xref: /linux/net/ipv4/fib_semantics.c (revision 6bab77ced3ffbce3d6c5b5bcce17da7c8a3f8266)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: semantics.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/uaccess.h>
13 #include <linux/bitops.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
17 #include <linux/mm.h>
18 #include <linux/string.h>
19 #include <linux/socket.h>
20 #include <linux/sockios.h>
21 #include <linux/errno.h>
22 #include <linux/in.h>
23 #include <linux/inet.h>
24 #include <linux/inetdevice.h>
25 #include <linux/netdevice.h>
26 #include <linux/if_arp.h>
27 #include <linux/proc_fs.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/netlink.h>
32 #include <linux/hash.h>
33 #include <linux/nospec.h>
34 
35 #include <net/arp.h>
36 #include <net/inet_dscp.h>
37 #include <net/ip.h>
38 #include <net/protocol.h>
39 #include <net/route.h>
40 #include <net/tcp.h>
41 #include <net/sock.h>
42 #include <net/ip_fib.h>
43 #include <net/ip6_fib.h>
44 #include <net/nexthop.h>
45 #include <net/netlink.h>
46 #include <net/rtnh.h>
47 #include <net/lwtunnel.h>
48 #include <net/fib_notifier.h>
49 #include <net/addrconf.h>
50 
51 #include "fib_lookup.h"
52 
53 /* for_nexthops and change_nexthops only used when nexthop object
54  * is not set in a fib_info. The logic within can reference fib_nh.
55  */
56 #ifdef CONFIG_IP_ROUTE_MULTIPATH
57 
58 #define for_nexthops(fi) {						\
59 	int nhsel; const struct fib_nh *nh;				\
60 	for (nhsel = 0, nh = (fi)->fib_nh;				\
61 	     nhsel < fib_info_num_path((fi));				\
62 	     nh++, nhsel++)
63 
64 #define change_nexthops(fi) {						\
65 	int nhsel; struct fib_nh *nexthop_nh;				\
66 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
67 	     nhsel < fib_info_num_path((fi));				\
68 	     nexthop_nh++, nhsel++)
69 
70 #else /* CONFIG_IP_ROUTE_MULTIPATH */
71 
72 /* Hope, that gcc will optimize it to get rid of dummy loop */
73 
74 #define for_nexthops(fi) {						\
75 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
76 	for (nhsel = 0; nhsel < 1; nhsel++)
77 
78 #define change_nexthops(fi) {						\
79 	int nhsel;							\
80 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
81 	for (nhsel = 0; nhsel < 1; nhsel++)
82 
83 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
84 
85 #define endfor_nexthops(fi) }
86 
87 
88 const struct fib_prop fib_props[RTN_MAX + 1] = {
89 	[RTN_UNSPEC] = {
90 		.error	= 0,
91 		.scope	= RT_SCOPE_NOWHERE,
92 	},
93 	[RTN_UNICAST] = {
94 		.error	= 0,
95 		.scope	= RT_SCOPE_UNIVERSE,
96 	},
97 	[RTN_LOCAL] = {
98 		.error	= 0,
99 		.scope	= RT_SCOPE_HOST,
100 	},
101 	[RTN_BROADCAST] = {
102 		.error	= 0,
103 		.scope	= RT_SCOPE_LINK,
104 	},
105 	[RTN_ANYCAST] = {
106 		.error	= 0,
107 		.scope	= RT_SCOPE_LINK,
108 	},
109 	[RTN_MULTICAST] = {
110 		.error	= 0,
111 		.scope	= RT_SCOPE_UNIVERSE,
112 	},
113 	[RTN_BLACKHOLE] = {
114 		.error	= -EINVAL,
115 		.scope	= RT_SCOPE_UNIVERSE,
116 	},
117 	[RTN_UNREACHABLE] = {
118 		.error	= -EHOSTUNREACH,
119 		.scope	= RT_SCOPE_UNIVERSE,
120 	},
121 	[RTN_PROHIBIT] = {
122 		.error	= -EACCES,
123 		.scope	= RT_SCOPE_UNIVERSE,
124 	},
125 	[RTN_THROW] = {
126 		.error	= -EAGAIN,
127 		.scope	= RT_SCOPE_UNIVERSE,
128 	},
129 	[RTN_NAT] = {
130 		.error	= -EINVAL,
131 		.scope	= RT_SCOPE_NOWHERE,
132 	},
133 	[RTN_XRESOLVE] = {
134 		.error	= -EINVAL,
135 		.scope	= RT_SCOPE_NOWHERE,
136 	},
137 };
138 
139 static void rt_fibinfo_free(struct rtable __rcu **rtp)
140 {
141 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
142 
143 	if (!rt)
144 		return;
145 
146 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
147 	 * because we waited an RCU grace period before calling
148 	 * free_fib_info_rcu()
149 	 */
150 
151 	dst_dev_put(&rt->dst);
152 	dst_release_immediate(&rt->dst);
153 }
154 
155 static void free_nh_exceptions(struct fib_nh_common *nhc)
156 {
157 	struct fnhe_hash_bucket *hash;
158 	int i;
159 
160 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
161 	if (!hash)
162 		return;
163 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
164 		struct fib_nh_exception *fnhe;
165 
166 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
167 		while (fnhe) {
168 			struct fib_nh_exception *next;
169 
170 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
171 
172 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
173 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
174 
175 			kfree(fnhe);
176 
177 			fnhe = next;
178 		}
179 	}
180 	kfree(hash);
181 }
182 
183 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
184 {
185 	int cpu;
186 
187 	if (!rtp)
188 		return;
189 
190 	for_each_possible_cpu(cpu) {
191 		struct rtable *rt;
192 
193 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
194 		if (rt) {
195 			dst_dev_put(&rt->dst);
196 			dst_release_immediate(&rt->dst);
197 		}
198 	}
199 	free_percpu(rtp);
200 }
201 
202 void fib_nh_common_release(struct fib_nh_common *nhc)
203 {
204 	netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
205 	lwtstate_put(nhc->nhc_lwtstate);
206 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
207 	rt_fibinfo_free(&nhc->nhc_rth_input);
208 	free_nh_exceptions(nhc);
209 }
210 EXPORT_SYMBOL_GPL(fib_nh_common_release);
211 
212 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
213 {
214 #ifdef CONFIG_IP_ROUTE_CLASSID
215 	if (fib_nh->nh_tclassid)
216 		atomic_dec(&net->ipv4.fib_num_tclassid_users);
217 #endif
218 	fib_nh_common_release(&fib_nh->nh_common);
219 }
220 
221 /* Release a nexthop info record */
222 static void free_fib_info_rcu(struct rcu_head *head)
223 {
224 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
225 
226 	if (fi->nh) {
227 		nexthop_put(fi->nh);
228 	} else {
229 		change_nexthops(fi) {
230 			fib_nh_release(fi->fib_net, nexthop_nh);
231 		} endfor_nexthops(fi);
232 	}
233 
234 	ip_fib_metrics_put(fi->fib_metrics);
235 
236 	kfree(fi);
237 }
238 
239 void free_fib_info(struct fib_info *fi)
240 {
241 	if (fi->fib_dead == 0) {
242 		pr_warn("Freeing alive fib_info %p\n", fi);
243 		return;
244 	}
245 
246 	call_rcu_hurry(&fi->rcu, free_fib_info_rcu);
247 }
248 EXPORT_SYMBOL_GPL(free_fib_info);
249 
250 void fib_release_info(struct fib_info *fi)
251 {
252 	ASSERT_RTNL();
253 	if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
254 		hlist_del(&fi->fib_hash);
255 		fi->fib_net->ipv4.fib_info_cnt--;
256 
257 		if (fi->fib_prefsrc)
