON(!grp);
909 		msg = ctrl_build_mcgrp_msg(family, grp, grp_id, 0, 0, event);
910 		break;
911 	default:
912 		return -EINVAL;
913 	}
914 
915 	if (IS_ERR(msg))
916 		return PTR_ERR(msg);
917 
918 	if (!family->netnsok) {
919 		genlmsg_multicast_netns(&genl_ctrl, &init_net, msg, 0,
920 					0, GFP_KERNEL);
921 	} else {
922 		rcu_read_lock();
923 		genlmsg_multicast_allns(&genl_ctrl, msg, 0,
924 					0, GFP_ATOMIC);
925 		rcu_read_unlock();
926 	}
927 
928 	return 0;
929 }
930 
931 static const struct genl_ops genl_ctrl_ops[] = {
932 	{
933 		.cmd		= CTRL_CMD_GETFAMILY,
934 		.doit		= ctrl_getfamily,
935 		.dumpit		= ctrl_dumpfamily,
936 		.policy		= ctrl_policy,
937 	},
938 };
939 
940 static const struct genl_multicast_group genl_ctrl_groups[] = {
941 	{ .name = "notify", },
942 };
943 
944 static struct genl_family genl_ctrl __ro_after_init = {
945 	.module = THIS_MODULE,
946 	.ops = genl_ctrl_ops,
947 	.n_ops = ARRAY_SIZE(genl_ctrl_ops),
948 	.mcgrps = genl_ctrl_groups,
949 	.n_mcgrps = ARRAY_SIZE(genl_ctrl_groups),
950 	.id = GENL_ID_CTRL,
951 	.name = "nlctrl",
952 	.version = 0x2,
953 	.maxattr = CTRL_ATTR_MAX,
954 	.netnsok = true,
955 };
956 
957 static int genl_bind(struct net *net, int group)
958 {
959 	struct genl_family *f;
960 	int err = -ENOENT;
961 	unsigned int id;
962 
963 	down_read(&cb_lock);
964 
965 	idr_for_each_entry(&genl_fam_idr, f, id) {
966 		if (group >= f->mcgrp_offset &&
967 		    group < f->mcgrp_offset + f->n_mcgrps) {
968 			int fam_grp = group - f->mcgrp_offset;
969 
970 			if (!f->netnsok && net != &init_net)
971 				err = -ENOENT;
972 			else if (f->mcast_bind)
973 				err = f->mcast_bind(net, fam_grp);
974 			else
975 				err = 0;
976 			break;
977 		}
978 	}
979 	up_read(&cb_lock);
980 
981 	return err;
982 }
983 
984 static void genl_unbind(struct net *net, int group)
985 {
986 	struct genl_family *f;
987 	unsigned int id;
988 
989 	down_read(&cb_lock);
990 
991 	idr_for_each_entry(&genl_fam_idr, f, id) {
992 		if (group >= f->mcgrp_offset &&
993 		    group < f->mcgrp_offset + f->n_mcgrps) {
994 			int fam_grp = group - f->mcgrp_offset;
995 
996 			if (f->mcast_unbind)
997 				f->mcast_unbind(net, fam_grp);
998 			break;
999 		}
1000 	}
1001 	up_read(&cb_lock);
1002 }
1003 
1004 static int __net_init genl_pernet_init(struct net *net)
1005 {
1006 	struct netlink_kernel_cfg cfg = {
1007 		.input		= genl_rcv,
1008 		.flags		= NL_CFG_F_NONROOT_RECV,
1009 		.bind		= genl_bind,
1010 		.unbind		= genl_unbind,
1011 	};
1012 
1013 	/* we'll bump the group number right afterwards */
1014 	net->genl_sock = netlink_kernel_create(net, NETLINK_GENERIC, &cfg);
1015 
1016 	if (!net->genl_sock && net_eq(net, &init_net))
1017 		panic("GENL: Cannot initialize generic netlink\n");
1018 
1019 	if (!net->genl_sock)
1020 		return -ENOMEM;
1021 
1022 	return 0;
1023 }
1024 
1025 static void __net_exit genl_pernet_exit(struct net *net)
