xref: /openbmc/linux/net/ipv4/nexthop.c (revision c1cf3d89)
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[NHA_MAX + 1] = {
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_GROUPS]		= { .type = NLA_FLAG },
35 	[NHA_MASTER]		= { .type = NLA_U32 },
36 	[NHA_FDB]		= { .type = NLA_FLAG },
37 };
38 
39 static bool nexthop_notifiers_is_empty(struct net *net)
40 {
41 	return !net->nexthop.notifier_chain.head;
42 }
43 
44 static void
45 __nh_notifier_single_info_init(struct nh_notifier_single_info *nh_info,
46 			       const struct nexthop *nh)
47 {
48 	struct nh_info *nhi = rtnl_dereference(nh->nh_info);
49 
50 	nh_info->dev = nhi->fib_nhc.nhc_dev;
51 	nh_info->gw_family = nhi->fib_nhc.nhc_gw_family;
52 	if (nh_info->gw_family == AF_INET)
53 		nh_info->ipv4 = nhi->fib_nhc.nhc_gw.ipv4;
54 	else if (nh_info->gw_family == AF_INET6)
55 		nh_info->ipv6 = nhi->fib_nhc.nhc_gw.ipv6;
56 
57 	nh_info->is_reject = nhi->reject_nh;
58 	nh_info->is_fdb = nhi->fdb_nh;
59 	nh_info->has_encap = !!nhi->fib_nhc.nhc_lwtstate;
60 }
61 
62 static int nh_notifier_single_info_init(struct nh_notifier_info *info,
63 					const struct nexthop *nh)
64 {
65 	info->nh = kzalloc(sizeof(*info->nh), GFP_KERNEL);
66 	if (!info->nh)
67 		return -ENOMEM;
68 
69 	__nh_notifier_single_info_init(info->nh, nh);
70 
71 	return 0;
72 }
73 
74 static void nh_notifier_single_info_fini(struct nh_notifier_info *info)
75 {
76 	kfree(info->nh);
77 }
78 
79 static int nh_notifier_grp_info_init(struct nh_notifier_info *info,
80 				     const struct nexthop *nh)
81 {
82 	struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
83 	u16 num_nh = nhg->num_nh;
84 	int i;
85 
86 	info->nh_grp = kzalloc(struct_size(info->nh_grp, nh_entries, num_nh),
87 			       GFP_KERNEL);
88 	if (!info->nh_grp)
89 		return -ENOMEM;
90 
91 	info->nh_grp->num_nh = num_nh;
92 	info->nh_grp->is_fdb = nhg->fdb_nh;
93 
94 	for (i = 0; i < num_nh; i++) {
95 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
96 
97 		info->nh_grp->nh_entries[i].id = nhge->nh->id;
98 		info->nh_grp->nh_entries[i].weight = nhge->weight;
99 		__nh_notifier_single_info_init(&info->nh_grp->nh_entries[i].nh,
100 					       nhge->nh);
101 	}
102 
103 	return 0;
104 }
105 
106 static void nh_notifier_grp_info_fini(struct nh_notifier_info *info)
107 {
108 	kfree(info->nh_grp);
109 }
110 
111 static int nh_notifier_info_init(struct nh_notifier_info *info,
112 				 const struct nexthop *nh)
113 {
114 	info->id = nh->id;
115 	info->is_grp = nh->is_group;
116 
117 	if (info->is_grp)
118 		return nh_notifier_grp_info_init(info, nh);
119 	else
120 		return nh_notifier_single_info_init(info, nh);
121 }
122 
123 static void nh_notifier_info_fini(struct nh_notifier_info *info)
124 {
125 	if (info->is_grp)
126 		nh_notifier_grp_info_fini(info);
127 	else
128 		nh_notifier_single_info_fini(info);
129 }
130 
131 static int call_nexthop_notifiers(struct net *net,
132 				  enum nexthop_event_type event_type,
133 				  struct nexthop *nh,
134 				  struct netlink_ext_ack *extack)
135 {
136 	struct nh_notifier_info info = {
137 		.net = net,
138 		.extack = extack,
139 	};
140 	int err;
141 
142 	ASSERT_RTNL();
143 
144 	if (nexthop_notifiers_is_empty(net))
145 		return 0;
146 
147 	err = nh_notifier_info_init(&info, nh);
148 	if (err) {
149 		NL_SET_ERR_MSG(extack, "Failed to initialize nexthop notifier info");
150 		return err;
151 	}
152 
153 	err = blocking_notifier_call_chain(&net->nexthop.notifier_chain,
154 					   event_type, &info);
155 	nh_notifier_info_fini(&info);
156 
157 	return notifier_to_errno(err);
158 }
159 
160 static int call_nexthop_notifier(struct notifier_block *nb, struct net *net,
161 				 enum nexthop_event_type event_type,
162 				 struct nexthop *nh,
163 				 struct netlink_ext_ack *extack)
164 {
165 	struct nh_notifier_info info = {
166 		.net = net,
167 		.extack = extack,
168 	};
169 	int err;
170 
171 	err = nh_notifier_info_init(&info, nh);
172 	if (err)
173 		return err;
174 
175 	err = nb->notifier_call(nb, event_type, &info);
176 	nh_notifier_info_fini(&info);
177 
178 	return notifier_to_errno(err);
179 }
180 
181 static unsigned int nh_dev_hashfn(unsigned int val)
182 {
183 	unsigned int mask = NH_DEV_HASHSIZE - 1;
184 
185 	return (val ^
186 		(val >> NH_DEV_HASHBITS) ^
187 		(val >> (NH_DEV_HASHBITS * 2))) & mask;
188 }
189 
190 static void nexthop_devhash_add(struct net *net, struct nh_info *nhi)
191 {
192 	struct net_device *dev = nhi->fib_nhc.nhc_dev;
193 	struct hlist_head *head;
194 	unsigned int hash;
195 
196 	WARN_ON(!dev);
197 
198 	hash = nh_dev_hashfn(dev->ifindex);
199 	head = &net->nexthop.devhash[hash];
200 	hlist_add_head(&nhi->dev_hash, head);
201 }
202 
203 static void nexthop_free_mpath(struct nexthop *nh)
204 {
205 	struct nh_group *nhg;
206 	int i;
207 
208 	nhg = rcu_dereference_raw(nh->nh_grp);
209 	for (i = 0; i < nhg->num_nh; ++i) {
210 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
211 
212 		WARN_ON(!list_empty(&nhge->nh_list));
213 		nexthop_put(nhge->nh);
214 	}
215 
216 	WARN_ON(nhg->spare == nhg);
217 
218 	kfree(nhg->spare);
219 	kfree(nhg);
220 }
221 
222 static void nexthop_free_single(struct nexthop *nh)
223 {
224 	struct nh_info *nhi;
225 
226 	nhi = rcu_dereference_raw(nh->nh_info);
227 	switch (nhi->family) {
228 	case AF_INET:
229 		fib_nh_release(nh->net, &nhi->fib_nh);
230 		break;
231 	case AF_INET6:
232 		ipv6_stub->fib6_nh_release(&nhi->fib6_nh);
233 		break;
234 	}
235 	kfree(nhi);
236 }
237 
238 void nexthop_free_rcu(struct rcu_head *head)
239 {
240 	struct nexthop *nh = container_of(head, struct nexthop, rcu);
241 
242 	if (nh->is_group)
243 		nexthop_free_mpath(nh);
244 	else
245 		nexthop_free_single(nh);
246 
247 	kfree(nh);
248 }
249 EXPORT_SYMBOL_GPL(nexthop_free_rcu);
250 
251 static struct nexthop *nexthop_alloc(void)
252 {
253 	struct nexthop *nh;
254 
255 	nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL);
256 	if (nh) {
257 		INIT_LIST_HEAD(&nh->fi_list);
258 		INIT_LIST_HEAD(&nh->f6i_list);
259 		INIT_LIST_HEAD(&nh->grp_list);
260 		INIT_LIST_HEAD(&nh->fdb_list);
261 	}
262 	return nh;
263 }
264 
265 static struct nh_group *nexthop_grp_alloc(u16 num_nh)
266 {
267 	struct nh_group *nhg;
268 
269 	nhg = kzalloc(struct_size(nhg, nh_entries, num_nh), GFP_KERNEL);
270 	if (nhg)
271 		nhg->num_nh = num_nh;
272 
273 	return nhg;
274 }
275 
276 static void nh_base_seq_inc(struct net *net)
277 {
278 	while (++net->nexthop.seq == 0)
279 		;
280 }
281 
282 /* no reference taken; rcu lock or rtnl must be held */
283 struct nexthop *nexthop_find_by_id(struct net *net, u32 id)
284 {
285 	struct rb_node **pp, *parent = NULL, *next;
286 
287 	pp = &net->nexthop.rb_root.rb_node;
288 	while (1) {
289 		struct nexthop *nh;
290 
291 		next = rcu_dereference_raw(*pp);
292 		if (!next)
293 			break;
294 		parent = next;
295 
296 		nh = rb_entry(parent, struct nexthop, rb_node);
297 		if (id < nh->id)
298 			pp = &next->rb_left;
299 		else if (id > nh->id)
300 			pp = &next->rb_right;
301 		else
302 			return nh;
303 	}
304 	return NULL;
305 }
306 EXPORT_SYMBOL_GPL(nexthop_find_by_id);
