xref: /linux/net/netlink/genetlink.c (revision dd093fb0)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NETLINK      Generic Netlink Family
4  *
5  * 		Authors:	Jamal Hadi Salim
6  * 				Thomas Graf <tgraf@suug.ch>
7  *				Johannes Berg <johannes@sipsolutions.net>
8  */
9 
10 #include <linux/module.h>
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/errno.h>
14 #include <linux/types.h>
15 #include <linux/socket.h>
16 #include <linux/string.h>
17 #include <linux/skbuff.h>
18 #include <linux/mutex.h>
19 #include <linux/bitmap.h>
20 #include <linux/rwsem.h>
21 #include <linux/idr.h>
22 #include <net/sock.h>
23 #include <net/genetlink.h>
24 
25 static DEFINE_MUTEX(genl_mutex); /* serialization of message processing */
26 static DECLARE_RWSEM(cb_lock);
27 
28 atomic_t genl_sk_destructing_cnt = ATOMIC_INIT(0);
29 DECLARE_WAIT_QUEUE_HEAD(genl_sk_destructing_waitq);
30 
31 void genl_lock(void)
32 {
33 	mutex_lock(&genl_mutex);
34 }
35 EXPORT_SYMBOL(genl_lock);
36 
37 void genl_unlock(void)
38 {
39 	mutex_unlock(&genl_mutex);
40 }
41 EXPORT_SYMBOL(genl_unlock);
42 
43 static void genl_lock_all(void)
44 {
45 	down_write(&cb_lock);
46 	genl_lock();
47 }
48 
49 static void genl_unlock_all(void)
50 {
51 	genl_unlock();
52 	up_write(&cb_lock);
53 }
54 
55 static DEFINE_IDR(genl_fam_idr);
56 
57 /*
58  * Bitmap of multicast groups that are currently in use.
59  *
60  * To avoid an allocation at boot of just one unsigned long,
61  * declare it global instead.
62  * Bit 0 is marked as already used since group 0 is invalid.
63  * Bit 1 is marked as already used since the drop-monitor code
64  * abuses the API and thinks it can statically use group 1.
65  * That group will typically conflict with other groups that
66  * any proper users use.
67  * Bit 16 is marked as used since it's used for generic netlink
68  * and the code no longer marks pre-reserved IDs as used.
69  * Bit 17 is marked as already used since the VFS quota code
70  * also abused this API and relied on family == group ID, we
71  * cater to that by giving it a static family and group ID.
72  * Bit 18 is marked as already used since the PMCRAID driver
73  * did the same thing as the VFS quota code (maybe copied?)
74  */
75 static unsigned long mc_group_start = 0x3 | BIT(GENL_ID_CTRL) |
76 				      BIT(GENL_ID_VFS_DQUOT) |
77 				      BIT(GENL_ID_PMCRAID);
78 static unsigned long *mc_groups = &mc_group_start;
79 static unsigned long mc_groups_longs = 1;
80 
81 /* We need the last attribute with non-zero ID therefore a 2-entry array */
82 static struct nla_policy genl_policy_reject_all[] = {
83 	{ .type = NLA_REJECT },
84 	{ .type = NLA_REJECT },
85 };
86 
87 static int genl_ctrl_event(int event, const struct genl_family *family,
88 			   const struct genl_multicast_group *grp,
89 			   int grp_id);
90 
91 static void
92 genl_op_fill_in_reject_policy(const struct genl_family *family,
93 			      struct genl_ops *op)
94 {
95 	BUILD_BUG_ON(ARRAY_SIZE(genl_policy_reject_all) - 1 != 1);
96 
97 	if (op->policy || op->cmd < family->resv_start_op)
98 		return;
99 
100 	op->policy = genl_policy_reject_all;
101 	op->maxattr = 1;
102 }
103 
104 static void
105 genl_op_fill_in_reject_policy_split(const struct genl_family *family,
106 				    struct genl_split_ops *op)
107 {
108 	if (op->policy)
109 		return;
110 
111 	op->policy = genl_policy_reject_all;
112 	op->maxattr = 1;
113 }
114 
115 static const struct genl_family *genl_family_find_byid(unsigned int id)
116 {
117 	return idr_find(&genl_fam_idr, id);
118 }
119 
120 static const struct genl_family *genl_family_find_byname(char *name)
121 {
122 	const struct genl_family *family;
123 	unsigned int id;
124 
125 	idr_for_each_entry(&genl_fam_idr, family, id)
126 		if (strcmp(family->name, name) == 0)
127 			return family;
128 
129 	return NULL;
130 }
131 
132 struct genl_op_iter {
133 	const struct genl_family *family;
134 	struct genl_split_ops doit;
135 	struct genl_split_ops dumpit;
136 	int cmd_idx;
137 	int entry_idx;
138 	u32 cmd;
139 	u8 flags;
140 };
141 
142 static void genl_op_from_full(const struct genl_family *family,
143 			      unsigned int i, struct genl_ops *op)
144 {
145 	*op = family->ops[i];
146 
147 	if (!op->maxattr)
148 		op->maxattr = family->maxattr;
149 	if (!op->policy)
150 		op->policy = family->policy;
151 
152 	genl_op_fill_in_reject_policy(family, op);
153 }
154 
155 static int genl_get_cmd_full(u32 cmd, const struct genl_family *family,
156 			     struct genl_ops *op)
157 {
158 	int i;
159 
160 	for (i = 0; i < family->n_ops; i++)
161 		if (family->ops[i].cmd == cmd) {
162 			genl_op_from_full(family, i, op);
163 			return 0;
164 		}
165 
166 	return -ENOENT;
167 }
168 
169 static void genl_op_from_small(const struct genl_family *family,
170 			       unsigned int i, struct genl_ops *op)
171 {
172 	memset(op, 0, sizeof(*op));
173 	op->doit	= family->small_ops[i].doit;
174 	op->dumpit	= family->small_ops[i].dumpit;
175 	op->cmd		= family->small_ops[i].cmd;
176 	op->internal_flags = family->small_ops[i].internal_flags;
177 	op->flags	= family->small_ops[i].flags;
178 	op->validate	= family->small_ops[i].validate;
179 
180 	op->maxattr = family->maxattr;
181 	op->policy = family->policy;
182 
183 	genl_op_fill_in_reject_policy(family, op);
184 }
185 
186 static int genl_get_cmd_small(u32 cmd, const struct genl_family *family,
187 			      struct genl_ops *op)
188 {
189 	int i;
190 
191 	for (i = 0; i < family->n_small_ops; i++)
192 		if (family->small_ops[i].cmd == cmd) {
193 			genl_op_from_small(family, i, op);
194 			return 0;
195 		}
196 
197 	return -ENOENT;
198 }
199 
200 static void genl_op_from_split(struct genl_op_iter *iter)
201 {
202 	const struct genl_family *family = iter->family;
203 	int i, cnt = 0;
204 
205 	i = iter->entry_idx - family->n_ops - family->n_small_ops;
206 
207 	if (family->split_ops[i + cnt].flags & GENL_CMD_CAP_DO) {
208 		iter->doit = family->split_ops[i + cnt];
209 		genl_op_fill_in_reject_policy_split(family, &iter->doit);
210 		cnt++;
211 	} else {
212 		memset(&iter->doit, 0, sizeof(iter->doit));
213 	}
214 
215 	if (i + cnt < family->n_split_ops &&
216 	    family->split_ops[i + cnt].flags & GENL_CMD_CAP_DUMP) {
217 		iter->dumpit = family->split_ops[i + cnt];
218 		genl_op_fill_in_reject_policy_split(family, &iter->dumpit);
219 		cnt++;
220 	} else {
221 		memset(&iter->dumpit, 0, sizeof(iter->dumpit));
222 	}
223 
224 	WARN_ON(!cnt);
225 	iter->entry_idx += cnt;
226 }
227 
228 static int
229 genl_get_cmd_split(u32 cmd, u8 flag, const struct genl_family *family,
230 		   struct genl_split_ops *op)
231 {
232 	int i;
233 
234 	for (i = 0; i < family->n_split_ops; i++)
235 		if (family->split_ops[i].cmd == cmd &&
236 		    family->split_ops[i].flags & flag) {
237 			*op = family->split_ops[i];
238 			return 0;
239 		}
240 
241 	return -ENOENT;
242 }
243 
244 static int
245 genl_cmd_full_to_split(struct genl_split_ops *op,
246 		       const struct genl_family *family,
247 		       const struct genl_ops *full, u8 flags)
248 {
249 	if ((flags & GENL_CMD_CAP_DO && !full->doit) ||
250 	    (flags & GENL_CMD_CAP_DUMP && !full->dumpit)) {
251 		memset(op, 0, sizeof(*op));
252 		return -ENOENT;
253 	}
254 
255 	if (flags & GENL_CMD_CAP_DUMP) {
256 		op->start	= full->start;
257 		op->dumpit	= full->dumpit;
258 		op->done	= full->done;
259 	} else {
260 		op->pre_doit	= family->pre_doit;
261 		op->doit	= full->doit;
262 		op->post_doit	= family->post_doit;
263 	}
264 
265 	if (flags & GENL_CMD_CAP_DUMP &&
266 	    full->validate & GENL_DONT_VALIDATE_DUMP) {
267 		op->policy	= NULL;
268 		op->maxattr	= 0;
269 	} else {
270 		op->policy	= full->policy;
271 		op->maxattr	= full->maxattr;
272 	}
273 
274 	op->cmd			= full->cmd;
275 	op->internal_flags	= full->internal_flags;
276 	op->flags		= full->flags;
277 	op->validate		= full->validate;
278 
279 	/* Make sure flags include the GENL_CMD_CAP_DO / GENL_CMD_CAP_DUMP */
280 	op->flags		|= flags;
281 
282 	return 0;
283 }
284 
285 /* Must make sure that op is initialized to 0 on failure */
286 static int
287 genl_get_cmd(u32 cmd, u8 flags, const struct genl_family *family,
288 	     struct genl_split_ops *op)
289 {
290 	struct genl_ops full;
291 	int err;
292 
293 	err = genl_get_cmd_full(cmd, family, &full);
294 	if (err == -ENOENT)
295 		err = genl_get_cmd_small(cmd, family, &full);
296 	/* Found one of legacy forms */
297 	if (err == 0)
298 		return genl_cmd_full_to_split(op, family, &full, flags);
299 
300 	err = genl_get_cmd_split(cmd, flags, family, op);
301 	if (err)
302 		memset(op, 0, sizeof(*op));
303 	return err;
304 }
305 
306 /* For policy dumping only, get ops of both do and dump.