1026 {
1027 	netlink_kernel_release(net->genl_sock);
1028 	net->genl_sock = NULL;
1029 }
1030 
1031 static struct pernet_operations genl_pernet_ops = {
1032 	.init = genl_pernet_init,
1033 	.exit = genl_pernet_exit,
1034 };
1035 
1036 static int __init genl_init(void)
1037 {
1038 	int err;
1039 
1040 	err = genl_register_family(&genl_ctrl);
1041 	if (err < 0)
1042 		goto problem;
1043 
1044 	err = register_pernet_subsys(&genl_pernet_ops);
1045 	if (err)
1046 		goto problem;
1047 
1048 	return 0;
1049 
1050 problem:
1051 	panic("GENL: Cannot register controller: %d\n", err);
1052 }
1053 
1054 subsys_initcall(genl_init);
1055 
1056 /**
1057  * genl_family_attrbuf - return family's attrbuf
1058  * @family: the family
1059  *
1060  * Return the family's attrbuf, while validating that it's
1061  * actually valid to access it.
1062  *
1063  * You cannot use this function with a family that has parallel_ops
1064  * and you can only use it within (pre/post) doit/dumpit callbacks.
1065  */
1066 struct nlattr **genl_family_attrbuf(const struct genl_family *family)
1067 {
1068 	if (!WARN_ON(family->parallel_ops))
1069 		lockdep_assert_held(&genl_mutex);
1070 
1071 	return family->attrbuf;
1072 }
1073 EXPORT_SYMBOL(genl_family_attrbuf);
1074 
1075 static int genlmsg_mcast(struct sk_buff *skb, u32 portid, unsigned long group,
1076 			 gfp_t flags)
1077 {
1078 	struct sk_buff *tmp;
1079 	struct net *net, *prev = NULL;
1080 	int err;
1081 
1082 	for_each_net_rcu(net) {
1083 		if (prev) {
1084 			tmp = skb_clone(skb, flags);
1085 			if (!tmp) {
1086 				err = -ENOMEM;
1087 				goto error;
1088 			}
1089 			err = nlmsg_multicast(prev->genl_sock, tmp,
1090 					      portid, group, flags);
1091 			if (err)
1092 				goto error;
1093 		}
1094 
1095 		prev = net;
1096 	}
1097 
1098 	return nlmsg_multicast(prev->genl_sock, skb, portid, group, flags);
1099  error:
1100 	kfree_skb(skb);
1101 	return err;
1102 }
1103 
1104 int genlmsg_multicast_allns(const struct genl_family *family,
1105 			    struct sk_buff *skb, u32 portid,
1106 			    unsigned int group, gfp_t flags)
1107 {
1108 	if (WARN_ON_ONCE(group >= family->n_mcgrps))
1109 		return -EINVAL;
1110 	group = family->mcgrp_offset + group;
1111 	return genlmsg_mcast(skb, portid, group, flags);
1112 }
1113 EXPORT_SYMBOL(genlmsg_multicast_allns);
1114 
1115 void genl_notify(const struct genl_family *family, struct sk_buff *skb,
1116 		 struct genl_info *info, u32 group, gfp_t flags)
1117 {
1118 	struct net *net = genl_info_net(info);
1119 	struct sock *sk = net->genl_sock;
1120 	int report = 0;
1121 
1122 	if (info->nlhdr)
1123 		report = nlmsg_report(info->nlhdr);
1124 
1125 	if (WARN_ON_ONCE(group >= family->n_mcgrps))
1126 		return;
1127 	group = family->mcgrp_offset + group;
1128 	nlmsg_notify(sk, skb, info->snd_portid, group, report, flags);
1129 }
1130 EXPORT_SYMBOL(genl_notify);
1131