xref: /openbmc/linux/net/ipv4/nexthop.c (revision 8ffdff6a)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Generic nexthop implementation
3  *
4  * Copyright (c) 2017-19 Cumulus Networks
5  * Copyright (c) 2017-19 David Ahern <dsa@cumulusnetworks.com>
6  */
7 
8 #include <linux/nexthop.h>
9 #include <linux/rtnetlink.h>
10 #include <linux/slab.h>
11 #include <net/arp.h>
12 #include <net/ipv6_stubs.h>
13 #include <net/lwtunnel.h>
14 #include <net/ndisc.h>
15 #include <net/nexthop.h>
16 #include <net/route.h>
17 #include <net/sock.h>
18 
19 static void remove_nexthop(struct net *net, struct nexthop *nh,
20 			   struct nl_info *nlinfo);
21 
22 #define NH_DEV_HASHBITS  8
23 #define NH_DEV_HASHSIZE (1U << NH_DEV_HASHBITS)
24 
25 static const struct nla_policy rtm_nh_policy_new[] = {
26 	[NHA_ID]		= { .type = NLA_U32 },
27 	[NHA_GROUP]		= { .type = NLA_BINARY },
28 	[NHA_GROUP_TYPE]	= { .type = NLA_U16 },
29 	[NHA_BLACKHOLE]		= { .type = NLA_FLAG },
30 	[NHA_OIF]		= { .type = NLA_U32 },
31 	[NHA_GATEWAY]		= { .type = NLA_BINARY },
32 	[NHA_ENCAP_TYPE]	= { .type = NLA_U16 },
33 	[NHA_ENCAP]		= { .type = NLA_NESTED },
34 	[NHA_FDB]		= { .type = NLA_FLAG },
35 };
36 
37 static const struct nla_policy rtm_nh_policy_get[] = {
38 	[NHA_ID]		= { .type = NLA_U32 },
39 };
40 
41 static const struct nla_policy rtm_nh_policy_dump[] = {
42 	[NHA_OIF]		= { .type = NLA_U32 },
43 	[NHA_GROUPS]		= { .type = NLA_FLAG },
44 	[NHA_MASTER]		= { .type = NLA_U32 },
45 	[NHA_FDB]		= { .type = NLA_FLAG },
46 };
47 
48 static bool nexthop_notifiers_is_empty(struct net *net)
49 {
50 	return !net->nexthop.notifier_chain.head;
51 }
52 
53 static void
54 __nh_notifier_single_info_init(struct nh_notifier_single_info *nh_info,
55 			       const struct nexthop *nh)
56 {
57 	struct nh_info *nhi = rtnl_dereference(nh->nh_info);
58 
59 	nh_info->dev = nhi->fib_nhc.nhc_dev;
60 	nh_info->gw_family = nhi->fib_nhc.nhc_gw_family;
61 	if (nh_info->gw_family == AF_INET)
62 		nh_info->ipv4 = nhi->fib_nhc.nhc_gw.ipv4;
63 	else if (nh_info->gw_family == AF_INET6)
64 		nh_info->ipv6 = nhi->fib_nhc.nhc_gw.ipv6;
65 
66 	nh_info->is_reject = nhi->reject_nh;
67 	nh_info->is_fdb = nhi->fdb_nh;
68 	nh_info->has_encap = !!nhi->fib_nhc.nhc_lwtstate;
69 }
70 
71 static int nh_notifier_single_info_init(struct nh_notifier_info *info,
72 					const struct nexthop *nh)
73 {
74 	info->type = NH_NOTIFIER_INFO_TYPE_SINGLE;
75 	info->nh = kzalloc(sizeof(*info->nh), GFP_KERNEL);
76 	if (!info->nh)
77 		return -ENOMEM;
78 
79 	__nh_notifier_single_info_init(info->nh, nh);
80 
81 	return 0;
82 }
83 
84 static void nh_notifier_single_info_fini(struct nh_notifier_info *info)
85 {
86 	kfree(info->nh);
87 }
88 
89 static int nh_notifier_mp_info_init(struct nh_notifier_info *info,
90 				    struct nh_group *nhg)
91 {
92 	u16 num_nh = nhg->num_nh;
93 	int i;
94 
95 	info->type = NH_NOTIFIER_INFO_TYPE_GRP;
96 	info->nh_grp = kzalloc(struct_size(info->nh_grp, nh_entries, num_nh),
97 			       GFP_KERNEL);
98 	if (!info->nh_grp)
99 		return -ENOMEM;
100 
101 	info->nh_grp->num_nh = num_nh;
102 	info->nh_grp->is_fdb = nhg->fdb_nh;
103 
104 	for (i = 0; i < num_nh; i++) {
105 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
106 
107 		info->nh_grp->nh_entries[i].id = nhge->nh->id;
108 		info->nh_grp->nh_entries[i].weight = nhge->weight;
109 		__nh_notifier_single_info_init(&info->nh_grp->nh_entries[i].nh,
110 					       nhge->nh);
111 	}
112 
113 	return 0;
114 }
115 
116 static int nh_notifier_grp_info_init(struct nh_notifier_info *info,
117 				     const struct nexthop *nh)
118 {
119 	struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
120 
121 	if (nhg->mpath)
122 		return nh_notifier_mp_info_init(info, nhg);
123 	return -EINVAL;
124 }
125 
126 static void nh_notifier_grp_info_fini(struct nh_notifier_info *info,
127 				      const struct nexthop *nh)
128 {
129 	struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
130 
131 	if (nhg->mpath)
132 		kfree(info->nh_grp);
133 }
134 
135 static int nh_notifier_info_init(struct nh_notifier_info *info,
136 				 const struct nexthop *nh)
137 {
138 	info->id = nh->id;
139 
140 	if (nh->is_group)
141 		return nh_notifier_grp_info_init(info, nh);
142 	else
143 		return nh_notifier_single_info_init(info, nh);
144 }
145 
146 static void nh_notifier_info_fini(struct nh_notifier_info *info,
147 				  const struct nexthop *nh)
148 {
149 	if (nh->is_group)
150 		nh_notifier_grp_info_fini(info, nh);
151 	else
152 		nh_notifier_single_info_fini(info);
153 }
154 
155 static int call_nexthop_notifiers(struct net *net,
156 				  enum nexthop_event_type event_type,
157 				  struct nexthop *nh,
158 				  struct netlink_ext_ack *extack)
159 {
160 	struct nh_notifier_info info = {
161 		.net = net,
162 		.extack = extack,
163 	};
164 	int err;
165 
166 	ASSERT_RTNL();
167 
168 	if (nexthop_notifiers_is_empty(net))
169 		return 0;
170 
171 	err = nh_notifier_info_init(&info, nh);
172 	if (err) {
173 		NL_SET_ERR_MSG(extack, "Failed to initialize nexthop notifier info");
174 		return err;
175 	}
176 
177 	err = blocking_notifier_call_chain(&net->nexthop.notifier_chain,
178 					   event_type, &info);
179 	nh_notifier_info_fini(&info, nh);
180 
181 	return notifier_to_errno(err);
182 }
183 
184 static int call_nexthop_notifier(struct notifier_block *nb, struct net *net,
185 				 enum nexthop_event_type event_type,
186 				 struct nexthop *nh,
187 				 struct netlink_ext_ack *extack)
188 {
189 	struct nh_notifier_info info = {
190 		.net = net,
191 		.extack = extack,
192 	};
193 	int err;
194 
195 	err = nh_notifier_info_init(&info, nh);
196 	if (err)
197 		return err;
198 
199 	err = nb->notifier_call(nb, event_type, &info);
200 	nh_notifier_info_fini(&info, nh);
201 
202 	return notifier_to_errno(err);
203 }
204 
205 static unsigned int nh_dev_hashfn(unsigned int val)
206 {
207 	unsigned int mask = NH_DEV_HASHSIZE - 1;
208 
209 	return (val ^
210 		(val >> NH_DEV_HASHBITS) ^
211 		(val >> (NH_DEV_HASHBITS * 2))) & mask;
212 }
213 
214 static void nexthop_devhash_add(struct net *net, struct nh_info *nhi)
215 {
216 	struct net_device *dev = nhi->fib_nhc.nhc_dev;
217 	struct hlist_head *head;
218 	unsigned int hash;
219 
220 	WARN_ON(!dev);
221 
222 	hash = nh_dev_hashfn(dev->ifindex);
223 	head = &net->nexthop.devhash[hash];
224 	hlist_add_head(&nhi->dev_hash, head);
225 }
226 
227 static void nexthop_free_group(struct nexthop *nh)
228 {
229 	struct nh_group *nhg;
230 	int i;
231 
232 	nhg = rcu_dereference_raw(nh->nh_grp);
233 	for (i = 0; i < nhg->num_nh; ++i) {
234 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
235 
236 		WARN_ON(!list_empty(&nhge->nh_list));
237 		nexthop_put(nhge->nh);
238 	}
239 
240 	WARN_ON(nhg->spare == nhg);
241 
242 	kfree(nhg->spare);
243 	kfree(nhg);
244 }
245 
246 static void nexthop_free_single(struct nexthop *nh)
247 {
248 	struct nh_info *nhi;
249 
250 	nhi = rcu_dereference_raw(nh->nh_info);
251 	switch (nhi->family) {
252 	case AF_INET:
253 		fib_nh_release(nh->net, &nhi->fib_nh);
254 		break;
255 	case AF_INET6:
256 		ipv6_stub->fib6_nh_release(&nhi->fib6_nh);
257 		break;
258 	}
259 	kfree(nhi);
260 }
261 
262 void nexthop_free_rcu(struct rcu_head *head)
263 {
264 	struct nexthop *nh = container_of(head, struct nexthop, rcu);
265 
266 	if (nh->is_group)
267 		nexthop_free_group(nh);
268 	else
269 		nexthop_free_single(nh);
270 
271 	kfree(nh);
272 }
273 EXPORT_SYMBOL_GPL(nexthop_free_rcu);
274 
275 static struct nexthop *nexthop_alloc(void)
276 {
277 	struct nexthop *nh;
278 
279 	nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL);
280 	if (nh) {
281 		INIT_LIST_HEAD(&nh->fi_list);
282 		INIT_LIST_HEAD(&nh->f6i_list);
283 		INIT_LIST_HEAD(&nh->grp_list);
284 		INIT_LIST_HEAD(&nh->fdb_list);