307  * Fail if both are missing, genl_get_cmd() will zero-init in case of failure.
308  */
309 static int
310 genl_get_cmd_both(u32 cmd, const struct genl_family *family,
311 		  struct genl_split_ops *doit, struct genl_split_ops *dumpit)
312 {
313 	int err1, err2;
314 
315 	err1 = genl_get_cmd(cmd, GENL_CMD_CAP_DO, family, doit);
316 	err2 = genl_get_cmd(cmd, GENL_CMD_CAP_DUMP, family, dumpit);
317 
318 	return err1 && err2 ? -ENOENT : 0;
319 }
320 
321 static bool
322 genl_op_iter_init(const struct genl_family *family, struct genl_op_iter *iter)
323 {
324 	iter->family = family;
325 	iter->cmd_idx = 0;
326 	iter->entry_idx = 0;
327 
328 	iter->flags = 0;
329 
330 	return iter->family->n_ops +
331 		iter->family->n_small_ops +
332 		iter->family->n_split_ops;
333 }
334 
335 static bool genl_op_iter_next(struct genl_op_iter *iter)
336 {
337 	const struct genl_family *family = iter->family;
338 	bool legacy_op = true;
339 	struct genl_ops op;
340 
341 	if (iter->entry_idx < family->n_ops) {
342 		genl_op_from_full(family, iter->entry_idx, &op);
343 	} else if (iter->entry_idx < family->n_ops + family->n_small_ops) {
344 		genl_op_from_small(family, iter->entry_idx - family->n_ops,
345 				   &op);
346 	} else if (iter->entry_idx <
347 		   family->n_ops + family->n_small_ops + family->n_split_ops) {
348 		legacy_op = false;
349 		/* updates entry_idx */
350 		genl_op_from_split(iter);
351 	} else {
352 		return false;
353 	}
354 
355 	iter->cmd_idx++;
356 
357 	if (legacy_op) {
358 		iter->entry_idx++;
359 
360 		genl_cmd_full_to_split(&iter->doit, family,
361 				       &op, GENL_CMD_CAP_DO);
362 		genl_cmd_full_to_split(&iter->dumpit, family,
363 				       &op, GENL_CMD_CAP_DUMP);
364 	}
365 
366 	iter->cmd = iter->doit.cmd | iter->dumpit.cmd;
367 	iter->flags = iter->doit.flags | iter->dumpit.flags;
368 
369 	return true;
370 }
371 
372 static void
373 genl_op_iter_copy(struct genl_op_iter *dst, struct genl_op_iter *src)
374 {
375 	*dst = *src;
376 }
377 
378 static unsigned int genl_op_iter_idx(struct genl_op_iter *iter)
379 {
380 	return iter->cmd_idx;
381 }
382 
383 static int genl_allocate_reserve_groups(int n_groups, int *first_id)
384 {
385 	unsigned long *new_groups;
386 	int start = 0;
387 	int i;
388 	int id;
389 	bool fits;
390 
391 	do {
392 		if (start == 0)
393 			id = find_first_zero_bit(mc_groups,
394 						 mc_groups_longs *
395 						 BITS_PER_LONG);
396 		else
397 			id = find_next_zero_bit(mc_groups,
398 						mc_groups_longs * BITS_PER_LONG,
399 						start);
400 
401 		fits = true;
402 		for (i = id;
403 		     i < min_t(int, id + n_groups,
404 			       mc_groups_longs * BITS_PER_LONG);
405 		     i++) {
406 			if (test_bit(i, mc_groups)) {
407 				start = i;
408 				fits = false;
409 				break;
410 			}
411 		}
412 
413 		if (id + n_groups > mc_groups_longs * BITS_PER_LONG) {
414 			unsigned long new_longs = mc_groups_longs +
415 						  BITS_TO_LONGS(n_groups);
416 			size_t nlen = new_longs * sizeof(unsigned long);
417 
418 			if (mc_groups == &mc_group_start) {
419 				new_groups = kzalloc(nlen, GFP_KERNEL);
420 				if (!new_groups)
421 					return -ENOMEM;
422 				mc_groups = new_groups;
423 				*mc_groups = mc_group_start;
424 			} else {
425 				new_groups = krealloc(mc_groups, nlen,
426 						      GFP_KERNEL);
427 				if (!new_groups)
428 					return -ENOMEM;
429 				mc_groups = new_groups;
430 				for (i = 0; i < BITS_TO_LONGS(n_groups); i++)
431 					mc_groups[mc_groups_longs + i] = 0;
432 			}
433 			mc_groups_longs = new_longs;
434 		}
435 	} while (!fits);
436 
437 	for (i = id; i < id + n_groups; i++)
438 		set_bit(i, mc_groups);
439 	*first_id = id;
440 	return 0;
441 }
442 
443 static struct genl_family genl_ctrl;
444 
445 static int genl_validate_assign_mc_groups(struct genl_family *family)
446 {
447 	int first_id;
448 	int n_groups = family->n_mcgrps;
449 	int err = 0, i;
450 	bool groups_allocated = false;
451 
452 	if (!n_groups)
453 		return 0;
454 
455 	for (i = 0; i < n_groups; i++) {
456 		const struct genl_multicast_group *grp = &family->mcgrps[i];
457 
458 		if (WARN_ON(grp->name[0] == '\0'))
459 			return -EINVAL;
460 		if (WARN_ON(memchr(grp->name, '\0', GENL_NAMSIZ) == NULL))
461 			return -EINVAL;
462 	}
463 
464 	/* special-case our own group and hacks */
465 	if (family == &genl_ctrl) {
466 		first_id = GENL_ID_CTRL;
467 		BUG_ON(n_groups != 1);
468 	} else if (strcmp(family->name, "NET_DM") == 0) {
469 		first_id = 1;
470 		BUG_ON(n_groups != 1);
471 	} else if (family->id == GENL_ID_VFS_DQUOT) {
472 		first_id = GENL_ID_VFS_DQUOT;
473 		BUG_ON(n_groups != 1);
474 	} else if (family->id == GENL_ID_PMCRAID) {
475 		first_id = GENL_ID_PMCRAID;
476 		BUG_ON(n_groups != 1);
477 	} else {
478 		groups_allocated = true;
479 		err = genl_allocate_reserve_groups(n_groups, &first_id);
480 		if (err)
481 			return err;
482 	}
483 
484 	family->mcgrp_offset = first_id;
485 
486 	/* if still initializing, can't and don't need to realloc bitmaps */
487 	if (!init_net.genl_sock)
488 		return 0;
489 
490 	if (family->netnsok) {
491 		struct net *net;
492 
493 		netlink_table_grab();
494 		rcu_read_lock();
495 		for_each_net_rcu(net) {
496 			err = __netlink_change_ngroups(net->genl_sock,
497 					mc_groups_longs * BITS_PER_LONG);
498 			if (err) {
499 				/*
500 				 * No need to roll back, can only fail if
501 				 * memory allocation fails and then the
502 				 * number of _possible_ groups has been
503 				 * increased on some sockets which is ok.