258 			hlist_del(&fi->fib_lhash);
259 		if (fi->nh) {
260 			list_del(&fi->nh_list);
261 		} else {
262 			change_nexthops(fi) {
263 				if (!nexthop_nh->fib_nh_dev)
264 					continue;
265 				hlist_del_rcu(&nexthop_nh->nh_hash);
266 			} endfor_nexthops(fi)
267 		}
268 		/* Paired with READ_ONCE() from fib_table_lookup() */
269 		WRITE_ONCE(fi->fib_dead, 1);
270 		fib_info_put(fi);
271 	}
272 }
273 
274 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
275 {
276 	const struct fib_nh *onh;
277 
278 	if (fi->nh || ofi->nh)
279 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
280 
281 	if (ofi->fib_nhs == 0)
282 		return 0;
283 
284 	for_nexthops(fi) {
285 		onh = fib_info_nh(ofi, nhsel);
286 
287 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
288 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
289 		    nh->fib_nh_scope != onh->fib_nh_scope ||
290 #ifdef CONFIG_IP_ROUTE_MULTIPATH
291 		    nh->fib_nh_weight != onh->fib_nh_weight ||
292 #endif
293 #ifdef CONFIG_IP_ROUTE_CLASSID
294 		    nh->nh_tclassid != onh->nh_tclassid ||
295 #endif
296 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
297 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
298 			return -1;
299 
300 		if (nh->fib_nh_gw_family == AF_INET &&
301 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
302 			return -1;
303 
304 		if (nh->fib_nh_gw_family == AF_INET6 &&
305 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
306 			return -1;
307 	} endfor_nexthops(fi);
308 	return 0;
309 }
310 
311 static struct hlist_head *fib_nh_head(struct net_device *dev)
312 {
313 	return &dev->fib_nh_head;
314 }
315 
316 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
317 				      u32 prefsrc, u32 priority)
318 {
319 	unsigned int val = init_val;
320 
321 	val ^= (protocol << 8) | scope;
322 	val ^= prefsrc;
323 	val ^= priority;
324 
325 	return val;
326 }
327 
328 static unsigned int fib_info_hashfn_result(const struct net *net,
329 					   unsigned int val)
330 {
331 	return hash_32(val ^ net_hash_mix(net), net->ipv4.fib_info_hash_bits);
332 }
333 
334 static struct hlist_head *fib_info_hash_bucket(struct fib_info *fi)
335 {
336 	struct net *net = fi->fib_net;
337 	unsigned int val;
338 
339 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
340 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
341 				fi->fib_priority);
342 
343 	if (fi->nh) {
344 		val ^= fi->nh->id;
345 	} else {
346 		for_nexthops(fi) {
347 			val ^= nh->fib_nh_oif;
348 		} endfor_nexthops(fi)
349 	}
350 
351 	return &net->ipv4.fib_info_hash[fib_info_hashfn_result(net, val)];
352 }
353 
354 static struct hlist_head *fib_info_laddrhash_bucket(const struct net *net,
355 						    __be32 val)
356 {
357 	unsigned int hash_bits = net->ipv4.fib_info_hash_bits;
358 	u32 slot;
359 
360 	slot = hash_32(net_hash_mix(net) ^ (__force u32)val, hash_bits);
361 
362 	return &net->ipv4.fib_info_hash[(1 << hash_bits) + slot];
363 }
364 
365 static struct hlist_head *fib_info_hash_alloc(unsigned int hash_bits)
366 {
367 	/* The second half is used for prefsrc */
368 	return kvcalloc((1 << hash_bits) * 2, sizeof(struct hlist_head),
369 			GFP_KERNEL);
370 }
371 
372 static void fib_info_hash_free(struct hlist_head *head)
373 {
374 	kvfree(head);
375 }
376 
377 static void fib_info_hash_grow(struct net *net)
378 {
379 	unsigned int old_size = 1 << net->ipv4.fib_info_hash_bits;
380 	struct hlist_head *new_info_hash, *old_info_hash;
381 	unsigned int i;
382 
383 	if (net->ipv4.fib_info_cnt < old_size)
384 		return;
385 
386 	new_info_hash = fib_info_hash_alloc(net->ipv4.fib_info_hash_bits + 1);
387 	if (!new_info_hash)
388 		return;
389 
390 	old_info_hash = net->ipv4.fib_info_hash;
391 	net->ipv4.fib_info_hash = new_info_hash;
392 	net->ipv4.fib_info_hash_bits += 1;
393 
394 	for (i = 0; i < old_size; i++) {
395 		struct hlist_head *head = &old_info_hash[i];
396 		struct hlist_node *n;
397 		struct fib_info *fi;
398 
399 		hlist_for_each_entry_safe(fi, n, head, fib_hash)
400 			hlist_add_head(&fi->fib_hash, fib_info_hash_bucket(fi));
401 	}
402 
403 	for (i = 0; i < old_size; i++) {
404 		struct hlist_head *lhead = &old_info_hash[old_size + i];
405 		struct hlist_node *n;
406 		struct fib_info *fi;
407 
408 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash)
409 			hlist_add_head(&fi->fib_lhash,
410 				       fib_info_laddrhash_bucket(fi->fib_net,
411 								 fi->fib_prefsrc));
412 	}
413 
414 	fib_info_hash_free(old_info_hash);
415 }
416 
417 /* no metrics, only nexthop id */
418 static struct fib_info *fib_find_info_nh(struct net *net,
419 					 const struct fib_config *cfg)
420 {
421 	struct hlist_head *head;
422 	struct fib_info *fi;
423 	unsigned int hash;
424 
425 	hash = fib_info_hashfn_1(cfg->fc_nh_id,
426 				 cfg->fc_protocol, cfg->fc_scope,
427 				 (__force u32)cfg->fc_prefsrc,
428 				 cfg->fc_priority);
429 	hash = fib_info_hashfn_result(net, hash);
430 	head = &net->ipv4.fib_info_hash[hash];
431 
432 	hlist_for_each_entry(fi, head, fib_hash) {
433 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
434 			continue;
435 
436 		if (cfg->fc_protocol == fi->fib_protocol &&
437 		    cfg->fc_scope == fi->fib_scope &&
438 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
439 		    cfg->fc_priority == fi->fib_priority &&
440 		    cfg->fc_type == fi->fib_type &&
441 		    cfg->fc_table == fi->fib_tb_id &&
442 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
443 			return fi;
444 	}
445 
446 	return NULL;
447 }
448 
449 static struct fib_info *fib_find_info(struct fib_info *nfi)
450 {
451 	struct hlist_head *head = fib_info_hash_bucket(nfi);
452 	struct fib_info *fi;
453 
454 	hlist_for_each_entry(fi, head, fib_hash) {
455 		if (fi->fib_nhs != nfi->fib_nhs)
456 			continue;
457 
458 		if (nfi->fib_protocol == fi->fib_protocol &&
459 		    nfi->fib_scope == fi->fib_scope &&
460 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
461 		    nfi->fib_priority == fi->fib_priority &&
462 		    nfi->fib_type == fi->fib_type &&
463 		    nfi->fib_tb_id == fi->fib_tb_id &&
464 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
465 			   sizeof(u32) * RTAX_MAX) == 0 &&
466 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
467 		    nh_comp(fi, nfi) == 0)
468 			return fi;
469 	}
470 
471 	return NULL;
472 }
473 
474 /* Check, that the gateway is already configured.
