xref: /linux/net/ipv4/nexthop.c (revision 7bb377107c72a40ab7505341f8626c8eb79a0cb7)
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 };
37 
38 static unsigned int nh_dev_hashfn(unsigned int val)
39 {
40 	unsigned int mask = NH_DEV_HASHSIZE - 1;
41 
42 	return (val ^
43 		(val >> NH_DEV_HASHBITS) ^
44 		(val >> (NH_DEV_HASHBITS * 2))) & mask;
45 }
46 
47 static void nexthop_devhash_add(struct net *net, struct nh_info *nhi)
48 {
49 	struct net_device *dev = nhi->fib_nhc.nhc_dev;
50 	struct hlist_head *head;
51 	unsigned int hash;
52 
53 	WARN_ON(!dev);
54 
55 	hash = nh_dev_hashfn(dev->ifindex);
56 	head = &net->nexthop.devhash[hash];
57 	hlist_add_head(&nhi->dev_hash, head);
58 }
59 
60 static void nexthop_free_mpath(struct nexthop *nh)
61 {
62 	struct nh_group *nhg;
63 	int i;
64 
65 	nhg = rcu_dereference_raw(nh->nh_grp);
66 	for (i = 0; i < nhg->num_nh; ++i)
67 		WARN_ON(nhg->nh_entries[i].nh);
68 
69 	kfree(nhg);
70 }
71 
72 static void nexthop_free_single(struct nexthop *nh)
73 {
74 	struct nh_info *nhi;
75 
76 	nhi = rcu_dereference_raw(nh->nh_info);
77 	switch (nhi->family) {
78 	case AF_INET:
79 		fib_nh_release(nh->net, &nhi->fib_nh);
80 		break;
81 	case AF_INET6:
82 		ipv6_stub->fib6_nh_release(&nhi->fib6_nh);
83 		break;
84 	}
85 	kfree(nhi);
86 }
87 
88 void nexthop_free_rcu(struct rcu_head *head)
89 {
90 	struct nexthop *nh = container_of(head, struct nexthop, rcu);
91 
92 	if (nh->is_group)
93 		nexthop_free_mpath(nh);
94 	else
95 		nexthop_free_single(nh);
96 
97 	kfree(nh);
98 }
99 EXPORT_SYMBOL_GPL(nexthop_free_rcu);
100 
101 static struct nexthop *nexthop_alloc(void)
102 {
103 	struct nexthop *nh;
104 
105 	nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL);
106 	if (nh) {
107 		INIT_LIST_HEAD(&nh->fi_list);
108 		INIT_LIST_HEAD(&nh->f6i_list);
109 		INIT_LIST_HEAD(&nh->grp_list);
110 	}
111 	return nh;
112 }
113 
114 static struct nh_group *nexthop_grp_alloc(u16 num_nh)
115 {
116 	size_t sz = offsetof(struct nexthop, nh_grp)
117 		    + sizeof(struct nh_group)
118 		    + sizeof(struct nh_grp_entry) * num_nh;
119 	struct nh_group *nhg;
120 
121 	nhg = kzalloc(sz, GFP_KERNEL);
122 	if (nhg)
123 		nhg->num_nh = num_nh;
124 
125 	return nhg;
126 }
127 
128 static void nh_base_seq_inc(struct net *net)
129 {
130 	while (++net->nexthop.seq == 0)
131 		;
132 }
133 
134 /* no reference taken; rcu lock or rtnl must be held */
135 struct nexthop *nexthop_find_by_id(struct net *net, u32 id)
136 {
137 	struct rb_node **pp, *parent = NULL, *next;
138 
139 	pp = &net->nexthop.rb_root.rb_node;
140 	while (1) {
141 		struct nexthop *nh;
142 
143 		next = rcu_dereference_raw(*pp);
144 		if (!next)
145 			break;
146 		parent = next;
147 
148 		nh = rb_entry(parent, struct nexthop, rb_node);
149 		if (id < nh->id)
150 			pp = &next->rb_left;
151 		else if (id > nh->id)
152 			pp = &next->rb_right;
153 		else
154 			return nh;
155 	}
156 	return NULL;
157 }
158 EXPORT_SYMBOL_GPL(nexthop_find_by_id);
159 
160 /* used for auto id allocation; called with rtnl held */
161 static u32 nh_find_unused_id(struct net *net)
162 {
163 	u32 id_start = net->nexthop.last_id_allocated;
164 
165 	while (1) {
166 		net->nexthop.last_id_allocated++;
167 		if (net->nexthop.last_id_allocated == id_start)
168 			break;
169 
170 		if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated))
171 			return net->nexthop.last_id_allocated;
172 	}
173 	return 0;
174 }
175 
176 static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
177 {
178 	struct nexthop_grp *p;
179 	size_t len = nhg->num_nh * sizeof(*p);
180 	struct nlattr *nla;
181 	u16 group_type = 0;
182 	int i;
183 
184 	if (nhg->mpath)
185 		group_type = NEXTHOP_GRP_TYPE_MPATH;
186 
187 	if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type))
188 		goto nla_put_failure;
189 
190 	nla = nla_reserve(skb, NHA_GROUP, len);
191 	if (!nla)
192 		goto nla_put_failure;
193 
194 	p = nla_data(nla);
195 	for (i = 0; i < nhg->num_nh; ++i) {
196 		p->id = nhg->nh_entries[i].nh->id;
197 		p->weight = nhg->nh_entries[i].weight - 1;
198 		p += 1;
199 	}
200 
201 	return 0;
202 
203 nla_put_failure:
204 	return -EMSGSIZE;
205 }
206 
207 static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh,
208 			int event, u32 portid, u32 seq, unsigned int nlflags)
209 {
210 	struct fib6_nh *fib6_nh;
211 	struct fib_nh *fib_nh;
212 	struct nlmsghdr *nlh;
213 	struct nh_info *nhi;
214 	struct nhmsg *nhm;
215 
216 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
217 	if (!nlh)
218 		return -EMSGSIZE;
219 
220 	nhm = nlmsg_data(nlh);
221 	nhm->nh_family = AF_UNSPEC;
222 	nhm->nh_flags = nh->nh_flags;
223 	nhm->nh_protocol = nh->protocol;
224 	nhm->nh_scope = 0;
225 	nhm->resvd = 0;
226 
227 	if (nla_put_u32(skb, NHA_ID, nh->id))
228 		goto nla_put_failure;
229 
230 	if (nh->is_group) {
231 		struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
232 
233 		if (nla_put_nh_group(skb, nhg))
234 			goto nla_put_failure;
235 		goto out;
236 	}
237 
238 	nhi = rtnl_dereference(nh->nh_info);
239 	nhm->nh_family = nhi->family;
240 	if (nhi->reject_nh) {
241 		if (nla_put_flag(skb, NHA_BLACKHOLE))
242 			goto nla_put_failure;
243 		goto out;
244 	} else {
245 		const struct net_device *dev;
246 
247 		dev = nhi->fib_nhc.nhc_dev;
248 		if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex))
249 			goto nla_put_failure;
250 	}
251 
252 	nhm->nh_scope = nhi->fib_nhc.nhc_scope;
253 	switch (nhi->family) {
254 	case AF_INET:
255 		fib_nh = &nhi->fib_nh;
256 		if (fib_nh->fib_nh_gw_family &&
257 		    nla_put_u32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4))
258 			goto nla_put_failure;
259 		break;
260 
261 	case AF_INET6:
262 		fib6_nh = &nhi->fib6_nh;
263 		if (fib6_nh->fib_nh_gw_family &&
264 		    nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6))
265 			goto nla_put_failure;
266 		break;
267 	}
268 
269 	if (nhi->fib_nhc.nhc_lwtstate &&
270 	    lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate,
271 				NHA_ENCAP, NHA_ENCAP_TYPE) < 0)
