xref: /linux/net/ipv4/fib_semantics.c (revision a6cdeeb16bff89c8486324f53577db058cbe81ba)
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 
33 #include <net/arp.h>
34 #include <net/ip.h>
35 #include <net/protocol.h>
36 #include <net/route.h>
37 #include <net/tcp.h>
38 #include <net/sock.h>
39 #include <net/ip_fib.h>
40 #include <net/ip6_fib.h>
41 #include <net/nexthop.h>
42 #include <net/netlink.h>
43 #include <net/rtnh.h>
44 #include <net/lwtunnel.h>
45 #include <net/fib_notifier.h>
46 #include <net/addrconf.h>
47 
48 #include "fib_lookup.h"
49 
50 static DEFINE_SPINLOCK(fib_info_lock);
51 static struct hlist_head *fib_info_hash;
52 static struct hlist_head *fib_info_laddrhash;
53 static unsigned int fib_info_hash_size;
54 static unsigned int fib_info_cnt;
55 
56 #define DEVINDEX_HASHBITS 8
57 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
58 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
59 
60 /* for_nexthops and change_nexthops only used when nexthop object
61  * is not set in a fib_info. The logic within can reference fib_nh.
62  */
63 #ifdef CONFIG_IP_ROUTE_MULTIPATH
64 
65 #define for_nexthops(fi) {						\
66 	int nhsel; const struct fib_nh *nh;				\
67 	for (nhsel = 0, nh = (fi)->fib_nh;				\
68 	     nhsel < fib_info_num_path((fi));				\
69 	     nh++, nhsel++)
70 
71 #define change_nexthops(fi) {						\
72 	int nhsel; struct fib_nh *nexthop_nh;				\
73 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
74 	     nhsel < fib_info_num_path((fi));				\
75 	     nexthop_nh++, nhsel++)
76 
77 #else /* CONFIG_IP_ROUTE_MULTIPATH */
78 
79 /* Hope, that gcc will optimize it to get rid of dummy loop */
80 
81 #define for_nexthops(fi) {						\
82 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
83 	for (nhsel = 0; nhsel < 1; nhsel++)
84 
85 #define change_nexthops(fi) {						\
86 	int nhsel;							\
87 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
88 	for (nhsel = 0; nhsel < 1; nhsel++)
89 
90 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
91 
92 #define endfor_nexthops(fi) }
93 
94 
95 const struct fib_prop fib_props[RTN_MAX + 1] = {
96 	[RTN_UNSPEC] = {
97 		.error	= 0,
98 		.scope	= RT_SCOPE_NOWHERE,
99 	},
100 	[RTN_UNICAST] = {
101 		.error	= 0,
102 		.scope	= RT_SCOPE_UNIVERSE,
103 	},
104 	[RTN_LOCAL] = {
105 		.error	= 0,
106 		.scope	= RT_SCOPE_HOST,
107 	},
108 	[RTN_BROADCAST] = {
109 		.error	= 0,
110 		.scope	= RT_SCOPE_LINK,
111 	},
112 	[RTN_ANYCAST] = {
113 		.error	= 0,
114 		.scope	= RT_SCOPE_LINK,
115 	},
116 	[RTN_MULTICAST] = {
117 		.error	= 0,
118 		.scope	= RT_SCOPE_UNIVERSE,
119 	},
120 	[RTN_BLACKHOLE] = {
121 		.error	= -EINVAL,
122 		.scope	= RT_SCOPE_UNIVERSE,
123 	},
124 	[RTN_UNREACHABLE] = {
125 		.error	= -EHOSTUNREACH,
126 		.scope	= RT_SCOPE_UNIVERSE,
127 	},
128 	[RTN_PROHIBIT] = {
129 		.error	= -EACCES,
130 		.scope	= RT_SCOPE_UNIVERSE,
131 	},
132 	[RTN_THROW] = {
133 		.error	= -EAGAIN,
134 		.scope	= RT_SCOPE_UNIVERSE,
135 	},
136 	[RTN_NAT] = {
137 		.error	= -EINVAL,
138 		.scope	= RT_SCOPE_NOWHERE,
139 	},
140 	[RTN_XRESOLVE] = {
141 		.error	= -EINVAL,
142 		.scope	= RT_SCOPE_NOWHERE,
143 	},
144 };
145 
146 static void rt_fibinfo_free(struct rtable __rcu **rtp)
147 {
148 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
149 
150 	if (!rt)
151 		return;
152 
153 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