285 	}
286 	return nh;
287 }
288 
289 static struct nh_group *nexthop_grp_alloc(u16 num_nh)
290 {
291 	struct nh_group *nhg;
292 
293 	nhg = kzalloc(struct_size(nhg, nh_entries, num_nh), GFP_KERNEL);
294 	if (nhg)
295 		nhg->num_nh = num_nh;
296 
297 	return nhg;
298 }
299 
300 static void nh_base_seq_inc(struct net *net)
301 {
302 	while (++net->nexthop.seq == 0)
303 		;
304 }
305 
306 /* no reference taken; rcu lock or rtnl must be held */
307 struct nexthop *nexthop_find_by_id(struct net *net, u32 id)
308 {
309 	struct rb_node **pp, *parent = NULL, *next;
310 
311 	pp = &net->nexthop.rb_root.rb_node;
312 	while (1) {
313 		struct nexthop *nh;
314 
315 		next = rcu_dereference_raw(*pp);
316 		if (!next)
317 			break;
318 		parent = next;
319 
320 		nh = rb_entry(parent, struct nexthop, rb_node);
321 		if (id < nh->id)
322 			pp = &next->rb_left;
323 		else if (id > nh->id)
324 			pp = &next->rb_right;
325 		else
326 			return nh;
327 	}
328 	return NULL;
329 }
330 EXPORT_SYMBOL_GPL(nexthop_find_by_id);
331 
332 /* used for auto id allocation; called with rtnl held */
333 static u32 nh_find_unused_id(struct net *net)
334 {
335 	u32 id_start = net->nexthop.last_id_allocated;
336 
337 	while (1) {
338 		net->nexthop.last_id_allocated++;
339 		if (net->nexthop.last_id_allocated == id_start)
340 			break;
341 
342 		if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated))
343 			return net->nexthop.last_id_allocated;
344 	}
345 	return 0;
346 }
347 
348 static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
349 {
350 	struct nexthop_grp *p;
351 	size_t len = nhg->num_nh * sizeof(*p);
352 	struct nlattr *nla;
353 	u16 group_type = 0;
354 	int i;
355 
356 	if (nhg->mpath)
357 		group_type = NEXTHOP_GRP_TYPE_MPATH;
358 
359 	if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type))
360 		goto nla_put_failure;
361 
362 	nla = nla_reserve(skb, NHA_GROUP, len);
363 	if (!nla)
364 		goto nla_put_failure;
365 
366 	p = nla_data(nla);
367 	for (i = 0; i < nhg->num_nh; ++i) {
368 		p->id = nhg->nh_entries[i].nh->id;
369 		p->weight = nhg->nh_entries[i].weight - 1;
370 		p += 1;
371 	}
372 
373 	return 0;
374 
375 nla_put_failure:
376 	return -EMSGSIZE;
377 }
378 
379 static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh,
380 			int event, u32 portid, u32 seq, unsigned int nlflags)
381 {
382 	struct fib6_nh *fib6_nh;
383 	struct fib_nh *fib_nh;
384 	struct nlmsghdr *nlh;
385 	struct nh_info *nhi;
386 	struct nhmsg *nhm;
387 
388 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
389 	if (!nlh)
390 		return -EMSGSIZE;
391 
392 	nhm = nlmsg_data(nlh);
393 	nhm->nh_family = AF_UNSPEC;
394 	nhm->nh_flags = nh->nh_flags;
395 	nhm->nh_protocol = nh->protocol;
396 	nhm->nh_scope = 0;
397 	nhm->resvd = 0;
398 
399 	if (nla_put_u32(skb, NHA_ID, nh->id))
400 		goto nla_put_failure;
401 
402 	if (nh->is_group) {
403 		struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
404 
405 		if (nhg->fdb_nh && nla_put_flag(skb, NHA_FDB))
406 			goto nla_put_failure;
407 		if (nla_put_nh_group(skb, nhg))
408 			goto nla_put_failure;
409 		goto out;
410 	}
411 
412 	nhi = rtnl_dereference(nh->nh_info);
413 	nhm->nh_family = nhi->family;
414 	if (nhi->reject_nh) {
415 		if (nla_put_flag(skb, NHA_BLACKHOLE))
416 			goto nla_put_failure;
417 		goto out;
418 	} else if (nhi->fdb_nh) {
419 		if (nla_put_flag(skb, NHA_FDB))
420 			goto nla_put_failure;
421 	} else {
422 		const struct net_device *dev;
423 
424 		dev = nhi->fib_nhc.nhc_dev;
425 		if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex))
426 			goto nla_put_failure;
427 	}
428 
429 	nhm->nh_scope = nhi->fib_nhc.nhc_scope;
430 	switch (nhi->family) {
431 	case AF_INET:
432 		fib_nh = &nhi->fib_nh;
433 		if (fib_nh->fib_nh_gw_family &&
434 		    nla_put_be32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4))
435 			goto nla_put_failure;
436 		break;
437 
438 	case AF_INET6:
439 		fib6_nh = &nhi->fib6_nh;
440 		if (fib6_nh->fib_nh_gw_family &&
441 		    nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6))
442 			goto nla_put_failure;
443 		break;
444 	}
445 
446 	if (nhi->fib_nhc.nhc_lwtstate &&
447 	    lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate,
448 				NHA_ENCAP, NHA_ENCAP_TYPE) < 0)
449 		goto nla_put_failure;
450 
451 out:
452 	nlmsg_end(skb, nlh);
453 	return 0;
454 
455 nla_put_failure:
456 	nlmsg_cancel(skb, nlh);
457 	return -EMSGSIZE;
458 }
459 
460 static size_t nh_nlmsg_size_grp(struct nexthop *nh)
461 {
462 	struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
463 	size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh;
464 
465 	return nla_total_size(sz) +
466 	       nla_total_size(2);  /* NHA_GROUP_TYPE */
467 }
468 
469 static size_t nh_nlmsg_size_single(struct nexthop *nh)
470 {
471 	struct nh_info *nhi = rtnl_dereference(nh->nh_info);
472 	size_t sz;
473 
474 	/* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE
475 	 * are mutually exclusive
476 	 */
477 	sz = nla_total_size(4);  /* NHA_OIF */
478 
479 	switch (nhi->family) {
480 	case AF_INET:
481 		if (nhi->fib_nh.fib_nh_gw_family)
482 			sz += nla_total_size(4);  /* NHA_GATEWAY */
483 		break;
484 
485 	case AF_INET6:
486 		/* NHA_GATEWAY */
487 		if (nhi->fib6_nh.fib_nh_gw_family)
488 			sz += nla_total_size(sizeof(const struct in6_addr));
489 		break;
490 	}
491 
492 	if (nhi->fib_nhc.nhc_lwtstate) {
493 		sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate);
494 		sz += nla_total_size(2);  /* NHA_ENCAP_TYPE */
495 	}
496 
497 	return sz;
498 }
499 
500 static size_t nh_nlmsg_size(struct nexthop *nh)
501 {
502 	size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg));
503 
504 	sz += nla_total_size(4); /* NHA_ID */
505 
506 	if (nh->is_group)
507 		sz += nh_nlmsg_size_grp(nh);
508 	else
509 		sz += nh_nlmsg_size_single(nh);
510 
511 	return sz;
512 }
513 
514 static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info)
515 {
516 	unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0;
517 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
518 	struct sk_buff *skb;
519 	int err = -ENOBUFS;
520 
521 	skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any());
522 	if (!skb)
523 		goto errout;
524 
525 	err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags);
526 	if (err < 0) {
527 		/* -EMSGSIZE implies BUG in nh_nlmsg_size() */
528 		WARN_ON(err == -EMSGSIZE);
529 		kfree_skb(skb);
530 		goto errout;
531 	}
532 
533 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP,
534 		    info->nlh, gfp_any());
535 	return;
536 errout:
537 	if (err < 0)
538 		rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err);
539 }
540 
541 static bool valid_group_nh(struct nexthop *nh, unsigned int npaths,
542 			   bool *is_fdb, struct netlink_ext_ack *extack)
543 {
544 	if (nh->is_group) {
545 		struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
546 
547 		/* nested multipath (group within a group) is not
548 		 * supported
549 		 */
550 		if (nhg->mpath) {
551 			NL_SET_ERR_MSG(extack,
552 				       "Multipath group can not be a nexthop within a group");
553 			return false;
554 		}
555 		*is_fdb = nhg->fdb_nh;
556 	} else {
557 		struct nh_info *nhi = rtnl_dereference(nh->nh_info);
558 
559 		if (nhi->reject_nh && npaths > 1) {
560 			NL_SET_ERR_MSG(extack,
561 				       "Blackhole nexthop can not be used in a group with more than 1 path");
562 			return false;
563 		}
564 		*is_fdb = nhi->fdb_nh;
565 	}
566 
567 	return true;
568 }
569 
570 static int nh_check_attr_fdb_group(struct nexthop *nh, u8 *nh_family,
571 				   struct netlink_ext_ack *extack)
572 {
573 	struct nh_info *nhi;
574 
575 	nhi = rtnl_dereference(nh->nh_info);
576 
577 	if (!nhi->fdb_nh) {
578 		NL_SET_ERR_MSG(extack, "FDB nexthop group can only have fdb nexthops");