307 
308 /* used for auto id allocation; called with rtnl held */
309 static u32 nh_find_unused_id(struct net *net)
310 {
311 	u32 id_start = net->nexthop.last_id_allocated;
312 
313 	while (1) {
314 		net->nexthop.last_id_allocated++;
315 		if (net->nexthop.last_id_allocated == id_start)
316 			break;
317 
318 		if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated))
319 			return net->nexthop.last_id_allocated;
320 	}
321 	return 0;
322 }
323 
324 static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
325 {
326 	struct nexthop_grp *p;
327 	size_t len = nhg->num_nh * sizeof(*p);
328 	struct nlattr *nla;
329 	u16 group_type = 0;
330 	int i;
331 
332 	if (nhg->mpath)
333 		group_type = NEXTHOP_GRP_TYPE_MPATH;
334 
335 	if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type))
336 		goto nla_put_failure;
337 
338 	nla = nla_reserve(skb, NHA_GROUP, len);
339 	if (!nla)
340 		goto nla_put_failure;
341 
342 	p = nla_data(nla);
343 	for (i = 0; i < nhg->num_nh; ++i) {
344 		p->id = nhg->nh_entries[i].nh->id;
345 		p->weight = nhg->nh_entries[i].weight - 1;
346 		p += 1;
347 	}
348 
349 	return 0;
350 
351 nla_put_failure:
352 	return -EMSGSIZE;
353 }
354 
355 static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh,
356 			int event, u32 portid, u32 seq, unsigned int nlflags)
357 {
358 	struct fib6_nh *fib6_nh;
359 	struct fib_nh *fib_nh;
360 	struct nlmsghdr *nlh;
361 	struct nh_info *nhi;
362 	struct nhmsg *nhm;
363 
364 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
365 	if (!nlh)
366 		return -EMSGSIZE;
367 
368 	nhm = nlmsg_data(nlh);
369 	nhm->nh_family = AF_UNSPEC;
370 	nhm->nh_flags = nh->nh_flags;
371 	nhm->nh_protocol = nh->protocol;
372 	nhm->nh_scope = 0;
373 	nhm->resvd = 0;
374 
375 	if (nla_put_u32(skb, NHA_ID, nh->id))
376 		goto nla_put_failure;
377 
378 	if (nh->is_group) {
379 		struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
380 
381 		if (nhg->fdb_nh && nla_put_flag(skb, NHA_FDB))
382 			goto nla_put_failure;
383 		if (nla_put_nh_group(skb, nhg))
384 			goto nla_put_failure;
385 		goto out;
386 	}
387 
388 	nhi = rtnl_dereference(nh->nh_info);
389 	nhm->nh_family = nhi->family;
390 	if (nhi->reject_nh) {
391 		if (nla_put_flag(skb, NHA_BLACKHOLE))
392 			goto nla_put_failure;
393 		goto out;
394 	} else if (nhi->fdb_nh) {
395 		if (nla_put_flag(skb, NHA_FDB))
396 			goto nla_put_failure;
397 	} else {
398 		const struct net_device *dev;
399 
400 		dev = nhi->fib_nhc.nhc_dev;
401 		if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex))
402 			goto nla_put_failure;
403 	}
404 
405 	nhm->nh_scope = nhi->fib_nhc.nhc_scope;
406 	switch (nhi->family) {
407 	case AF_INET:
408 		fib_nh = &nhi->fib_nh;
409 		if (fib_nh->fib_nh_gw_family &&
410 		    nla_put_be32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4))
411 			goto nla_put_failure;
412 		break;
413 
414 	case AF_INET6:
415 		fib6_nh = &nhi->fib6_nh;
416 		if (fib6_nh->fib_nh_gw_family &&
417 		    nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6))
418 			goto nla_put_failure;
419 		break;
420 	}
421 
422 	if (nhi->fib_nhc.nhc_lwtstate &&
423 	    lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate,
424 				NHA_ENCAP, NHA_ENCAP_TYPE) < 0)
425 		goto nla_put_failure;
426 
427 out:
428 	nlmsg_end(skb, nlh);
429 	return 0;
430 
431 nla_put_failure:
432 	nlmsg_cancel(skb, nlh);
433 	return -EMSGSIZE;
434 }
435 
436 static size_t nh_nlmsg_size_grp(struct nexthop *nh)
437 {
438 	struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
439 	size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh;
440 
441 	return nla_total_size(sz) +
442 	       nla_total_size(2);  /* NHA_GROUP_TYPE */
443 }
444 
445 static size_t nh_nlmsg_size_single(struct nexthop *nh)
446 {
447 	struct nh_info *nhi = rtnl_dereference(nh->nh_info);
448 	size_t sz;
449 
450 	/* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE
451 	 * are mutually exclusive
452 	 */
453 	sz = nla_total_size(4);  /* NHA_OIF */
454 
455 	switch (nhi->family) {
456 	case AF_INET:
457 		if (nhi->fib_nh.fib_nh_gw_family)
458 			sz += nla_total_size(4);  /* NHA_GATEWAY */
459 		break;
460 
461 	case AF_INET6:
462 		/* NHA_GATEWAY */
463 		if (nhi->fib6_nh.fib_nh_gw_family)
464 			sz += nla_total_size(sizeof(const struct in6_addr));
465 		break;
466 	}
467 
468 	if (nhi->fib_nhc.nhc_lwtstate) {
469 		sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate);
470 		sz += nla_total_size(2);  /* NHA_ENCAP_TYPE */
471 	}
472 
473 	return sz;
474 }
475 
476 static size_t nh_nlmsg_size(struct nexthop *nh)
477 {
478 	size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg));
479 
480 	sz += nla_total_size(4); /* NHA_ID */
481 
482 	if (nh->is_group)
483 		sz += nh_nlmsg_size_grp(nh);
484 	else
485 		sz += nh_nlmsg_size_single(nh);
486 
487 	return sz;
488 }
489 
490 static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info)
491 {
492 	unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0;
493 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
494 	struct sk_buff *skb;
495 	int err = -ENOBUFS;
496 
497 	skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any());
498 	if (!skb)
499 		goto errout;
500 
501 	err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags);
502 	if (err < 0) {
503 		/* -EMSGSIZE implies BUG in nh_nlmsg_size() */
504 		WARN_ON(err == -EMSGSIZE);
505 		kfree_skb(skb);
506 		goto errout;
507 	}
508 
509 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP,
510 		    info->nlh, gfp_any());
511 	return;
512 errout:
513 	if (err < 0)
514 		rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err);
515 }
516 
517 static bool valid_group_nh(struct nexthop *nh, unsigned int npaths,
518 			   bool *is_fdb, struct netlink_ext_ack *extack)
519 {
520 	if (nh->is_group) {
521 		struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
522 
523 		/* nested multipath (group within a group) is not
524 		 * supported
525 		 */
526 		if (nhg->mpath) {
527 			NL_SET_ERR_MSG(extack,
528 				       "Multipath group can not be a nexthop within a group");
529 			return false;
530 		}
531 		*is_fdb = nhg->fdb_nh;
532 	} else {
533 		struct nh_info *nhi = rtnl_dereference(nh->nh_info);
534 
535 		if (nhi->reject_nh && npaths > 1) {
536 			NL_SET_ERR_MSG(extack,
537 				       "Blackhole nexthop can not be used in a group with more than 1 path");
538 			return false;
539 		}
540 		*is_fdb = nhi->fdb_nh;
541 	}
542 
543 	return true;
544 }
545 
546 static int nh_check_attr_fdb_group(struct nexthop *nh, u8 *nh_family,
547 				   struct netlink_ext_ack *extack)
548 {
549 	struct nh_info *nhi;
550 
551 	nhi = rtnl_dereference(nh->nh_info);
552 
553 	if (!nhi->fdb_nh) {
554 		NL_SET_ERR_MSG(extack, "FDB nexthop group can only have fdb nexthops");
555 		return -EINVAL;
556 	}
557 
558 	if (*nh_family == AF_UNSPEC) {
559 		*nh_family = nhi->family;
560 	} else if (*nh_family != nhi->family) {
561 		NL_SET_ERR_MSG(extack, "FDB nexthop group cannot have mixed family nexthops");
562 		return -EINVAL;
563 	}
564 
565 	return 0;
566 }
567 
568 static int nh_check_attr_group(struct net *net, struct nlattr *tb[],