475  * Used only by redirect accept routine, under rcu_read_lock();
476  */
477 int ip_fib_check_default(__be32 gw, struct net_device *dev)
478 {
479 	struct hlist_head *head;
480 	struct fib_nh *nh;
481 
482 	head = fib_nh_head(dev);
483 
484 	hlist_for_each_entry_rcu(nh, head, nh_hash) {
485 		DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev);
486 		if (nh->fib_nh_gw4 == gw &&
487 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
488 			return 0;
489 		}
490 	}
491 
492 	return -1;
493 }
494 
495 size_t fib_nlmsg_size(struct fib_info *fi)
496 {
497 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
498 			 + nla_total_size(4) /* RTA_TABLE */
499 			 + nla_total_size(4) /* RTA_DST */
500 			 + nla_total_size(4) /* RTA_PRIORITY */
501 			 + nla_total_size(4) /* RTA_PREFSRC */
502 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
503 	unsigned int nhs = fib_info_num_path(fi);
504 
505 	/* space for nested metrics */
506 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
507 
508 	if (fi->nh)
509 		payload += nla_total_size(4); /* RTA_NH_ID */
510 
511 	if (nhs) {
512 		size_t nh_encapsize = 0;
513 		/* Also handles the special case nhs == 1 */
514 
515 		/* each nexthop is packed in an attribute */
516 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
517 		unsigned int i;
518 
519 		/* may contain flow and gateway attribute */
520 		nhsize += 2 * nla_total_size(4);
521 
522 		/* grab encap info */
523 		for (i = 0; i < fib_info_num_path(fi); i++) {
524 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
525 
526 			if (nhc->nhc_lwtstate) {
527 				/* RTA_ENCAP_TYPE */
528 				nh_encapsize += lwtunnel_get_encap_size(
529 						nhc->nhc_lwtstate);
530 				/* RTA_ENCAP */
531 				nh_encapsize +=  nla_total_size(2);
532 			}
533 		}
534 
535 		/* all nexthops are packed in a nested attribute */
536 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
537 
538 	}
539 
540 	return payload;
541 }
542 
543 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
544 	       int dst_len, u32 tb_id, const struct nl_info *info,
545 	       unsigned int nlm_flags)
546 {
547 	struct fib_rt_info fri;
548 	struct sk_buff *skb;
549 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
550 	int err = -ENOBUFS;
551 
552 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
553 	if (!skb)
554 		goto errout;
555 
556 	fri.fi = fa->fa_info;
557 	fri.tb_id = tb_id;
558 	fri.dst = key;
559 	fri.dst_len = dst_len;
560 	fri.dscp = fa->fa_dscp;
561 	fri.type = fa->fa_type;
562 	fri.offload = READ_ONCE(fa->offload);
563 	fri.trap = READ_ONCE(fa->trap);
564 	fri.offload_failed = READ_ONCE(fa->offload_failed);
565 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
566 	if (err < 0) {
567 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
568 		WARN_ON(err == -EMSGSIZE);
569 		kfree_skb(skb);
570 		goto errout;
571 	}
572 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
573 		    info->nlh, GFP_KERNEL);
574 	return;
575 errout:
576 	rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
577 }
578 
579 static int fib_detect_death(struct fib_info *fi, int order,
580 			    struct fib_info **last_resort, int *last_idx,
581 			    int dflt)
582 {
583 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
584 	struct neighbour *n;
585 	int state = NUD_NONE;
586 
587 	if (likely(nhc->nhc_gw_family == AF_INET))
588 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
589 	else if (nhc->nhc_gw_family == AF_INET6)
590 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
591 				 nhc->nhc_dev);
592 	else
593 		n = NULL;
594 
595 	if (n) {
596 		state = READ_ONCE(n->nud_state);
597 		neigh_release(n);
598 	} else {
599 		return 0;
600 	}
601 	if (state == NUD_REACHABLE)
602 		return 0;
603 	if ((state & NUD_VALID) && order != dflt)
604 		return 0;
605 	if ((state & NUD_VALID) ||
606 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
607 		*last_resort = fi;
608 		*last_idx = order;
609 	}
610 	return 1;
611 }
612 
613 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
614 		       struct nlattr *encap, u16 encap_type,
615 		       void *cfg, gfp_t gfp_flags,
616 		       struct netlink_ext_ack *extack)
617 {
618 	int err;
619 
620 	if (!nhc->nhc_pcpu_rth_output) {
621 		nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
622 							    gfp_flags);
623 		if (!nhc->nhc_pcpu_rth_output)
624 			return -ENOMEM;
625 	}
626 
627 	if (encap) {
628 		struct lwtunnel_state *lwtstate;
629 
630 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
631 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
632 			err = -EINVAL;
633 			goto lwt_failure;
634 		}
635 		err = lwtunnel_build_state(net, encap_type, encap,
636 					   nhc->nhc_family, cfg, &lwtstate,
637 					   extack);
638 		if (err)
639 			goto lwt_failure;
640 
641 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
642 	}
643 
644 	return 0;
645 
646 lwt_failure:
647 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
648 	nhc->nhc_pcpu_rth_output = NULL;
649 	return err;
650 }
651 EXPORT_SYMBOL_GPL(fib_nh_common_init);
652 
653 int fib_nh_init(struct net *net, struct fib_nh *nh,
654 		struct fib_config *cfg, int nh_weight,
655 		struct netlink_ext_ack *extack)
656 {
657 	int err;
658 
659 	nh->fib_nh_family = AF_INET;
660 
661 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
662 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
663 	if (err)
664 		return err;
665 
666 	nh->fib_nh_oif = cfg->fc_oif;
667 	nh->fib_nh_gw_family = cfg->fc_gw_family;
668 	if (cfg->fc_gw_family == AF_INET)
669 		nh->fib_nh_gw4 = cfg->fc_gw4;
670 	else if (cfg->fc_gw_family == AF_INET6)
671 		nh->fib_nh_gw6 = cfg->fc_gw6;
672 
673 	nh->fib_nh_flags = cfg->fc_flags;
674 
675 #ifdef CONFIG_IP_ROUTE_CLASSID
676 	nh->nh_tclassid = cfg->fc_flow;
677 	if (nh->nh_tclassid)
678 		atomic_inc(&net->ipv4.fib_num_tclassid_users);
679 #endif
680 #ifdef CONFIG_IP_ROUTE_MULTIPATH
681 	nh->fib_nh_weight = nh_weight;
682 #endif
683 	return 0;
684 }
685 
686 #ifdef CONFIG_IP_ROUTE_MULTIPATH
687 
688 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
689 			      struct netlink_ext_ack *extack)
690 {
691 	int nhs = 0;
692 
693 	while (rtnh_ok(rtnh, remaining)) {
694 		nhs++;
695 		rtnh = rtnh_next(rtnh, &remaining);
696 	}
697 
698 	/* leftover implies invalid nexthop configuration, discard it */
699 	if (remaining > 0) {
700 		NL_SET_ERR_MSG(extack,
701 			       "Invalid nexthop configuration - extra data after nexthops");
702 		nhs = 0;
703 	}
704 
705 	return nhs;
706 }
707 
708 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
709 			    struct netlink_ext_ack *extack)
710 {
711 	if (nla_len(nla) < sizeof(*gw)) {
712 		NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
713 		return -EINVAL;
714 	}
715 
716 	*gw = nla_get_in_addr(nla);
717 
718 	return 0;
719 }
720 
721 /* only called when fib_nh is integrated into fib_info */
722 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
723 		       int remaining, struct fib_config *cfg,
724 		       struct netlink_ext_ack *extack)
725 {
726 	struct net *net = fi->fib_net;
727 	struct fib_config fib_cfg;
728 	struct fib_nh *nh;
729 	int ret;
730 
731 	change_nexthops(fi) {
732 		int attrlen;
733 
734 		memset(&fib_cfg, 0, sizeof(fib_cfg));
735 
736 		if (!rtnh_ok(rtnh, remaining)) {
737 			NL_SET_ERR_MSG(extack,
738 				       "Invalid nexthop configuration - extra data after nexthop");
739 			return -EINVAL;
740 		}
741 
742 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
743 			NL_SET_ERR_MSG(extack,
744 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
745 			return -EINVAL;
746 		}
747 
748 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
749 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
750 
751 		attrlen = rtnh_attrlen(rtnh);
752 		if (attrlen > 0) {
753 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
754 
755 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
756 			nlav = nla_find(attrs, attrlen, RTA_VIA);
757 			if (nla && nlav) {
758 				NL_SET_ERR_MSG(extack,
759 					       "Nexthop configuration can not contain both GATEWAY and VIA");
760 				return -EINVAL;
761 			}
762 			if (nla) {
763 				ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
764 						       extack);
765 				if (ret)
766 					goto errout;
767 
768 				if (fib_cfg.fc_gw4)
769 					fib_cfg.fc_gw_family = AF_INET;
770 			} else if (nlav) {
771 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
772 				if (ret)
773 					goto errout;
774 			}
775 
776 			nla = nla_find(attrs, attrlen, RTA_FLOW);
777 			if (nla) {
778 				if (nla_len(nla) < sizeof(u32)) {
779 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
780 					return -EINVAL;
781 				}
782 				fib_cfg.fc_flow = nla_get_u32(nla);
783 			}
784 
785 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
786 			/* RTA_ENCAP_TYPE length checked in
787 			 * lwtunnel_valid_encap_type_attr
788 			 */
789 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
790 			if (nla)
791 				fib_cfg.fc_encap_type = nla_get_u16(nla);
792 		}
793 
794 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
795 				  rtnh->rtnh_hops + 1, extack);
796 		if (ret)
797 			goto errout;
798 
799 		rtnh = rtnh_next(rtnh, &remaining);
800 	} endfor_nexthops(fi);
801 
802 	ret = -EINVAL;
803 	nh = fib_info_nh(fi, 0);
804 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
805 		NL_SET_ERR_MSG(extack,
806 			       "Nexthop device index does not match RTA_OIF");