579 		return -EINVAL;
580 	}
581 
582 	if (*nh_family == AF_UNSPEC) {
583 		*nh_family = nhi->family;
584 	} else if (*nh_family != nhi->family) {
585 		NL_SET_ERR_MSG(extack, "FDB nexthop group cannot have mixed family nexthops");
586 		return -EINVAL;
587 	}
588 
589 	return 0;
590 }
591 
592 static int nh_check_attr_group(struct net *net,
593 			       struct nlattr *tb[], size_t tb_size,
594 			       struct netlink_ext_ack *extack)
595 {
596 	unsigned int len = nla_len(tb[NHA_GROUP]);
597 	u8 nh_family = AF_UNSPEC;
598 	struct nexthop_grp *nhg;
599 	unsigned int i, j;
600 	u8 nhg_fdb = 0;
601 
602 	if (!len || len & (sizeof(struct nexthop_grp) - 1)) {
603 		NL_SET_ERR_MSG(extack,
604 			       "Invalid length for nexthop group attribute");
605 		return -EINVAL;
606 	}
607 
608 	/* convert len to number of nexthop ids */
609 	len /= sizeof(*nhg);
610 
611 	nhg = nla_data(tb[NHA_GROUP]);
612 	for (i = 0; i < len; ++i) {
613 		if (nhg[i].resvd1 || nhg[i].resvd2) {
614 			NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0");
615 			return -EINVAL;
616 		}
617 		if (nhg[i].weight > 254) {
618 			NL_SET_ERR_MSG(extack, "Invalid value for weight");
619 			return -EINVAL;
620 		}
621 		for (j = i + 1; j < len; ++j) {
622 			if (nhg[i].id == nhg[j].id) {
623 				NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group");
624 				return -EINVAL;
625 			}
626 		}
627 	}
628 
629 	if (tb[NHA_FDB])
630 		nhg_fdb = 1;
631 	nhg = nla_data(tb[NHA_GROUP]);
632 	for (i = 0; i < len; ++i) {
633 		struct nexthop *nh;
634 		bool is_fdb_nh;
635 
636 		nh = nexthop_find_by_id(net, nhg[i].id);
637 		if (!nh) {
638 			NL_SET_ERR_MSG(extack, "Invalid nexthop id");
639 			return -EINVAL;
640 		}
641 		if (!valid_group_nh(nh, len, &is_fdb_nh, extack))
642 			return -EINVAL;
643 
644 		if (nhg_fdb && nh_check_attr_fdb_group(nh, &nh_family, extack))
645 			return -EINVAL;
646 
647 		if (!nhg_fdb && is_fdb_nh) {
648 			NL_SET_ERR_MSG(extack, "Non FDB nexthop group cannot have fdb nexthops");
649 			return -EINVAL;
650 		}
651 	}
652 	for (i = NHA_GROUP_TYPE + 1; i < tb_size; ++i) {
653 		if (!tb[i])
654 			continue;
655 		if (i == NHA_FDB)
656 			continue;
657 		NL_SET_ERR_MSG(extack,
658 			       "No other attributes can be set in nexthop groups");
659 		return -EINVAL;
660 	}
661 
662 	return 0;
663 }
664 
665 static bool ipv6_good_nh(const struct fib6_nh *nh)
666 {
667 	int state = NUD_REACHABLE;
668 	struct neighbour *n;
669 
670 	rcu_read_lock_bh();
671 
672 	n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6);
673 	if (n)
674 		state = n->nud_state;
675 
676 	rcu_read_unlock_bh();
677 
678 	return !!(state & NUD_VALID);
679 }
680 
681 static bool ipv4_good_nh(const struct fib_nh *nh)
682 {
683 	int state = NUD_REACHABLE;
684 	struct neighbour *n;
685 
686 	rcu_read_lock_bh();
687 
688 	n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
689 				      (__force u32)nh->fib_nh_gw4);
690 	if (n)
691 		state = n->nud_state;
692 
693 	rcu_read_unlock_bh();
694 
695 	return !!(state & NUD_VALID);
696 }
697 
698 static struct nexthop *nexthop_select_path_mp(struct nh_group *nhg, int hash)
699 {
700 	struct nexthop *rc = NULL;
701 	int i;
702 
703 	for (i = 0; i < nhg->num_nh; ++i) {
704 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
705 		struct nh_info *nhi;
706 
707 		if (hash > atomic_read(&nhge->mpath.upper_bound))
708 			continue;
709 
710 		nhi = rcu_dereference(nhge->nh->nh_info);
711 		if (nhi->fdb_nh)
712 			return nhge->nh;
713 
714 		/* nexthops always check if it is good and does
715 		 * not rely on a sysctl for this behavior
716 		 */
717 		switch (nhi->family) {
718 		case AF_INET:
719 			if (ipv4_good_nh(&nhi->fib_nh))
720 				return nhge->nh;
721 			break;
722 		case AF_INET6:
723 			if (ipv6_good_nh(&nhi->fib6_nh))
724 				return nhge->nh;
725 			break;
726 		}
727 
728 		if (!rc)
729 			rc = nhge->nh;
730 	}
731 
732 	return rc;
733 }
734 
735 struct nexthop *nexthop_select_path(struct nexthop *nh, int hash)
736 {
737 	struct nh_group *nhg;
738 
739 	if (!nh->is_group)
740 		return nh;
741 
742 	nhg = rcu_dereference(nh->nh_grp);
743 	if (nhg->mpath)
744 		return nexthop_select_path_mp(nhg, hash);
745 
746 	/* Unreachable. */
747 	return NULL;
748 }
749 EXPORT_SYMBOL_GPL(nexthop_select_path);
750 
751 int nexthop_for_each_fib6_nh(struct nexthop *nh,
752 			     int (*cb)(struct fib6_nh *nh, void *arg),
753 			     void *arg)
754 {
755 	struct nh_info *nhi;
756 	int err;
757 
758 	if (nh->is_group) {
759 		struct nh_group *nhg;
760 		int i;
761 
762 		nhg = rcu_dereference_rtnl(nh->nh_grp);
763 		for (i = 0; i < nhg->num_nh; i++) {
764 			struct nh_grp_entry *nhge = &nhg->nh_entries[i];
765 
766 			nhi = rcu_dereference_rtnl(nhge->nh->nh_info);
767 			err = cb(&nhi->fib6_nh, arg);
768 			if (err)
769 				return err;
770 		}
771 	} else {
772 		nhi = rcu_dereference_rtnl(nh->nh_info);
773 		err = cb(&nhi->fib6_nh, arg);
774 		if (err)
775 			return err;
776 	}
777 
778 	return 0;
779 }
780 EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh);
781 
782 static int check_src_addr(const struct in6_addr *saddr,
783 			  struct netlink_ext_ack *extack)
784 {
785 	if (!ipv6_addr_any(saddr)) {
786 		NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects");
787 		return -EINVAL;
788 	}
789 	return 0;
790 }
791 
792 int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
793 		       struct netlink_ext_ack *extack)
794 {
795 	struct nh_info *nhi;
796 	bool is_fdb_nh;
797 
798 	/* fib6_src is unique to a fib6_info and limits the ability to cache
799 	 * routes in fib6_nh within a nexthop that is potentially shared
800 	 * across multiple fib entries. If the config wants to use source
801 	 * routing it can not use nexthop objects. mlxsw also does not allow
802 	 * fib6_src on routes.
803 	 */
804 	if (cfg && check_src_addr(&cfg->fc_src, extack) < 0)
805 		return -EINVAL;
806 
807 	if (nh->is_group) {
808 		struct nh_group *nhg;
809 
810 		nhg = rtnl_dereference(nh->nh_grp);
811 		if (nhg->has_v4)
812 			goto no_v4_nh;
813 		is_fdb_nh = nhg->fdb_nh;
814 	} else {
815 		nhi = rtnl_dereference(nh->nh_info);
816 		if (nhi->family == AF_INET)
817 			goto no_v4_nh;
818 		is_fdb_nh = nhi->fdb_nh;
819 	}
820 
821 	if (is_fdb_nh) {
822 		NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
823 		return -EINVAL;
824 	}
825 
826 	return 0;
827 no_v4_nh:
828 	NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop");
829 	return -EINVAL;
830 }
831 EXPORT_SYMBOL_GPL(fib6_check_nexthop);
832 
833 /* if existing nexthop has ipv6 routes linked to it, need
834  * to verify this new spec works with ipv6
835  */
836 static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new,
837 			      struct netlink_ext_ack *extack)
838 {
839 	struct fib6_info *f6i;
840 
841 	if (list_empty(&old->f6i_list))
842 		return 0;
843 
844 	list_for_each_entry(f6i, &old->f6i_list, nh_list) {
845 		if (check_src_addr(&f6i->fib6_src.addr, extack) < 0)
846 			return -EINVAL;
847 	}
848 
849 	return fib6_check_nexthop(new, NULL, extack);
850 }
851 
852 static int nexthop_check_scope(struct nh_info *nhi, u8 scope,
853 			       struct netlink_ext_ack *extack)
854 {
855 	if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) {
856 		NL_SET_ERR_MSG(extack,
857 			       "Route with host scope can not have a gateway");
858 		return -EINVAL;
859 	}
860 
861 	if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) {
862 		NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop");
863 		return -EINVAL;
864 	}
865 
866 	return 0;
867 }
868 
869 /* Invoked by fib add code to verify nexthop by id is ok with
870  * config for prefix; parts of fib_check_nh not done when nexthop
871  * object is used.