569 			       struct netlink_ext_ack *extack)
570 {
571 	unsigned int len = nla_len(tb[NHA_GROUP]);
572 	u8 nh_family = AF_UNSPEC;
573 	struct nexthop_grp *nhg;
574 	unsigned int i, j;
575 	u8 nhg_fdb = 0;
576 
577 	if (!len || len & (sizeof(struct nexthop_grp) - 1)) {
578 		NL_SET_ERR_MSG(extack,
579 			       "Invalid length for nexthop group attribute");
580 		return -EINVAL;
581 	}
582 
583 	/* convert len to number of nexthop ids */
584 	len /= sizeof(*nhg);
585 
586 	nhg = nla_data(tb[NHA_GROUP]);
587 	for (i = 0; i < len; ++i) {
588 		if (nhg[i].resvd1 || nhg[i].resvd2) {
589 			NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0");
590 			return -EINVAL;
591 		}
592 		if (nhg[i].weight > 254) {
593 			NL_SET_ERR_MSG(extack, "Invalid value for weight");
594 			return -EINVAL;
595 		}
596 		for (j = i + 1; j < len; ++j) {
597 			if (nhg[i].id == nhg[j].id) {
598 				NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group");
599 				return -EINVAL;
600 			}
601 		}
602 	}
603 
604 	if (tb[NHA_FDB])
605 		nhg_fdb = 1;
606 	nhg = nla_data(tb[NHA_GROUP]);
607 	for (i = 0; i < len; ++i) {
608 		struct nexthop *nh;
609 		bool is_fdb_nh;
610 
611 		nh = nexthop_find_by_id(net, nhg[i].id);
612 		if (!nh) {
613 			NL_SET_ERR_MSG(extack, "Invalid nexthop id");
614 			return -EINVAL;
615 		}
616 		if (!valid_group_nh(nh, len, &is_fdb_nh, extack))
617 			return -EINVAL;
618 
619 		if (nhg_fdb && nh_check_attr_fdb_group(nh, &nh_family, extack))
620 			return -EINVAL;
621 
622 		if (!nhg_fdb && is_fdb_nh) {
623 			NL_SET_ERR_MSG(extack, "Non FDB nexthop group cannot have fdb nexthops");
624 			return -EINVAL;
625 		}
626 	}
627 	for (i = NHA_GROUP_TYPE + 1; i < __NHA_MAX; ++i) {
628 		if (!tb[i])
629 			continue;
630 		if (i == NHA_FDB)
631 			continue;
632 		NL_SET_ERR_MSG(extack,
633 			       "No other attributes can be set in nexthop groups");
634 		return -EINVAL;
635 	}
636 
637 	return 0;
638 }
639 
640 static bool ipv6_good_nh(const struct fib6_nh *nh)
641 {
642 	int state = NUD_REACHABLE;
643 	struct neighbour *n;
644 
645 	rcu_read_lock_bh();
646 
647 	n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6);
648 	if (n)
649 		state = n->nud_state;
650 
651 	rcu_read_unlock_bh();
652 
653 	return !!(state & NUD_VALID);
654 }
655 
656 static bool ipv4_good_nh(const struct fib_nh *nh)
657 {
658 	int state = NUD_REACHABLE;
659 	struct neighbour *n;
660 
661 	rcu_read_lock_bh();
662 
663 	n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
664 				      (__force u32)nh->fib_nh_gw4);
665 	if (n)
666 		state = n->nud_state;
667 
668 	rcu_read_unlock_bh();
669 
670 	return !!(state & NUD_VALID);
671 }
672 
673 struct nexthop *nexthop_select_path(struct nexthop *nh, int hash)
674 {
675 	struct nexthop *rc = NULL;
676 	struct nh_group *nhg;
677 	int i;
678 
679 	if (!nh->is_group)
680 		return nh;
681 
682 	nhg = rcu_dereference(nh->nh_grp);
683 	for (i = 0; i < nhg->num_nh; ++i) {
684 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
685 		struct nh_info *nhi;
686 
687 		if (hash > atomic_read(&nhge->upper_bound))
688 			continue;
689 
690 		nhi = rcu_dereference(nhge->nh->nh_info);
691 		if (nhi->fdb_nh)
692 			return nhge->nh;
693 
694 		/* nexthops always check if it is good and does
695 		 * not rely on a sysctl for this behavior
696 		 */
697 		switch (nhi->family) {
698 		case AF_INET:
699 			if (ipv4_good_nh(&nhi->fib_nh))
700 				return nhge->nh;
701 			break;
702 		case AF_INET6:
703 			if (ipv6_good_nh(&nhi->fib6_nh))
704 				return nhge->nh;
705 			break;
706 		}
707 
708 		if (!rc)
709 			rc = nhge->nh;
710 	}
711 
712 	return rc;
713 }
714 EXPORT_SYMBOL_GPL(nexthop_select_path);
715 
716 int nexthop_for_each_fib6_nh(struct nexthop *nh,
717 			     int (*cb)(struct fib6_nh *nh, void *arg),
718 			     void *arg)
719 {
720 	struct nh_info *nhi;
721 	int err;
722 
723 	if (nh->is_group) {
724 		struct nh_group *nhg;
725 		int i;
726 
727 		nhg = rcu_dereference_rtnl(nh->nh_grp);
728 		for (i = 0; i < nhg->num_nh; i++) {
729 			struct nh_grp_entry *nhge = &nhg->nh_entries[i];
730 
731 			nhi = rcu_dereference_rtnl(nhge->nh->nh_info);
732 			err = cb(&nhi->fib6_nh, arg);
733 			if (err)
734 				return err;
735 		}
736 	} else {
737 		nhi = rcu_dereference_rtnl(nh->nh_info);
738 		err = cb(&nhi->fib6_nh, arg);
739 		if (err)
740 			return err;
741 	}
742 
743 	return 0;
744 }
745 EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh);
746 
747 static int check_src_addr(const struct in6_addr *saddr,
748 			  struct netlink_ext_ack *extack)
749 {
750 	if (!ipv6_addr_any(saddr)) {
751 		NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects");
752 		return -EINVAL;
753 	}
754 	return 0;
755 }
756 
757 int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
758 		       struct netlink_ext_ack *extack)
759 {
760 	struct nh_info *nhi;
761 	bool is_fdb_nh;
762 
763 	/* fib6_src is unique to a fib6_info and limits the ability to cache
764 	 * routes in fib6_nh within a nexthop that is potentially shared
765 	 * across multiple fib entries. If the config wants to use source
766 	 * routing it can not use nexthop objects. mlxsw also does not allow
767 	 * fib6_src on routes.
768 	 */
769 	if (cfg && check_src_addr(&cfg->fc_src, extack) < 0)
770 		return -EINVAL;
771 
772 	if (nh->is_group) {
773 		struct nh_group *nhg;
774 
775 		nhg = rtnl_dereference(nh->nh_grp);
776 		if (nhg->has_v4)
777 			goto no_v4_nh;
778 		is_fdb_nh = nhg->fdb_nh;
779 	} else {
780 		nhi = rtnl_dereference(nh->nh_info);
781 		if (nhi->family == AF_INET)
782 			goto no_v4_nh;
783 		is_fdb_nh = nhi->fdb_nh;
784 	}
785 
786 	if (is_fdb_nh) {
787 		NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
788 		return -EINVAL;
789 	}
790 
791 	return 0;
792 no_v4_nh:
793 	NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop");
794 	return -EINVAL;
795 }
796 EXPORT_SYMBOL_GPL(fib6_check_nexthop);
797 
798 /* if existing nexthop has ipv6 routes linked to it, need
799  * to verify this new spec works with ipv6
800  */
801 static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new,
802 			      struct netlink_ext_ack *extack)
803 {
804 	struct fib6_info *f6i;
805 
806 	if (list_empty(&old->f6i_list))
807 		return 0;
808 
809 	list_for_each_entry(f6i, &old->f6i_list, nh_list) {
810 		if (check_src_addr(&f6i->fib6_src.addr, extack) < 0)
811 			return -EINVAL;
812 	}
813 
814 	return fib6_check_nexthop(new, NULL, extack);
815 }
816 
817 static int nexthop_check_scope(struct nh_info *nhi, u8 scope,
818 			       struct netlink_ext_ack *extack)
819 {
820 	if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) {
821 		NL_SET_ERR_MSG(extack,
822 			       "Route with host scope can not have a gateway");
823 		return -EINVAL;
824 	}
825 
826 	if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) {
827 		NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop");
828 		return -EINVAL;
829 	}
830 
831 	return 0;
832 }
833 
834 /* Invoked by fib add code to verify nexthop by id is ok with
835  * config for prefix; parts of fib_check_nh not done when nexthop
836  * object is used.