272 		goto nla_put_failure;
273 
274 out:
275 	nlmsg_end(skb, nlh);
276 	return 0;
277 
278 nla_put_failure:
279 	return -EMSGSIZE;
280 }
281 
282 static size_t nh_nlmsg_size_grp(struct nexthop *nh)
283 {
284 	struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
285 	size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh;
286 
287 	return nla_total_size(sz) +
288 	       nla_total_size(2);  /* NHA_GROUP_TYPE */
289 }
290 
291 static size_t nh_nlmsg_size_single(struct nexthop *nh)
292 {
293 	struct nh_info *nhi = rtnl_dereference(nh->nh_info);
294 	size_t sz;
295 
296 	/* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE
297 	 * are mutually exclusive
298 	 */
299 	sz = nla_total_size(4);  /* NHA_OIF */
300 
301 	switch (nhi->family) {
302 	case AF_INET:
303 		if (nhi->fib_nh.fib_nh_gw_family)
304 			sz += nla_total_size(4);  /* NHA_GATEWAY */
305 		break;
306 
307 	case AF_INET6:
308 		/* NHA_GATEWAY */
309 		if (nhi->fib6_nh.fib_nh_gw_family)
310 			sz += nla_total_size(sizeof(const struct in6_addr));
311 		break;
312 	}
313 
314 	if (nhi->fib_nhc.nhc_lwtstate) {
315 		sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate);
316 		sz += nla_total_size(2);  /* NHA_ENCAP_TYPE */
317 	}
318 
319 	return sz;
320 }
321 
322 static size_t nh_nlmsg_size(struct nexthop *nh)
323 {
324 	size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg));
325 
326 	sz += nla_total_size(4); /* NHA_ID */
327 
328 	if (nh->is_group)
329 		sz += nh_nlmsg_size_grp(nh);
330 	else
331 		sz += nh_nlmsg_size_single(nh);
332 
333 	return sz;
334 }
335 
336 static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info)
337 {
338 	unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0;
339 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
340 	struct sk_buff *skb;
341 	int err = -ENOBUFS;
342 
343 	skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any());
344 	if (!skb)
345 		goto errout;
346 
347 	err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags);
348 	if (err < 0) {
349 		/* -EMSGSIZE implies BUG in nh_nlmsg_size() */
350 		WARN_ON(err == -EMSGSIZE);
351 		kfree_skb(skb);
352 		goto errout;
353 	}
354 
355 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP,
356 		    info->nlh, gfp_any());
357 	return;
358 errout:
359 	if (err < 0)
360 		rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err);
361 }
362 
363 static bool valid_group_nh(struct nexthop *nh, unsigned int npaths,
364 			   struct netlink_ext_ack *extack)
365 {
366 	if (nh->is_group) {
367 		struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
368 
369 		/* nested multipath (group within a group) is not
370 		 * supported
371 		 */
372 		if (nhg->mpath) {
373 			NL_SET_ERR_MSG(extack,
374 				       "Multipath group can not be a nexthop within a group");
375 			return false;
376 		}
377 	} else {
378 		struct nh_info *nhi = rtnl_dereference(nh->nh_info);
379 
380 		if (nhi->reject_nh && npaths > 1) {
381 			NL_SET_ERR_MSG(extack,
382 				       "Blackhole nexthop can not be used in a group with more than 1 path");
383 			return false;
384 		}
385 	}
386 
387 	return true;
388 }
389 
390 static int nh_check_attr_group(struct net *net, struct nlattr *tb[],
391 			       struct netlink_ext_ack *extack)
392 {
393 	unsigned int len = nla_len(tb[NHA_GROUP]);
394 	struct nexthop_grp *nhg;
395 	unsigned int i, j;
396 
397 	if (len & (sizeof(struct nexthop_grp) - 1)) {
398 		NL_SET_ERR_MSG(extack,
399 			       "Invalid length for nexthop group attribute");
400 		return -EINVAL;
401 	}
402 
403 	/* convert len to number of nexthop ids */
404 	len /= sizeof(*nhg);
405 
406 	nhg = nla_data(tb[NHA_GROUP]);
407 	for (i = 0; i < len; ++i) {
408 		if (nhg[i].resvd1 || nhg[i].resvd2) {
409 			NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0");
410 			return -EINVAL;
411 		}
412 		if (nhg[i].weight > 254) {
413 			NL_SET_ERR_MSG(extack, "Invalid value for weight");
414 			return -EINVAL;
415 		}
416 		for (j = i + 1; j < len; ++j) {
417 			if (nhg[i].id == nhg[j].id) {
418 				NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group");
419 				return -EINVAL;
420 			}
421 		}
422 	}
423 
424 	nhg = nla_data(tb[NHA_GROUP]);
425 	for (i = 0; i < len; ++i) {
426 		struct nexthop *nh;
427 
428 		nh = nexthop_find_by_id(net, nhg[i].id);
429 		if (!nh) {
430 			NL_SET_ERR_MSG(extack, "Invalid nexthop id");
431 			return -EINVAL;
432 		}
433 		if (!valid_group_nh(nh, len, extack))
434 			return -EINVAL;
435 	}
436 	for (i = NHA_GROUP + 1; i < __NHA_MAX; ++i) {
437 		if (!tb[i])
438 			continue;
439 
440 		NL_SET_ERR_MSG(extack,
441 			       "No other attributes can be set in nexthop groups");
442 		return -EINVAL;
443 	}
444 
445 	return 0;
446 }
447 
448 static bool ipv6_good_nh(const struct fib6_nh *nh)
449 {
450 	int state = NUD_REACHABLE;
451 	struct neighbour *n;
452 
453 	rcu_read_lock_bh();
454 
455 	n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6);
456 	if (n)
457 		state = n->nud_state;
458 
459 	rcu_read_unlock_bh();
460 
461 	return !!(state & NUD_VALID);
462 }
463 
464 static bool ipv4_good_nh(const struct fib_nh *nh)
465 {
466 	int state = NUD_REACHABLE;
467 	struct neighbour *n;
468 
469 	rcu_read_lock_bh();
470 
471 	n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
472 				      (__force u32)nh->fib_nh_gw4);
473 	if (n)
474 		state = n->nud_state;
475 
476 	rcu_read_unlock_bh();
477 
478 	return !!(state & NUD_VALID);
479 }
480 
481 struct nexthop *nexthop_select_path(struct nexthop *nh, int hash)
482 {
483 	struct nexthop *rc = NULL;
484 	struct nh_group *nhg;
485 	int i;
486 
487 	if (!nh->is_group)
488 		return nh;
489 
490 	nhg = rcu_dereference(nh->nh_grp);
491 	for (i = 0; i < nhg->num_nh; ++i) {
492 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
493 		struct nh_info *nhi;
494 
495 		if (hash > atomic_read(&nhge->upper_bound))
496 			continue;
497 
498 		/* nexthops always check if it is good and does
499 		 * not rely on a sysctl for this behavior