504 				 */
505 				break;
506 			}
507 		}
508 		rcu_read_unlock();
509 		netlink_table_ungrab();
510 	} else {
511 		err = netlink_change_ngroups(init_net.genl_sock,
512 					     mc_groups_longs * BITS_PER_LONG);
513 	}
514 
515 	if (groups_allocated && err) {
516 		for (i = 0; i < family->n_mcgrps; i++)
517 			clear_bit(family->mcgrp_offset + i, mc_groups);
518 	}
519 
520 	return err;
521 }
522 
523 static void genl_unregister_mc_groups(const struct genl_family *family)
524 {
525 	struct net *net;
526 	int i;
527 
528 	netlink_table_grab();
529 	rcu_read_lock();
530 	for_each_net_rcu(net) {
531 		for (i = 0; i < family->n_mcgrps; i++)
532 			__netlink_clear_multicast_users(
533 				net->genl_sock, family->mcgrp_offset + i);
534 	}
535 	rcu_read_unlock();
536 	netlink_table_ungrab();
537 
538 	for (i = 0; i < family->n_mcgrps; i++) {
539 		int grp_id = family->mcgrp_offset + i;
540 
541 		if (grp_id != 1)
542 			clear_bit(grp_id, mc_groups);
543 		genl_ctrl_event(CTRL_CMD_DELMCAST_GRP, family,
544 				&family->mcgrps[i], grp_id);
545 	}
546 }
547 
548 static bool genl_split_op_check(const struct genl_split_ops *op)
549 {
550 	if (WARN_ON(hweight8(op->flags & (GENL_CMD_CAP_DO |
551 					  GENL_CMD_CAP_DUMP)) != 1))
552 		return true;
553 	return false;
554 }
555 
556 static int genl_validate_ops(const struct genl_family *family)
557 {
558 	struct genl_op_iter i, j;
559 	unsigned int s;
560 
561 	if (WARN_ON(family->n_ops && !family->ops) ||
562 	    WARN_ON(family->n_small_ops && !family->small_ops) ||
563 	    WARN_ON(family->n_split_ops && !family->split_ops))
564 		return -EINVAL;
565 
566 	for (genl_op_iter_init(family, &i); genl_op_iter_next(&i); ) {
567 		if (!(i.flags & (GENL_CMD_CAP_DO | GENL_CMD_CAP_DUMP)))
568 			return -EINVAL;
569 
570 		if (WARN_ON(i.cmd >= family->resv_start_op &&
571 			    (i.doit.validate || i.dumpit.validate)))
572 			return -EINVAL;
573 
574 		genl_op_iter_copy(&j, &i);
575 		while (genl_op_iter_next(&j)) {
576 			if (i.cmd == j.cmd)
577 				return -EINVAL;
578 		}
579 	}
580 
581 	if (family->n_split_ops) {
582 		if (genl_split_op_check(&family->split_ops[0]))
583 			return -EINVAL;
584 	}
585 
586 	for (s = 1; s < family->n_split_ops; s++) {
587 		const struct genl_split_ops *a, *b;
588 
589 		a = &family->split_ops[s - 1];
590 		b = &family->split_ops[s];
591 
592 		if (genl_split_op_check(b))
593 			return -EINVAL;
594 
595 		/* Check sort order */
596 		if (a->cmd < b->cmd)
597 			continue;
598 
599 		if (a->internal_flags != b->internal_flags ||
600 		    ((a->flags ^ b->flags) & ~(GENL_CMD_CAP_DO |
601 					       GENL_CMD_CAP_DUMP))) {
602 			WARN_ON(1);
603 			return -EINVAL;
604 		}
605 
606 		if ((a->flags & GENL_CMD_CAP_DO) &&
607 		    (b->flags & GENL_CMD_CAP_DUMP))
608 			continue;
609 
610 		WARN_ON(1);
611 		return -EINVAL;
612 	}
613 
614 	return 0;
615 }
616 
617 /**
618  * genl_register_family - register a generic netlink family
619  * @family: generic netlink family
620  *
621  * Registers the specified family after validating it first. Only one
622  * family may be registered with the same family name or identifier.
623  *
624  * The family's ops, multicast groups and module pointer must already
625  * be assigned.
626  *
627  * Return 0 on success or a negative error code.
628  */
629 int genl_register_family(struct genl_family *family)
630 {
631 	int err, i;
632 	int start = GENL_START_ALLOC, end = GENL_MAX_ID;
633 
634 	err = genl_validate_ops(family);
635 	if (err)
636 		return err;
637 
638 	genl_lock_all();
639 
640 	if (genl_family_find_byname(family->name)) {
641 		err = -EEXIST;
642 		goto errout_locked;
643 	}
644 
645 	/*
646 	 * Sadly, a few cases need to be special-cased
647 	 * due to them having previously abused the API
648 	 * and having used their family ID also as their
649 	 * multicast group ID, so we use reserved IDs
650 	 * for both to be sure we can do that mapping.
651 	 */
652 	if (family == &genl_ctrl) {
653 		/* and this needs to be special for initial family lookups */
654 		start = end = GENL_ID_CTRL;
655 	} else if (strcmp(family->name, "pmcraid") == 0) {
656 		start = end = GENL_ID_PMCRAID;
657 	} else if (strcmp(family->name, "VFS_DQUOT") == 0) {
658 		start = end = GENL_ID_VFS_DQUOT;
659 	}
660 
661 	family->id = idr_alloc_cyclic(&genl_fam_idr, family,
662 				      start, end + 1, GFP_KERNEL);
663 	if (family->id < 0) {
664 		err = family->id;
665 		goto errout_locked;
666 	}
667 
668 	err = genl_validate_assign_mc_groups(family);
669 	if (err)
670 		goto errout_remove;
671 
672 	genl_unlock_all();
673 
674 	/* send all events */
675 	genl_ctrl_event(CTRL_CMD_NEWFAMILY, family, NULL, 0);
676 	for (i = 0; i < family->n_mcgrps; i++)
677 		genl_ctrl_event(CTRL_CMD_NEWMCAST_GRP, family,
678 				&family->mcgrps[i], family->mcgrp_offset + i);
679 
680 	return 0;
681 
682 errout_remove:
683 	idr_remove(&genl_fam_idr, family->id);
684 errout_locked:
685 	genl_unlock_all();
686 	return err;
687 }
688 EXPORT_SYMBOL(genl_register_family);
689 
690 /**
691  * genl_unregister_family - unregister generic netlink family
692  * @family: generic netlink family
693  *
694  * Unregisters the specified family.
695  *
696  * Returns 0 on success or a negative error code.