837  */
838 int fib_check_nexthop(struct nexthop *nh, u8 scope,
839 		      struct netlink_ext_ack *extack)
840 {
841 	struct nh_info *nhi;
842 	int err = 0;
843 
844 	if (nh->is_group) {
845 		struct nh_group *nhg;
846 
847 		nhg = rtnl_dereference(nh->nh_grp);
848 		if (nhg->fdb_nh) {
849 			NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
850 			err = -EINVAL;
851 			goto out;
852 		}
853 
854 		if (scope == RT_SCOPE_HOST) {
855 			NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops");
856 			err = -EINVAL;
857 			goto out;
858 		}
859 
860 		/* all nexthops in a group have the same scope */
861 		nhi = rtnl_dereference(nhg->nh_entries[0].nh->nh_info);
862 		err = nexthop_check_scope(nhi, scope, extack);
863 	} else {
864 		nhi = rtnl_dereference(nh->nh_info);
865 		if (nhi->fdb_nh) {
866 			NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
867 			err = -EINVAL;
868 			goto out;
869 		}
870 		err = nexthop_check_scope(nhi, scope, extack);
871 	}
872 
873 out:
874 	return err;
875 }
876 
877 static int fib_check_nh_list(struct nexthop *old, struct nexthop *new,
878 			     struct netlink_ext_ack *extack)
879 {
880 	struct fib_info *fi;
881 
882 	list_for_each_entry(fi, &old->fi_list, nh_list) {
883 		int err;
884 
885 		err = fib_check_nexthop(new, fi->fib_scope, extack);
886 		if (err)
887 			return err;
888 	}
889 	return 0;
890 }
891 
892 static void nh_group_rebalance(struct nh_group *nhg)
893 {
894 	int total = 0;
895 	int w = 0;
896 	int i;
897 
898 	for (i = 0; i < nhg->num_nh; ++i)
899 		total += nhg->nh_entries[i].weight;
900 
901 	for (i = 0; i < nhg->num_nh; ++i) {
902 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
903 		int upper_bound;
904 
905 		w += nhge->weight;
906 		upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1;
907 		atomic_set(&nhge->upper_bound, upper_bound);
908 	}
909 }
910 
911 static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge,
912 				struct nl_info *nlinfo)
913 {
914 	struct nh_grp_entry *nhges, *new_nhges;
915 	struct nexthop *nhp = nhge->nh_parent;
916 	struct netlink_ext_ack extack;
917 	struct nexthop *nh = nhge->nh;
918 	struct nh_group *nhg, *newg;
919 	int i, j, err;
920 
921 	WARN_ON(!nh);
922 
923 	nhg = rtnl_dereference(nhp->nh_grp);
924 	newg = nhg->spare;
925 
926 	/* last entry, keep it visible and remove the parent */
927 	if (nhg->num_nh == 1) {
928 		remove_nexthop(net, nhp, nlinfo);
929 		return;
930 	}
931 
932 	newg->has_v4 = false;
933 	newg->mpath = nhg->mpath;
934 	newg->fdb_nh = nhg->fdb_nh;
935 	newg->num_nh = nhg->num_nh;
936 
937 	/* copy old entries to new except the one getting removed */
938 	nhges = nhg->nh_entries;
939 	new_nhges = newg->nh_entries;
940 	for (i = 0, j = 0; i < nhg->num_nh; ++i) {
941 		struct nh_info *nhi;
942 
943 		/* current nexthop getting removed */
944 		if (nhg->nh_entries[i].nh == nh) {
945 			newg->num_nh--;
946 			continue;
947 		}
948 
949 		nhi = rtnl_dereference(nhges[i].nh->nh_info);
950 		if (nhi->family == AF_INET)
951 			newg->has_v4 = true;
952 
953 		list_del(&nhges[i].nh_list);
954 		new_nhges[j].nh_parent = nhges[i].nh_parent;
955 		new_nhges[j].nh = nhges[i].nh;
956 		new_nhges[j].weight = nhges[i].weight;
957 		list_add(&new_nhges[j].nh_list, &new_nhges[j].nh->grp_list);
958 		j++;
959 	}
960 
961 	nh_group_rebalance(newg);
962 	rcu_assign_pointer(nhp->nh_grp, newg);
963 
964 	list_del(&nhge->nh_list);
965 	nexthop_put(nhge->nh);
966 
967 	err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp, &extack);
968 	if (err)
969 		pr_err("%s\n", extack._msg);
970 
971 	if (nlinfo)
972 		nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo);
973 }
974 
975 static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh,
976 				       struct nl_info *nlinfo)
977 {
978 	struct nh_grp_entry *nhge, *tmp;
979 
980 	list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list)
981 		remove_nh_grp_entry(net, nhge, nlinfo);
982 
983 	/* make sure all see the newly published array before releasing rtnl */
984 	synchronize_net();
985 }
986 
987 static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo)
988 {
989 	struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
990 	int i, num_nh = nhg->num_nh;
991 
992 	for (i = 0; i < num_nh; ++i) {
993 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
994 
995 		if (WARN_ON(!nhge->nh))
996 			continue;
997 
998 		list_del_init(&nhge->nh_list);
999 	}
1000 }
1001 
1002 /* not called for nexthop replace */
1003 static void __remove_nexthop_fib(struct net *net, struct nexthop *nh)
1004 {
1005 	struct fib6_info *f6i, *tmp;
1006 	bool do_flush = false;
1007 	struct fib_info *fi;
1008 
1009 	list_for_each_entry(fi, &nh->fi_list, nh_list) {
1010 		fi->fib_flags |= RTNH_F_DEAD;
1011 		do_flush = true;
1012 	}
1013 	if (do_flush)
1014 		fib_flush(net);
1015 
1016 	/* ip6_del_rt removes the entry from this list hence the _safe */
1017 	list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) {
1018 		/* __ip6_del_rt does a release, so do a hold here */
1019 		fib6_info_hold(f6i);
1020 		ipv6_stub->ip6_del_rt(net, f6i,
1021 				      !net->ipv4.sysctl_nexthop_compat_mode);
1022 	}
1023 }
1024 
1025 static void __remove_nexthop(struct net *net, struct nexthop *nh,
1026 			     struct nl_info *nlinfo)
1027 {
1028 	__remove_nexthop_fib(net, nh);
1029 
1030 	if (nh->is_group) {
1031 		remove_nexthop_group(nh, nlinfo);
1032 	} else {
1033 		struct nh_info *nhi;
1034 
1035 		nhi = rtnl_dereference(nh->nh_info);
1036 		if (nhi->fib_nhc.nhc_dev)
1037 			hlist_del(&nhi->dev_hash);
1038 
1039 		remove_nexthop_from_groups(net, nh, nlinfo);
1040 	}
1041 }
1042 
1043 static void remove_nexthop(struct net *net, struct nexthop *nh,
1044 			   struct nl_info *nlinfo)
1045 {
1046 	call_nexthop_notifiers(net, NEXTHOP_EVENT_DEL, nh, NULL);
1047 
1048 	/* remove from the tree */
1049 	rb_erase(&nh->rb_node, &net->nexthop.rb_root);
1050 
1051 	if (nlinfo)
1052 		nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo);
1053 
1054 	__remove_nexthop(net, nh, nlinfo);
1055 	nh_base_seq_inc(net);
1056 
1057 	nexthop_put(nh);
1058 }
1059 
1060 /* if any FIB entries reference this nexthop, any dst entries
1061  * need to be regenerated
1062  */
1063 static void nh_rt_cache_flush(struct net *net, struct nexthop *nh)
1064 {
1065 	struct fib6_info *f6i;
1066 
1067 	if (!list_empty(&nh->fi_list))
1068 		rt_cache_flush(net);
1069 
1070 	list_for_each_entry(f6i, &nh->f6i_list, nh_list)
1071 		ipv6_stub->fib6_update_sernum(net, f6i);
1072 }
1073 
1074 static int replace_nexthop_grp(struct net *net, struct nexthop *old,
1075 			       struct nexthop *new,
1076 			       struct netlink_ext_ack *extack)
1077 {
1078 	struct nh_group *oldg, *newg;
1079 	int i, err;
1080 
1081 	if (!new->is_group) {
1082 		NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop.");
1083 		return -EINVAL;
1084 	}
1085 
1086 	err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack);
1087 	if (err)
1088 		return err;
1089 
1090 	oldg = rtnl_dereference(old->nh_grp);
1091 	newg = rtnl_dereference(new->nh_grp);
1092 
1093 	/* update parents - used by nexthop code for cleanup */
1094 	for (i = 0; i < newg->num_nh; i++)
1095 		newg->nh_entries[i].nh_parent = old;
1096 
1097 	rcu_assign_pointer(old->nh_grp, newg);
1098 
1099 	for (i = 0; i < oldg->num_nh; i++)
1100 		oldg->nh_entries[i].nh_parent = new;
1101 
1102 	rcu_assign_pointer(new->nh_grp, oldg);
1103 
1104 	return 0;
1105 }
1106 
1107 static void nh_group_v4_update(struct nh_group *nhg)
1108 {
1109 	struct nh_grp_entry *nhges;
1110 	bool has_v4 = false;
1111 	int i;
1112 
1113 	nhges = nhg->nh_entries;
1114 	for (i = 0; i < nhg->num_nh; i++) {
1115 		struct nh_info *nhi;
1116 
1117 		nhi = rtnl_dereference(nhges[i].nh->nh_info);
1118 		if (nhi->family == AF_INET)
1119 			has_v4 = true;
1120 	}
1121 	nhg->has_v4 = has_v4;
1122 }
1123 
1124 static int replace_nexthop_single(struct net *net, struct nexthop *old,
1125 				  struct nexthop *new,
1126 				  struct netlink_ext_ack *extack)
1127 {
1128 	u8 old_protocol, old_nh_flags;
1129 	struct nh_info *oldi, *newi;
1130 	struct nh_grp_entry *nhge;
1131 	int err;
1132 
1133 	if (new->is_group) {
1134 		NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
1135 		return -EINVAL;
1136 	}
1137 
1138 	err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack);
1139 	if (err)
1140 		return err;
1141 
1142 	/* Hardware flags were set on 'old' as 'new' is not in the red-black
1143 	 * tree. Therefore, inherit the flags from 'old' to 'new'.
1144 	 */
1145 	new->nh_flags |= old->nh_flags & (RTNH_F_OFFLOAD | RTNH_F_TRAP);
1146 
1147 	oldi = rtnl_dereference(old->nh_info);
1148 	newi = rtnl_dereference(new->nh_info);
1149 
1150 	newi->nh_parent = old;
1151 	oldi->nh_parent = new;
1152 
1153 	old_protocol = old->protocol;
1154 	old_nh_flags = old->nh_flags;
1155 
1156 	old->protocol = new->protocol;
1157 	old->nh_flags = new->nh_flags;
1158 
1159 	rcu_assign_pointer(old->nh_info, newi);
1160 	rcu_assign_pointer(new->nh_info, oldi);
1161 
1162 	/* Send a replace notification for all the groups using the nexthop. */
1163 	list_for_each_entry(nhge, &old->grp_list, nh_list) {
1164 		struct nexthop *nhp = nhge->nh_parent;
1165 
1166 		err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp,
1167 					     extack);
1168 		if (err)
1169 			goto err_notify;
1170 	}
1171 
1172 	/* When replacing an IPv4 nexthop with an IPv6 nexthop, potentially
1173 	 * update IPv4 indication in all the groups using the nexthop.