500 		 */
501 		nhi = rcu_dereference(nhge->nh->nh_info);
502 		switch (nhi->family) {
503 		case AF_INET:
504 			if (ipv4_good_nh(&nhi->fib_nh))
505 				return nhge->nh;
506 			break;
507 		case AF_INET6:
508 			if (ipv6_good_nh(&nhi->fib6_nh))
509 				return nhge->nh;
510 			break;
511 		}
512 
513 		if (!rc)
514 			rc = nhge->nh;
515 	}
516 
517 	return rc;
518 }
519 EXPORT_SYMBOL_GPL(nexthop_select_path);
520 
521 int nexthop_for_each_fib6_nh(struct nexthop *nh,
522 			     int (*cb)(struct fib6_nh *nh, void *arg),
523 			     void *arg)
524 {
525 	struct nh_info *nhi;
526 	int err;
527 
528 	if (nh->is_group) {
529 		struct nh_group *nhg;
530 		int i;
531 
532 		nhg = rcu_dereference_rtnl(nh->nh_grp);
533 		for (i = 0; i < nhg->num_nh; i++) {
534 			struct nh_grp_entry *nhge = &nhg->nh_entries[i];
535 
536 			nhi = rcu_dereference_rtnl(nhge->nh->nh_info);
537 			err = cb(&nhi->fib6_nh, arg);
538 			if (err)
539 				return err;
540 		}
541 	} else {
542 		nhi = rcu_dereference_rtnl(nh->nh_info);
543 		err = cb(&nhi->fib6_nh, arg);
544 		if (err)
545 			return err;
546 	}
547 
548 	return 0;
549 }
550 EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh);
551 
552 static int check_src_addr(const struct in6_addr *saddr,
553 			  struct netlink_ext_ack *extack)
554 {
555 	if (!ipv6_addr_any(saddr)) {
556 		NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects");
557 		return -EINVAL;
558 	}
559 	return 0;
560 }
561 
562 int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
563 		       struct netlink_ext_ack *extack)
564 {
565 	struct nh_info *nhi;
566 
567 	/* fib6_src is unique to a fib6_info and limits the ability to cache
568 	 * routes in fib6_nh within a nexthop that is potentially shared
569 	 * across multiple fib entries. If the config wants to use source
570 	 * routing it can not use nexthop objects. mlxsw also does not allow
571 	 * fib6_src on routes.
572 	 */
573 	if (cfg && check_src_addr(&cfg->fc_src, extack) < 0)
574 		return -EINVAL;
575 
576 	if (nh->is_group) {
577 		struct nh_group *nhg;
578 
579 		nhg = rtnl_dereference(nh->nh_grp);
580 		if (nhg->has_v4)
581 			goto no_v4_nh;
582 	} else {
583 		nhi = rtnl_dereference(nh->nh_info);
584 		if (nhi->family == AF_INET)
585 			goto no_v4_nh;
586 	}
587 
588 	return 0;
589 no_v4_nh:
590 	NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop");
591 	return -EINVAL;
592 }
593 EXPORT_SYMBOL_GPL(fib6_check_nexthop);
594 
595 /* if existing nexthop has ipv6 routes linked to it, need
596  * to verify this new spec works with ipv6
597  */
598 static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new,
599 			      struct netlink_ext_ack *extack)
600 {
601 	struct fib6_info *f6i;
602 
603 	if (list_empty(&old->f6i_list))
604 		return 0;
605 
606 	list_for_each_entry(f6i, &old->f6i_list, nh_list) {
607 		if (check_src_addr(&f6i->fib6_src.addr, extack) < 0)
608 			return -EINVAL;
609 	}
610 
611 	return fib6_check_nexthop(new, NULL, extack);
612 }
613 
614 static int nexthop_check_scope(struct nexthop *nh, u8 scope,
615 			       struct netlink_ext_ack *extack)
616 {
617 	struct nh_info *nhi;
618 
619 	nhi = rtnl_dereference(nh->nh_info);
620 	if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) {
621 		NL_SET_ERR_MSG(extack,
622 			       "Route with host scope can not have a gateway");
623 		return -EINVAL;
624 	}
625 
626 	if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) {
627 		NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop");
628 		return -EINVAL;
629 	}
630 
631 	return 0;
632 }
633 
634 /* Invoked by fib add code to verify nexthop by id is ok with
635  * config for prefix; parts of fib_check_nh not done when nexthop
636  * object is used.
637  */
638 int fib_check_nexthop(struct nexthop *nh, u8 scope,
639 		      struct netlink_ext_ack *extack)
640 {
641 	int err = 0;
642 
643 	if (nh->is_group) {
644 		struct nh_group *nhg;
645 
646 		if (scope == RT_SCOPE_HOST) {
647 			NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops");
648 			err = -EINVAL;
649 			goto out;
650 		}
651 
652 		nhg = rtnl_dereference(nh->nh_grp);
653 		/* all nexthops in a group have the same scope */
654 		err = nexthop_check_scope(nhg->nh_entries[0].nh, scope, extack);
655 	} else {
656 		err = nexthop_check_scope(nh, scope, extack);
657 	}
658 out:
659 	return err;
660 }
661 
662 static int fib_check_nh_list(struct nexthop *old, struct nexthop *new,
663 			     struct netlink_ext_ack *extack)
664 {
665 	struct fib_info *fi;
666 
667 	list_for_each_entry(fi, &old->fi_list, nh_list) {
668 		int err;
669 
670 		err = fib_check_nexthop(new, fi->fib_scope, extack);
671 		if (err)
672 			return err;
673 	}
674 	return 0;
675 }
676 
677 static void nh_group_rebalance(struct nh_group *nhg)
678 {
679 	int total = 0;
680 	int w = 0;
681 	int i;
682 
683 	for (i = 0; i < nhg->num_nh; ++i)
684 		total += nhg->nh_entries[i].weight;
685 
686 	for (i = 0; i < nhg->num_nh; ++i) {
687 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
688 		int upper_bound;
689 
690 		w += nhge->weight;
691 		upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1;
692 		atomic_set(&nhge->upper_bound, upper_bound);
693 	}
694 }
695 
696 static void remove_nh_grp_entry(struct nh_grp_entry *nhge,
697 				struct nh_group *nhg,
698 				struct nl_info *nlinfo)
699 {
700 	struct nexthop *nh = nhge->nh;
701 	struct nh_grp_entry *nhges;
702 	bool found = false;
703 	int i;
704 
705 	WARN_ON(!nh);
706 
707 	nhges = nhg->nh_entries;
708 	for (i = 0; i < nhg->num_nh; ++i) {
709 		if (found) {
710 			nhges[i-1].nh = nhges[i].nh;
711 			nhges[i-1].weight = nhges[i].weight;
712 			list_del(&nhges[i].nh_list);
713 			list_add(&nhges[i-1].nh_list, &nhges[i-1].nh->grp_list);
714 		} else if (nhg->nh_entries[i].nh == nh) {
715 			found = true;
716 		}
717 	}
718 
719 	if (WARN_ON(!found))
720 		return;
721 
722 	nhg->num_nh--;
723 	nhg->nh_entries[nhg->num_nh].nh = NULL;
724 
725 	nh_group_rebalance(nhg);
726 
727 	nexthop_put(nh);