807 		goto errout;
808 	}
809 	if (cfg->fc_gw_family) {
810 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
811 		    (cfg->fc_gw_family == AF_INET &&
812 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
813 		    (cfg->fc_gw_family == AF_INET6 &&
814 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
815 			NL_SET_ERR_MSG(extack,
816 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
817 			goto errout;
818 		}
819 	}
820 #ifdef CONFIG_IP_ROUTE_CLASSID
821 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
822 		NL_SET_ERR_MSG(extack,
823 			       "Nexthop class id does not match RTA_FLOW");
824 		goto errout;
825 	}
826 #endif
827 	ret = 0;
828 errout:
829 	return ret;
830 }
831 
832 /* only called when fib_nh is integrated into fib_info */
833 static void fib_rebalance(struct fib_info *fi)
834 {
835 	int total;
836 	int w;
837 
838 	if (fib_info_num_path(fi) < 2)
839 		return;
840 
841 	total = 0;
842 	for_nexthops(fi) {
843 		if (nh->fib_nh_flags & RTNH_F_DEAD)
844 			continue;
845 
846 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
847 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
848 			continue;
849 
850 		total += nh->fib_nh_weight;
851 	} endfor_nexthops(fi);
852 
853 	w = 0;
854 	change_nexthops(fi) {
855 		int upper_bound;
856 
857 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
858 			upper_bound = -1;
859 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
860 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
861 			upper_bound = -1;
862 		} else {
863 			w += nexthop_nh->fib_nh_weight;
864 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
865 							    total) - 1;
866 		}
867 
868 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
869 	} endfor_nexthops(fi);
870 }
871 #else /* CONFIG_IP_ROUTE_MULTIPATH */
872 
873 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
874 		       int remaining, struct fib_config *cfg,
875 		       struct netlink_ext_ack *extack)
876 {
877 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
878 
879 	return -EINVAL;
880 }
881 
882 #define fib_rebalance(fi) do { } while (0)
883 
884 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
885 
886 static int fib_encap_match(struct net *net, u16 encap_type,
887 			   struct nlattr *encap,
888 			   const struct fib_nh *nh,
889 			   const struct fib_config *cfg,
890 			   struct netlink_ext_ack *extack)
891 {
892 	struct lwtunnel_state *lwtstate;
893 	int ret, result = 0;
894 
895 	if (encap_type == LWTUNNEL_ENCAP_NONE)
896 		return 0;
897 
898 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
899 				   cfg, &lwtstate, extack);
900 	if (!ret) {
901 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
902 		lwtstate_free(lwtstate);
903 	}
904 
905 	return result;
906 }
907 
908 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
909 		 struct netlink_ext_ack *extack)
910 {
911 #ifdef CONFIG_IP_ROUTE_MULTIPATH
912 	struct rtnexthop *rtnh;
913 	int remaining;
914 #endif
915 
916 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
917 		return 1;
918 
919 	if (cfg->fc_nh_id) {
920 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
921 			return 0;
922 		return 1;
923 	}
924 
925 	if (fi->nh) {
926 		if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
927 			return 1;
928 		return 0;
929 	}
930 
931 	if (cfg->fc_oif || cfg->fc_gw_family) {
932 		struct fib_nh *nh;
933 
934 		nh = fib_info_nh(fi, 0);
935 		if (cfg->fc_encap) {
936 			if (fib_encap_match(net, cfg->fc_encap_type,
937 					    cfg->fc_encap, nh, cfg, extack))
938 				return 1;
939 		}
940 #ifdef CONFIG_IP_ROUTE_CLASSID
941 		if (cfg->fc_flow &&
942 		    cfg->fc_flow != nh->nh_tclassid)
943 			return 1;
944 #endif
945 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
946 		    (cfg->fc_gw_family &&
947 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
948 			return 1;
949 
950 		if (cfg->fc_gw_family == AF_INET &&
951 		    cfg->fc_gw4 != nh->fib_nh_gw4)
952 			return 1;
953 
954 		if (cfg->fc_gw_family == AF_INET6 &&
955 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
956 			return 1;
957 
958 		return 0;
959 	}
960 
961 #ifdef CONFIG_IP_ROUTE_MULTIPATH
962 	if (!cfg->fc_mp)
963 		return 0;
964 
965 	rtnh = cfg->fc_mp;
966 	remaining = cfg->fc_mp_len;
967 
968 	for_nexthops(fi) {
969 		int attrlen;
970 
971 		if (!rtnh_ok(rtnh, remaining))
972 			return -EINVAL;
973 
974 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
975 			return 1;
976 
977 		attrlen = rtnh_attrlen(rtnh);
978 		if (attrlen > 0) {
979 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
980 			int err;
981 
982 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
983 			nlav = nla_find(attrs, attrlen, RTA_VIA);
984 			if (nla && nlav) {
985 				NL_SET_ERR_MSG(extack,
986 					       "Nexthop configuration can not contain both GATEWAY and VIA");
987 				return -EINVAL;
988 			}
989 
990 			if (nla) {
991 				__be32 gw;
992 
993 				err = fib_gw_from_attr(&gw, nla, extack);
994 				if (err)
995 					return err;
996 
997 				if (nh->fib_nh_gw_family != AF_INET ||
998 				    gw != nh->fib_nh_gw4)
999 					return 1;
1000 			} else if (nlav) {
1001 				struct fib_config cfg2;
1002 
1003 				err = fib_gw_from_via(&cfg2, nlav, extack);
1004 				if (err)
1005 					return err;
1006 
1007 				switch (nh->fib_nh_gw_family) {
1008 				case AF_INET:
1009 					if (cfg2.fc_gw_family != AF_INET ||
1010 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
1011 						return 1;
1012 					break;
1013 				case AF_INET6:
1014 					if (cfg2.fc_gw_family != AF_INET6 ||
1015 					    ipv6_addr_cmp(&cfg2.fc_gw6,
1016 							  &nh->fib_nh_gw6))
1017 						return 1;
1018 					break;
1019 				}
1020 			}
1021 
1022 #ifdef CONFIG_IP_ROUTE_CLASSID
1023 			nla = nla_find(attrs, attrlen, RTA_FLOW);
1024 			if (nla) {
1025 				if (nla_len(nla) < sizeof(u32)) {
1026 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
1027 					return -EINVAL;
1028 				}
1029 				if (nla_get_u32(nla) != nh->nh_tclassid)
1030 					return 1;
1031 			}
1032 #endif
1033 		}
1034 
1035 		rtnh = rtnh_next(rtnh, &remaining);
1036 	} endfor_nexthops(fi);
1037 #endif
1038 	return 0;
1039 }
1040 
1041 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1042 {
1043 	struct nlattr *nla;
1044 	int remaining;
1045 
1046 	if (!cfg->fc_mx)
1047 		return true;
1048 
1049 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1050 		int type = nla_type(nla);
1051 		u32 fi_val, val;
1052 
1053 		if (!type)
1054 			continue;
1055 		if (type > RTAX_MAX)
1056 			return false;
1057 
1058 		type = array_index_nospec(type, RTAX_MAX + 1);
1059 		if (type == RTAX_CC_ALGO) {
1060 			char tmp[TCP_CA_NAME_MAX];
1061 			bool ecn_ca = false;
1062 
1063 			nla_strscpy(tmp, nla, sizeof(tmp));
1064 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
1065 		} else {
1066 			if (nla_len(nla) != sizeof(u32))
1067 				return false;
1068 			val = nla_get_u32(nla);
1069 		}
1070 
1071 		fi_val = fi->fib_metrics->metrics[type - 1];
1072 		if (type == RTAX_FEATURES)
1073 			fi_val &= ~DST_FEATURE_ECN_CA;
1074 
1075 		if (fi_val != val)
1076 			return false;
1077 	}
1078 
1079 	return true;
1080 }
1081 
1082 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1083 			      u32 table, struct netlink_ext_ack *extack)
1084 {
1085 	struct fib6_config cfg = {
1086 		.fc_table = table,
1087 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1088 		.fc_ifindex = nh->fib_nh_oif,
1089 		.fc_gateway = nh->fib_nh_gw6,
1090 	};
1091 	struct fib6_nh fib6_nh = {};
1092 	int err;
1093 
1094 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1095 	if (!err) {
1096 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1097 		netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1098 			    GFP_KERNEL);
1099 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1100 		nh->fib_nh_scope = RT_SCOPE_LINK;
1101 
1102 		ipv6_stub->fib6_nh_release(&fib6_nh);
1103 	}
1104 
1105 	return err;
1106 }
1107 
1108 /*
1109  * Picture
1110  * -------
1111  *
1112  * Semantics of nexthop is very messy by historical reasons.
1113  * We have to take into account, that:
1114  * a) gateway can be actually local interface address,
1115  *    so that gatewayed route is direct.
1116  * b) gateway must be on-link address, possibly
1117  *    described not by an ifaddr, but also by a direct route.
1118  * c) If both gateway and interface are specified, they should not
1119  *    contradict.
1120  * d) If we use tunnel routes, gateway could be not on-link.
1121  *
1122  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1123  * results in pretty ugly and hairy code with obscure logic.
1124  *
1125  * I chose to generalized it instead, so that the size
1126  * of code does not increase practically, but it becomes
1127  * much more general.
1128  * Every prefix is assigned a "scope" value: "host" is local address,
1129  * "link" is direct route,
1130  * [ ... "site" ... "interior" ... ]
1131  * and "universe" is true gateway route with global meaning.
1132  *
1133  * Every prefix refers to a set of "nexthop"s (gw, oif),
1134  * where gw must have narrower scope. This recursion stops
1135  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1136  * which means that gw is forced to be on link.
1137  *
1138  * Code is still hairy, but now it is apparently logically
1139  * consistent and very flexible. F.e. as by-product it allows
1140  * to co-exists in peace independent exterior and interior
1141  * routing processes.