1174 	 */
1175 	if (oldi->family == AF_INET && newi->family == AF_INET6) {
1176 		list_for_each_entry(nhge, &old->grp_list, nh_list) {
1177 			struct nexthop *nhp = nhge->nh_parent;
1178 			struct nh_group *nhg;
1179 
1180 			nhg = rtnl_dereference(nhp->nh_grp);
1181 			nh_group_v4_update(nhg);
1182 		}
1183 	}
1184 
1185 	return 0;
1186 
1187 err_notify:
1188 	rcu_assign_pointer(new->nh_info, newi);
1189 	rcu_assign_pointer(old->nh_info, oldi);
1190 	old->nh_flags = old_nh_flags;
1191 	old->protocol = old_protocol;
1192 	oldi->nh_parent = old;
1193 	newi->nh_parent = new;
1194 	list_for_each_entry_continue_reverse(nhge, &old->grp_list, nh_list) {
1195 		struct nexthop *nhp = nhge->nh_parent;
1196 
1197 		call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp, extack);
1198 	}
1199 	call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, old, extack);
1200 	return err;
1201 }
1202 
1203 static void __nexthop_replace_notify(struct net *net, struct nexthop *nh,
1204 				     struct nl_info *info)
1205 {
1206 	struct fib6_info *f6i;
1207 
1208 	if (!list_empty(&nh->fi_list)) {
1209 		struct fib_info *fi;
1210 
1211 		/* expectation is a few fib_info per nexthop and then
1212 		 * a lot of routes per fib_info. So mark the fib_info
1213 		 * and then walk the fib tables once
1214 		 */
1215 		list_for_each_entry(fi, &nh->fi_list, nh_list)
1216 			fi->nh_updated = true;
1217 
1218 		fib_info_notify_update(net, info);
1219 
1220 		list_for_each_entry(fi, &nh->fi_list, nh_list)
1221 			fi->nh_updated = false;
1222 	}
1223 
1224 	list_for_each_entry(f6i, &nh->f6i_list, nh_list)
1225 		ipv6_stub->fib6_rt_update(net, f6i, info);
1226 }
1227 
1228 /* send RTM_NEWROUTE with REPLACE flag set for all FIB entries
1229  * linked to this nexthop and for all groups that the nexthop
1230  * is a member of
1231  */
1232 static void nexthop_replace_notify(struct net *net, struct nexthop *nh,
1233 				   struct nl_info *info)
1234 {
1235 	struct nh_grp_entry *nhge;
1236 
1237 	__nexthop_replace_notify(net, nh, info);
1238 
1239 	list_for_each_entry(nhge, &nh->grp_list, nh_list)
1240 		__nexthop_replace_notify(net, nhge->nh_parent, info);
1241 }
1242 
1243 static int replace_nexthop(struct net *net, struct nexthop *old,
1244 			   struct nexthop *new, struct netlink_ext_ack *extack)
1245 {
1246 	bool new_is_reject = false;
1247 	struct nh_grp_entry *nhge;
1248 	int err;
1249 
1250 	/* check that existing FIB entries are ok with the
1251 	 * new nexthop definition
1252 	 */
1253 	err = fib_check_nh_list(old, new, extack);
1254 	if (err)
1255 		return err;
1256 
1257 	err = fib6_check_nh_list(old, new, extack);
1258 	if (err)
1259 		return err;
1260 
1261 	if (!new->is_group) {
1262 		struct nh_info *nhi = rtnl_dereference(new->nh_info);
1263 
1264 		new_is_reject = nhi->reject_nh;
1265 	}
1266 
1267 	list_for_each_entry(nhge, &old->grp_list, nh_list) {
1268 		/* if new nexthop is a blackhole, any groups using this
1269 		 * nexthop cannot have more than 1 path
1270 		 */
1271 		if (new_is_reject &&
1272 		    nexthop_num_path(nhge->nh_parent) > 1) {
1273 			NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path");
1274 			return -EINVAL;
1275 		}
1276 
1277 		err = fib_check_nh_list(nhge->nh_parent, new, extack);
1278 		if (err)
1279 			return err;
1280 
1281 		err = fib6_check_nh_list(nhge->nh_parent, new, extack);
1282 		if (err)
1283 			return err;
1284 	}
1285 
1286 	if (old->is_group)
1287 		err = replace_nexthop_grp(net, old, new, extack);
1288 	else
1289 		err = replace_nexthop_single(net, old, new, extack);
1290 
1291 	if (!err) {
1292 		nh_rt_cache_flush(net, old);
1293 
1294 		__remove_nexthop(net, new, NULL);
1295 		nexthop_put(new);
1296 	}
1297 
1298 	return err;
1299 }
1300 
1301 /* called with rtnl_lock held */
1302 static int insert_nexthop(struct net *net, struct nexthop *new_nh,
1303 			  struct nh_config *cfg, struct netlink_ext_ack *extack)
1304 {
1305 	struct rb_node **pp, *parent = NULL, *next;
1306 	struct rb_root *root = &net->nexthop.rb_root;
1307 	bool replace = !!(cfg->nlflags & NLM_F_REPLACE);
1308 	bool create = !!(cfg->nlflags & NLM_F_CREATE);
1309 	u32 new_id = new_nh->id;
1310 	int replace_notify = 0;
1311 	int rc = -EEXIST;
1312 
1313 	pp = &root->rb_node;
1314 	while (1) {
1315 		struct nexthop *nh;
1316 
1317 		next = *pp;
1318 		if (!next)
1319 			break;
1320 
1321 		parent = next;
1322 
1323 		nh = rb_entry(parent, struct nexthop, rb_node);
1324 		if (new_id < nh->id) {
1325 			pp = &next->rb_left;
1326 		} else if (new_id > nh->id) {
1327 			pp = &next->rb_right;
1328 		} else if (replace) {
1329 			rc = replace_nexthop(net, nh, new_nh, extack);
1330 			if (!rc) {
1331 				new_nh = nh; /* send notification with old nh */
1332 				replace_notify = 1;
1333 			}
1334 			goto out;
1335 		} else {
1336 			/* id already exists and not a replace */
1337 			goto out;
1338 		}
1339 	}
1340 
1341 	if (replace && !create) {
1342 		NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists");
1343 		rc = -ENOENT;
1344 		goto out;
1345 	}
1346 
1347 	rb_link_node_rcu(&new_nh->rb_node, parent, pp);
1348 	rb_insert_color(&new_nh->rb_node, root);
1349 
1350 	rc = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new_nh, extack);
1351 	if (rc)
1352 		rb_erase(&new_nh->rb_node, &net->nexthop.rb_root);
1353 
1354 out:
1355 	if (!rc) {
1356 		nh_base_seq_inc(net);
1357 		nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo);
1358 		if (replace_notify && net->ipv4.sysctl_nexthop_compat_mode)
1359 			nexthop_replace_notify(net, new_nh, &cfg->nlinfo);
1360 	}
1361 
1362 	return rc;
1363 }
1364 
1365 /* rtnl */
1366 /* remove all nexthops tied to a device being deleted */
1367 static void nexthop_flush_dev(struct net_device *dev)
1368 {
1369 	unsigned int hash = nh_dev_hashfn(dev->ifindex);
1370 	struct net *net = dev_net(dev);
1371 	struct hlist_head *head = &net->nexthop.devhash[hash];
1372 	struct hlist_node *n;
1373 	struct nh_info *nhi;
1374 
1375 	hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1376 		if (nhi->fib_nhc.nhc_dev != dev)
1377 			continue;
1378 
1379 		remove_nexthop(net, nhi->nh_parent, NULL);
1380 	}
1381 }
1382 
1383 /* rtnl; called when net namespace is deleted */
1384 static void flush_all_nexthops(struct net *net)
1385 {
1386 	struct rb_root *root = &net->nexthop.rb_root;
1387 	struct rb_node *node;
1388 	struct nexthop *nh;
1389 
1390 	while ((node = rb_first(root))) {
1391 		nh = rb_entry(node, struct nexthop, rb_node);
1392 		remove_nexthop(net, nh, NULL);
1393 		cond_resched();
1394 	}
1395 }
1396 
1397 static struct nexthop *nexthop_create_group(struct net *net,
1398 					    struct nh_config *cfg)
1399 {
1400 	struct nlattr *grps_attr = cfg->nh_grp;
1401 	struct nexthop_grp *entry = nla_data(grps_attr);
1402 	u16 num_nh = nla_len(grps_attr) / sizeof(*entry);
1403 	struct nh_group *nhg;
1404 	struct nexthop *nh;
1405 	int i;
1406 
1407 	if (WARN_ON(!num_nh))
1408 		return ERR_PTR(-EINVAL);
1409 
1410 	nh = nexthop_alloc();
1411 	if (!nh)
1412 		return ERR_PTR(-ENOMEM);
1413 
1414 	nh->is_group = 1;
1415 
1416 	nhg = nexthop_grp_alloc(num_nh);
1417 	if (!nhg) {
1418 		kfree(nh);
1419 		return ERR_PTR(-ENOMEM);
1420 	}
1421 
1422 	/* spare group used for removals */
1423 	nhg->spare = nexthop_grp_alloc(num_nh);
1424 	if (!nhg->spare) {
1425 		kfree(nhg);
1426 		kfree(nh);
1427 		return ERR_PTR(-ENOMEM);
1428 	}
1429 	nhg->spare->spare = nhg;
1430 