872  */
873 int fib_check_nexthop(struct nexthop *nh, u8 scope,
874 		      struct netlink_ext_ack *extack)
875 {
876 	struct nh_info *nhi;
877 	int err = 0;
878 
879 	if (nh->is_group) {
880 		struct nh_group *nhg;
881 
882 		nhg = rtnl_dereference(nh->nh_grp);
883 		if (nhg->fdb_nh) {
884 			NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
885 			err = -EINVAL;
886 			goto out;
887 		}
888 
889 		if (scope == RT_SCOPE_HOST) {
890 			NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops");
891 			err = -EINVAL;
892 			goto out;
893 		}
894 
895 		/* all nexthops in a group have the same scope */
896 		nhi = rtnl_dereference(nhg->nh_entries[0].nh->nh_info);
897 		err = nexthop_check_scope(nhi, scope, extack);
898 	} else {
899 		nhi = rtnl_dereference(nh->nh_info);
900 		if (nhi->fdb_nh) {
901 			NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
902 			err = -EINVAL;
903 			goto out;
904 		}
905 		err = nexthop_check_scope(nhi, scope, extack);
906 	}
907 
908 out:
909 	return err;
910 }
911 
912 static int fib_check_nh_list(struct nexthop *old, struct nexthop *new,
913 			     struct netlink_ext_ack *extack)
914 {
915 	struct fib_info *fi;
916 
917 	list_for_each_entry(fi, &old->fi_list, nh_list) {
918 		int err;
919 
920 		err = fib_check_nexthop(new, fi->fib_scope, extack);
921 		if (err)
922 			return err;
923 	}
924 	return 0;
925 }
926 
927 static void nh_group_rebalance(struct nh_group *nhg)
928 {
929 	int total = 0;
930 	int w = 0;
931 	int i;
932 
933 	for (i = 0; i < nhg->num_nh; ++i)
934 		total += nhg->nh_entries[i].weight;
935 
936 	for (i = 0; i < nhg->num_nh; ++i) {
937 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
938 		int upper_bound;
939 
940 		w += nhge->weight;
941 		upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1;
942 		atomic_set(&nhge->mpath.upper_bound, upper_bound);
943 	}
944 }
945 
946 static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge,
947 				struct nl_info *nlinfo)
948 {
949 	struct nh_grp_entry *nhges, *new_nhges;
950 	struct nexthop *nhp = nhge->nh_parent;
951 	struct netlink_ext_ack extack;
952 	struct nexthop *nh = nhge->nh;
953 	struct nh_group *nhg, *newg;
954 	int i, j, err;
955 
956 	WARN_ON(!nh);
957 
958 	nhg = rtnl_dereference(nhp->nh_grp);
959 	newg = nhg->spare;
960 
961 	/* last entry, keep it visible and remove the parent */
962 	if (nhg->num_nh == 1) {
963 		remove_nexthop(net, nhp, nlinfo);
964 		return;
965 	}
966 
967 	newg->has_v4 = false;
968 	newg->mpath = nhg->mpath;
969 	newg->fdb_nh = nhg->fdb_nh;
970 	newg->num_nh = nhg->num_nh;
971 
972 	/* copy old entries to new except the one getting removed */
973 	nhges = nhg->nh_entries;
974 	new_nhges = newg->nh_entries;
975 	for (i = 0, j = 0; i < nhg->num_nh; ++i) {
976 		struct nh_info *nhi;
977 
978 		/* current nexthop getting removed */
979 		if (nhg->nh_entries[i].nh == nh) {
980 			newg->num_nh--;
981 			continue;
982 		}
983 
984 		nhi = rtnl_dereference(nhges[i].nh->nh_info);
985 		if (nhi->family == AF_INET)
986 			newg->has_v4 = true;
987 
988 		list_del(&nhges[i].nh_list);
989 		new_nhges[j].nh_parent = nhges[i].nh_parent;
990 		new_nhges[j].nh = nhges[i].nh;
991 		new_nhges[j].weight = nhges[i].weight;
992 		list_add(&new_nhges[j].nh_list, &new_nhges[j].nh->grp_list);
993 		j++;
994 	}
995 
996 	nh_group_rebalance(newg);
997 	rcu_assign_pointer(nhp->nh_grp, newg);
998 
999 	list_del(&nhge->nh_list);
1000 	nexthop_put(nhge->nh);
1001 
1002 	err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp, &extack);
1003 	if (err)
1004 		pr_err("%s\n", extack._msg);
1005 
1006 	if (nlinfo)
1007 		nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo);
1008 }
1009 
1010 static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh,
1011 				       struct nl_info *nlinfo)
1012 {
1013 	struct nh_grp_entry *nhge, *tmp;
1014 
1015 	list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list)
1016 		remove_nh_grp_entry(net, nhge, nlinfo);
1017 
1018 	/* make sure all see the newly published array before releasing rtnl */
1019 	synchronize_net();
1020 }
1021 
1022 static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo)
1023 {
1024 	struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
1025 	int i, num_nh = nhg->num_nh;
1026 
1027 	for (i = 0; i < num_nh; ++i) {
1028 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
1029 
1030 		if (WARN_ON(!nhge->nh))
1031 			continue;
1032 
1033 		list_del_init(&nhge->nh_list);
1034 	}
1035 }
1036 
1037 /* not called for nexthop replace */
1038 static void __remove_nexthop_fib(struct net *net, struct nexthop *nh)
1039 {
1040 	struct fib6_info *f6i, *tmp;
1041 	bool do_flush = false;
1042 	struct fib_info *fi;
1043 
1044 	list_for_each_entry(fi, &nh->fi_list, nh_list) {
1045 		fi->fib_flags |= RTNH_F_DEAD;
1046 		do_flush = true;
1047 	}
1048 	if (do_flush)
1049 		fib_flush(net);
1050 
1051 	/* ip6_del_rt removes the entry from this list hence the _safe */
1052 	list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) {
1053 		/* __ip6_del_rt does a release, so do a hold here */
1054 		fib6_info_hold(f6i);
1055 		ipv6_stub->ip6_del_rt(net, f6i,
1056 				      !net->ipv4.sysctl_nexthop_compat_mode);
1057 	}
1058 }
1059 
1060 static void __remove_nexthop(struct net *net, struct nexthop *nh,
1061 			     struct nl_info *nlinfo)
1062 {
1063 	__remove_nexthop_fib(net, nh);
1064 
1065 	if (nh->is_group) {
1066 		remove_nexthop_group(nh, nlinfo);
1067 	} else {
1068 		struct nh_info *nhi;
1069 
1070 		nhi = rtnl_dereference(nh->nh_info);
1071 		if (nhi->fib_nhc.nhc_dev)
1072 			hlist_del(&nhi->dev_hash);
1073 
1074 		remove_nexthop_from_groups(net, nh, nlinfo);
1075 	}
1076 }
1077 
1078 static void remove_nexthop(struct net *net, struct nexthop *nh,
1079 			   struct nl_info *nlinfo)
1080 {
1081 	call_nexthop_notifiers(net, NEXTHOP_EVENT_DEL, nh, NULL);
1082 
1083 	/* remove from the tree */
1084 	rb_erase(&nh->rb_node, &net->nexthop.rb_root);
1085 
1086 	if (nlinfo)
1087 		nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo);
1088 
1089 	__remove_nexthop(net, nh, nlinfo);
1090 	nh_base_seq_inc(net);
1091 
1092 	nexthop_put(nh);
1093 }
1094 
1095 /* if any FIB entries reference this nexthop, any dst entries
1096  * need to be regenerated
1097  */
1098 static void nh_rt_cache_flush(struct net *net, struct nexthop *nh)
1099 {
1100 	struct fib6_info *f6i;
1101 
1102 	if (!list_empty(&nh->fi_list))
1103 		rt_cache_flush(net);
1104 
1105 	list_for_each_entry(f6i, &nh->f6i_list, nh_list)
1106 		ipv6_stub->fib6_update_sernum(net, f6i);
1107 }
1108 
1109 static int replace_nexthop_grp(struct net *net, struct nexthop *old,
1110 			       struct nexthop *new,
1111 			       struct netlink_ext_ack *extack)
1112 {
1113 	struct nh_group *oldg, *newg;
1114 	int i, err;
1115 
1116 	if (!new->is_group) {
1117 		NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop.");
1118 		return -EINVAL;
1119 	}
1120 
1121 	err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack);
1122 	if (err)
1123 		return err;
1124 
1125 	oldg = rtnl_dereference(old->nh_grp);
1126 	newg = rtnl_dereference(new->nh_grp);
1127 
1128 	/* update parents - used by nexthop code for cleanup */
1129 	for (i = 0; i < newg->num_nh; i++)
1130 		newg->nh_entries[i].nh_parent = old;
1131 
1132 	rcu_assign_pointer(old->nh_grp, newg);
1133 
1134 	for (i = 0; i < oldg->num_nh; i++)
1135 		oldg->nh_entries[i].nh_parent = new;
1136 
1137 	rcu_assign_pointer(new->nh_grp, oldg);
1138 
1139 	return 0;
1140 }
1141 
1142 static void nh_group_v4_update(struct nh_group *nhg)
1143 {
1144 	struct nh_grp_entry *nhges;
1145 	bool has_v4 = false;
1146 	int i;
1147 
1148 	nhges = nhg->nh_entries;
1149 	for (i = 0; i < nhg->num_nh; i++) {
1150 		struct nh_info *nhi;
1151 
1152 		nhi = rtnl_dereference(nhges[i].nh->nh_info);
1153 		if (nhi->family == AF_INET)
1154 			has_v4 = true;
1155 	}
1156 	nhg->has_v4 = has_v4;
1157 }
1158 
1159 static int replace_nexthop_single(struct net *net, struct nexthop *old,
1160 				  struct nexthop *new,
1161 				  struct netlink_ext_ack *extack)
1162 {
1163 	u8 old_protocol, old_nh_flags;
1164 	struct nh_info *oldi, *newi;
1165 	struct nh_grp_entry *nhge;
1166 	int err;
1167 
1168 	if (new->is_group) {
1169 		NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
1170 		return -EINVAL;
1171 	}
1172 
1173 	err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack);
1174 	if (err)
1175 		return err;
1176 
1177 	/* Hardware flags were set on 'old' as 'new' is not in the red-black
1178 	 * tree. Therefore, inherit the flags from 'old' to 'new'.
1179 	 */
1180 	new->nh_flags |= old->nh_flags & (RTNH_F_OFFLOAD | RTNH_F_TRAP);
1181 
1182 	oldi = rtnl_dereference(old->nh_info);
1183 	newi = rtnl_dereference(new->nh_info);
1184 
1185 	newi->nh_parent = old;
1186 	oldi->nh_parent = new;
1187 
1188 	old_protocol = old->protocol;
1189 	old_nh_flags = old->nh_flags;
1190 
1191 	old->protocol = new->protocol;
1192 	old->nh_flags = new->nh_flags;
1193 
1194 	rcu_assign_pointer(old->nh_info, newi);
1195 	rcu_assign_pointer(new->nh_info, oldi);
1196 
1197 	/* Send a replace notification for all the groups using the nexthop. */
1198 	list_for_each_entry(nhge, &old->grp_list, nh_list) {
1199 		struct nexthop *nhp = nhge->nh_parent;
1200 
1201 		err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp,
1202 					     extack);
1203 		if (err)
1204 			goto err_notify;
1205 	}
1206 
1207 	/* When replacing an IPv4 nexthop with an IPv6 nexthop, potentially
1208 	 * update IPv4 indication in all the groups using the nexthop.