1142  *
1143  * Normally it looks as following.
1144  *
1145  * {universe prefix}  -> (gw, oif) [scope link]
1146  *		  |
1147  *		  |-> {link prefix} -> (gw, oif) [scope local]
1148  *					|
1149  *					|-> {local prefix} (terminal node)
1150  */
1151 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1152 			      u8 scope, struct netlink_ext_ack *extack)
1153 {
1154 	struct net_device *dev;
1155 	struct fib_result res;
1156 	int err = 0;
1157 
1158 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1159 		unsigned int addr_type;
1160 
1161 		if (scope >= RT_SCOPE_LINK) {
1162 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1163 			return -EINVAL;
1164 		}
1165 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1166 		if (!dev) {
1167 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1168 			return -ENODEV;
1169 		}
1170 		if (!(dev->flags & IFF_UP)) {
1171 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1172 			return -ENETDOWN;
1173 		}
1174 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1175 		if (addr_type != RTN_UNICAST) {
1176 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1177 			return -EINVAL;
1178 		}
1179 		if (!netif_carrier_ok(dev))
1180 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1181 		nh->fib_nh_dev = dev;
1182 		netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1183 		nh->fib_nh_scope = RT_SCOPE_LINK;
1184 		return 0;
1185 	}
1186 	rcu_read_lock();
1187 	{
1188 		struct fib_table *tbl = NULL;
1189 		struct flowi4 fl4 = {
1190 			.daddr = nh->fib_nh_gw4,
1191 			.flowi4_scope = scope + 1,
1192 			.flowi4_oif = nh->fib_nh_oif,
1193 			.flowi4_iif = LOOPBACK_IFINDEX,
1194 		};
1195 
1196 		/* It is not necessary, but requires a bit of thinking */
1197 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1198 			fl4.flowi4_scope = RT_SCOPE_LINK;
1199 
1200 		if (table && table != RT_TABLE_MAIN)
1201 			tbl = fib_get_table(net, table);
1202 
1203 		if (tbl)
1204 			err = fib_table_lookup(tbl, &fl4, &res,
1205 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1206 					       FIB_LOOKUP_NOREF);
1207 
1208 		/* on error or if no table given do full lookup. This
1209 		 * is needed for example when nexthops are in the local
1210 		 * table rather than the given table
1211 		 */
1212 		if (!tbl || err) {
1213 			err = fib_lookup(net, &fl4, &res,
1214 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1215 		}
1216 
1217 		if (err) {
1218 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1219 			goto out;
1220 		}
1221 	}
1222 
1223 	err = -EINVAL;
1224 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1225 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1226 		goto out;
1227 	}
1228 	nh->fib_nh_scope = res.scope;
1229 	nh->fib_nh_oif = FIB_RES_OIF(res);
1230 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1231 	if (!dev) {
1232 		NL_SET_ERR_MSG(extack,
1233 			       "No egress device for nexthop gateway");
1234 		goto out;
1235 	}
1236 	netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1237 	if (!netif_carrier_ok(dev))
1238 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1239 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1240 out:
1241 	rcu_read_unlock();
1242 	return err;
1243 }
1244 
1245 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1246 			      struct netlink_ext_ack *extack)
1247 {
1248 	struct in_device *in_dev;
1249 	int err;
1250 
1251 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1252 		NL_SET_ERR_MSG(extack,
1253 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1254 		return -EINVAL;
1255 	}
1256 
1257 	rcu_read_lock();
1258 
1259 	err = -ENODEV;
1260 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1261 	if (!in_dev)
1262 		goto out;
1263 	err = -ENETDOWN;
1264 	if (!(in_dev->dev->flags & IFF_UP)) {
1265 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1266 		goto out;
1267 	}
1268 
1269 	nh->fib_nh_dev = in_dev->dev;
1270 	netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1271 	nh->fib_nh_scope = RT_SCOPE_HOST;
1272 	if (!netif_carrier_ok(nh->fib_nh_dev))
1273 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1274 	err = 0;
1275 out:
1276 	rcu_read_unlock();
1277 	return err;
1278 }
1279 
1280 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1281 		 struct netlink_ext_ack *extack)
1282 {
1283 	int err;
1284 
1285 	if (nh->fib_nh_gw_family == AF_INET)
1286 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1287 	else if (nh->fib_nh_gw_family == AF_INET6)
1288 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1289 	else
1290 		err = fib_check_nh_nongw(net, nh, extack);
1291 
1292 	return err;
1293 }
1294 
1295 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1296 				 unsigned char scope)
1297 {
1298 	struct fib_nh *nh;
1299 	__be32 saddr;
1300 
1301 	if (nhc->nhc_family != AF_INET)
1302 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1303 
1304 	nh = container_of(nhc, struct fib_nh, nh_common);
1305 	saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1306 
1307 	WRITE_ONCE(nh->nh_saddr, saddr);
1308 	WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid));
1309 
1310 	return saddr;
1311 }
1312 
1313 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1314 {
1315 	struct fib_nh_common *nhc = res->nhc;
1316 
1317 	if (res->fi->fib_prefsrc)
1318 		return res->fi->fib_prefsrc;
1319 
1320 	if (nhc->nhc_family == AF_INET) {
1321 		struct fib_nh *nh;
1322 
1323 		nh = container_of(nhc, struct fib_nh, nh_common);
1324 		if (READ_ONCE(nh->nh_saddr_genid) ==
1325 		    atomic_read(&net->ipv4.dev_addr_genid))
1326 			return READ_ONCE(nh->nh_saddr);
1327 	}
1328 
1329 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1330 }
1331 
1332 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1333 {
1334 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1335 	    fib_prefsrc != cfg->fc_dst) {
1336 		u32 tb_id = cfg->fc_table;
1337 		int rc;
1338 
1339 		if (tb_id == RT_TABLE_MAIN)
1340 			tb_id = RT_TABLE_LOCAL;
1341 
1342 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1343 					  fib_prefsrc, tb_id);
1344 
1345 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1346 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1347 						  fib_prefsrc, RT_TABLE_LOCAL);
1348 		}
1349 
1350 		if (rc != RTN_LOCAL)
1351 			return false;
1352 	}
1353 	return true;
1354 }
1355 
1356 struct fib_info *fib_create_info(struct fib_config *cfg,
1357 				 struct netlink_ext_ack *extack)
1358 {
1359 	int err;
1360 	struct fib_info *fi = NULL;
1361 	struct nexthop *nh = NULL;
1362 	struct fib_info *ofi;
1363 	int nhs = 1;
1364 	struct net *net = cfg->fc_nlinfo.nl_net;
1365 
1366 	ASSERT_RTNL();
1367 	if (cfg->fc_type > RTN_MAX)
1368 		goto err_inval;
1369 
1370 	/* Fast check to catch the most weird cases */
1371 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1372 		NL_SET_ERR_MSG(extack, "Invalid scope");
1373 		goto err_inval;
1374 	}
1375 
1376 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1377 		NL_SET_ERR_MSG(extack,
1378 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1379 		goto err_inval;
1380 	}
1381 
1382 	if (cfg->fc_nh_id) {
1383 		if (!cfg->fc_mx) {
1384 			fi = fib_find_info_nh(net, cfg);
1385 			if (fi) {
1386 				refcount_inc(&fi->fib_treeref);
1387 				return fi;
1388 			}
1389 		}
1390 
1391 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1392 		if (!nh) {
1393 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1394 			goto err_inval;
1395 		}
1396 		nhs = 0;
1397 	}
1398 
1399 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1400 	if (cfg->fc_mp) {
1401 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1402 		if (nhs == 0)
1403 			goto err_inval;
1404 	}
1405 #endif
1406 
1407 	fib_info_hash_grow(net);
1408 
1409 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1410 	if (!fi) {
1411 		err = -ENOBUFS;
1412 		goto failure;
1413 	}
1414 
1415 	fi->fib_metrics = ip_fib_metrics_init(cfg->fc_mx, cfg->fc_mx_len, extack);
1416 	if (IS_ERR(fi->fib_metrics)) {
1417 		err = PTR_ERR(fi->fib_metrics);
1418 		kfree(fi);
1419 		return ERR_PTR(err);
1420 	}
1421 
1422 	fi->fib_net = net;
1423 	fi->fib_protocol = cfg->fc_protocol;
1424 	fi->fib_scope = cfg->fc_scope;
1425 	fi->fib_flags = cfg->fc_flags;
1426 	fi->fib_priority = cfg->fc_priority;
1427 	fi->fib_prefsrc = cfg->fc_prefsrc;
1428 	fi->fib_type = cfg->fc_type;
1429 	fi->fib_tb_id = cfg->fc_table;
1430 
1431 	fi->fib_nhs = nhs;
1432 	if (nh) {
1433 		if (!nexthop_get(nh)) {
1434 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1435 			err = -EINVAL;
1436 		} else {
1437 			err = 0;
1438 			fi->nh = nh;
1439 		}
1440 	} else {
1441 		change_nexthops(fi) {
1442 			nexthop_nh->nh_parent = fi;
1443 		} endfor_nexthops(fi)
1444 
1445 		if (cfg->fc_mp)
1446 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1447 					  extack);
1448 		else
1449 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1450 	}
1451 
1452 	if (err != 0)
1453 		goto failure;
1454 
1455 	if (fib_props[cfg->fc_type].error) {