728 
729 	if (nlinfo)
730 		nexthop_notify(RTM_NEWNEXTHOP, nhge->nh_parent, nlinfo);
731 }
732 
733 static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh,
734 				       struct nl_info *nlinfo)
735 {
736 	struct nh_grp_entry *nhge, *tmp;
737 
738 	list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list) {
739 		struct nh_group *nhg;
740 
741 		list_del(&nhge->nh_list);
742 		nhg = rtnl_dereference(nhge->nh_parent->nh_grp);
743 		remove_nh_grp_entry(nhge, nhg, nlinfo);
744 
745 		/* if this group has no more entries then remove it */
746 		if (!nhg->num_nh)
747 			remove_nexthop(net, nhge->nh_parent, nlinfo);
748 	}
749 }
750 
751 static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo)
752 {
753 	struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
754 	int i, num_nh = nhg->num_nh;
755 
756 	for (i = 0; i < num_nh; ++i) {
757 		struct nh_grp_entry *nhge = &nhg->nh_entries[i];
758 
759 		if (WARN_ON(!nhge->nh))
760 			continue;
761 
762 		list_del(&nhge->nh_list);
763 		nexthop_put(nhge->nh);
764 		nhge->nh = NULL;
765 		nhg->num_nh--;
766 	}
767 }
768 
769 /* not called for nexthop replace */
770 static void __remove_nexthop_fib(struct net *net, struct nexthop *nh)
771 {
772 	struct fib6_info *f6i, *tmp;
773 	bool do_flush = false;
774 	struct fib_info *fi;
775 
776 	list_for_each_entry(fi, &nh->fi_list, nh_list) {
777 		fi->fib_flags |= RTNH_F_DEAD;
778 		do_flush = true;
779 	}
780 	if (do_flush)
781 		fib_flush(net);
782 
783 	/* ip6_del_rt removes the entry from this list hence the _safe */
784 	list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) {
785 		/* __ip6_del_rt does a release, so do a hold here */
786 		fib6_info_hold(f6i);
787 		ipv6_stub->ip6_del_rt(net, f6i,
788 				      !net->ipv4.sysctl_nexthop_compat_mode);
789 	}
790 }
791 
792 static void __remove_nexthop(struct net *net, struct nexthop *nh,
793 			     struct nl_info *nlinfo)
794 {
795 	__remove_nexthop_fib(net, nh);
796 
797 	if (nh->is_group) {
798 		remove_nexthop_group(nh, nlinfo);
799 	} else {
800 		struct nh_info *nhi;
801 
802 		nhi = rtnl_dereference(nh->nh_info);
803 		if (nhi->fib_nhc.nhc_dev)
804 			hlist_del(&nhi->dev_hash);
805 
806 		remove_nexthop_from_groups(net, nh, nlinfo);
807 	}
808 }
809 
810 static void remove_nexthop(struct net *net, struct nexthop *nh,
811 			   struct nl_info *nlinfo)
812 {
813 	/* remove from the tree */
814 	rb_erase(&nh->rb_node, &net->nexthop.rb_root);
815 
816 	if (nlinfo)
817 		nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo);
818 
819 	__remove_nexthop(net, nh, nlinfo);
820 	nh_base_seq_inc(net);
821 
822 	nexthop_put(nh);
823 }
824 
825 /* if any FIB entries reference this nexthop, any dst entries
826  * need to be regenerated
827  */
828 static void nh_rt_cache_flush(struct net *net, struct nexthop *nh)
829 {
830 	struct fib6_info *f6i;
831 
832 	if (!list_empty(&nh->fi_list))
833 		rt_cache_flush(net);
834 
835 	list_for_each_entry(f6i, &nh->f6i_list, nh_list)
836 		ipv6_stub->fib6_update_sernum(net, f6i);
837 }
838 
839 static int replace_nexthop_grp(struct net *net, struct nexthop *old,
840 			       struct nexthop *new,
841 			       struct netlink_ext_ack *extack)
842 {
843 	struct nh_group *oldg, *newg;
844 	int i;
845 
846 	if (!new->is_group) {
847 		NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop.");
848 		return -EINVAL;
849 	}
850 
851 	oldg = rtnl_dereference(old->nh_grp);
852 	newg = rtnl_dereference(new->nh_grp);
853 
854 	/* update parents - used by nexthop code for cleanup */
855 	for (i = 0; i < newg->num_nh; i++)
856 		newg->nh_entries[i].nh_parent = old;
857 
858 	rcu_assign_pointer(old->nh_grp, newg);
859 
860 	for (i = 0; i < oldg->num_nh; i++)
861 		oldg->nh_entries[i].nh_parent = new;
862 
863 	rcu_assign_pointer(new->nh_grp, oldg);
864 
865 	return 0;
866 }
867 
868 static int replace_nexthop_single(struct net *net, struct nexthop *old,
869 				  struct nexthop *new,
870 				  struct netlink_ext_ack *extack)
871 {
872 	struct nh_info *oldi, *newi;
873 
874 	if (new->is_group) {
875 		NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
876 		return -EINVAL;
877 	}
878 
879 	oldi = rtnl_dereference(old->nh_info);
880 	newi = rtnl_dereference(new->nh_info);
881 
882 	newi->nh_parent = old;
883 	oldi->nh_parent = new;
884 
885 	old->protocol = new->protocol;
886 	old->nh_flags = new->nh_flags;
887 
888 	rcu_assign_pointer(old->nh_info, newi);
889 	rcu_assign_pointer(new->nh_info, oldi);
890 
891 	return 0;
892 }
893 
894 static void __nexthop_replace_notify(struct net *net, struct nexthop *nh,
895 				     struct nl_info *info)
896 {
897 	struct fib6_info *f6i;
898 
899 	if (!list_empty(&nh->fi_list)) {
900 		struct fib_info *fi;
901 
902 		/* expectation is a few fib_info per nexthop and then
903 		 * a lot of routes per fib_info. So mark the fib_info
904 		 * and then walk the fib tables once
905 		 */
906 		list_for_each_entry(fi, &nh->fi_list, nh_list)
907 			fi->nh_updated = true;
908 
909 		fib_info_notify_update(net, info);
910 
911 		list_for_each_entry(fi, &nh->fi_list, nh_list)
912 			fi->nh_updated = false;
913 	}
914 
915 	list_for_each_entry(f6i, &nh->f6i_list, nh_list)
916 		ipv6_stub->fib6_rt_update(net, f6i, info);
917 }
918 
919 /* send RTM_NEWROUTE with REPLACE flag set for all FIB entries
920  * linked to this nexthop and for all groups that the nexthop
921  * is a member of
922  */
923 static void nexthop_replace_notify(struct net *net, struct nexthop *nh,
924 				   struct nl_info *info)
925 {
926 	struct nh_grp_entry *nhge;
927 
928 	__nexthop_replace_notify(net, nh, info);
929 
930 	list_for_each_entry(nhge, &nh->grp_list, nh_list)
931 		__nexthop_replace_notify(net, nhge->nh_parent, info);
932 }
933 
934 static int replace_nexthop(struct net *net, struct nexthop *old,
935 			   struct nexthop *new, struct netlink_ext_ack *extack)
936 {
937 	bool new_is_reject = false;
938 	struct nh_grp_entry *nhge;
939 	int err;
940 
941 	/* check that existing FIB entries are ok with the