1209 	 */
1210 	if (oldi->family == AF_INET && newi->family == AF_INET6) {
1211 		list_for_each_entry(nhge, &old->grp_list, nh_list) {
1212 			struct nexthop *nhp = nhge->nh_parent;
1213 			struct nh_group *nhg;
1214 
1215 			nhg = rtnl_dereference(nhp->nh_grp);
1216 			nh_group_v4_update(nhg);
1217 		}
1218 	}
1219 
1220 	return 0;
1221 
1222 err_notify:
1223 	rcu_assign_pointer(new->nh_info, newi);
1224 	rcu_assign_pointer(old->nh_info, oldi);
1225 	old->nh_flags = old_nh_flags;
1226 	old->protocol = old_protocol;
1227 	oldi->nh_parent = old;
1228 	newi->nh_parent = new;
1229 	list_for_each_entry_continue_reverse(nhge, &old->grp_list, nh_list) {
1230 		struct nexthop *nhp = nhge->nh_parent;
1231 
1232 		call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp, extack);
1233 	}
1234 	call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, old, extack);
1235 	return err;
1236 }
1237 
1238 static void __nexthop_replace_notify(struct net *net, struct nexthop *nh,
1239 				     struct nl_info *info)
1240 {
1241 	struct fib6_info *f6i;
1242 
1243 	if (!list_empty(&nh->fi_list)) {
1244 		struct fib_info *fi;
1245 
1246 		/* expectation is a few fib_info per nexthop and then
1247 		 * a lot of routes per fib_info. So mark the fib_info
1248 		 * and then walk the fib tables once
1249 		 */
1250 		list_for_each_entry(fi, &nh->fi_list, nh_list)
1251 			fi->nh_updated = true;
1252 
1253 		fib_info_notify_update(net, info);
1254 
1255 		list_for_each_entry(fi, &nh->fi_list, nh_list)
1256 			fi->nh_updated = false;
1257 	}
1258 
1259 	list_for_each_entry(f6i, &nh->f6i_list, nh_list)
1260 		ipv6_stub->fib6_rt_update(net, f6i, info);
1261 }
1262 
1263 /* send RTM_NEWROUTE with REPLACE flag set for all FIB entries
1264  * linked to this nexthop and for all groups that the nexthop
1265  * is a member of
1266  */
1267 static void nexthop_replace_notify(struct net *net, struct nexthop *nh,
1268 				   struct nl_info *info)
1269 {
1270 	struct nh_grp_entry *nhge;
1271 
1272 	__nexthop_replace_notify(net, nh, info);
1273 
1274 	list_for_each_entry(nhge, &nh->grp_list, nh_list)
1275 		__nexthop_replace_notify(net, nhge->nh_parent, info);
1276 }
1277 
1278 static int replace_nexthop(struct net *net, struct nexthop *old,
1279 			   struct nexthop *new, struct netlink_ext_ack *extack)
1280 {
1281 	bool new_is_reject = false;
1282 	struct nh_grp_entry *nhge;
1283 	int err;
1284 
1285 	/* check that existing FIB entries are ok with the
1286 	 * new nexthop definition
1287 	 */
1288 	err = fib_check_nh_list(old, new, extack);
1289 	if (err)
1290 		return err;
1291 
1292 	err = fib6_check_nh_list(old, new, extack);
1293 	if (err)
1294 		return err;
1295 
1296 	if (!new->is_group) {
1297 		struct nh_info *nhi = rtnl_dereference(new->nh_info);
1298 
1299 		new_is_reject = nhi->reject_nh;
1300 	}
1301 
1302 	list_for_each_entry(nhge, &old->grp_list, nh_list) {
1303 		/* if new nexthop is a blackhole, any groups using this
1304 		 * nexthop cannot have more than 1 path
1305 		 */
1306 		if (new_is_reject &&
1307 		    nexthop_num_path(nhge->nh_parent) > 1) {
1308 			NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path");
1309 			return -EINVAL;
1310 		}
1311 
1312 		err = fib_check_nh_list(nhge->nh_parent, new, extack);
1313 		if (err)
1314 			return err;
1315 
1316 		err = fib6_check_nh_list(nhge->nh_parent, new, extack);
1317 		if (err)
1318 			return err;
1319 	}
1320 
1321 	if (old->is_group)
1322 		err = replace_nexthop_grp(net, old, new, extack);
1323 	else
1324 		err = replace_nexthop_single(net, old, new, extack);
1325 
1326 	if (!err) {
1327 		nh_rt_cache_flush(net, old);
1328 
1329 		__remove_nexthop(net, new, NULL);
1330 		nexthop_put(new);
1331 	}
1332 
1333 	return err;
1334 }
1335 
1336 /* called with rtnl_lock held */
1337 static int insert_nexthop(struct net *net, struct nexthop *new_nh,
1338 			  struct nh_config *cfg, struct netlink_ext_ack *extack)
1339 {
1340 	struct rb_node **pp, *parent = NULL, *next;
1341 	struct rb_root *root = &net->nexthop.rb_root;
1342 	bool replace = !!(cfg->nlflags & NLM_F_REPLACE);
1343 	bool create = !!(cfg->nlflags & NLM_F_CREATE);
1344 	u32 new_id = new_nh->id;
1345 	int replace_notify = 0;
1346 	int rc = -EEXIST;
1347 
1348 	pp = &root->rb_node;
1349 	while (1) {
1350 		struct nexthop *nh;
1351 
1352 		next = *pp;
1353 		if (!next)
1354 			break;
1355 
1356 		parent = next;
1357 
1358 		nh = rb_entry(parent, struct nexthop, rb_node);
1359 		if (new_id < nh->id) {
1360 			pp = &next->rb_left;
1361 		} else if (new_id > nh->id) {
1362 			pp = &next->rb_right;
1363 		} else if (replace) {
1364 			rc = replace_nexthop(net, nh, new_nh, extack);
1365 			if (!rc) {
1366 				new_nh = nh; /* send notification with old nh */
1367 				replace_notify = 1;
1368 			}
1369 			goto out;
1370 		} else {
1371 			/* id already exists and not a replace */
1372 			goto out;
1373 		}
1374 	}
1375 
1376 	if (replace && !create) {
1377 		NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists");
1378 		rc = -ENOENT;
1379 		goto out;
1380 	}
1381 
1382 	rb_link_node_rcu(&new_nh->rb_node, parent, pp);
1383 	rb_insert_color(&new_nh->rb_node, root);
1384 
1385 	rc = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new_nh, extack);
1386 	if (rc)
1387 		rb_erase(&new_nh->rb_node, &net->nexthop.rb_root);
1388 
1389 out:
1390 	if (!rc) {
1391 		nh_base_seq_inc(net);
1392 		nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo);
1393 		if (replace_notify && net->ipv4.sysctl_nexthop_compat_mode)
1394 			nexthop_replace_notify(net, new_nh, &cfg->nlinfo);
1395 	}
1396 
1397 	return rc;
1398 }
1399 
1400 /* rtnl */
1401 /* remove all nexthops tied to a device being deleted */
1402 static void nexthop_flush_dev(struct net_device *dev, unsigned long event)
1403 {
1404 	unsigned int hash = nh_dev_hashfn(dev->ifindex);
1405 	struct net *net = dev_net(dev);
1406 	struct hlist_head *head = &net->nexthop.devhash[hash];
1407 	struct hlist_node *n;
1408 	struct nh_info *nhi;
1409 
1410 	hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1411 		if (nhi->fib_nhc.nhc_dev != dev)
1412 			continue;
1413 
1414 		if (nhi->reject_nh &&
1415 		    (event == NETDEV_DOWN || event == NETDEV_CHANGE))
1416 			continue;
1417 
1418 		remove_nexthop(net, nhi->nh_parent, NULL);
1419 	}
1420 }
1421 
1422 /* rtnl; called when net namespace is deleted */
1423 static void flush_all_nexthops(struct net *net)
1424 {
1425 	struct rb_root *root = &net->nexthop.rb_root;
1426 	struct rb_node *node;
1427 	struct nexthop *nh;
1428 
1429 	while ((node = rb_first(root))) {
1430 		nh = rb_entry(node, struct nexthop, rb_node);
1431 		remove_nexthop(net, nh, NULL);
1432 		cond_resched();
1433 	}
1434 }
1435 
1436 static struct nexthop *nexthop_create_group(struct net *net,
1437 					    struct nh_config *cfg)
1438 {
1439 	struct nlattr *grps_attr = cfg->nh_grp;
1440 	struct nexthop_grp *entry = nla_data(grps_attr);
1441 	u16 num_nh = nla_len(grps_attr) / sizeof(*entry);
1442 	struct nh_group *nhg;
1443 	struct nexthop *nh;
1444 	int i;
1445 
1446 	if (WARN_ON(!num_nh))
1447 		return ERR_PTR(-EINVAL);
1448 
1449 	nh = nexthop_alloc();
1450 	if (!nh)
1451 		return ERR_PTR(-ENOMEM);
1452 
1453 	nh->is_group = 1;
1454 
1455 	nhg = nexthop_grp_alloc(num_nh);
1456 	if (!nhg) {
1457 		kfree(nh);
1458 		return ERR_PTR(-ENOMEM);
1459 	}
1460 
1461 	/* spare group used for removals */
1462 	nhg->spare = nexthop_grp_alloc(num_nh);
1463 	if (!nhg->spare) {
1464 		kfree(nhg);
1465 		kfree(nh);
1466 		return ERR_PTR(-ENOMEM);
1467 	}
1468 	nhg->spare->spare = nhg;
1469 
1470 	for (i = 0; i < nhg->num_nh; ++i) {
1471 		struct nexthop *nhe;
1472 		struct nh_info *nhi;
1473 
1474 		nhe = nexthop_find_by_id(net, entry[i].id);
1475 		if (!nexthop_get(nhe))
1476 			goto out_no_nh;
1477 