697  */
698 int genl_unregister_family(const struct genl_family *family)
699 {
700 	genl_lock_all();
701 
702 	if (!genl_family_find_byid(family->id)) {
703 		genl_unlock_all();
704 		return -ENOENT;
705 	}
706 
707 	genl_unregister_mc_groups(family);
708 
709 	idr_remove(&genl_fam_idr, family->id);
710 
711 	up_write(&cb_lock);
712 	wait_event(genl_sk_destructing_waitq,
713 		   atomic_read(&genl_sk_destructing_cnt) == 0);
714 	genl_unlock();
715 
716 	genl_ctrl_event(CTRL_CMD_DELFAMILY, family, NULL, 0);
717 
718 	return 0;
719 }
720 EXPORT_SYMBOL(genl_unregister_family);
721 
722 /**
723  * genlmsg_put - Add generic netlink header to netlink message
724  * @skb: socket buffer holding the message
725  * @portid: netlink portid the message is addressed to
726  * @seq: sequence number (usually the one of the sender)
727  * @family: generic netlink family
728  * @flags: netlink message flags
729  * @cmd: generic netlink command
730  *
731  * Returns pointer to user specific header
732  */
733 void *genlmsg_put(struct sk_buff *skb, u32 portid, u32 seq,
734 		  const struct genl_family *family, int flags, u8 cmd)
735 {
736 	struct nlmsghdr *nlh;
737 	struct genlmsghdr *hdr;
738 
739 	nlh = nlmsg_put(skb, portid, seq, family->id, GENL_HDRLEN +
740 			family->hdrsize, flags);
741 	if (nlh == NULL)
742 		return NULL;
743 
744 	hdr = nlmsg_data(nlh);
745 	hdr->cmd = cmd;
746 	hdr->version = family->version;
747 	hdr->reserved = 0;
748 
749 	return (char *) hdr + GENL_HDRLEN;
750 }
751 EXPORT_SYMBOL(genlmsg_put);
752 
753 static struct genl_dumpit_info *genl_dumpit_info_alloc(void)
754 {
755 	return kmalloc(sizeof(struct genl_dumpit_info), GFP_KERNEL);
756 }
757 
758 static void genl_dumpit_info_free(const struct genl_dumpit_info *info)
759 {
760 	kfree(info);
761 }
762 
763 static struct nlattr **
764 genl_family_rcv_msg_attrs_parse(const struct genl_family *family,
765 				struct nlmsghdr *nlh,
766 				struct netlink_ext_ack *extack,
767 				const struct genl_split_ops *ops,
768 				int hdrlen,
769 				enum genl_validate_flags no_strict_flag)
770 {
771 	enum netlink_validation validate = ops->validate & no_strict_flag ?
772 					   NL_VALIDATE_LIBERAL :
773 					   NL_VALIDATE_STRICT;
774 	struct nlattr **attrbuf;
775 	int err;
776 
777 	if (!ops->maxattr)
778 		return NULL;
779 
780 	attrbuf = kmalloc_array(ops->maxattr + 1,
781 				sizeof(struct nlattr *), GFP_KERNEL);
782 	if (!attrbuf)
783 		return ERR_PTR(-ENOMEM);
784 
785 	err = __nlmsg_parse(nlh, hdrlen, attrbuf, ops->maxattr, ops->policy,
786 			    validate, extack);
787 	if (err) {
788 		kfree(attrbuf);
789 		return ERR_PTR(err);
790 	}
791 	return attrbuf;
792 }
793 
794 static void genl_family_rcv_msg_attrs_free(struct nlattr **attrbuf)
795 {
796 	kfree(attrbuf);
797 }
798 
799 struct genl_start_context {
800 	const struct genl_family *family;
801 	struct nlmsghdr *nlh;
802 	struct netlink_ext_ack *extack;
803 	const struct genl_split_ops *ops;
804 	int hdrlen;
805 };
806 
807 static int genl_start(struct netlink_callback *cb)
808 {
809 	struct genl_start_context *ctx = cb->data;
810 	const struct genl_split_ops *ops;
811 	struct genl_dumpit_info *info;
812 	struct nlattr **attrs = NULL;
813 	int rc = 0;
814 
815 	ops = ctx->ops;
816 	if (!(ops->validate & GENL_DONT_VALIDATE_DUMP) &&
817 	    ctx->nlh->nlmsg_len < nlmsg_msg_size(ctx->hdrlen))
818 		return -EINVAL;
819 
820 	attrs = genl_family_rcv_msg_attrs_parse(ctx->family, ctx->nlh, ctx->extack,
821 						ops, ctx->hdrlen,
822 						GENL_DONT_VALIDATE_DUMP_STRICT);
823 	if (IS_ERR(attrs))
824 		return PTR_ERR(attrs);
825 
826 	info = genl_dumpit_info_alloc();
827 	if (!info) {
828 		genl_family_rcv_msg_attrs_free(attrs);
829 		return -ENOMEM;
830 	}
831 	info->family = ctx->family;
832 	info->op = *ops;
833 	info->attrs = attrs;
834 
835 	cb->data = info;
836 	if (ops->start) {
837 		if (!ctx->family->parallel_ops)
838 			genl_lock();
839 		rc = ops->start(cb);
840 		if (!ctx->family->parallel_ops)
841 			genl_unlock();
842 	}
843 
844 	if (rc) {
845 		genl_family_rcv_msg_attrs_free(info->attrs);
846 		genl_dumpit_info_free(info);
847 		cb->data = NULL;
848 	}
849 	return rc;
850 }
851 
852 static int genl_lock_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
853 {
854 	const struct genl_split_ops *ops = &genl_dumpit_info(cb)->op;
855 	int rc;
856 
857 	genl_lock();
858 	rc = ops->dumpit(skb, cb);
859 	genl_unlock();
860 	return rc;
861 }
862 
863 static int genl_lock_done(struct netlink_callback *cb)
864 {
865 	const struct genl_dumpit_info *info = genl_dumpit_info(cb);
866 	const struct genl_split_ops *ops = &info->op;
867 	int rc = 0;
868 
869 	if (ops->done) {
870 		genl_lock();
871 		rc = ops->done(cb);
872 		genl_unlock();
873 	}
874 	genl_family_rcv_msg_attrs_free(info->attrs);
875 	genl_dumpit_info_free(info);
876 	return rc;
877 }
878 
879 static int genl_parallel_done(struct netlink_callback *cb)
880 {
881 	const struct genl_dumpit_info *info = genl_dumpit_info(cb);
882 	const struct genl_split_ops *ops = &info->op;
883 	int rc = 0;
884 
885 	if (ops->done)
886 		rc = ops->done(cb);
887 	genl_family_rcv_msg_attrs_free(info->attrs);
888 	genl_dumpit_info_free(info);
889 	return rc;
890 }
891 
892 static int genl_family_rcv_msg_dumpit(const struct genl_family *family,
893 				      struct sk_buff *skb,
894 				      struct nlmsghdr *nlh,
895 				      struct netlink_ext_ack *extack,
896 				      const struct genl_split_ops *ops,
897 				      int hdrlen, struct net *net)
898 {
899 	struct genl_start_context ctx;
900 	int err;
901 
902 	ctx.family = family;
903 	ctx.nlh = nlh;
904 	ctx.extack = extack;
905 	ctx.ops = ops;
906 	ctx.hdrlen = hdrlen;
907 
908 	if (!family->parallel_ops) {
909 		struct netlink_dump_control c = {
910 			.module = family->module,
911 			.data = &ctx,
912 			.start = genl_start,
913 			.dump = genl_lock_dumpit,
914 			.done = genl_lock_done,
915 		};
916 
917 		genl_unlock();
918 		err = __netlink_dump_start(net->genl_sock, skb, nlh, &c);
919 		genl_lock();
920 	} else {
921 		struct netlink_dump_control c = {
922 			.module = family->module,
923 			.data = &ctx,
924 			.start = genl_start,
925 			.dump = ops->dumpit,
926 			.done = genl_parallel_done,
927 		};
928 
929 		err = __netlink_dump_start(net->genl_sock, skb, nlh, &c);
930 	}
931 
932 	return err;
933 }
934 
935 static int genl_family_rcv_msg_doit(const struct genl_family *family,
936 				    struct sk_buff *skb,
937 				    struct nlmsghdr *nlh,
938 				    struct netlink_ext_ack *extack,
939 				    const struct genl_split_ops *ops,
940 				    int hdrlen, struct net *net)
941 {
942 	struct nlattr **attrbuf;
943 	struct genl_info info;
944 	int err;
945 
946 	attrbuf = genl_family_rcv_msg_attrs_parse(family, nlh, extack,
947 						  ops, hdrlen,
948 						  GENL_DONT_VALIDATE_STRICT);
949 	if (IS_ERR(attrbuf))
950 		return PTR_ERR(attrbuf);
951 
952 	info.snd_seq = nlh->nlmsg_seq;
953 	info.snd_portid = NETLINK_CB(skb).portid;
954 	info.nlhdr = nlh;
955 	info.genlhdr = nlmsg_data(nlh);
956 	info.userhdr = nlmsg_data(nlh) + GENL_HDRLEN;
957 	info.attrs = attrbuf;
958 	info.extack = extack;
959 	genl_info_net_set(&info, net);
960 	memset(&info.user_ptr, 0, sizeof(info.user_ptr));
961 
962 	if (ops->pre_doit) {
963 		err = ops->pre_doit(ops, skb, &info);
964 		if (err)
965 			goto out;
966 	}
967 
968 	err = ops->doit(skb, &info);
969 
970 	if (ops->post_doit)
971 		ops->post_doit(ops, skb, &info);
972 
973 out:
974 	genl_family_rcv_msg_attrs_free(attrbuf);
975 
976 	return err;
977 }
978 
979 static int genl_header_check(const struct genl_family *family,
980 			     struct nlmsghdr *nlh, struct genlmsghdr *hdr,
981 			     struct netlink_ext_ack *extack)
982 {
983 	u16 flags;
984 
985 	/* Only for commands added after we started validating */
986 	if (hdr->cmd < family->resv_start_op)
987 		return 0;
988 
989 	if (hdr->reserved) {
990 		NL_SET_ERR_MSG(extack, "genlmsghdr.reserved field is not 0");
991 		return -EINVAL;
992 	}
993 
994 	/* Old netlink flags have pretty loose semantics, allow only the flags
995 	 * consumed by the core where we can enforce the meaning.