942 	 * new nexthop definition
943 	 */
944 	err = fib_check_nh_list(old, new, extack);
945 	if (err)
946 		return err;
947 
948 	err = fib6_check_nh_list(old, new, extack);
949 	if (err)
950 		return err;
951 
952 	if (!new->is_group) {
953 		struct nh_info *nhi = rtnl_dereference(new->nh_info);
954 
955 		new_is_reject = nhi->reject_nh;
956 	}
957 
958 	list_for_each_entry(nhge, &old->grp_list, nh_list) {
959 		/* if new nexthop is a blackhole, any groups using this
960 		 * nexthop cannot have more than 1 path
961 		 */
962 		if (new_is_reject &&
963 		    nexthop_num_path(nhge->nh_parent) > 1) {
964 			NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path");
965 			return -EINVAL;
966 		}
967 
968 		err = fib_check_nh_list(nhge->nh_parent, new, extack);
969 		if (err)
970 			return err;
971 
972 		err = fib6_check_nh_list(nhge->nh_parent, new, extack);
973 		if (err)
974 			return err;
975 	}
976 
977 	if (old->is_group)
978 		err = replace_nexthop_grp(net, old, new, extack);
979 	else
980 		err = replace_nexthop_single(net, old, new, extack);
981 
982 	if (!err) {
983 		nh_rt_cache_flush(net, old);
984 
985 		__remove_nexthop(net, new, NULL);
986 		nexthop_put(new);
987 	}
988 
989 	return err;
990 }
991 
992 /* called with rtnl_lock held */
993 static int insert_nexthop(struct net *net, struct nexthop *new_nh,
994 			  struct nh_config *cfg, struct netlink_ext_ack *extack)
995 {
996 	struct rb_node **pp, *parent = NULL, *next;
997 	struct rb_root *root = &net->nexthop.rb_root;
998 	bool replace = !!(cfg->nlflags & NLM_F_REPLACE);
999 	bool create = !!(cfg->nlflags & NLM_F_CREATE);
1000 	u32 new_id = new_nh->id;
1001 	int replace_notify = 0;
1002 	int rc = -EEXIST;
1003 
1004 	pp = &root->rb_node;
1005 	while (1) {
1006 		struct nexthop *nh;
1007 
1008 		next = rtnl_dereference(*pp);
1009 		if (!next)
1010 			break;
1011 
1012 		parent = next;
1013 
1014 		nh = rb_entry(parent, struct nexthop, rb_node);
1015 		if (new_id < nh->id) {
1016 			pp = &next->rb_left;
1017 		} else if (new_id > nh->id) {
1018 			pp = &next->rb_right;
1019 		} else if (replace) {
1020 			rc = replace_nexthop(net, nh, new_nh, extack);
1021 			if (!rc) {
1022 				new_nh = nh; /* send notification with old nh */
1023 				replace_notify = 1;
1024 			}
1025 			goto out;
1026 		} else {
1027 			/* id already exists and not a replace */
1028 			goto out;
1029 		}
1030 	}
1031 
1032 	if (replace && !create) {
1033 		NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists");
1034 		rc = -ENOENT;
1035 		goto out;
1036 	}
1037 
1038 	rb_link_node_rcu(&new_nh->rb_node, parent, pp);
1039 	rb_insert_color(&new_nh->rb_node, root);
1040 	rc = 0;
1041 out:
1042 	if (!rc) {
1043 		nh_base_seq_inc(net);
1044 		nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo);
1045 		if (replace_notify && net->ipv4.sysctl_nexthop_compat_mode)
1046 			nexthop_replace_notify(net, new_nh, &cfg->nlinfo);
1047 	}
1048 
1049 	return rc;
1050 }
1051 
1052 /* rtnl */
1053 /* remove all nexthops tied to a device being deleted */
1054 static void nexthop_flush_dev(struct net_device *dev)
1055 {
1056 	unsigned int hash = nh_dev_hashfn(dev->ifindex);
1057 	struct net *net = dev_net(dev);
1058 	struct hlist_head *head = &net->nexthop.devhash[hash];
1059 	struct hlist_node *n;
1060 	struct nh_info *nhi;
1061 
1062 	hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1063 		if (nhi->fib_nhc.nhc_dev != dev)
1064 			continue;
1065 
1066 		remove_nexthop(net, nhi->nh_parent, NULL);
1067 	}
1068 }
1069 
1070 /* rtnl; called when net namespace is deleted */
1071 static void flush_all_nexthops(struct net *net)
1072 {
1073 	struct rb_root *root = &net->nexthop.rb_root;
1074 	struct rb_node *node;
1075 	struct nexthop *nh;
1076 
1077 	while ((node = rb_first(root))) {
1078 		nh = rb_entry(node, struct nexthop, rb_node);
1079 		remove_nexthop(net, nh, NULL);
1080 		cond_resched();
1081 	}
1082 }
1083 
1084 static struct nexthop *nexthop_create_group(struct net *net,
1085 					    struct nh_config *cfg)
1086 {
1087 	struct nlattr *grps_attr = cfg->nh_grp;
1088 	struct nexthop_grp *entry = nla_data(grps_attr);
1089 	struct nh_group *nhg;
1090 	struct nexthop *nh;
1091 	int i;
1092 
1093 	nh = nexthop_alloc();
1094 	if (!nh)
1095 		return ERR_PTR(-ENOMEM);
1096 
1097 	nh->is_group = 1;
1098 
1099 	nhg = nexthop_grp_alloc(nla_len(grps_attr) / sizeof(*entry));
1100 	if (!nhg) {
1101 		kfree(nh);
1102 		return ERR_PTR(-ENOMEM);
1103 	}
1104 
1105 	for (i = 0; i < nhg->num_nh; ++i) {
1106 		struct nexthop *nhe;
1107 		struct nh_info *nhi;
1108 
1109 		nhe = nexthop_find_by_id(net, entry[i].id);
1110 		if (!nexthop_get(nhe))
1111 			goto out_no_nh;
1112 
1113 		nhi = rtnl_dereference(nhe->nh_info);
1114 		if (nhi->family == AF_INET)
1115 			nhg->has_v4 = true;
1116 
1117 		nhg->nh_entries[i].nh = nhe;
1118 		nhg->nh_entries[i].weight = entry[i].weight + 1;
1119 		list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);
1120 		nhg->nh_entries[i].nh_parent = nh;
1121 	}
1122 
1123 	if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) {
1124 		nhg->mpath = 1;
1125 		nh_group_rebalance(nhg);
1126 	}
1127 
1128 	rcu_assign_pointer(nh->nh_grp, nhg);
1129 
1130 	return nh;
1131 
1132 out_no_nh:
1133 	for (; i >= 0; --i)
1134 		nexthop_put(nhg->nh_entries[i].nh);
1135 
1136 	kfree(nhg);
1137 	kfree(nh);
1138 
1139 	return ERR_PTR(-ENOENT);
1140 }
1141 
1142 static int nh_create_ipv4(struct net *net, struct nexthop *nh,
1143 			  struct nh_info *nhi, struct nh_config *cfg,
1144 			  struct netlink_ext_ack *extack)
1145 {
1146 	struct fib_nh *fib_nh = &nhi->fib_nh;
1147 	struct fib_config fib_cfg = {
1148 		.fc_oif   = cfg->nh_ifindex,
1149 		.fc_gw4   = cfg->gw.ipv4,
1150 		.fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0,
1151 		.fc_flags = cfg->nh_flags,
1152 		.fc_encap = cfg->nh_encap,
1153 		.fc_encap_type = cfg->nh_encap_type,
1154 	};
1155 	u32 tb_id = l3mdev_fib_table(cfg->dev);
1156 	int err;
1157 