1478 		nhi = rtnl_dereference(nhe->nh_info);
1479 		if (nhi->family == AF_INET)
1480 			nhg->has_v4 = true;
1481 
1482 		nhg->nh_entries[i].nh = nhe;
1483 		nhg->nh_entries[i].weight = entry[i].weight + 1;
1484 		list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);
1485 		nhg->nh_entries[i].nh_parent = nh;
1486 	}
1487 
1488 	if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH)
1489 		nhg->mpath = 1;
1490 
1491 	WARN_ON_ONCE(nhg->mpath != 1);
1492 
1493 	if (nhg->mpath)
1494 		nh_group_rebalance(nhg);
1495 
1496 	if (cfg->nh_fdb)
1497 		nhg->fdb_nh = 1;
1498 
1499 	rcu_assign_pointer(nh->nh_grp, nhg);
1500 
1501 	return nh;
1502 
1503 out_no_nh:
1504 	for (i--; i >= 0; --i) {
1505 		list_del(&nhg->nh_entries[i].nh_list);
1506 		nexthop_put(nhg->nh_entries[i].nh);
1507 	}
1508 
1509 	kfree(nhg->spare);
1510 	kfree(nhg);
1511 	kfree(nh);
1512 
1513 	return ERR_PTR(-ENOENT);
1514 }
1515 
1516 static int nh_create_ipv4(struct net *net, struct nexthop *nh,
1517 			  struct nh_info *nhi, struct nh_config *cfg,
1518 			  struct netlink_ext_ack *extack)
1519 {
1520 	struct fib_nh *fib_nh = &nhi->fib_nh;
1521 	struct fib_config fib_cfg = {
1522 		.fc_oif   = cfg->nh_ifindex,
1523 		.fc_gw4   = cfg->gw.ipv4,
1524 		.fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0,
1525 		.fc_flags = cfg->nh_flags,
1526 		.fc_encap = cfg->nh_encap,
1527 		.fc_encap_type = cfg->nh_encap_type,
1528 	};
1529 	u32 tb_id = (cfg->dev ? l3mdev_fib_table(cfg->dev) : RT_TABLE_MAIN);
1530 	int err;
1531 
1532 	err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack);
1533 	if (err) {
1534 		fib_nh_release(net, fib_nh);
1535 		goto out;
1536 	}
1537 
1538 	if (nhi->fdb_nh)
1539 		goto out;
1540 
1541 	/* sets nh_dev if successful */
1542 	err = fib_check_nh(net, fib_nh, tb_id, 0, extack);
1543 	if (!err) {
1544 		nh->nh_flags = fib_nh->fib_nh_flags;
1545 		fib_info_update_nhc_saddr(net, &fib_nh->nh_common,
1546 					  fib_nh->fib_nh_scope);
1547 	} else {
1548 		fib_nh_release(net, fib_nh);
1549 	}
1550 out:
1551 	return err;
1552 }
1553 
1554 static int nh_create_ipv6(struct net *net,  struct nexthop *nh,
1555 			  struct nh_info *nhi, struct nh_config *cfg,
1556 			  struct netlink_ext_ack *extack)
1557 {
1558 	struct fib6_nh *fib6_nh = &nhi->fib6_nh;
1559 	struct fib6_config fib6_cfg = {
1560 		.fc_table = l3mdev_fib_table(cfg->dev),
1561 		.fc_ifindex = cfg->nh_ifindex,
1562 		.fc_gateway = cfg->gw.ipv6,
1563 		.fc_flags = cfg->nh_flags,
1564 		.fc_encap = cfg->nh_encap,
1565 		.fc_encap_type = cfg->nh_encap_type,
1566 		.fc_is_fdb = cfg->nh_fdb,
1567 	};
1568 	int err;
1569 
1570 	if (!ipv6_addr_any(&cfg->gw.ipv6))
1571 		fib6_cfg.fc_flags |= RTF_GATEWAY;
1572 
1573 	/* sets nh_dev if successful */
1574 	err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL,
1575 				      extack);
1576 	if (err)
1577 		ipv6_stub->fib6_nh_release(fib6_nh);
1578 	else
1579 		nh->nh_flags = fib6_nh->fib_nh_flags;
1580 
1581 	return err;
1582 }
1583 
1584 static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg,
1585 				      struct netlink_ext_ack *extack)
1586 {
1587 	struct nh_info *nhi;
1588 	struct nexthop *nh;
1589 	int err = 0;
1590 
1591 	nh = nexthop_alloc();
1592 	if (!nh)
1593 		return ERR_PTR(-ENOMEM);
1594 
1595 	nhi = kzalloc(sizeof(*nhi), GFP_KERNEL);
1596 	if (!nhi) {
1597 		kfree(nh);
1598 		return ERR_PTR(-ENOMEM);
1599 	}
1600 
1601 	nh->nh_flags = cfg->nh_flags;
1602 	nh->net = net;
1603 
1604 	nhi->nh_parent = nh;
1605 	nhi->family = cfg->nh_family;
1606 	nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK;
1607 
1608 	if (cfg->nh_fdb)
1609 		nhi->fdb_nh = 1;
1610 
1611 	if (cfg->nh_blackhole) {
1612 		nhi->reject_nh = 1;
1613 		cfg->nh_ifindex = net->loopback_dev->ifindex;
1614 	}
1615 
1616 	switch (cfg->nh_family) {
1617 	case AF_INET:
1618 		err = nh_create_ipv4(net, nh, nhi, cfg, extack);
1619 		break;
1620 	case AF_INET6:
1621 		err = nh_create_ipv6(net, nh, nhi, cfg, extack);
1622 		break;
1623 	}
1624 
1625 	if (err) {
1626 		kfree(nhi);
1627 		kfree(nh);
1628 		return ERR_PTR(err);
1629 	}
1630 
1631 	/* add the entry to the device based hash */
1632 	if (!nhi->fdb_nh)
1633 		nexthop_devhash_add(net, nhi);
1634 
1635 	rcu_assign_pointer(nh->nh_info, nhi);
1636 
1637 	return nh;
1638 }
1639 
1640 /* called with rtnl lock held */
1641 static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg,
1642 				   struct netlink_ext_ack *extack)
1643 {
1644 	struct nexthop *nh;
1645 	int err;
1646 
1647 	if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) {
1648 		NL_SET_ERR_MSG(extack, "Replace requires nexthop id");
1649 		return ERR_PTR(-EINVAL);
1650 	}
1651 
1652 	if (!cfg->nh_id) {
1653 		cfg->nh_id = nh_find_unused_id(net);
1654 		if (!cfg->nh_id) {
1655 			NL_SET_ERR_MSG(extack, "No unused id");
1656 			return ERR_PTR(-EINVAL);
1657 		}
1658 	}
1659 
1660 	if (cfg->nh_grp)
1661 		nh = nexthop_create_group(net, cfg);
1662 	else
1663 		nh = nexthop_create(net, cfg, extack);
1664 
1665 	if (IS_ERR(nh))
1666 		return nh;
1667 
1668 	refcount_set(&nh->refcnt, 1);
1669 	nh->id = cfg->nh_id;
1670 	nh->protocol = cfg->nh_protocol;
1671 	nh->net = net;
1672 
1673 	err = insert_nexthop(net, nh, cfg, extack);
1674 	if (err) {
1675 		__remove_nexthop(net, nh, NULL);
1676 		nexthop_put(nh);
1677 		nh = ERR_PTR(err);
1678 	}
1679 
1680 	return nh;
1681 }
1682 
1683 static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
1684 			    struct nlmsghdr *nlh, struct nh_config *cfg,
1685 			    struct netlink_ext_ack *extack)
1686 {
1687 	struct nhmsg *nhm = nlmsg_data(nlh);
1688 	struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_new)];
1689 	int err;
1690 
1691 	err = nlmsg_parse(nlh, sizeof(*nhm), tb,
1692 			  ARRAY_SIZE(rtm_nh_policy_new) - 1,
1693 			  rtm_nh_policy_new, extack);
1694 	if (err < 0)
1695 		return err;
1696 
1697 	err = -EINVAL;
1698 	if (nhm->resvd || nhm->nh_scope) {
1699 		NL_SET_ERR_MSG(extack, "Invalid values in ancillary header");
1700 		goto out;
1701 	}
1702 	if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) {
1703 		NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header");
1704 		goto out;
1705 	}
1706 
1707 	switch (nhm->nh_family) {
1708 	case AF_INET:
1709 	case AF_INET6:
1710 		break;
1711 	case AF_UNSPEC:
1712 		if (tb[NHA_GROUP])
1713 			break;
1714 		fallthrough;
1715 	default:
1716 		NL_SET_ERR_MSG(extack, "Invalid address family");
1717 		goto out;
1718 	}
1719 
1720 	memset(cfg, 0, sizeof(*cfg));
1721 	cfg->nlflags = nlh->nlmsg_flags;
1722 	cfg->nlinfo.portid = NETLINK_CB(skb).portid;
1723 	cfg->nlinfo.nlh = nlh;
1724 	cfg->nlinfo.nl_net = net;
1725 
1726 	cfg->nh_family = nhm->nh_family;
1727 	cfg->nh_protocol = nhm->nh_protocol;
1728 	cfg->nh_flags = nhm->nh_flags;
1729 
1730 	if (tb[NHA_ID])
1731 		cfg->nh_id = nla_get_u32(tb[NHA_ID]);
1732 
1733 	if (tb[NHA_FDB]) {
1734 		if (tb[NHA_OIF] || tb[NHA_BLACKHOLE] ||
1735 		    tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE]) {
1736 			NL_SET_ERR_MSG(extack, "Fdb attribute can not be used with encap, oif or blackhole");
1737 			goto out;
1738 		}
1739 		if (nhm->nh_flags) {
1740 			NL_SET_ERR_MSG(extack, "Unsupported nexthop flags in ancillary header");
1741 			goto out;
1742 		}
1743 		cfg->nh_fdb = nla_get_flag(tb[NHA_FDB]);
1744 	}
1745 
1746 	if (tb[NHA_GROUP]) {
1747 		if (nhm->nh_family != AF_UNSPEC) {
1748 			NL_SET_ERR_MSG(extack, "Invalid family for group");
1749 			goto out;
1750 		}
1751 		cfg->nh_grp = tb[NHA_GROUP];
1752 
1753 		cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH;
1754 		if (tb[NHA_GROUP_TYPE])
1755 			cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]);
1756 
1757 		if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) {
1758 			NL_SET_ERR_MSG(extack, "Invalid group type");
1759 			goto out;