1456 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1457 			NL_SET_ERR_MSG(extack,
1458 				       "Gateway, device and multipath can not be specified for this route type");
1459 			goto err_inval;
1460 		}
1461 		goto link_it;
1462 	} else {
1463 		switch (cfg->fc_type) {
1464 		case RTN_UNICAST:
1465 		case RTN_LOCAL:
1466 		case RTN_BROADCAST:
1467 		case RTN_ANYCAST:
1468 		case RTN_MULTICAST:
1469 			break;
1470 		default:
1471 			NL_SET_ERR_MSG(extack, "Invalid route type");
1472 			goto err_inval;
1473 		}
1474 	}
1475 
1476 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1477 		NL_SET_ERR_MSG(extack, "Invalid scope");
1478 		goto err_inval;
1479 	}
1480 
1481 	if (fi->nh) {
1482 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1483 		if (err)
1484 			goto failure;
1485 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1486 		struct fib_nh *nh = fi->fib_nh;
1487 
1488 		/* Local address is added. */
1489 		if (nhs != 1) {
1490 			NL_SET_ERR_MSG(extack,
1491 				       "Route with host scope can not have multiple nexthops");
1492 			goto err_inval;
1493 		}
1494 		if (nh->fib_nh_gw_family) {
1495 			NL_SET_ERR_MSG(extack,
1496 				       "Route with host scope can not have a gateway");
1497 			goto err_inval;
1498 		}
1499 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1500 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1501 		err = -ENODEV;
1502 		if (!nh->fib_nh_dev)
1503 			goto failure;
1504 		netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1505 				     GFP_KERNEL);
1506 	} else {
1507 		int linkdown = 0;
1508 
1509 		change_nexthops(fi) {
1510 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1511 					   cfg->fc_table, cfg->fc_scope,
1512 					   extack);
1513 			if (err != 0)
1514 				goto failure;
1515 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1516 				linkdown++;
1517 		} endfor_nexthops(fi)
1518 		if (linkdown == fi->fib_nhs)
1519 			fi->fib_flags |= RTNH_F_LINKDOWN;
1520 	}
1521 
1522 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1523 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1524 		goto err_inval;
1525 	}
1526 
1527 	if (!fi->nh) {
1528 		change_nexthops(fi) {
1529 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1530 						  fi->fib_scope);
1531 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1532 				fi->fib_nh_is_v6 = true;
1533 		} endfor_nexthops(fi)
1534 
1535 		fib_rebalance(fi);
1536 	}
1537 
1538 link_it:
1539 	ofi = fib_find_info(fi);
1540 	if (ofi) {
1541 		/* fib_table_lookup() should not see @fi yet. */
1542 		fi->fib_dead = 1;
1543 		free_fib_info(fi);
1544 		refcount_inc(&ofi->fib_treeref);
1545 		return ofi;
1546 	}
1547 
1548 	refcount_set(&fi->fib_treeref, 1);
1549 	refcount_set(&fi->fib_clntref, 1);
1550 
1551 	net->ipv4.fib_info_cnt++;
1552 	hlist_add_head(&fi->fib_hash, fib_info_hash_bucket(fi));
1553 
1554 	if (fi->fib_prefsrc) {
1555 		struct hlist_head *head;
1556 
1557 		head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1558 		hlist_add_head(&fi->fib_lhash, head);
1559 	}
1560 	if (fi->nh) {
1561 		list_add(&fi->nh_list, &nh->fi_list);
1562 	} else {
1563 		change_nexthops(fi) {
1564 			struct hlist_head *head;
1565 
1566 			if (!nexthop_nh->fib_nh_dev)
1567 				continue;
1568 			head = fib_nh_head(nexthop_nh->fib_nh_dev);
1569 			hlist_add_head_rcu(&nexthop_nh->nh_hash, head);
1570 		} endfor_nexthops(fi)
1571 	}
1572 	return fi;
1573 
1574 err_inval:
1575 	err = -EINVAL;
1576 
1577 failure:
1578 	if (fi) {
1579 		/* fib_table_lookup() should not see @fi yet. */
1580 		fi->fib_dead = 1;
1581 		free_fib_info(fi);
1582 	}
1583 
1584 	return ERR_PTR(err);
1585 }
1586 
1587 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1588 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1589 {
1590 	if (nhc->nhc_flags & RTNH_F_DEAD)
1591 		*flags |= RTNH_F_DEAD;
1592 
1593 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1594 		*flags |= RTNH_F_LINKDOWN;
1595 
1596 		rcu_read_lock();
1597 		switch (nhc->nhc_family) {
1598 		case AF_INET:
1599 			if (ip_ignore_linkdown(nhc->nhc_dev))
1600 				*flags |= RTNH_F_DEAD;
1601 			break;
1602 		case AF_INET6:
1603 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1604 				*flags |= RTNH_F_DEAD;
1605 			break;
1606 		}
1607 		rcu_read_unlock();
1608 	}
1609 
1610 	switch (nhc->nhc_gw_family) {
1611 	case AF_INET:
1612 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1613 			goto nla_put_failure;
1614 		break;
1615 	case AF_INET6:
1616 		/* if gateway family does not match nexthop family
1617 		 * gateway is encoded as RTA_VIA
1618 		 */
1619 		if (rt_family != nhc->nhc_gw_family) {
1620 			int alen = sizeof(struct in6_addr);
1621 			struct nlattr *nla;
1622 			struct rtvia *via;
1623 
1624 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1625 			if (!nla)
1626 				goto nla_put_failure;
1627 
1628 			via = nla_data(nla);
1629 			via->rtvia_family = AF_INET6;
1630 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1631 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1632 					    &nhc->nhc_gw.ipv6) < 0) {
1633 			goto nla_put_failure;
1634 		}
1635 		break;
1636 	}
1637 
1638 	*flags |= (nhc->nhc_flags &
1639 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1640 
1641 	if (!skip_oif && nhc->nhc_dev &&
1642 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1643 		goto nla_put_failure;
1644 
1645 	if (nhc->nhc_lwtstate &&
1646 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1647 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1648 		goto nla_put_failure;
1649 
1650 	return 0;
1651 
1652 nla_put_failure:
1653 	return -EMSGSIZE;
1654 }
1655 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1656 
1657 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1658 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1659 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1660 {
1661 	const struct net_device *dev = nhc->nhc_dev;
1662 	struct rtnexthop *rtnh;
1663 	unsigned char flags = 0;
1664 
1665 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1666 	if (!rtnh)
1667 		goto nla_put_failure;
1668 
1669 	rtnh->rtnh_hops = nh_weight - 1;
1670 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1671 
1672 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1673 		goto nla_put_failure;
1674 
1675 	rtnh->rtnh_flags = flags;
1676 
1677 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1678 		goto nla_put_failure;
1679 
1680 	/* length of rtnetlink header + attributes */
1681 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1682 
1683 	return 0;
1684 
1685 nla_put_failure:
1686 	return -EMSGSIZE;
1687 }
1688 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1689 #endif
1690 
1691 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1692 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1693 {
1694 	struct nlattr *mp;
1695 
1696 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1697 	if (!mp)
1698 		goto nla_put_failure;
1699 
1700 	if (unlikely(fi->nh)) {
1701 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1702 			goto nla_put_failure;
1703 		goto mp_end;
1704 	}
1705 
1706 	for_nexthops(fi) {
1707 		u32 nh_tclassid = 0;
1708 #ifdef CONFIG_IP_ROUTE_CLASSID
1709 		nh_tclassid = nh->nh_tclassid;
1710 #endif
1711 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1712 				    AF_INET, nh_tclassid) < 0)
1713 			goto nla_put_failure;
1714 	} endfor_nexthops(fi);
1715 
1716 mp_end:
1717 	nla_nest_end(skb, mp);
1718 
1719 	return 0;
1720 
1721 nla_put_failure:
1722 	return -EMSGSIZE;
1723 }
1724 #else
1725 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1726 {
1727 	return 0;
1728 }
1729 #endif
1730 
1731 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1732 		  const struct fib_rt_info *fri, unsigned int flags)
1733 {
1734 	unsigned int nhs = fib_info_num_path(fri->fi);
1735 	struct fib_info *fi = fri->fi;
1736 	u32 tb_id = fri->tb_id;
1737 	struct nlmsghdr *nlh;
1738 	struct rtmsg *rtm;
1739 
1740 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1741 	if (!nlh)
1742 		return -EMSGSIZE;
1743 
1744 	rtm = nlmsg_data(nlh);
1745 	rtm->rtm_family = AF_INET;
1746 	rtm->rtm_dst_len = fri->dst_len;
1747 	rtm->rtm_src_len = 0;
1748 	rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1749 	if (tb_id < 256)
1750 		rtm->rtm_table = tb_id;
1751 	else
1752 		rtm->rtm_table = RT_TABLE_COMPAT;
1753 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1754 		goto nla_put_failure;
1755 	rtm->rtm_type = fri->type;
1756 	rtm->rtm_flags = fi->fib_flags;
1757 	rtm->rtm_scope = fi->fib_scope;
1758 	rtm->rtm_protocol = fi->fib_protocol;
1759 
1760 	if (rtm->rtm_dst_len &&
1761 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1762 		goto nla_put_failure;
1763 	if (fi->fib_priority &&
1764 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1765 		goto nla_put_failure;
1766 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1767 		goto nla_put_failure;
1768 
1769 	if (fi->fib_prefsrc &&
1770 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1771 		goto nla_put_failure;
1772 
1773 	if (fi->nh) {