996 	 */
997 	flags = nlh->nlmsg_flags;
998 	if ((flags & NLM_F_DUMP) == NLM_F_DUMP) /* DUMP is 2 bits */
999 		flags &= ~NLM_F_DUMP;
1000 	if (flags & ~(NLM_F_REQUEST | NLM_F_ACK | NLM_F_ECHO)) {
1001 		NL_SET_ERR_MSG(extack,
1002 			       "ambiguous or reserved bits set in nlmsg_flags");
1003 		return -EINVAL;
1004 	}
1005 
1006 	return 0;
1007 }
1008 
1009 static int genl_family_rcv_msg(const struct genl_family *family,
1010 			       struct sk_buff *skb,
1011 			       struct nlmsghdr *nlh,
1012 			       struct netlink_ext_ack *extack)
1013 {
1014 	struct net *net = sock_net(skb->sk);
1015 	struct genlmsghdr *hdr = nlmsg_data(nlh);
1016 	struct genl_split_ops op;
1017 	int hdrlen;
1018 	u8 flags;
1019 
1020 	/* this family doesn't exist in this netns */
1021 	if (!family->netnsok && !net_eq(net, &init_net))
1022 		return -ENOENT;
1023 
1024 	hdrlen = GENL_HDRLEN + family->hdrsize;
1025 	if (nlh->nlmsg_len < nlmsg_msg_size(hdrlen))
1026 		return -EINVAL;
1027 
1028 	if (genl_header_check(family, nlh, hdr, extack))
1029 		return -EINVAL;
1030 
1031 	flags = (nlh->nlmsg_flags & NLM_F_DUMP) == NLM_F_DUMP ?
1032 		GENL_CMD_CAP_DUMP : GENL_CMD_CAP_DO;
1033 	if (genl_get_cmd(hdr->cmd, flags, family, &op))
1034 		return -EOPNOTSUPP;
1035 
1036 	if ((op.flags & GENL_ADMIN_PERM) &&
1037 	    !netlink_capable(skb, CAP_NET_ADMIN))
1038 		return -EPERM;
1039 
1040 	if ((op.flags & GENL_UNS_ADMIN_PERM) &&
1041 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1042 		return -EPERM;
1043 
1044 	if (flags & GENL_CMD_CAP_DUMP)
1045 		return genl_family_rcv_msg_dumpit(family, skb, nlh, extack,
1046 						  &op, hdrlen, net);
1047 	else
1048 		return genl_family_rcv_msg_doit(family, skb, nlh, extack,
1049 						&op, hdrlen, net);
1050 }
1051 
1052 static int genl_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
1053 			struct netlink_ext_ack *extack)
1054 {
1055 	const struct genl_family *family;
1056 	int err;
1057 
1058 	family = genl_family_find_byid(nlh->nlmsg_type);
1059 	if (family == NULL)
1060 		return -ENOENT;
1061 
1062 	if (!family->parallel_ops)
1063 		genl_lock();
1064 
1065 	err = genl_family_rcv_msg(family, skb, nlh, extack);
1066 
1067 	if (!family->parallel_ops)
1068 		genl_unlock();
1069 
1070 	return err;
1071 }
1072 
1073 static void genl_rcv(struct sk_buff *skb)
1074 {
1075 	down_read(&cb_lock);
1076 	netlink_rcv_skb(skb, &genl_rcv_msg);
1077 	up_read(&cb_lock);
1078 }
1079 
1080 /**************************************************************************
1081  * Controller
1082  **************************************************************************/
1083 
1084 static struct genl_family genl_ctrl;
1085 
1086 static int ctrl_fill_info(const struct genl_family *family, u32 portid, u32 seq,
1087 			  u32 flags, struct sk_buff *skb, u8 cmd)
1088 {
1089 	struct genl_op_iter i;
1090 	void *hdr;
1091 
1092 	hdr = genlmsg_put(skb, portid, seq, &genl_ctrl, flags, cmd);
1093 	if (hdr == NULL)
1094 		return -1;
1095 
1096 	if (nla_put_string(skb, CTRL_ATTR_FAMILY_NAME, family->name) ||
1097 	    nla_put_u16(skb, CTRL_ATTR_FAMILY_ID, family->id) ||
1098 	    nla_put_u32(skb, CTRL_ATTR_VERSION, family->version) ||
1099 	    nla_put_u32(skb, CTRL_ATTR_HDRSIZE, family->hdrsize) ||
1100 	    nla_put_u32(skb, CTRL_ATTR_MAXATTR, family->maxattr))
1101 		goto nla_put_failure;
1102 
1103 	if (genl_op_iter_init(family, &i)) {
1104 		struct nlattr *nla_ops;
1105 
1106 		nla_ops = nla_nest_start_noflag(skb, CTRL_ATTR_OPS);
1107 		if (nla_ops == NULL)
1108 			goto nla_put_failure;
1109 
1110 		while (genl_op_iter_next(&i)) {
1111 			struct nlattr *nest;
1112 			u32 op_flags;
1113 
1114 			op_flags = i.flags;
1115 			if (i.doit.policy || i.dumpit.policy)
1116 				op_flags |= GENL_CMD_CAP_HASPOL;
1117 
1118 			nest = nla_nest_start_noflag(skb, genl_op_iter_idx(&i));
1119 			if (nest == NULL)
1120 				goto nla_put_failure;
1121 
1122 			if (nla_put_u32(skb, CTRL_ATTR_OP_ID, i.cmd) ||
1123 			    nla_put_u32(skb, CTRL_ATTR_OP_FLAGS, op_flags))
1124 				goto nla_put_failure;
1125 
1126 			nla_nest_end(skb, nest);
1127 		}
1128 
1129 		nla_nest_end(skb, nla_ops);
1130 	}
1131 
1132 	if (family->n_mcgrps) {
1133 		struct nlattr *nla_grps;
1134 		int i;
1135 
1136 		nla_grps = nla_nest_start_noflag(skb, CTRL_ATTR_MCAST_GROUPS);
1137 		if (nla_grps == NULL)
1138 			goto nla_put_failure;
1139 
1140 		for (i = 0; i < family->n_mcgrps; i++) {
1141 			struct nlattr *nest;
1142 			const struct genl_multicast_group *grp;
1143 
1144 			grp = &family->mcgrps[i];
1145 
1146 			nest = nla_nest_start_noflag(skb, i + 1);
1147 			if (nest == NULL)
1148 				goto nla_put_failure;
1149 
1150 			if (nla_put_u32(skb, CTRL_ATTR_MCAST_GRP_ID,
1151 					family->mcgrp_offset + i) ||
1152 			    nla_put_string(skb, CTRL_ATTR_MCAST_GRP_NAME,
1153 					   grp->name))
1154 				goto nla_put_failure;
1155 
1156 			nla_nest_end(skb, nest);
1157 		}
1158 		nla_nest_end(skb, nla_grps);
1159 	}
1160 
1161 	genlmsg_end(skb, hdr);
1162 	return 0;
1163 
1164 nla_put_failure:
1165 	genlmsg_cancel(skb, hdr);
1166 	return -EMSGSIZE;
1167 }
1168 
1169 static int ctrl_fill_mcgrp_info(const struct genl_family *family,
1170 				const struct genl_multicast_group *grp,
1171 				int grp_id, u32 portid, u32 seq, u32 flags,
1172 				struct sk_buff *skb, u8 cmd)
1173 {
1174 	void *hdr;
1175 	struct nlattr *nla_grps;
1176 	struct nlattr *nest;
1177 
1178 	hdr = genlmsg_put(skb, portid, seq, &genl_ctrl, flags, cmd);
1179 	if (hdr == NULL)
1180 		return -1;
1181 
1182 	if (nla_put_string(skb, CTRL_ATTR_FAMILY_NAME, family->name) ||
1183 	    nla_put_u16(skb, CTRL_ATTR_FAMILY_ID, family->id))