1760 		}
1761 		err = nh_check_attr_group(net, tb, ARRAY_SIZE(tb), extack);
1762 
1763 		/* no other attributes should be set */
1764 		goto out;
1765 	}
1766 
1767 	if (tb[NHA_BLACKHOLE]) {
1768 		if (tb[NHA_GATEWAY] || tb[NHA_OIF] ||
1769 		    tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE] || tb[NHA_FDB]) {
1770 			NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway, oif, encap or fdb");
1771 			goto out;
1772 		}
1773 
1774 		cfg->nh_blackhole = 1;
1775 		err = 0;
1776 		goto out;
1777 	}
1778 
1779 	if (!cfg->nh_fdb && !tb[NHA_OIF]) {
1780 		NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole and non-fdb nexthops");
1781 		goto out;
1782 	}
1783 
1784 	if (!cfg->nh_fdb && tb[NHA_OIF]) {
1785 		cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]);
1786 		if (cfg->nh_ifindex)
1787 			cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex);
1788 
1789 		if (!cfg->dev) {
1790 			NL_SET_ERR_MSG(extack, "Invalid device index");
1791 			goto out;
1792 		} else if (!(cfg->dev->flags & IFF_UP)) {
1793 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1794 			err = -ENETDOWN;
1795 			goto out;
1796 		} else if (!netif_carrier_ok(cfg->dev)) {
1797 			NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down");
1798 			err = -ENETDOWN;
1799 			goto out;
1800 		}
1801 	}
1802 
1803 	err = -EINVAL;
1804 	if (tb[NHA_GATEWAY]) {
1805 		struct nlattr *gwa = tb[NHA_GATEWAY];
1806 
1807 		switch (cfg->nh_family) {
1808 		case AF_INET:
1809 			if (nla_len(gwa) != sizeof(u32)) {
1810 				NL_SET_ERR_MSG(extack, "Invalid gateway");
1811 				goto out;
1812 			}
1813 			cfg->gw.ipv4 = nla_get_be32(gwa);
1814 			break;
1815 		case AF_INET6:
1816 			if (nla_len(gwa) != sizeof(struct in6_addr)) {
1817 				NL_SET_ERR_MSG(extack, "Invalid gateway");
1818 				goto out;
1819 			}
1820 			cfg->gw.ipv6 = nla_get_in6_addr(gwa);
1821 			break;
1822 		default:
1823 			NL_SET_ERR_MSG(extack,
1824 				       "Unknown address family for gateway");
1825 			goto out;
1826 		}
1827 	} else {
1828 		/* device only nexthop (no gateway) */
1829 		if (cfg->nh_flags & RTNH_F_ONLINK) {
1830 			NL_SET_ERR_MSG(extack,
1831 				       "ONLINK flag can not be set for nexthop without a gateway");
1832 			goto out;
1833 		}
1834 	}
1835 
1836 	if (tb[NHA_ENCAP]) {
1837 		cfg->nh_encap = tb[NHA_ENCAP];
1838 
1839 		if (!tb[NHA_ENCAP_TYPE]) {
1840 			NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing");
1841 			goto out;
1842 		}
1843 
1844 		cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]);
1845 		err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack);
1846 		if (err < 0)
1847 			goto out;
1848 
1849 	} else if (tb[NHA_ENCAP_TYPE]) {
1850 		NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing");
1851 		goto out;
1852 	}
1853 
1854 
1855 	err = 0;
1856 out:
1857 	return err;
1858 }
1859 
1860 /* rtnl */
1861 static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1862 			   struct netlink_ext_ack *extack)
1863 {
1864 	struct net *net = sock_net(skb->sk);
1865 	struct nh_config cfg;
1866 	struct nexthop *nh;
1867 	int err;
1868 
1869 	err = rtm_to_nh_config(net, skb, nlh, &cfg, extack);
1870 	if (!err) {
1871 		nh = nexthop_add(net, &cfg, extack);
1872 		if (IS_ERR(nh))
1873 			err = PTR_ERR(nh);
1874 	}
1875 
1876 	return err;
1877 }
1878 
1879 static int __nh_valid_get_del_req(const struct nlmsghdr *nlh,
1880 				  struct nlattr **tb, u32 *id,
1881 				  struct netlink_ext_ack *extack)
1882 {
1883 	struct nhmsg *nhm = nlmsg_data(nlh);
1884 
1885 	if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1886 		NL_SET_ERR_MSG(extack, "Invalid values in header");
1887 		return -EINVAL;
1888 	}
1889 
1890 	if (!tb[NHA_ID]) {
1891 		NL_SET_ERR_MSG(extack, "Nexthop id is missing");
1892 		return -EINVAL;
1893 	}
1894 
1895 	*id = nla_get_u32(tb[NHA_ID]);
1896 	if (!(*id)) {
1897 		NL_SET_ERR_MSG(extack, "Invalid nexthop id");
1898 		return -EINVAL;
1899 	}
1900 
1901 	return 0;
1902 }
1903 
1904 static int nh_valid_get_del_req(const struct nlmsghdr *nlh, u32 *id,
1905 				struct netlink_ext_ack *extack)
1906 {
1907 	struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get)];
1908 	int err;
1909 
1910 	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
1911 			  ARRAY_SIZE(rtm_nh_policy_get) - 1,
1912 			  rtm_nh_policy_get, extack);
1913 	if (err < 0)
1914 		return err;
1915 
1916 	return __nh_valid_get_del_req(nlh, tb, id, extack);
1917 }
1918 
1919 /* rtnl */
1920 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1921 			   struct netlink_ext_ack *extack)
1922 {
1923 	struct net *net = sock_net(skb->sk);
1924 	struct nl_info nlinfo = {
1925 		.nlh = nlh,
1926 		.nl_net = net,
1927 		.portid = NETLINK_CB(skb).portid,
1928 	};
1929 	struct nexthop *nh;
1930 	int err;
1931 	u32 id;
1932 
1933 	err = nh_valid_get_del_req(nlh, &id, extack);
1934 	if (err)
1935 		return err;
1936 
1937 	nh = nexthop_find_by_id(net, id);
1938 	if (!nh)
1939 		return -ENOENT;
1940 
1941 	remove_nexthop(net, nh, &nlinfo);
1942 
1943 	return 0;
1944 }
1945 
1946 /* rtnl */
1947 static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1948 			   struct netlink_ext_ack *extack)
1949 {
1950 	struct net *net = sock_net(in_skb->sk);
1951 	struct sk_buff *skb = NULL;
1952 	struct nexthop *nh;
1953 	int err;
1954 	u32 id;
1955 
1956 	err = nh_valid_get_del_req(nlh, &id, extack);
1957 	if (err)
1958 		return err;
1959 
1960 	err = -ENOBUFS;
1961 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1962 	if (!skb)
1963 		goto out;
1964 
1965 	err = -ENOENT;
1966 	nh = nexthop_find_by_id(net, id);
1967 	if (!nh)
1968 		goto errout_free;
1969 
1970 	err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid,
1971 			   nlh->nlmsg_seq, 0);
1972 	if (err < 0) {
1973 		WARN_ON(err == -EMSGSIZE);
1974 		goto errout_free;
1975 	}
1976 
1977 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1978 out:
1979 	return err;
1980 errout_free:
1981 	kfree_skb(skb);
1982 	goto out;
1983 }
1984 
1985 struct nh_dump_filter {
1986 	int dev_idx;
1987 	int master_idx;
1988 	bool group_filter;
1989 	bool fdb_filter;
1990 };
1991 
1992 static bool nh_dump_filtered(struct nexthop *nh,
1993 			     struct nh_dump_filter *filter, u8 family)
1994 {
1995 	const struct net_device *dev;
1996 	const struct nh_info *nhi;
1997 
1998 	if (filter->group_filter && !nh->is_group)
1999 		return true;
2000 
2001 	if (!filter->dev_idx && !filter->master_idx && !family)
2002 		return false;
2003 
2004 	if (nh->is_group)
2005 		return true;
2006 
2007 	nhi = rtnl_dereference(nh->nh_info);
2008 	if (family && nhi->family != family)
2009 		return true;
2010 
2011 	dev = nhi->fib_nhc.nhc_dev;
2012 	if (filter->dev_idx && (!dev || dev->ifindex != filter->dev_idx))
2013 		return true;
2014 
2015 	if (filter->master_idx) {
2016 		struct net_device *master;
2017 
2018 		if (!dev)
2019 			return true;
2020 
2021 		master = netdev_master_upper_dev_get((struct net_device *)dev);
2022 		if (!master || master->ifindex != filter->master_idx)
2023 			return true;
2024 	}
2025 
2026 	return false;
2027 }
2028 
2029 static int __nh_valid_dump_req(const struct nlmsghdr *nlh, struct nlattr **tb,
2030 			       struct nh_dump_filter *filter,
2031 			       struct netlink_ext_ack *extack)
2032 {
2033 	struct nhmsg *nhm;
2034 	u32 idx;
2035 
2036 	if (tb[NHA_OIF]) {
2037 		idx = nla_get_u32(tb[NHA_OIF]);
2038 		if (idx > INT_MAX) {
2039 			NL_SET_ERR_MSG(extack, "Invalid device index");
2040 			return -EINVAL;
2041 		}
2042 		filter->dev_idx = idx;
2043 	}
2044 	if (tb[NHA_MASTER]) {
2045 		idx = nla_get_u32(tb[NHA_MASTER]);
2046 		if (idx > INT_MAX) {
2047 			NL_SET_ERR_MSG(extack, "Invalid master device index");
2048 			return -EINVAL;
2049 		}
2050 		filter->master_idx = idx;
2051 	}