1774 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1775 			goto nla_put_failure;
1776 		if (nexthop_is_blackhole(fi->nh))
1777 			rtm->rtm_type = RTN_BLACKHOLE;
1778 		if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1779 			goto offload;
1780 	}
1781 
1782 	if (nhs == 1) {
1783 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1784 		unsigned char flags = 0;
1785 
1786 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1787 			goto nla_put_failure;
1788 
1789 		rtm->rtm_flags = flags;
1790 #ifdef CONFIG_IP_ROUTE_CLASSID
1791 		if (nhc->nhc_family == AF_INET) {
1792 			struct fib_nh *nh;
1793 
1794 			nh = container_of(nhc, struct fib_nh, nh_common);
1795 			if (nh->nh_tclassid &&
1796 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1797 				goto nla_put_failure;
1798 		}
1799 #endif
1800 	} else {
1801 		if (fib_add_multipath(skb, fi) < 0)
1802 			goto nla_put_failure;
1803 	}
1804 
1805 offload:
1806 	if (fri->offload)
1807 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1808 	if (fri->trap)
1809 		rtm->rtm_flags |= RTM_F_TRAP;
1810 	if (fri->offload_failed)
1811 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1812 
1813 	nlmsg_end(skb, nlh);
1814 	return 0;
1815 
1816 nla_put_failure:
1817 	nlmsg_cancel(skb, nlh);
1818 	return -EMSGSIZE;
1819 }
1820 
1821 /*
1822  * Update FIB if:
1823  * - local address disappeared -> we must delete all the entries
1824  *   referring to it.
1825  * - device went down -> we must shutdown all nexthops going via it.
1826  */
1827 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1828 {
1829 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1830 	struct net *net = dev_net(dev);
1831 	struct hlist_head *head;
1832 	struct fib_info *fi;
1833 	int ret = 0;
1834 
1835 	if (!local)
1836 		return 0;
1837 
1838 	head = fib_info_laddrhash_bucket(net, local);
1839 	hlist_for_each_entry(fi, head, fib_lhash) {
1840 		if (!net_eq(fi->fib_net, net) ||
1841 		    fi->fib_tb_id != tb_id)
1842 			continue;
1843 		if (fi->fib_prefsrc == local) {
1844 			fi->fib_flags |= RTNH_F_DEAD;
1845 			fi->pfsrc_removed = true;
1846 			ret++;
1847 		}
1848 	}
1849 	return ret;
1850 }
1851 
1852 static int call_fib_nh_notifiers(struct fib_nh *nh,
1853 				 enum fib_event_type event_type)
1854 {
1855 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1856 	struct fib_nh_notifier_info info = {
1857 		.fib_nh = nh,
1858 	};
1859 
1860 	switch (event_type) {
1861 	case FIB_EVENT_NH_ADD:
1862 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1863 			break;
1864 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1865 			break;
1866 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1867 					   &info.info);
1868 	case FIB_EVENT_NH_DEL:
1869 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1870 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1871 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1872 						   event_type, &info.info);
1873 		break;
1874 	default:
1875 		break;
1876 	}
1877 
1878 	return NOTIFY_DONE;
1879 }
1880 
1881 /* Update the PMTU of exceptions when:
1882  * - the new MTU of the first hop becomes smaller than the PMTU
1883  * - the old MTU was the same as the PMTU, and it limited discovery of
1884  *   larger MTUs on the path. With that limit raised, we can now
1885  *   discover larger MTUs
1886  * A special case is locked exceptions, for which the PMTU is smaller
1887  * than the minimal accepted PMTU:
1888  * - if the new MTU is greater than the PMTU, don't make any change
1889  * - otherwise, unlock and set PMTU
1890  */
1891 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1892 {
1893 	struct fnhe_hash_bucket *bucket;
1894 	int i;
1895 
1896 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1897 	if (!bucket)
1898 		return;
1899 
1900 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1901 		struct fib_nh_exception *fnhe;
1902 
1903 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1904 		     fnhe;
1905 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1906 			if (fnhe->fnhe_mtu_locked) {
1907 				if (new <= fnhe->fnhe_pmtu) {
1908 					fnhe->fnhe_pmtu = new;
1909 					fnhe->fnhe_mtu_locked = false;
1910 				}
1911 			} else if (new < fnhe->fnhe_pmtu ||
1912 				   orig == fnhe->fnhe_pmtu) {
1913 				fnhe->fnhe_pmtu = new;
1914 			}
1915 		}
1916 	}
1917 }
1918 
1919 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1920 {
1921 	struct hlist_head *head = fib_nh_head(dev);
1922 	struct fib_nh *nh;
1923 
1924 	hlist_for_each_entry(nh, head, nh_hash) {
1925 		DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev);
1926 		fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1927 	}
1928 }
1929 
1930 /* Event              force Flags           Description
1931  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1932  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1933  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1934  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1935  *
1936  * only used when fib_nh is built into fib_info
1937  */
1938 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1939 {
1940 	struct hlist_head *head = fib_nh_head(dev);
1941 	struct fib_info *prev_fi = NULL;
1942 	int scope = RT_SCOPE_NOWHERE;
1943 	struct fib_nh *nh;
1944 	int ret = 0;
1945 
1946 	if (force)
1947 		scope = -1;
1948 
1949 	hlist_for_each_entry(nh, head, nh_hash) {
1950 		struct fib_info *fi = nh->nh_parent;
1951 		int dead;
1952 
1953 		BUG_ON(!fi->fib_nhs);
1954 		DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev);
1955 		if (fi == prev_fi)
1956 			continue;
1957 		prev_fi = fi;
1958 		dead = 0;
1959 		change_nexthops(fi) {
1960 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1961 				dead++;
1962 			else if (nexthop_nh->fib_nh_dev == dev &&
1963 				 nexthop_nh->fib_nh_scope != scope) {
1964 				switch (event) {
1965 				case NETDEV_DOWN:
1966 				case NETDEV_UNREGISTER:
1967 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1968 					fallthrough;
1969 				case NETDEV_CHANGE:
1970 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1971 					break;
1972 				}
1973 				call_fib_nh_notifiers(nexthop_nh,
1974 						      FIB_EVENT_NH_DEL);
1975 				dead++;
1976 			}
1977 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1978 			if (event == NETDEV_UNREGISTER &&
1979 			    nexthop_nh->fib_nh_dev == dev) {
1980 				dead = fi->fib_nhs;
1981 				break;
1982 			}
1983 #endif
1984 		} endfor_nexthops(fi)
1985 		if (dead == fi->fib_nhs) {
1986 			switch (event) {
1987 			case NETDEV_DOWN:
1988 			case NETDEV_UNREGISTER:
1989 				fi->fib_flags |= RTNH_F_DEAD;
1990 				fallthrough;
1991 			case NETDEV_CHANGE:
1992 				fi->fib_flags |= RTNH_F_LINKDOWN;
1993 				break;
1994 			}
1995 			ret++;
1996 		}
1997 
1998 		fib_rebalance(fi);
1999 	}
2000 
2001 	return ret;
2002 }
2003 
2004 /* Must be invoked inside of an RCU protected region.  */
2005 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2006 {
2007 	struct fib_info *fi = NULL, *last_resort = NULL;
2008 	struct hlist_head *fa_head = res->fa_head;
2009 	struct fib_table *tb = res->table;
2010 	u8 slen = 32 - res->prefixlen;
2011 	int order = -1, last_idx = -1;
2012 	struct fib_alias *fa, *fa1 = NULL;
2013 	u32 last_prio = res->fi->fib_priority;
2014 	dscp_t last_dscp = 0;
2015 
2016 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2017 		struct fib_info *next_fi = fa->fa_info;
2018 		struct fib_nh_common *nhc;
2019 
2020 		if (fa->fa_slen != slen)
2021 			continue;
2022 		if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp))
2023 			continue;
2024 		if (fa->tb_id != tb->tb_id)
2025 			continue;
2026 		if (next_fi->fib_priority > last_prio &&
2027 		    fa->fa_dscp == last_dscp) {
2028 			if (last_dscp)
2029 				continue;
2030 			break;
2031 		}
2032 		if (next_fi->fib_flags & RTNH_F_DEAD)
2033 			continue;
2034 		last_dscp = fa->fa_dscp;
2035 		last_prio = next_fi->fib_priority;
2036 
2037 		if (next_fi->fib_scope != res->scope ||
2038 		    fa->fa_type != RTN_UNICAST)
2039 			continue;
2040 
2041 		nhc = fib_info_nhc(next_fi, 0);
2042 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2043 			continue;
2044 
2045 		fib_alias_accessed(fa);
2046 
2047 		if (!fi) {
2048 			if (next_fi != res->fi)
2049 				break;
2050 			fa1 = fa;
2051 		} else if (!fib_detect_death(fi, order, &last_resort,
2052 					     &last_idx, fa1->fa_default)) {
2053 			fib_result_assign(res, fi);
2054 			fa1->fa_default = order;
2055 			goto out;
2056 		}
2057 		fi = next_fi;
2058 		order++;
2059 	}
2060 
2061 	if (order <= 0 || !fi) {
2062 		if (fa1)
2063 			fa1->fa_default = -1;
2064 		goto out;
2065 	}
2066 
2067 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2068 			      fa1->fa_default)) {
2069 		fib_result_assign(res, fi);
2070 		fa1->fa_default = order;
2071 		goto out;
2072 	}
2073 
2074 	if (last_idx >= 0)
2075 		fib_result_assign(res, last_resort);
2076 	fa1->fa_default = last_idx;
2077 out:
2078 	return;
2079 }
2080 
2081 /*
2082  * Dead device goes up. We wake up dead nexthops.