1184 		goto nla_put_failure;
1185 
1186 	nla_grps = nla_nest_start_noflag(skb, CTRL_ATTR_MCAST_GROUPS);
1187 	if (nla_grps == NULL)
1188 		goto nla_put_failure;
1189 
1190 	nest = nla_nest_start_noflag(skb, 1);
1191 	if (nest == NULL)
1192 		goto nla_put_failure;
1193 
1194 	if (nla_put_u32(skb, CTRL_ATTR_MCAST_GRP_ID, grp_id) ||
1195 	    nla_put_string(skb, CTRL_ATTR_MCAST_GRP_NAME,
1196 			   grp->name))
1197 		goto nla_put_failure;
1198 
1199 	nla_nest_end(skb, nest);
1200 	nla_nest_end(skb, nla_grps);
1201 
1202 	genlmsg_end(skb, hdr);
1203 	return 0;
1204 
1205 nla_put_failure:
1206 	genlmsg_cancel(skb, hdr);
1207 	return -EMSGSIZE;
1208 }
1209 
1210 static int ctrl_dumpfamily(struct sk_buff *skb, struct netlink_callback *cb)
1211 {
1212 	int n = 0;
1213 	struct genl_family *rt;
1214 	struct net *net = sock_net(skb->sk);
1215 	int fams_to_skip = cb->args[0];
1216 	unsigned int id;
1217 
1218 	idr_for_each_entry(&genl_fam_idr, rt, id) {
1219 		if (!rt->netnsok && !net_eq(net, &init_net))
1220 			continue;
1221 
1222 		if (n++ < fams_to_skip)
1223 			continue;
1224 
1225 		if (ctrl_fill_info(rt, NETLINK_CB(cb->skb).portid,
1226 				   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1227 				   skb, CTRL_CMD_NEWFAMILY) < 0) {
1228 			n--;
1229 			break;
1230 		}
1231 	}
1232 
1233 	cb->args[0] = n;
1234 	return skb->len;
1235 }
1236 
1237 static struct sk_buff *ctrl_build_family_msg(const struct genl_family *family,
1238 					     u32 portid, int seq, u8 cmd)
1239 {
1240 	struct sk_buff *skb;
1241 	int err;
1242 
1243 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1244 	if (skb == NULL)
1245 		return ERR_PTR(-ENOBUFS);
1246 
1247 	err = ctrl_fill_info(family, portid, seq, 0, skb, cmd);
1248 	if (err < 0) {
1249 		nlmsg_free(skb);
1250 		return ERR_PTR(err);
1251 	}
1252 
1253 	return skb;
1254 }
1255 
1256 static struct sk_buff *
1257 ctrl_build_mcgrp_msg(const struct genl_family *family,
1258 		     const struct genl_multicast_group *grp,
1259 		     int grp_id, u32 portid, int seq, u8 cmd)
1260 {
1261 	struct sk_buff *skb;
1262 	int err;
1263 
1264 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1265 	if (skb == NULL)
1266 		return ERR_PTR(-ENOBUFS);
1267 
1268 	err = ctrl_fill_mcgrp_info(family, grp, grp_id, portid,
1269 				   seq, 0, skb, cmd);
1270 	if (err < 0) {
1271 		nlmsg_free(skb);
1272 		return ERR_PTR(err);
1273 	}
1274 
1275 	return skb;
1276 }
1277 
1278 static const struct nla_policy ctrl_policy_family[] = {
1279 	[CTRL_ATTR_FAMILY_ID]	= { .type = NLA_U16 },
1280 	[CTRL_ATTR_FAMILY_NAME]	= { .type = NLA_NUL_STRING,
1281 				    .len = GENL_NAMSIZ - 1 },
1282 };
1283 
1284 static int ctrl_getfamily(struct sk_buff *skb, struct genl_info *info)
1285 {
1286 	struct sk_buff *msg;
1287 	const struct genl_family *res = NULL;
1288 	int err = -EINVAL;
1289 
1290 	if (info->attrs[CTRL_ATTR_FAMILY_ID]) {
1291 		u16 id = nla_get_u16(info->attrs[CTRL_ATTR_FAMILY_ID]);
1292 		res = genl_family_find_byid(id);
1293 		err = -ENOENT;
1294 	}
1295 
1296 	if (info->attrs[CTRL_ATTR_FAMILY_NAME]) {
1297 		char *name;
1298 
1299 		name = nla_data(info->attrs[CTRL_ATTR_FAMILY_NAME]);
1300 		res = genl_family_find_byname(name);
1301 #ifdef CONFIG_MODULES
1302 		if (res == NULL) {
1303 			genl_unlock();
1304 			up_read(&cb_lock);
1305 			request_module("net-pf-%d-proto-%d-family-%s",
1306 				       PF_NETLINK, NETLINK_GENERIC, name);
1307 			down_read(&cb_lock);
1308 			genl_lock();
1309 			res = genl_family_find_byname(name);
1310 		}
1311 #endif
1312 		err = -ENOENT;
1313 	}
1314 
1315 	if (res == NULL)
1316 		return err;
1317 
1318 	if (!res->netnsok && !net_eq(genl_info_net(info), &init_net)) {
1319 		/* family doesn't exist here */
1320 		return -ENOENT;
1321 	}
1322 
1323 	msg = ctrl_build_family_msg(res, info->snd_portid, info->snd_seq,
1324 				    CTRL_CMD_NEWFAMILY);
1325 	if (IS_ERR(msg))
1326 		return PTR_ERR(msg);
1327 
1328 	return genlmsg_reply(msg, info);
1329 }
1330 
1331 static int genl_ctrl_event(int event, const struct genl_family *family,
1332 			   const struct genl_multicast_group *grp,
1333 			   int grp_id)
1334 {
1335 	struct sk_buff *msg;
1336 
1337 	/* genl is still initialising */
1338 	if (!init_net.genl_sock)
1339 		return 0;
1340 
1341 	switch (event) {
1342 	case CTRL_CMD_NEWFAMILY:
1343 	case CTRL_CMD_DELFAMILY:
1344 		WARN_ON(grp);
1345 		msg = ctrl_build_family_msg(family, 0, 0, event);
1346 		break;
1347 	case CTRL_CMD_NEWMCAST_GRP:
1348 	case CTRL_CMD_DELMCAST_GRP:
1349 		BUG_ON(!grp);
1350 		msg = ctrl_build_mcgrp_msg(family, grp, grp_id, 0, 0, event);
1351 		break;
1352 	default:
1353 		return -EINVAL;
1354 	}
1355 
1356 	if (IS_ERR(msg))
1357 		return PTR_ERR(msg);
1358 
1359 	if (!family->netnsok) {
1360 		genlmsg_multicast_netns(&genl_ctrl, &init_net, msg, 0,
1361 					0, GFP_KERNEL);
1362 	} else {
1363 		rcu_read_lock();
1364 		genlmsg_multicast_allns(&genl_ctrl, msg, 0,
1365 					0, GFP_ATOMIC);
1366 		rcu_read_unlock();
1367 	}
1368 
1369 	return 0;
1370 }
1371 
1372 struct ctrl_dump_policy_ctx {
1373 	struct netlink_policy_dump_state *state;
1374 	const struct genl_family *rt;
1375 	struct genl_op_iter *op_iter;
1376 	u32 op;
1377 	u16 fam_id;
1378 	u8 dump_map:1,
1379 	   single_op:1;
1380 };
1381 
1382 static const struct nla_policy ctrl_policy_policy[] = {
1383 	[CTRL_ATTR_FAMILY_ID]	= { .type = NLA_U16 },
1384 	[CTRL_ATTR_FAMILY_NAME]	= { .type = NLA_NUL_STRING,
1385 				    .len = GENL_NAMSIZ - 1 },
1386 	[CTRL_ATTR_OP]		= { .type = NLA_U32 },
1387 };
1388 
1389 static int ctrl_dumppolicy_start(struct netlink_callback *cb)
1390 {
1391 	const struct genl_dumpit_info *info = genl_dumpit_info(cb);