1431 	for (i = 0; i < nhg->num_nh; ++i) {
1432 		struct nexthop *nhe;
1433 		struct nh_info *nhi;
1434 
1435 		nhe = nexthop_find_by_id(net, entry[i].id);
1436 		if (!nexthop_get(nhe))
1437 			goto out_no_nh;
1438 
1439 		nhi = rtnl_dereference(nhe->nh_info);
1440 		if (nhi->family == AF_INET)
1441 			nhg->has_v4 = true;
1442 
1443 		nhg->nh_entries[i].nh = nhe;
1444 		nhg->nh_entries[i].weight = entry[i].weight + 1;
1445 		list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);
1446 		nhg->nh_entries[i].nh_parent = nh;
1447 	}
1448 
1449 	if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) {
1450 		nhg->mpath = 1;
1451 		nh_group_rebalance(nhg);
1452 	}
1453 
1454 	if (cfg->nh_fdb)
1455 		nhg->fdb_nh = 1;
1456 
1457 	rcu_assign_pointer(nh->nh_grp, nhg);
1458 
1459 	return nh;
1460 
1461 out_no_nh:
1462 	for (i--; i >= 0; --i) {
1463 		list_del(&nhg->nh_entries[i].nh_list);
1464 		nexthop_put(nhg->nh_entries[i].nh);
1465 	}
1466 
1467 	kfree(nhg->spare);
1468 	kfree(nhg);
1469 	kfree(nh);
1470 
1471 	return ERR_PTR(-ENOENT);
1472 }
1473 
1474 static int nh_create_ipv4(struct net *net, struct nexthop *nh,
1475 			  struct nh_info *nhi, struct nh_config *cfg,
1476 			  struct netlink_ext_ack *extack)
1477 {
1478 	struct fib_nh *fib_nh = &nhi->fib_nh;
1479 	struct fib_config fib_cfg = {
1480 		.fc_oif   = cfg->nh_ifindex,
1481 		.fc_gw4   = cfg->gw.ipv4,
1482 		.fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0,
1483 		.fc_flags = cfg->nh_flags,
1484 		.fc_encap = cfg->nh_encap,
1485 		.fc_encap_type = cfg->nh_encap_type,
1486 	};
1487 	u32 tb_id = (cfg->dev ? l3mdev_fib_table(cfg->dev) : RT_TABLE_MAIN);
1488 	int err;
1489 
1490 	err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack);
1491 	if (err) {
1492 		fib_nh_release(net, fib_nh);
1493 		goto out;
1494 	}
1495 
1496 	if (nhi->fdb_nh)
1497 		goto out;
1498 
1499 	/* sets nh_dev if successful */
1500 	err = fib_check_nh(net, fib_nh, tb_id, 0, extack);
1501 	if (!err) {
1502 		nh->nh_flags = fib_nh->fib_nh_flags;
1503 		fib_info_update_nhc_saddr(net, &fib_nh->nh_common,
1504 					  fib_nh->fib_nh_scope);
1505 	} else {
1506 		fib_nh_release(net, fib_nh);
1507 	}
1508 out:
1509 	return err;
1510 }
1511 
1512 static int nh_create_ipv6(struct net *net,  struct nexthop *nh,
1513 			  struct nh_info *nhi, struct nh_config *cfg,
1514 			  struct netlink_ext_ack *extack)
1515 {
1516 	struct fib6_nh *fib6_nh = &nhi->fib6_nh;
1517 	struct fib6_config fib6_cfg = {
1518 		.fc_table = l3mdev_fib_table(cfg->dev),
1519 		.fc_ifindex = cfg->nh_ifindex,
1520 		.fc_gateway = cfg->gw.ipv6,
1521 		.fc_flags = cfg->nh_flags,
1522 		.fc_encap = cfg->nh_encap,
1523 		.fc_encap_type = cfg->nh_encap_type,
1524 		.fc_is_fdb = cfg->nh_fdb,
1525 	};
1526 	int err;
1527 
1528 	if (!ipv6_addr_any(&cfg->gw.ipv6))
1529 		fib6_cfg.fc_flags |= RTF_GATEWAY;
1530 
1531 	/* sets nh_dev if successful */
1532 	err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL,
1533 				      extack);
1534 	if (err)
1535 		ipv6_stub->fib6_nh_release(fib6_nh);
1536 	else
1537 		nh->nh_flags = fib6_nh->fib_nh_flags;
1538 
1539 	return err;
1540 }
1541 
1542 static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg,
1543 				      struct netlink_ext_ack *extack)
1544 {
1545 	struct nh_info *nhi;
1546 	struct nexthop *nh;
1547 	int err = 0;
1548 
1549 	nh = nexthop_alloc();
1550 	if (!nh)
1551 		return ERR_PTR(-ENOMEM);
1552 
1553 	nhi = kzalloc(sizeof(*nhi), GFP_KERNEL);
1554 	if (!nhi) {
1555 		kfree(nh);
1556 		return ERR_PTR(-ENOMEM);
1557 	}
1558 
1559 	nh->nh_flags = cfg->nh_flags;
1560 	nh->net = net;
1561 
1562 	nhi->nh_parent = nh;
1563 	nhi->family = cfg->nh_family;
1564 	nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK;
1565 
1566 	if (cfg->nh_fdb)
1567 		nhi->fdb_nh = 1;
1568 
1569 	if (cfg->nh_blackhole) {
1570 		nhi->reject_nh = 1;
1571 		cfg->nh_ifindex = net->loopback_dev->ifindex;
1572 	}
1573 
1574 	switch (cfg->nh_family) {
1575 	case AF_INET:
1576 		err = nh_create_ipv4(net, nh, nhi, cfg, extack);
1577 		break;
1578 	case AF_INET6:
1579 		err = nh_create_ipv6(net, nh, nhi, cfg, extack);
1580 		break;
1581 	}
1582 
1583 	if (err) {
1584 		kfree(nhi);
1585 		kfree(nh);
1586 		return ERR_PTR(err);
1587 	}
1588 
1589 	/* add the entry to the device based hash */
1590 	if (!nhi->fdb_nh)
1591 		nexthop_devhash_add(net, nhi);
1592 
1593 	rcu_assign_pointer(nh->nh_info, nhi);
1594 
1595 	return nh;
1596 }
1597 
1598 /* called with rtnl lock held */
1599 static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg,
1600 				   struct netlink_ext_ack *extack)
1601 {
1602 	struct nexthop *nh;
1603 	int err;
1604 
1605 	if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) {
1606 		NL_SET_ERR_MSG(extack, "Replace requires nexthop id");
1607 		return ERR_PTR(-EINVAL);
1608 	}
1609 
1610 	if (!cfg->nh_id) {
1611 		cfg->nh_id = nh_find_unused_id(net);
1612 		if (!cfg->nh_id) {
1613 			NL_SET_ERR_MSG(extack, "No unused id");
1614 			return ERR_PTR(-EINVAL);
1615 		}
1616 	}
1617 
1618 	if (cfg->nh_grp)
1619 		nh = nexthop_create_group(net, cfg);
1620 	else
1621 		nh = nexthop_create(net, cfg, extack);
1622 
1623 	if (IS_ERR(nh))
1624 		return nh;
1625 
1626 	refcount_set(&nh->refcnt, 1);
1627 	nh->id = cfg->nh_id;
1628 	nh->protocol = cfg->nh_protocol;
1629 	nh->net = net;
1630 
1631 	err = insert_nexthop(net, nh, cfg, extack);
1632 	if (err) {
1633 		__remove_nexthop(net, nh, NULL);
1634 		nexthop_put(nh);
1635 		nh = ERR_PTR(err);
1636 	}
1637 
1638 	return nh;
1639 }
1640 
1641 static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
1642 			    struct nlmsghdr *nlh, struct nh_config *cfg,
1643 			    struct netlink_ext_ack *extack)
1644 {
1645 	struct nhmsg *nhm = nlmsg_data(nlh);
1646 	struct nlattr *tb[NHA_MAX + 1];
1647 	int err;
1648 
1649 	err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1650 			  extack);
1651 	if (err < 0)
1652 		return err;
1653 
1654 	err = -EINVAL;
1655 	if (nhm->resvd || nhm->nh_scope) {
1656 		NL_SET_ERR_MSG(extack, "Invalid values in ancillary header");
1657 		goto out;
1658 	}
1659 	if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) {
1660 		NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header");
1661 		goto out;
1662 	}
1663 
1664 	switch (nhm->nh_family) {
1665 	case AF_INET:
1666 	case AF_INET6:
1667 		break;
1668 	case AF_UNSPEC:
1669 		if (tb[NHA_GROUP])
1670 			break;
1671 		fallthrough;
1672 	default:
1673 		NL_SET_ERR_MSG(extack, "Invalid address family");
1674 		goto out;
1675 	}
1676 
1677 	if (tb[NHA_GROUPS] || tb[NHA_MASTER]) {
1678 		NL_SET_ERR_MSG(extack, "Invalid attributes in request");
1679 		goto out;
1680 	}
1681 
1682 	memset(cfg, 0, sizeof(*cfg));
1683 	cfg->nlflags = nlh->nlmsg_flags;
1684 	cfg->nlinfo.portid = NETLINK_CB(skb).portid;
1685 	cfg->nlinfo.nlh = nlh;
1686 	cfg->nlinfo.nl_net = net;
1687 
1688 	cfg->nh_family = nhm->nh_family;
1689 	cfg->nh_protocol = nhm->nh_protocol;
1690 	cfg->nh_flags = nhm->nh_flags;
1691 
1692 	if (tb[NHA_ID])
1693 		cfg->nh_id = nla_get_u32(tb[NHA_ID]);
1694 
1695 	if (tb[NHA_FDB]) {
1696 		if (tb[NHA_OIF] || tb[NHA_BLACKHOLE] ||
1697 		    tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE]) {
1698 			NL_SET_ERR_MSG(extack, "Fdb attribute can not be used with encap, oif or blackhole");
1699 			goto out;
1700 		}
1701 		if (nhm->nh_flags) {
1702 			NL_SET_ERR_MSG(extack, "Unsupported nexthop flags in ancillary header");
1703 			goto out;
1704 		}