1158 	err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack);
1159 	if (err) {
1160 		fib_nh_release(net, fib_nh);
1161 		goto out;
1162 	}
1163 
1164 	/* sets nh_dev if successful */
1165 	err = fib_check_nh(net, fib_nh, tb_id, 0, extack);
1166 	if (!err) {
1167 		nh->nh_flags = fib_nh->fib_nh_flags;
1168 		fib_info_update_nhc_saddr(net, &fib_nh->nh_common,
1169 					  fib_nh->fib_nh_scope);
1170 	} else {
1171 		fib_nh_release(net, fib_nh);
1172 	}
1173 out:
1174 	return err;
1175 }
1176 
1177 static int nh_create_ipv6(struct net *net,  struct nexthop *nh,
1178 			  struct nh_info *nhi, struct nh_config *cfg,
1179 			  struct netlink_ext_ack *extack)
1180 {
1181 	struct fib6_nh *fib6_nh = &nhi->fib6_nh;
1182 	struct fib6_config fib6_cfg = {
1183 		.fc_table = l3mdev_fib_table(cfg->dev),
1184 		.fc_ifindex = cfg->nh_ifindex,
1185 		.fc_gateway = cfg->gw.ipv6,
1186 		.fc_flags = cfg->nh_flags,
1187 		.fc_encap = cfg->nh_encap,
1188 		.fc_encap_type = cfg->nh_encap_type,
1189 	};
1190 	int err;
1191 
1192 	if (!ipv6_addr_any(&cfg->gw.ipv6))
1193 		fib6_cfg.fc_flags |= RTF_GATEWAY;
1194 
1195 	/* sets nh_dev if successful */
1196 	err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL,
1197 				      extack);
1198 	if (err)
1199 		ipv6_stub->fib6_nh_release(fib6_nh);
1200 	else
1201 		nh->nh_flags = fib6_nh->fib_nh_flags;
1202 
1203 	return err;
1204 }
1205 
1206 static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg,
1207 				      struct netlink_ext_ack *extack)
1208 {
1209 	struct nh_info *nhi;
1210 	struct nexthop *nh;
1211 	int err = 0;
1212 
1213 	nh = nexthop_alloc();
1214 	if (!nh)
1215 		return ERR_PTR(-ENOMEM);
1216 
1217 	nhi = kzalloc(sizeof(*nhi), GFP_KERNEL);
1218 	if (!nhi) {
1219 		kfree(nh);
1220 		return ERR_PTR(-ENOMEM);
1221 	}
1222 
1223 	nh->nh_flags = cfg->nh_flags;
1224 	nh->net = net;
1225 
1226 	nhi->nh_parent = nh;
1227 	nhi->family = cfg->nh_family;
1228 	nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK;
1229 
1230 	if (cfg->nh_blackhole) {
1231 		nhi->reject_nh = 1;
1232 		cfg->nh_ifindex = net->loopback_dev->ifindex;
1233 	}
1234 
1235 	switch (cfg->nh_family) {
1236 	case AF_INET:
1237 		err = nh_create_ipv4(net, nh, nhi, cfg, extack);
1238 		break;
1239 	case AF_INET6:
1240 		err = nh_create_ipv6(net, nh, nhi, cfg, extack);
1241 		break;
1242 	}
1243 
1244 	if (err) {
1245 		kfree(nhi);
1246 		kfree(nh);
1247 		return ERR_PTR(err);
1248 	}
1249 
1250 	/* add the entry to the device based hash */
1251 	nexthop_devhash_add(net, nhi);
1252 
1253 	rcu_assign_pointer(nh->nh_info, nhi);
1254 
1255 	return nh;
1256 }
1257 
1258 /* called with rtnl lock held */
1259 static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg,
1260 				   struct netlink_ext_ack *extack)
1261 {
1262 	struct nexthop *nh;
1263 	int err;
1264 
1265 	if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) {
1266 		NL_SET_ERR_MSG(extack, "Replace requires nexthop id");
1267 		return ERR_PTR(-EINVAL);
1268 	}
1269 
1270 	if (!cfg->nh_id) {
1271 		cfg->nh_id = nh_find_unused_id(net);
1272 		if (!cfg->nh_id) {
1273 			NL_SET_ERR_MSG(extack, "No unused id");
1274 			return ERR_PTR(-EINVAL);
1275 		}
1276 	}
1277 
1278 	if (cfg->nh_grp)
1279 		nh = nexthop_create_group(net, cfg);
1280 	else
1281 		nh = nexthop_create(net, cfg, extack);
1282 
1283 	if (IS_ERR(nh))
1284 		return nh;
1285 
1286 	refcount_set(&nh->refcnt, 1);
1287 	nh->id = cfg->nh_id;
1288 	nh->protocol = cfg->nh_protocol;
1289 	nh->net = net;
1290 
1291 	err = insert_nexthop(net, nh, cfg, extack);
1292 	if (err) {
1293 		__remove_nexthop(net, nh, NULL);
1294 		nexthop_put(nh);
1295 		nh = ERR_PTR(err);
1296 	}
1297 
1298 	return nh;
1299 }
1300 
1301 static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
1302 			    struct nlmsghdr *nlh, struct nh_config *cfg,
1303 			    struct netlink_ext_ack *extack)
1304 {
1305 	struct nhmsg *nhm = nlmsg_data(nlh);
1306 	struct nlattr *tb[NHA_MAX + 1];
1307 	int err;
1308 
1309 	err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1310 			  extack);
1311 	if (err < 0)
1312 		return err;
1313 
1314 	err = -EINVAL;
1315 	if (nhm->resvd || nhm->nh_scope) {
1316 		NL_SET_ERR_MSG(extack, "Invalid values in ancillary header");
1317 		goto out;
1318 	}
1319 	if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) {
1320 		NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header");
1321 		goto out;
1322 	}
1323 
1324 	switch (nhm->nh_family) {
1325 	case AF_INET:
1326 	case AF_INET6:
1327 		break;
1328 	case AF_UNSPEC:
1329 		if (tb[NHA_GROUP])
1330 			break;
1331 		fallthrough;
1332 	default:
1333 		NL_SET_ERR_MSG(extack, "Invalid address family");
1334 		goto out;
1335 	}
1336 
1337 	if (tb[NHA_GROUPS] || tb[NHA_MASTER]) {
1338 		NL_SET_ERR_MSG(extack, "Invalid attributes in request");
1339 		goto out;
1340 	}
1341 
1342 	memset(cfg, 0, sizeof(*cfg));
1343 	cfg->nlflags = nlh->nlmsg_flags;
1344 	cfg->nlinfo.portid = NETLINK_CB(skb).portid;
1345 	cfg->nlinfo.nlh = nlh;
1346 	cfg->nlinfo.nl_net = net;
1347 
1348 	cfg->nh_family = nhm->nh_family;
1349 	cfg->nh_protocol = nhm->nh_protocol;
1350 	cfg->nh_flags = nhm->nh_flags;
1351 
1352 	if (tb[NHA_ID])
1353 		cfg->nh_id = nla_get_u32(tb[NHA_ID]);
1354 
1355 	if (tb[NHA_GROUP]) {
1356 		if (nhm->nh_family != AF_UNSPEC) {
1357 			NL_SET_ERR_MSG(extack, "Invalid family for group");
1358 			goto out;
1359 		}
1360 		cfg->nh_grp = tb[NHA_GROUP];
1361 
1362 		cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH;
1363 		if (tb[NHA_GROUP_TYPE])
1364 			cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]);
1365 
1366 		if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) {
1367 			NL_SET_ERR_MSG(extack, "Invalid group type");
1368 			goto out;
1369 		}
1370 		err = nh_check_attr_group(net, tb, extack);
1371 
1372 		/* no other attributes should be set */
1373 		goto out;
1374 	}
1375 
1376 	if (tb[NHA_BLACKHOLE]) {
1377 		if (tb[NHA_GATEWAY] || tb[NHA_OIF] ||
1378 		    tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE]) {