1392 	struct ctrl_dump_policy_ctx *ctx = (void *)cb->ctx;
1393 	struct nlattr **tb = info->attrs;
1394 	const struct genl_family *rt;
1395 	struct genl_op_iter i;
1396 	int err;
1397 
1398 	BUILD_BUG_ON(sizeof(*ctx) > sizeof(cb->ctx));
1399 
1400 	if (!tb[CTRL_ATTR_FAMILY_ID] && !tb[CTRL_ATTR_FAMILY_NAME])
1401 		return -EINVAL;
1402 
1403 	if (tb[CTRL_ATTR_FAMILY_ID]) {
1404 		ctx->fam_id = nla_get_u16(tb[CTRL_ATTR_FAMILY_ID]);
1405 	} else {
1406 		rt = genl_family_find_byname(
1407 			nla_data(tb[CTRL_ATTR_FAMILY_NAME]));
1408 		if (!rt)
1409 			return -ENOENT;
1410 		ctx->fam_id = rt->id;
1411 	}
1412 
1413 	rt = genl_family_find_byid(ctx->fam_id);
1414 	if (!rt)
1415 		return -ENOENT;
1416 
1417 	ctx->rt = rt;
1418 
1419 	if (tb[CTRL_ATTR_OP]) {
1420 		struct genl_split_ops doit, dump;
1421 
1422 		ctx->single_op = true;
1423 		ctx->op = nla_get_u32(tb[CTRL_ATTR_OP]);
1424 
1425 		err = genl_get_cmd_both(ctx->op, rt, &doit, &dump);
1426 		if (err) {
1427 			NL_SET_BAD_ATTR(cb->extack, tb[CTRL_ATTR_OP]);
1428 			return err;
1429 		}
1430 
1431 		if (doit.policy) {
1432 			err = netlink_policy_dump_add_policy(&ctx->state,
1433 							     doit.policy,
1434 							     doit.maxattr);
1435 			if (err)
1436 				goto err_free_state;
1437 		}
1438 		if (dump.policy) {
1439 			err = netlink_policy_dump_add_policy(&ctx->state,
1440 							     dump.policy,
1441 							     dump.maxattr);
1442 			if (err)
1443 				goto err_free_state;
1444 		}
1445 
1446 		if (!ctx->state)
1447 			return -ENODATA;
1448 
1449 		ctx->dump_map = 1;
1450 		return 0;
1451 	}
1452 
1453 	ctx->op_iter = kmalloc(sizeof(*ctx->op_iter), GFP_KERNEL);
1454 	if (!ctx->op_iter)
1455 		return -ENOMEM;
1456 
1457 	genl_op_iter_init(rt, ctx->op_iter);
1458 	ctx->dump_map = genl_op_iter_next(ctx->op_iter);
1459 
1460 	for (genl_op_iter_init(rt, &i); genl_op_iter_next(&i); ) {
1461 		if (i.doit.policy) {
1462 			err = netlink_policy_dump_add_policy(&ctx->state,
1463 							     i.doit.policy,
1464 							     i.doit.maxattr);
1465 			if (err)
1466 				goto err_free_state;
1467 		}
1468 		if (i.dumpit.policy) {
1469 			err = netlink_policy_dump_add_policy(&ctx->state,
1470 							     i.dumpit.policy,
1471 							     i.dumpit.maxattr);
1472 			if (err)
1473 				goto err_free_state;
1474 		}
1475 	}
1476 
1477 	if (!ctx->state) {
1478 		err = -ENODATA;
1479 		goto err_free_op_iter;
1480 	}
1481 	return 0;
1482 
1483 err_free_state:
1484 	netlink_policy_dump_free(ctx->state);
1485 err_free_op_iter:
1486 	kfree(ctx->op_iter);
1487 	return err;
1488 }
1489 
1490 static void *ctrl_dumppolicy_prep(struct sk_buff *skb,
1491 				  struct netlink_callback *cb)
1492 {
1493 	struct ctrl_dump_policy_ctx *ctx = (void *)cb->ctx;
1494 	void *hdr;
1495 
1496 	hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid,
1497 			  cb->nlh->nlmsg_seq, &genl_ctrl,
1498 			  NLM_F_MULTI, CTRL_CMD_GETPOLICY);
1499 	if (!hdr)
1500 		return NULL;
1501 
1502 	if (nla_put_u16(skb, CTRL_ATTR_FAMILY_ID, ctx->fam_id))
1503 		return NULL;
1504 
1505 	return hdr;
1506 }
1507 
1508 static int ctrl_dumppolicy_put_op(struct sk_buff *skb,
1509 				  struct netlink_callback *cb,
1510 				  struct genl_split_ops *doit,
1511 				  struct genl_split_ops *dumpit)
1512 {
1513 	struct ctrl_dump_policy_ctx *ctx = (void *)cb->ctx;
1514 	struct nlattr *nest_pol, *nest_op;
1515 	void *hdr;
1516 	int idx;
1517 
1518 	/* skip if we have nothing to show */
1519 	if (!doit->policy && !dumpit->policy)
1520 		return 0;
1521 
1522 	hdr = ctrl_dumppolicy_prep(skb, cb);
1523 	if (!hdr)
1524 		return -ENOBUFS;
1525 
1526 	nest_pol = nla_nest_start(skb, CTRL_ATTR_OP_POLICY);
1527 	if (!nest_pol)
1528 		goto err;
1529 
1530 	nest_op = nla_nest_start(skb, doit->cmd);
1531 	if (!nest_op)
1532 		goto err;
1533 
1534 	if (doit->policy) {
1535 		idx = netlink_policy_dump_get_policy_idx(ctx->state,
1536 							 doit->policy,
1537 							 doit->maxattr);
1538 
1539 		if (nla_put_u32(skb, CTRL_ATTR_POLICY_DO, idx))
1540 			goto err;
1541 	}
1542 	if (dumpit->policy) {
1543 		idx = netlink_policy_dump_get_policy_idx(ctx->state,
1544 							 dumpit->policy,
1545 							 dumpit->maxattr);
1546 
1547 		if (nla_put_u32(skb, CTRL_ATTR_POLICY_DUMP, idx))
1548 			goto err;
1549 	}
1550 
1551 	nla_nest_end(skb, nest_op);
1552 	nla_nest_end(skb, nest_pol);
1553 	genlmsg_end(skb, hdr);
1554 
1555 	return 0;
1556 err:
1557 	genlmsg_cancel(skb, hdr);
1558 	return -ENOBUFS;
1559 }
1560 
1561 static int ctrl_dumppolicy(struct sk_buff *skb, struct netlink_callback *cb)
1562 {
1563 	struct ctrl_dump_policy_ctx *ctx = (void *)cb->ctx;
1564 	void *hdr;
1565 
1566 	if (ctx->dump_map) {
1567 		if (ctx->single_op) {
1568 			struct genl_split_ops doit, dumpit;
1569 
1570 			if (WARN_ON(genl_get_cmd_both(ctx->op, ctx->rt,
1571 						      &doit, &dumpit)))
1572 				return -ENOENT;
1573 
1574 			if (ctrl_dumppolicy_put_op(skb, cb, &doit, &dumpit))
1575 				return skb->len;
1576 
1577 			/* done with the per-op policy index list */
1578 			ctx->dump_map = 0;
1579 		}
1580 
1581 		while (ctx->dump_map) {
1582 			if (ctrl_dumppolicy_put_op(skb, cb,
1583 						   &ctx->op_iter->doit,
1584 						   &ctx->op_iter->dumpit))
1585 				return skb->len;
1586 
1587 			ctx->dump_map = genl_op_iter_next(ctx->op_iter);
1588 		}
1589 	}
1590 
1591 	while (netlink_policy_dump_loop(ctx->state)) {
1592 		struct nlattr *nest;
1593 
1594 		hdr = ctrl_dumppolicy_prep(skb, cb);
1595 		if (!hdr)
1596 			goto nla_put_failure;
1597 
1598 		nest = nla_nest_start(skb, CTRL_ATTR_POLICY);
1599 		if (!nest)
1600 			goto nla_put_failure;
1601 
1602 		if (netlink_policy_dump_write(skb, ctx->state))