2083  * It takes sense only on multipath routes.
2084  *
2085  * only used when fib_nh is built into fib_info
2086  */
2087 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2088 {
2089 	struct fib_info *prev_fi;
2090 	struct hlist_head *head;
2091 	struct fib_nh *nh;
2092 	int ret;
2093 
2094 	if (!(dev->flags & IFF_UP))
2095 		return 0;
2096 
2097 	if (nh_flags & RTNH_F_DEAD) {
2098 		unsigned int flags = dev_get_flags(dev);
2099 
2100 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2101 			nh_flags |= RTNH_F_LINKDOWN;
2102 	}
2103 
2104 	prev_fi = NULL;
2105 	head = fib_nh_head(dev);
2106 	ret = 0;
2107 
2108 	hlist_for_each_entry(nh, head, nh_hash) {
2109 		struct fib_info *fi = nh->nh_parent;
2110 		int alive;
2111 
2112 		BUG_ON(!fi->fib_nhs);
2113 		DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev);
2114 		if (fi == prev_fi)
2115 			continue;
2116 
2117 		prev_fi = fi;
2118 		alive = 0;
2119 		change_nexthops(fi) {
2120 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2121 				alive++;
2122 				continue;
2123 			}
2124 			if (!nexthop_nh->fib_nh_dev ||
2125 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2126 				continue;
2127 			if (nexthop_nh->fib_nh_dev != dev ||
2128 			    !__in_dev_get_rtnl(dev))
2129 				continue;
2130 			alive++;
2131 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2132 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2133 		} endfor_nexthops(fi)
2134 
2135 		if (alive > 0) {
2136 			fi->fib_flags &= ~nh_flags;
2137 			ret++;
2138 		}
2139 
2140 		fib_rebalance(fi);
2141 	}
2142 
2143 	return ret;
2144 }
2145 
2146 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2147 static bool fib_good_nh(const struct fib_nh *nh)
2148 {
2149 	int state = NUD_REACHABLE;
2150 
2151 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2152 		struct neighbour *n;
2153 
2154 		rcu_read_lock();
2155 
2156 		if (likely(nh->fib_nh_gw_family == AF_INET))
2157 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2158 						   (__force u32)nh->fib_nh_gw4);
2159 		else if (nh->fib_nh_gw_family == AF_INET6)
2160 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2161 							   &nh->fib_nh_gw6);
2162 		else
2163 			n = NULL;
2164 		if (n)
2165 			state = READ_ONCE(n->nud_state);
2166 
2167 		rcu_read_unlock();
2168 	}
2169 
2170 	return !!(state & NUD_VALID);
2171 }
2172 
2173 void fib_select_multipath(struct fib_result *res, int hash,
2174 			  const struct flowi4 *fl4)
2175 {
2176 	struct fib_info *fi = res->fi;
2177 	struct net *net = fi->fib_net;
2178 	bool found = false;
2179 	bool use_neigh;
2180 	__be32 saddr;
2181 
2182 	if (unlikely(res->fi->nh)) {
2183 		nexthop_path_fib_result(res, hash);
2184 		return;
2185 	}
2186 
2187 	use_neigh = READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh);
2188 	saddr = fl4 ? fl4->saddr : 0;
2189 
2190 	change_nexthops(fi) {
2191 		int nh_upper_bound;
2192 
2193 		/* Nexthops without a carrier are assigned an upper bound of
2194 		 * minus one when "ignore_routes_with_linkdown" is set.
2195 		 */
2196 		nh_upper_bound = atomic_read(&nexthop_nh->fib_nh_upper_bound);
2197 		if (nh_upper_bound == -1 ||
2198 		    (use_neigh && !fib_good_nh(nexthop_nh)))
2199 			continue;
2200 
2201 		if (!found) {
2202 			res->nh_sel = nhsel;
2203 			res->nhc = &nexthop_nh->nh_common;
2204 			found = !saddr || nexthop_nh->nh_saddr == saddr;
2205 		}
2206 
2207 		if (hash > nh_upper_bound)
2208 			continue;
2209 
2210 		if (!saddr || nexthop_nh->nh_saddr == saddr) {
2211 			res->nh_sel = nhsel;
2212 			res->nhc = &nexthop_nh->nh_common;
2213 			return;
2214 		}
2215 
2216 		if (found)
2217 			return;
2218 
2219 	} endfor_nexthops(fi);
2220 }
2221 #endif
2222 
2223 void fib_select_path(struct net *net, struct fib_result *res,
2224 		     struct flowi4 *fl4, const struct sk_buff *skb)
2225 {
2226 	if (fl4->flowi4_oif)
2227 		goto check_saddr;
2228 
2229 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2230 	if (fib_info_num_path(res->fi) > 1) {
2231 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2232 
2233 		fib_select_multipath(res, h, fl4);
2234 	}
2235 	else
2236 #endif
2237 	if (!res->prefixlen &&
2238 	    res->table->tb_num_default > 1 &&
2239 	    res->type == RTN_UNICAST)
2240 		fib_select_default(fl4, res);
2241 
2242 check_saddr:
2243 	if (!fl4->saddr) {
2244 		struct net_device *l3mdev;
2245 
2246 		l3mdev = dev_get_by_index_rcu(net, fl4->flowi4_l3mdev);
2247 
2248 		if (!l3mdev ||
2249 		    l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) == l3mdev)
2250 			fl4->saddr = fib_result_prefsrc(net, res);
2251 		else
2252 			fl4->saddr = inet_select_addr(l3mdev, 0, RT_SCOPE_LINK);
2253 	}
2254 }
2255 
2256 int __net_init fib4_semantics_init(struct net *net)
2257 {
2258 	unsigned int hash_bits = 4;
2259 
2260 	net->ipv4.fib_info_hash = fib_info_hash_alloc(hash_bits);
2261 	if (!net->ipv4.fib_info_hash)
2262 		return -ENOMEM;
2263 
2264 	net->ipv4.fib_info_hash_bits = hash_bits;
2265 	net->ipv4.fib_info_cnt = 0;
2266 
2267 	return 0;
2268 }
2269 
2270 void __net_exit fib4_semantics_exit(struct net *net)
2271 {
2272 	fib_info_hash_free(net->ipv4.fib_info_hash);
2273 }
2274