1705 		cfg->nh_fdb = nla_get_flag(tb[NHA_FDB]);
1706 	}
1707 
1708 	if (tb[NHA_GROUP]) {
1709 		if (nhm->nh_family != AF_UNSPEC) {
1710 			NL_SET_ERR_MSG(extack, "Invalid family for group");
1711 			goto out;
1712 		}
1713 		cfg->nh_grp = tb[NHA_GROUP];
1714 
1715 		cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH;
1716 		if (tb[NHA_GROUP_TYPE])
1717 			cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]);
1718 
1719 		if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) {
1720 			NL_SET_ERR_MSG(extack, "Invalid group type");
1721 			goto out;
1722 		}
1723 		err = nh_check_attr_group(net, tb, extack);
1724 
1725 		/* no other attributes should be set */
1726 		goto out;
1727 	}
1728 
1729 	if (tb[NHA_BLACKHOLE]) {
1730 		if (tb[NHA_GATEWAY] || tb[NHA_OIF] ||
1731 		    tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE] || tb[NHA_FDB]) {
1732 			NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway, oif, encap or fdb");
1733 			goto out;
1734 		}
1735 
1736 		cfg->nh_blackhole = 1;
1737 		err = 0;
1738 		goto out;
1739 	}
1740 
1741 	if (!cfg->nh_fdb && !tb[NHA_OIF]) {
1742 		NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole and non-fdb nexthops");
1743 		goto out;
1744 	}
1745 
1746 	if (!cfg->nh_fdb && tb[NHA_OIF]) {
1747 		cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]);
1748 		if (cfg->nh_ifindex)
1749 			cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex);
1750 
1751 		if (!cfg->dev) {
1752 			NL_SET_ERR_MSG(extack, "Invalid device index");
1753 			goto out;
1754 		} else if (!(cfg->dev->flags & IFF_UP)) {
1755 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1756 			err = -ENETDOWN;
1757 			goto out;
1758 		} else if (!netif_carrier_ok(cfg->dev)) {
1759 			NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down");
1760 			err = -ENETDOWN;
1761 			goto out;
1762 		}
1763 	}
1764 
1765 	err = -EINVAL;
1766 	if (tb[NHA_GATEWAY]) {
1767 		struct nlattr *gwa = tb[NHA_GATEWAY];
1768 
1769 		switch (cfg->nh_family) {
1770 		case AF_INET:
1771 			if (nla_len(gwa) != sizeof(u32)) {
1772 				NL_SET_ERR_MSG(extack, "Invalid gateway");
1773 				goto out;
1774 			}
1775 			cfg->gw.ipv4 = nla_get_be32(gwa);
1776 			break;
1777 		case AF_INET6:
1778 			if (nla_len(gwa) != sizeof(struct in6_addr)) {
1779 				NL_SET_ERR_MSG(extack, "Invalid gateway");
1780 				goto out;
1781 			}
1782 			cfg->gw.ipv6 = nla_get_in6_addr(gwa);
1783 			break;
1784 		default:
1785 			NL_SET_ERR_MSG(extack,
1786 				       "Unknown address family for gateway");
1787 			goto out;
1788 		}
1789 	} else {
1790 		/* device only nexthop (no gateway) */
1791 		if (cfg->nh_flags & RTNH_F_ONLINK) {
1792 			NL_SET_ERR_MSG(extack,
1793 				       "ONLINK flag can not be set for nexthop without a gateway");
1794 			goto out;
1795 		}
1796 	}
1797 
1798 	if (tb[NHA_ENCAP]) {
1799 		cfg->nh_encap = tb[NHA_ENCAP];
1800 
1801 		if (!tb[NHA_ENCAP_TYPE]) {
1802 			NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing");
1803 			goto out;
1804 		}
1805 
1806 		cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]);
1807 		err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack);
1808 		if (err < 0)
1809 			goto out;
1810 
1811 	} else if (tb[NHA_ENCAP_TYPE]) {
1812 		NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing");
1813 		goto out;
1814 	}
1815 
1816 
1817 	err = 0;
1818 out:
1819 	return err;
1820 }
1821 
1822 /* rtnl */
1823 static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1824 			   struct netlink_ext_ack *extack)
1825 {
1826 	struct net *net = sock_net(skb->sk);
1827 	struct nh_config cfg;
1828 	struct nexthop *nh;
1829 	int err;
1830 
1831 	err = rtm_to_nh_config(net, skb, nlh, &cfg, extack);
1832 	if (!err) {
1833 		nh = nexthop_add(net, &cfg, extack);
1834 		if (IS_ERR(nh))
1835 			err = PTR_ERR(nh);
1836 	}
1837 
1838 	return err;
1839 }
1840 
1841 static int nh_valid_get_del_req(struct nlmsghdr *nlh, u32 *id,
1842 				struct netlink_ext_ack *extack)
1843 {
1844 	struct nhmsg *nhm = nlmsg_data(nlh);
1845 	struct nlattr *tb[NHA_MAX + 1];
1846 	int err, i;
1847 
1848 	err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1849 			  extack);
1850 	if (err < 0)
1851 		return err;
1852 
1853 	err = -EINVAL;
1854 	for (i = 0; i < __NHA_MAX; ++i) {
1855 		if (!tb[i])
1856 			continue;
1857 
1858 		switch (i) {
1859 		case NHA_ID:
1860 			break;
1861 		default:
1862 			NL_SET_ERR_MSG_ATTR(extack, tb[i],
1863 					    "Unexpected attribute in request");
1864 			goto out;
1865 		}
1866 	}
1867 	if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1868 		NL_SET_ERR_MSG(extack, "Invalid values in header");
1869 		goto out;
1870 	}
1871 
1872 	if (!tb[NHA_ID]) {
1873 		NL_SET_ERR_MSG(extack, "Nexthop id is missing");
1874 		goto out;
1875 	}
1876 
1877 	*id = nla_get_u32(tb[NHA_ID]);
1878 	if (!(*id))
1879 		NL_SET_ERR_MSG(extack, "Invalid nexthop id");
1880 	else
1881 		err = 0;
1882 out:
1883 	return err;
1884 }
1885 
1886 /* rtnl */
1887 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1888 			   struct netlink_ext_ack *extack)
1889 {
1890 	struct net *net = sock_net(skb->sk);
1891 	struct nl_info nlinfo = {
1892 		.nlh = nlh,
1893 		.nl_net = net,
1894 		.portid = NETLINK_CB(skb).portid,
1895 	};
1896 	struct nexthop *nh;
1897 	int err;
1898 	u32 id;
1899 
1900 	err = nh_valid_get_del_req(nlh, &id, extack);
1901 	if (err)
1902 		return err;
1903 
1904 	nh = nexthop_find_by_id(net, id);
1905 	if (!nh)
1906 		return -ENOENT;
1907 
1908 	remove_nexthop(net, nh, &nlinfo);
1909 
1910 	return 0;
1911 }
1912 
1913 /* rtnl */
1914 static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1915 			   struct netlink_ext_ack *extack)
1916 {
1917 	struct net *net = sock_net(in_skb->sk);
1918 	struct sk_buff *skb = NULL;
1919 	struct nexthop *nh;
1920 	int err;
1921 	u32 id;
1922 
1923 	err = nh_valid_get_del_req(nlh, &id, extack);
1924 	if (err)
1925 		return err;
1926 
1927 	err = -ENOBUFS;
1928 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1929 	if (!skb)
1930 		goto out;
1931 
1932 	err = -ENOENT;
1933 	nh = nexthop_find_by_id(net, id);
1934 	if (!nh)
1935 		goto errout_free;
1936 
1937 	err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid,
1938 			   nlh->nlmsg_seq, 0);
1939 	if (err < 0) {
1940 		WARN_ON(err == -EMSGSIZE);
1941 		goto errout_free;
1942 	}
1943 
1944 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1945 out:
1946 	return err;
1947 errout_free:
1948 	kfree_skb(skb);
1949 	goto out;
1950 }
1951 
1952 static bool nh_dump_filtered(struct nexthop *nh, int dev_idx, int master_idx,
1953 			     bool group_filter, u8 family)
1954 {
1955 	const struct net_device *dev;
1956 	const struct nh_info *nhi;
1957 
1958 	if (group_filter && !nh->is_group)
1959 		return true;
1960 
1961 	if (!dev_idx && !master_idx && !family)
1962 		return false;
1963 
1964 	if (nh->is_group)
1965 		return true;
1966 
1967 	nhi = rtnl_dereference(nh->nh_info);
1968 	if (family && nhi->family != family)
1969 		return true;
1970 
1971 	dev = nhi->fib_nhc.nhc_dev;
1972 	if (dev_idx && (!dev || dev->ifindex != dev_idx))
1973 		return true;
1974 
1975 	if (master_idx) {
1976 		struct net_device *master;
1977 
1978 		if (!dev)
1979 			return true;
1980 
1981 		master = netdev_master_upper_dev_get((struct net_device *)dev);
1982 		if (!master || master->ifindex != master_idx)
1983 			return true;
1984 	}
1985 
1986 	return false;
1987 }
1988 