1603 			goto nla_put_failure;
1604 
1605 		nla_nest_end(skb, nest);
1606 
1607 		genlmsg_end(skb, hdr);
1608 	}
1609 
1610 	return skb->len;
1611 
1612 nla_put_failure:
1613 	genlmsg_cancel(skb, hdr);
1614 	return skb->len;
1615 }
1616 
1617 static int ctrl_dumppolicy_done(struct netlink_callback *cb)
1618 {
1619 	struct ctrl_dump_policy_ctx *ctx = (void *)cb->ctx;
1620 
1621 	kfree(ctx->op_iter);
1622 	netlink_policy_dump_free(ctx->state);
1623 	return 0;
1624 }
1625 
1626 static const struct genl_split_ops genl_ctrl_ops[] = {
1627 	{
1628 		.cmd		= CTRL_CMD_GETFAMILY,
1629 		.validate	= GENL_DONT_VALIDATE_STRICT,
1630 		.policy		= ctrl_policy_family,
1631 		.maxattr	= ARRAY_SIZE(ctrl_policy_family) - 1,
1632 		.doit		= ctrl_getfamily,
1633 		.flags		= GENL_CMD_CAP_DO,
1634 	},
1635 	{
1636 		.cmd		= CTRL_CMD_GETFAMILY,
1637 		.validate	= GENL_DONT_VALIDATE_DUMP,
1638 		.policy		= ctrl_policy_family,
1639 		.maxattr	= ARRAY_SIZE(ctrl_policy_family) - 1,
1640 		.dumpit		= ctrl_dumpfamily,
1641 		.flags		= GENL_CMD_CAP_DUMP,
1642 	},
1643 	{
1644 		.cmd		= CTRL_CMD_GETPOLICY,
1645 		.policy		= ctrl_policy_policy,
1646 		.maxattr	= ARRAY_SIZE(ctrl_policy_policy) - 1,
1647 		.start		= ctrl_dumppolicy_start,
1648 		.dumpit		= ctrl_dumppolicy,
1649 		.done		= ctrl_dumppolicy_done,
1650 		.flags		= GENL_CMD_CAP_DUMP,
1651 	},
1652 };
1653 
1654 static const struct genl_multicast_group genl_ctrl_groups[] = {
1655 	{ .name = "notify", },
1656 };
1657 
1658 static struct genl_family genl_ctrl __ro_after_init = {
1659 	.module = THIS_MODULE,
1660 	.split_ops = genl_ctrl_ops,
1661 	.n_split_ops = ARRAY_SIZE(genl_ctrl_ops),
1662 	.resv_start_op = CTRL_CMD_GETPOLICY + 1,
1663 	.mcgrps = genl_ctrl_groups,
1664 	.n_mcgrps = ARRAY_SIZE(genl_ctrl_groups),
1665 	.id = GENL_ID_CTRL,
1666 	.name = "nlctrl",
1667 	.version = 0x2,
1668 	.netnsok = true,
1669 };
1670 
1671 static int genl_bind(struct net *net, int group)
1672 {
1673 	const struct genl_family *family;
1674 	unsigned int id;
1675 	int ret = 0;
1676 
1677 	down_read(&cb_lock);
1678 
1679 	idr_for_each_entry(&genl_fam_idr, family, id) {
1680 		const struct genl_multicast_group *grp;
1681 		int i;
1682 
1683 		if (family->n_mcgrps == 0)
1684 			continue;
1685 
1686 		i = group - family->mcgrp_offset;
1687 		if (i < 0 || i >= family->n_mcgrps)
1688 			continue;
1689 
1690 		grp = &family->mcgrps[i];
1691 		if ((grp->flags & GENL_UNS_ADMIN_PERM) &&
1692 		    !ns_capable(net->user_ns, CAP_NET_ADMIN))
1693 			ret = -EPERM;
1694 
1695 		break;
1696 	}
1697 
1698 	up_read(&cb_lock);
1699 	return ret;
1700 }
1701 
1702 static int __net_init genl_pernet_init(struct net *net)
1703 {
1704 	struct netlink_kernel_cfg cfg = {
1705 		.input		= genl_rcv,
1706 		.flags		= NL_CFG_F_NONROOT_RECV,
1707 		.bind		= genl_bind,
1708 	};
1709 
1710 	/* we'll bump the group number right afterwards */
1711 	net->genl_sock = netlink_kernel_create(net, NETLINK_GENERIC, &cfg);
1712 
1713 	if (!net->genl_sock && net_eq(net, &init_net))
1714 		panic("GENL: Cannot initialize generic netlink\n");
1715 
1716 	if (!net->genl_sock)
1717 		return -ENOMEM;
1718 
1719 	return 0;
1720 }
1721 
1722 static void __net_exit genl_pernet_exit(struct net *net)
1723 {
1724 	netlink_kernel_release(net->genl_sock);
1725 	net->genl_sock = NULL;
1726 }
1727 
1728 static struct pernet_operations genl_pernet_ops = {
1729 	.init = genl_pernet_init,
1730 	.exit = genl_pernet_exit,
1731 };
1732 
1733 static int __init genl_init(void)
1734 {
1735 	int err;
1736 
1737 	err = genl_register_family(&genl_ctrl);
1738 	if (err < 0)
1739 		goto problem;
1740 
1741 	err = register_pernet_subsys(&genl_pernet_ops);
1742 	if (err)
1743 		goto problem;
1744 
1745 	return 0;
1746 
1747 problem:
1748 	panic("GENL: Cannot register controller: %d\n", err);
1749 }
1750 
1751 core_initcall(genl_init);
1752 
1753 static int genlmsg_mcast(struct sk_buff *skb, u32 portid, unsigned long group,
1754 			 gfp_t flags)
1755 {
1756 	struct sk_buff *tmp;
1757 	struct net *net, *prev = NULL;
1758 	bool delivered = false;
1759 	int err;
1760 
1761 	for_each_net_rcu(net) {
1762 		if (prev) {
1763 			tmp = skb_clone(skb, flags);
1764 			if (!tmp) {
1765 				err = -ENOMEM;
1766 				goto error;
1767 			}
1768 			err = nlmsg_multicast(prev->genl_sock, tmp,
1769 					      portid, group, flags);
1770 			if (!err)
1771 				delivered = true;
1772 			else if (err != -ESRCH)
1773 				goto error;
1774 		}
1775 
1776 		prev = net;
1777 	}
1778 
1779 	err = nlmsg_multicast(prev->genl_sock, skb, portid, group, flags);
1780 	if (!err)
1781 		delivered = true;
1782 	else if (err != -ESRCH)
1783 		return err;
1784 	return delivered ? 0 : -ESRCH;
1785  error:
1786 	kfree_skb(skb);
1787 	return err;
1788 }
1789 
1790 int genlmsg_multicast_allns(const struct genl_family *family,
1791 			    struct sk_buff *skb, u32 portid,
1792 			    unsigned int group, gfp_t flags)
1793 {
1794 	if (WARN_ON_ONCE(group >= family->n_mcgrps))
1795 		return -EINVAL;
1796 
1797 	group = family->mcgrp_offset + group;
1798 	return genlmsg_mcast(skb, portid, group, flags);
1799 }
1800 EXPORT_SYMBOL(genlmsg_multicast_allns);
1801 
1802 void genl_notify(const struct genl_family *family, struct sk_buff *skb,
1803 		 struct genl_info *info, u32 group, gfp_t flags)
1804 {
1805 	struct net *net = genl_info_net(info);
1806 	struct sock *sk = net->genl_sock;
1807 
1808 	if (WARN_ON_ONCE(group >= family->n_mcgrps))
1809 		return;
1810 
1811 	group = family->mcgrp_offset + group;
1812 	nlmsg_notify(sk, skb, info->snd_portid, group,
1813 		     nlmsg_report(info->nlhdr), flags);
1814 }
1815 EXPORT_SYMBOL(genl_notify);
1816