1379 			NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway or oif");
1380 			goto out;
1381 		}
1382 
1383 		cfg->nh_blackhole = 1;
1384 		err = 0;
1385 		goto out;
1386 	}
1387 
1388 	if (!tb[NHA_OIF]) {
1389 		NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole nexthops");
1390 		goto out;
1391 	}
1392 
1393 	cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]);
1394 	if (cfg->nh_ifindex)
1395 		cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex);
1396 
1397 	if (!cfg->dev) {
1398 		NL_SET_ERR_MSG(extack, "Invalid device index");
1399 		goto out;
1400 	} else if (!(cfg->dev->flags & IFF_UP)) {
1401 		NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1402 		err = -ENETDOWN;
1403 		goto out;
1404 	} else if (!netif_carrier_ok(cfg->dev)) {
1405 		NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down");
1406 		err = -ENETDOWN;
1407 		goto out;
1408 	}
1409 
1410 	err = -EINVAL;
1411 	if (tb[NHA_GATEWAY]) {
1412 		struct nlattr *gwa = tb[NHA_GATEWAY];
1413 
1414 		switch (cfg->nh_family) {
1415 		case AF_INET:
1416 			if (nla_len(gwa) != sizeof(u32)) {
1417 				NL_SET_ERR_MSG(extack, "Invalid gateway");
1418 				goto out;
1419 			}
1420 			cfg->gw.ipv4 = nla_get_be32(gwa);
1421 			break;
1422 		case AF_INET6:
1423 			if (nla_len(gwa) != sizeof(struct in6_addr)) {
1424 				NL_SET_ERR_MSG(extack, "Invalid gateway");
1425 				goto out;
1426 			}
1427 			cfg->gw.ipv6 = nla_get_in6_addr(gwa);
1428 			break;
1429 		default:
1430 			NL_SET_ERR_MSG(extack,
1431 				       "Unknown address family for gateway");
1432 			goto out;
1433 		}
1434 	} else {
1435 		/* device only nexthop (no gateway) */
1436 		if (cfg->nh_flags & RTNH_F_ONLINK) {
1437 			NL_SET_ERR_MSG(extack,
1438 				       "ONLINK flag can not be set for nexthop without a gateway");
1439 			goto out;
1440 		}
1441 	}
1442 
1443 	if (tb[NHA_ENCAP]) {
1444 		cfg->nh_encap = tb[NHA_ENCAP];
1445 
1446 		if (!tb[NHA_ENCAP_TYPE]) {
1447 			NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing");
1448 			goto out;
1449 		}
1450 
1451 		cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]);
1452 		err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack);
1453 		if (err < 0)
1454 			goto out;
1455 
1456 	} else if (tb[NHA_ENCAP_TYPE]) {
1457 		NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing");
1458 		goto out;
1459 	}
1460 
1461 
1462 	err = 0;
1463 out:
1464 	return err;
1465 }
1466 
1467 /* rtnl */
1468 static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1469 			   struct netlink_ext_ack *extack)
1470 {
1471 	struct net *net = sock_net(skb->sk);
1472 	struct nh_config cfg;
1473 	struct nexthop *nh;
1474 	int err;
1475 
1476 	err = rtm_to_nh_config(net, skb, nlh, &cfg, extack);
1477 	if (!err) {
1478 		nh = nexthop_add(net, &cfg, extack);
1479 		if (IS_ERR(nh))
1480 			err = PTR_ERR(nh);
1481 	}
1482 
1483 	return err;
1484 }
1485 
1486 static int nh_valid_get_del_req(struct nlmsghdr *nlh, u32 *id,
1487 				struct netlink_ext_ack *extack)
1488 {
1489 	struct nhmsg *nhm = nlmsg_data(nlh);
1490 	struct nlattr *tb[NHA_MAX + 1];
1491 	int err, i;
1492 
1493 	err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1494 			  extack);
1495 	if (err < 0)
1496 		return err;
1497 
1498 	err = -EINVAL;
1499 	for (i = 0; i < __NHA_MAX; ++i) {
1500 		if (!tb[i])
1501 			continue;
1502 
1503 		switch (i) {
1504 		case NHA_ID:
1505 			break;
1506 		default:
1507 			NL_SET_ERR_MSG_ATTR(extack, tb[i],
1508 					    "Unexpected attribute in request");
1509 			goto out;
1510 		}
1511 	}
1512 	if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1513 		NL_SET_ERR_MSG(extack, "Invalid values in header");
1514 		goto out;
1515 	}
1516 
1517 	if (!tb[NHA_ID]) {
1518 		NL_SET_ERR_MSG(extack, "Nexthop id is missing");
1519 		goto out;
1520 	}
1521 
1522 	*id = nla_get_u32(tb[NHA_ID]);
1523 	if (!(*id))
1524 		NL_SET_ERR_MSG(extack, "Invalid nexthop id");
1525 	else
1526 		err = 0;
1527 out:
1528 	return err;
1529 }
1530 
1531 /* rtnl */
1532 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1533 			   struct netlink_ext_ack *extack)
1534 {
1535 	struct net *net = sock_net(skb->sk);
1536 	struct nl_info nlinfo = {
1537 		.nlh = nlh,
1538 		.nl_net = net,
1539 		.portid = NETLINK_CB(skb).portid,
1540 	};
1541 	struct nexthop *nh;
1542 	int err;
1543 	u32 id;
1544 
1545 	err = nh_valid_get_del_req(nlh, &id, extack);
1546 	if (err)
1547 		return err;
1548 
1549 	nh = nexthop_find_by_id(net, id);
1550 	if (!nh)
1551 		return -ENOENT;
1552 
1553 	remove_nexthop(net, nh, &nlinfo);
1554 
1555 	return 0;
1556 }
1557 
1558 /* rtnl */
1559 static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1560 			   struct netlink_ext_ack *extack)
1561 {
1562 	struct net *net = sock_net(in_skb->sk);
1563 	struct sk_buff *skb = NULL;
1564 	struct nexthop *nh;
1565 	int err;
1566 	u32 id;
1567 
1568 	err = nh_valid_get_del_req(nlh, &id, extack);
1569 	if (err)
1570 		return err;
1571 
1572 	err = -ENOBUFS;
1573 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1574 	if (!skb)
1575 		goto out;
1576 
1577 	err = -ENOENT;
1578 	nh = nexthop_find_by_id(net, id);
1579 	if (!nh)
1580 		goto errout_free;
1581 
1582 	err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid,
1583 			   nlh->nlmsg_seq, 0);
1584 	if (err < 0) {
1585 		WARN_ON(err == -EMSGSIZE);
1586 		goto errout_free;
1587 	}
1588 
1589 	err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1590 out:
1591 	return err;
1592 errout_free:
1593 	kfree_skb(skb);
1594 	goto out;
1595 }
1596 
1597 static bool nh_dump_filtered(struct nexthop *nh, int dev_idx, int master_idx,
1598 			     bool group_filter, u8 family)
1599 {
1600 	const struct net_device *dev;
1601 	const struct nh_info *nhi;
1602 
1603 	if (group_filter && !nh->is_group)
1604 		return true;
1605 
1606 	if (!dev_idx && !master_idx && !family)
1607 		return false;
1608 
1609 	if (nh->is_group)
1610 		return true;
1611 
1612 	nhi = rtnl_dereference(nh->nh_info);