2052 	filter->group_filter = nla_get_flag(tb[NHA_GROUPS]);
2053 	filter->fdb_filter = nla_get_flag(tb[NHA_FDB]);
2054 
2055 	nhm = nlmsg_data(nlh);
2056 	if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
2057 		NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request");
2058 		return -EINVAL;
2059 	}
2060 
2061 	return 0;
2062 }
2063 
2064 static int nh_valid_dump_req(const struct nlmsghdr *nlh,
2065 			     struct nh_dump_filter *filter,
2066 			     struct netlink_callback *cb)
2067 {
2068 	struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump)];
2069 	int err;
2070 
2071 	err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
2072 			  ARRAY_SIZE(rtm_nh_policy_dump) - 1,
2073 			  rtm_nh_policy_dump, cb->extack);
2074 	if (err < 0)
2075 		return err;
2076 
2077 	return __nh_valid_dump_req(nlh, tb, filter, cb->extack);
2078 }
2079 
2080 struct rtm_dump_nh_ctx {
2081 	u32 idx;
2082 };
2083 
2084 static struct rtm_dump_nh_ctx *
2085 rtm_dump_nh_ctx(struct netlink_callback *cb)
2086 {
2087 	struct rtm_dump_nh_ctx *ctx = (void *)cb->ctx;
2088 
2089 	BUILD_BUG_ON(sizeof(*ctx) > sizeof(cb->ctx));
2090 	return ctx;
2091 }
2092 
2093 static int rtm_dump_walk_nexthops(struct sk_buff *skb,
2094 				  struct netlink_callback *cb,
2095 				  struct rb_root *root,
2096 				  struct rtm_dump_nh_ctx *ctx,
2097 				  int (*nh_cb)(struct sk_buff *skb,
2098 					       struct netlink_callback *cb,
2099 					       struct nexthop *nh, void *data),
2100 				  void *data)
2101 {
2102 	struct rb_node *node;
2103 	int idx = 0, s_idx;
2104 	int err;
2105 
2106 	s_idx = ctx->idx;
2107 	for (node = rb_first(root); node; node = rb_next(node)) {
2108 		struct nexthop *nh;
2109 
2110 		if (idx < s_idx)
2111 			goto cont;
2112 
2113 		nh = rb_entry(node, struct nexthop, rb_node);
2114 		ctx->idx = idx;
2115 		err = nh_cb(skb, cb, nh, data);
2116 		if (err)
2117 			return err;
2118 cont:
2119 		idx++;
2120 	}
2121 
2122 	ctx->idx = idx;
2123 	return 0;
2124 }
2125 
2126 static int rtm_dump_nexthop_cb(struct sk_buff *skb, struct netlink_callback *cb,
2127 			       struct nexthop *nh, void *data)
2128 {
2129 	struct nhmsg *nhm = nlmsg_data(cb->nlh);
2130 	struct nh_dump_filter *filter = data;
2131 
2132 	if (nh_dump_filtered(nh, filter, nhm->nh_family))
2133 		return 0;
2134 
2135 	return nh_fill_node(skb, nh, RTM_NEWNEXTHOP,
2136 			    NETLINK_CB(cb->skb).portid,
2137 			    cb->nlh->nlmsg_seq, NLM_F_MULTI);
2138 }
2139 
2140 /* rtnl */
2141 static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb)
2142 {
2143 	struct rtm_dump_nh_ctx *ctx = rtm_dump_nh_ctx(cb);
2144 	struct net *net = sock_net(skb->sk);
2145 	struct rb_root *root = &net->nexthop.rb_root;
2146 	struct nh_dump_filter filter = {};
2147 	int err;
2148 
2149 	err = nh_valid_dump_req(cb->nlh, &filter, cb);
2150 	if (err < 0)
2151 		return err;
2152 
2153 	err = rtm_dump_walk_nexthops(skb, cb, root, ctx,
2154 				     &rtm_dump_nexthop_cb, &filter);
2155 	if (err < 0) {
2156 		if (likely(skb->len))
2157 			goto out;
2158 		goto out_err;
2159 	}
2160 
2161 out:
2162 	err = skb->len;
2163 out_err:
2164 	cb->seq = net->nexthop.seq;
2165 	nl_dump_check_consistent(cb, nlmsg_hdr(skb));
2166 	return err;
2167 }
2168 
2169 static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu)
2170 {
2171 	unsigned int hash = nh_dev_hashfn(dev->ifindex);
2172 	struct net *net = dev_net(dev);
2173 	struct hlist_head *head = &net->nexthop.devhash[hash];
2174 	struct hlist_node *n;
2175 	struct nh_info *nhi;
2176 
2177 	hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
2178 		if (nhi->fib_nhc.nhc_dev == dev) {
2179 			if (nhi->family == AF_INET)
2180 				fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu,
2181 						   orig_mtu);
2182 		}
2183 	}
2184 }
2185 
2186 /* rtnl */
2187 static int nh_netdev_event(struct notifier_block *this,
2188 			   unsigned long event, void *ptr)
2189 {
2190 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2191 	struct netdev_notifier_info_ext *info_ext;
2192 
2193 	switch (event) {
2194 	case NETDEV_DOWN:
2195 	case NETDEV_UNREGISTER:
2196 		nexthop_flush_dev(dev, event);
2197 		break;
2198 	case NETDEV_CHANGE:
2199 		if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP)))
2200 			nexthop_flush_dev(dev, event);
2201 		break;
2202 	case NETDEV_CHANGEMTU:
2203 		info_ext = ptr;
2204 		nexthop_sync_mtu(dev, info_ext->ext.mtu);
2205 		rt_cache_flush(dev_net(dev));
2206 		break;
2207 	}
2208 	return NOTIFY_DONE;
2209 }
2210 
2211 static struct notifier_block nh_netdev_notifier = {
2212 	.notifier_call = nh_netdev_event,
2213 };
2214 
2215 static int nexthops_dump(struct net *net, struct notifier_block *nb,
2216 			 struct netlink_ext_ack *extack)
2217 {
2218 	struct rb_root *root = &net->nexthop.rb_root;
2219 	struct rb_node *node;
2220 	int err = 0;
2221 
2222 	for (node = rb_first(root); node; node = rb_next(node)) {
2223 		struct nexthop *nh;
2224 
2225 		nh = rb_entry(node, struct nexthop, rb_node);
2226 		err = call_nexthop_notifier(nb, net, NEXTHOP_EVENT_REPLACE, nh,
2227 					    extack);
2228 		if (err)
2229 			break;
2230 	}
2231 
2232 	return err;
2233 }
2234 
2235 int register_nexthop_notifier(struct net *net, struct notifier_block *nb,
2236 			      struct netlink_ext_ack *extack)
2237 {
2238 	int err;
2239 
2240 	rtnl_lock();
2241 	err = nexthops_dump(net, nb, extack);
2242 	if (err)
2243 		goto unlock;
2244 	err = blocking_notifier_chain_register(&net->nexthop.notifier_chain,
2245 					       nb);
2246 unlock:
2247 	rtnl_unlock();
2248 	return err;
2249 }
2250 EXPORT_SYMBOL(register_nexthop_notifier);
2251 
2252 int unregister_nexthop_notifier(struct net *net, struct notifier_block *nb)
2253 {
2254 	return blocking_notifier_chain_unregister(&net->nexthop.notifier_chain,
2255 						  nb);
2256 }
2257 EXPORT_SYMBOL(unregister_nexthop_notifier);
2258 
2259 void nexthop_set_hw_flags(struct net *net, u32 id, bool offload, bool trap)
2260 {
2261 	struct nexthop *nexthop;
2262 
2263 	rcu_read_lock();
2264 
2265 	nexthop = nexthop_find_by_id(net, id);
2266 	if (!nexthop)
2267 		goto out;
2268 
2269 	nexthop->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP);
2270 	if (offload)
2271 		nexthop->nh_flags |= RTNH_F_OFFLOAD;
2272 	if (trap)
2273 		nexthop->nh_flags |= RTNH_F_TRAP;
2274 
2275 out:
2276 	rcu_read_unlock();
2277 }
2278 EXPORT_SYMBOL(nexthop_set_hw_flags);
2279 
2280 static void __net_exit nexthop_net_exit(struct net *net)
2281 {
2282 	rtnl_lock();
2283 	flush_all_nexthops(net);
2284 	rtnl_unlock();
2285 	kfree(net->nexthop.devhash);
2286 }
2287 
2288 static int __net_init nexthop_net_init(struct net *net)
2289 {
2290 	size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE;
2291 
2292 	net->nexthop.rb_root = RB_ROOT;
2293 	net->nexthop.devhash = kzalloc(sz, GFP_KERNEL);
2294 	if (!net->nexthop.devhash)
2295 		return -ENOMEM;
2296 	BLOCKING_INIT_NOTIFIER_HEAD(&net->nexthop.notifier_chain);
2297 
2298 	return 0;
2299 }
2300 
2301 static struct pernet_operations nexthop_net_ops = {
2302 	.init = nexthop_net_init,
2303 	.exit = nexthop_net_exit,
2304 };
2305 
2306 static int __init nexthop_init(void)
2307 {
2308 	register_pernet_subsys(&nexthop_net_ops);
2309 
2310 	register_netdevice_notifier(&nh_netdev_notifier);
2311 
2312 	rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
2313 	rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0);
2314 	rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop,
2315 		      rtm_dump_nexthop, 0);
2316 
2317 	rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
2318 	rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
2319 
2320 	rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
2321 	rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
2322 
2323 	return 0;
2324 }
2325 subsys_initcall(nexthop_init);
2326