1989 static int nh_valid_dump_req(const struct nlmsghdr *nlh, int *dev_idx,
1990 			     int *master_idx, bool *group_filter,
1991 			     bool *fdb_filter, struct netlink_callback *cb)
1992 {
1993 	struct netlink_ext_ack *extack = cb->extack;
1994 	struct nlattr *tb[NHA_MAX + 1];
1995 	struct nhmsg *nhm;
1996 	int err, i;
1997 	u32 idx;
1998 
1999 	err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
2000 			  NULL);
2001 	if (err < 0)
2002 		return err;
2003 
2004 	for (i = 0; i <= NHA_MAX; ++i) {
2005 		if (!tb[i])
2006 			continue;
2007 
2008 		switch (i) {
2009 		case NHA_OIF:
2010 			idx = nla_get_u32(tb[i]);
2011 			if (idx > INT_MAX) {
2012 				NL_SET_ERR_MSG(extack, "Invalid device index");
2013 				return -EINVAL;
2014 			}
2015 			*dev_idx = idx;
2016 			break;
2017 		case NHA_MASTER:
2018 			idx = nla_get_u32(tb[i]);
2019 			if (idx > INT_MAX) {
2020 				NL_SET_ERR_MSG(extack, "Invalid master device index");
2021 				return -EINVAL;
2022 			}
2023 			*master_idx = idx;
2024 			break;
2025 		case NHA_GROUPS:
2026 			*group_filter = true;
2027 			break;
2028 		case NHA_FDB:
2029 			*fdb_filter = true;
2030 			break;
2031 		default:
2032 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
2033 			return -EINVAL;
2034 		}
2035 	}
2036 
2037 	nhm = nlmsg_data(nlh);
2038 	if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
2039 		NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request");
2040 		return -EINVAL;
2041 	}
2042 
2043 	return 0;
2044 }
2045 
2046 /* rtnl */
2047 static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb)
2048 {
2049 	bool group_filter = false, fdb_filter = false;
2050 	struct nhmsg *nhm = nlmsg_data(cb->nlh);
2051 	int dev_filter_idx = 0, master_idx = 0;
2052 	struct net *net = sock_net(skb->sk);
2053 	struct rb_root *root = &net->nexthop.rb_root;
2054 	struct rb_node *node;
2055 	int idx = 0, s_idx;
2056 	int err;
2057 
2058 	err = nh_valid_dump_req(cb->nlh, &dev_filter_idx, &master_idx,
2059 				&group_filter, &fdb_filter, cb);
2060 	if (err < 0)
2061 		return err;
2062 
2063 	s_idx = cb->args[0];
2064 	for (node = rb_first(root); node; node = rb_next(node)) {
2065 		struct nexthop *nh;
2066 
2067 		if (idx < s_idx)
2068 			goto cont;
2069 
2070 		nh = rb_entry(node, struct nexthop, rb_node);
2071 		if (nh_dump_filtered(nh, dev_filter_idx, master_idx,
2072 				     group_filter, nhm->nh_family))
2073 			goto cont;
2074 
2075 		err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP,
2076 				   NETLINK_CB(cb->skb).portid,
2077 				   cb->nlh->nlmsg_seq, NLM_F_MULTI);
2078 		if (err < 0) {
2079 			if (likely(skb->len))
2080 				goto out;
2081 
2082 			goto out_err;
2083 		}
2084 cont:
2085 		idx++;
2086 	}
2087 
2088 out:
2089 	err = skb->len;
2090 out_err:
2091 	cb->args[0] = idx;
2092 	cb->seq = net->nexthop.seq;
2093 	nl_dump_check_consistent(cb, nlmsg_hdr(skb));
2094 
2095 	return err;
2096 }
2097 
2098 static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu)
2099 {
2100 	unsigned int hash = nh_dev_hashfn(dev->ifindex);
2101 	struct net *net = dev_net(dev);
2102 	struct hlist_head *head = &net->nexthop.devhash[hash];
2103 	struct hlist_node *n;
2104 	struct nh_info *nhi;
2105 
2106 	hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
2107 		if (nhi->fib_nhc.nhc_dev == dev) {
2108 			if (nhi->family == AF_INET)
2109 				fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu,
2110 						   orig_mtu);
2111 		}
2112 	}
2113 }
2114 
2115 /* rtnl */
2116 static int nh_netdev_event(struct notifier_block *this,
2117 			   unsigned long event, void *ptr)
2118 {
2119 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2120 	struct netdev_notifier_info_ext *info_ext;
2121 
2122 	switch (event) {
2123 	case NETDEV_DOWN:
2124 	case NETDEV_UNREGISTER:
2125 		nexthop_flush_dev(dev);
2126 		break;
2127 	case NETDEV_CHANGE:
2128 		if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP)))
2129 			nexthop_flush_dev(dev);
2130 		break;
2131 	case NETDEV_CHANGEMTU:
2132 		info_ext = ptr;
2133 		nexthop_sync_mtu(dev, info_ext->ext.mtu);
2134 		rt_cache_flush(dev_net(dev));
2135 		break;
2136 	}
2137 	return NOTIFY_DONE;
2138 }
2139 
2140 static struct notifier_block nh_netdev_notifier = {
2141 	.notifier_call = nh_netdev_event,
2142 };
2143 
2144 static int nexthops_dump(struct net *net, struct notifier_block *nb,
2145 			 struct netlink_ext_ack *extack)
2146 {
2147 	struct rb_root *root = &net->nexthop.rb_root;
2148 	struct rb_node *node;
2149 	int err = 0;
2150 
2151 	for (node = rb_first(root); node; node = rb_next(node)) {
2152 		struct nexthop *nh;
2153 
2154 		nh = rb_entry(node, struct nexthop, rb_node);
2155 		err = call_nexthop_notifier(nb, net, NEXTHOP_EVENT_REPLACE, nh,
2156 					    extack);
2157 		if (err)
2158 			break;
2159 	}
2160 
2161 	return err;
2162 }
2163 
2164 int register_nexthop_notifier(struct net *net, struct notifier_block *nb,
2165 			      struct netlink_ext_ack *extack)
2166 {
2167 	int err;
2168 
2169 	rtnl_lock();
2170 	err = nexthops_dump(net, nb, extack);
2171 	if (err)
2172 		goto unlock;
2173 	err = blocking_notifier_chain_register(&net->nexthop.notifier_chain,
2174 					       nb);
2175 unlock:
2176 	rtnl_unlock();
2177 	return err;
2178 }
2179 EXPORT_SYMBOL(register_nexthop_notifier);
2180 
2181 int unregister_nexthop_notifier(struct net *net, struct notifier_block *nb)
2182 {
2183 	return blocking_notifier_chain_unregister(&net->nexthop.notifier_chain,
2184 						  nb);
2185 }
2186 EXPORT_SYMBOL(unregister_nexthop_notifier);
2187 
2188 void nexthop_set_hw_flags(struct net *net, u32 id, bool offload, bool trap)
2189 {
2190 	struct nexthop *nexthop;
2191 
2192 	rcu_read_lock();
2193 
2194 	nexthop = nexthop_find_by_id(net, id);
2195 	if (!nexthop)
2196 		goto out;
2197 
2198 	nexthop->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP);
2199 	if (offload)
2200 		nexthop->nh_flags |= RTNH_F_OFFLOAD;
2201 	if (trap)
2202 		nexthop->nh_flags |= RTNH_F_TRAP;
2203 
2204 out:
2205 	rcu_read_unlock();
2206 }
2207 EXPORT_SYMBOL(nexthop_set_hw_flags);
2208 
2209 static void __net_exit nexthop_net_exit(struct net *net)
2210 {
2211 	rtnl_lock();
2212 	flush_all_nexthops(net);
2213 	rtnl_unlock();
2214 	kfree(net->nexthop.devhash);
2215 }
2216 
2217 static int __net_init nexthop_net_init(struct net *net)
2218 {
2219 	size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE;
2220 
2221 	net->nexthop.rb_root = RB_ROOT;
2222 	net->nexthop.devhash = kzalloc(sz, GFP_KERNEL);
2223 	if (!net->nexthop.devhash)
2224 		return -ENOMEM;
2225 	BLOCKING_INIT_NOTIFIER_HEAD(&net->nexthop.notifier_chain);
2226 
2227 	return 0;
2228 }
2229 
2230 static struct pernet_operations nexthop_net_ops = {
2231 	.init = nexthop_net_init,
2232 	.exit = nexthop_net_exit,
2233 };
2234 
2235 static int __init nexthop_init(void)
2236 {
2237 	register_pernet_subsys(&nexthop_net_ops);
2238 
2239 	register_netdevice_notifier(&nh_netdev_notifier);
2240 
2241 	rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
2242 	rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0);
2243 	rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop,
2244 		      rtm_dump_nexthop, 0);
2245 
2246 	rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
2247 	rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
2248 
2249 	rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
2250 	rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
2251 
2252 	return 0;
2253 }
2254 subsys_initcall(nexthop_init);
2255