1613 	if (family && nhi->family != family)
1614 		return true;
1615 
1616 	dev = nhi->fib_nhc.nhc_dev;
1617 	if (dev_idx && (!dev || dev->ifindex != dev_idx))
1618 		return true;
1619 
1620 	if (master_idx) {
1621 		struct net_device *master;
1622 
1623 		if (!dev)
1624 			return true;
1625 
1626 		master = netdev_master_upper_dev_get((struct net_device *)dev);
1627 		if (!master || master->ifindex != master_idx)
1628 			return true;
1629 	}
1630 
1631 	return false;
1632 }
1633 
1634 static int nh_valid_dump_req(const struct nlmsghdr *nlh, int *dev_idx,
1635 			     int *master_idx, bool *group_filter,
1636 			     struct netlink_callback *cb)
1637 {
1638 	struct netlink_ext_ack *extack = cb->extack;
1639 	struct nlattr *tb[NHA_MAX + 1];
1640 	struct nhmsg *nhm;
1641 	int err, i;
1642 	u32 idx;
1643 
1644 	err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1645 			  NULL);
1646 	if (err < 0)
1647 		return err;
1648 
1649 	for (i = 0; i <= NHA_MAX; ++i) {
1650 		if (!tb[i])
1651 			continue;
1652 
1653 		switch (i) {
1654 		case NHA_OIF:
1655 			idx = nla_get_u32(tb[i]);
1656 			if (idx > INT_MAX) {
1657 				NL_SET_ERR_MSG(extack, "Invalid device index");
1658 				return -EINVAL;
1659 			}
1660 			*dev_idx = idx;
1661 			break;
1662 		case NHA_MASTER:
1663 			idx = nla_get_u32(tb[i]);
1664 			if (idx > INT_MAX) {
1665 				NL_SET_ERR_MSG(extack, "Invalid master device index");
1666 				return -EINVAL;
1667 			}
1668 			*master_idx = idx;
1669 			break;
1670 		case NHA_GROUPS:
1671 			*group_filter = true;
1672 			break;
1673 		default:
1674 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
1675 			return -EINVAL;
1676 		}
1677 	}
1678 
1679 	nhm = nlmsg_data(nlh);
1680 	if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1681 		NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request");
1682 		return -EINVAL;
1683 	}
1684 
1685 	return 0;
1686 }
1687 
1688 /* rtnl */
1689 static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb)
1690 {
1691 	struct nhmsg *nhm = nlmsg_data(cb->nlh);
1692 	int dev_filter_idx = 0, master_idx = 0;
1693 	struct net *net = sock_net(skb->sk);
1694 	struct rb_root *root = &net->nexthop.rb_root;
1695 	bool group_filter = false;
1696 	struct rb_node *node;
1697 	int idx = 0, s_idx;
1698 	int err;
1699 
1700 	err = nh_valid_dump_req(cb->nlh, &dev_filter_idx, &master_idx,
1701 				&group_filter, cb);
1702 	if (err < 0)
1703 		return err;
1704 
1705 	s_idx = cb->args[0];
1706 	for (node = rb_first(root); node; node = rb_next(node)) {
1707 		struct nexthop *nh;
1708 
1709 		if (idx < s_idx)
1710 			goto cont;
1711 
1712 		nh = rb_entry(node, struct nexthop, rb_node);
1713 		if (nh_dump_filtered(nh, dev_filter_idx, master_idx,
1714 				     group_filter, nhm->nh_family))
1715 			goto cont;
1716 
1717 		err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP,
1718 				   NETLINK_CB(cb->skb).portid,
1719 				   cb->nlh->nlmsg_seq, NLM_F_MULTI);
1720 		if (err < 0) {
1721 			if (likely(skb->len))
1722 				goto out;
1723 
1724 			goto out_err;
1725 		}
1726 cont:
1727 		idx++;
1728 	}
1729 
1730 out:
1731 	err = skb->len;
1732 out_err:
1733 	cb->args[0] = idx;
1734 	cb->seq = net->nexthop.seq;
1735 	nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1736 
1737 	return err;
1738 }
1739 
1740 static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu)
1741 {
1742 	unsigned int hash = nh_dev_hashfn(dev->ifindex);
1743 	struct net *net = dev_net(dev);
1744 	struct hlist_head *head = &net->nexthop.devhash[hash];
1745 	struct hlist_node *n;
1746 	struct nh_info *nhi;
1747 
1748 	hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1749 		if (nhi->fib_nhc.nhc_dev == dev) {
1750 			if (nhi->family == AF_INET)
1751 				fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu,
1752 						   orig_mtu);
1753 		}
1754 	}
1755 }
1756 
1757 /* rtnl */
1758 static int nh_netdev_event(struct notifier_block *this,
1759 			   unsigned long event, void *ptr)
1760 {
1761 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1762 	struct netdev_notifier_info_ext *info_ext;
1763 
1764 	switch (event) {
1765 	case NETDEV_DOWN:
1766 	case NETDEV_UNREGISTER:
1767 		nexthop_flush_dev(dev);
1768 		break;
1769 	case NETDEV_CHANGE:
1770 		if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP)))
1771 			nexthop_flush_dev(dev);
1772 		break;
1773 	case NETDEV_CHANGEMTU:
1774 		info_ext = ptr;
1775 		nexthop_sync_mtu(dev, info_ext->ext.mtu);
1776 		rt_cache_flush(dev_net(dev));
1777 		break;
1778 	}
1779 	return NOTIFY_DONE;
1780 }
1781 
1782 static struct notifier_block nh_netdev_notifier = {
1783 	.notifier_call = nh_netdev_event,
1784 };
1785 
1786 static void __net_exit nexthop_net_exit(struct net *net)
1787 {
1788 	rtnl_lock();
1789 	flush_all_nexthops(net);
1790 	rtnl_unlock();
1791 	kfree(net->nexthop.devhash);
1792 }
1793 
1794 static int __net_init nexthop_net_init(struct net *net)
1795 {
1796 	size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE;
1797 
1798 	net->nexthop.rb_root = RB_ROOT;
1799 	net->nexthop.devhash = kzalloc(sz, GFP_KERNEL);
1800 	if (!net->nexthop.devhash)
1801 		return -ENOMEM;
1802 
1803 	return 0;
1804 }
1805 
1806 static struct pernet_operations nexthop_net_ops = {
1807 	.init = nexthop_net_init,
1808 	.exit = nexthop_net_exit,
1809 };
1810 
1811 static int __init nexthop_init(void)
1812 {
1813 	register_pernet_subsys(&nexthop_net_ops);
1814 
1815 	register_netdevice_notifier(&nh_netdev_notifier);
1816 
1817 	rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1818 	rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0);
1819 	rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop,
1820 		      rtm_dump_nexthop, 0);
1821 
1822 	rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1823 	rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
1824 
1825 	rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1826 	rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
1827 
1828 	return 0;
1829 }
1830 subsys_initcall(nexthop_init);
1831