154 	 * because we waited an RCU grace period before calling
155 	 * free_fib_info_rcu()
156 	 */
157 
158 	dst_dev_put(&rt->dst);
159 	dst_release_immediate(&rt->dst);
160 }
161 
162 static void free_nh_exceptions(struct fib_nh_common *nhc)
163 {
164 	struct fnhe_hash_bucket *hash;
165 	int i;
166 
167 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
168 	if (!hash)
169 		return;
170 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
171 		struct fib_nh_exception *fnhe;
172 
173 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
174 		while (fnhe) {
175 			struct fib_nh_exception *next;
176 
177 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
178 
179 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
180 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
181 
182 			kfree(fnhe);
183 
184 			fnhe = next;
185 		}
186 	}
187 	kfree(hash);
188 }
189 
190 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
191 {
192 	int cpu;
193 
194 	if (!rtp)
195 		return;
196 
197 	for_each_possible_cpu(cpu) {
198 		struct rtable *rt;
199 
200 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
201 		if (rt) {
202 			dst_dev_put(&rt->dst);
203 			dst_release_immediate(&rt->dst);
204 		}
205 	}
206 	free_percpu(rtp);
207 }
208 
209 void fib_nh_common_release(struct fib_nh_common *nhc)
210 {
211 	if (nhc->nhc_dev)
212 		dev_put(nhc->nhc_dev);
213 
214 	lwtstate_put(nhc->nhc_lwtstate);
215 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
216 	rt_fibinfo_free(&nhc->nhc_rth_input);
217 	free_nh_exceptions(nhc);
218 }
219 EXPORT_SYMBOL_GPL(fib_nh_common_release);
220 
221 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
222 {
223 #ifdef CONFIG_IP_ROUTE_CLASSID
224 	if (fib_nh->nh_tclassid)
225 		net->ipv4.fib_num_tclassid_users--;
226 #endif
227 	fib_nh_common_release(&fib_nh->nh_common);
228 }
229 
230 /* Release a nexthop info record */
231 static void free_fib_info_rcu(struct rcu_head *head)
232 {
233 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
234 
235 	if (fi->nh) {
236 		nexthop_put(fi->nh);
237 	} else {
238 		change_nexthops(fi) {
239 			fib_nh_release(fi->fib_net, nexthop_nh);
240 		} endfor_nexthops(fi);
241 	}
242 
243 	ip_fib_metrics_put(fi->fib_metrics);
244 
245 	kfree(fi);
246 }
247 
248 void free_fib_info(struct fib_info *fi)
249 {
250 	if (fi->fib_dead == 0) {
251 		pr_warn("Freeing alive fib_info %p\n", fi);
252 		return;
253 	}
254 	fib_info_cnt--;
255 
256 	call_rcu(&fi->rcu, free_fib_info_rcu);
257 }
258 EXPORT_SYMBOL_GPL(free_fib_info);
259 
260 void fib_release_info(struct fib_info *fi)
261 {
262 	spin_lock_bh(&fib_info_lock);
263 	if (fi && --fi->fib_treeref == 0) {
264 		hlist_del(&fi->fib_hash);
265 		if (fi->fib_prefsrc)
266 			hlist_del(&fi->fib_lhash);
267 		if (fi->nh) {
268 			list_del(&fi->nh_list);
269 		} else {
270 			change_nexthops(fi) {
271 				if (!nexthop_nh->fib_nh_dev)
272 					continue;
273 				hlist_del(&nexthop_nh->nh_hash);
274 			} endfor_nexthops(fi)
275 		}
276 		fi->fib_dead = 1;
277 		fib_info_put(fi);
278 	}
279 	spin_unlock_bh(&fib_info_lock);
280 }
281 
282 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
283 {
284 	const struct fib_nh *onh;
285 
286 	if (fi->nh || ofi->nh)
287 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
288 
289 	if (ofi->fib_nhs == 0)
290 		return 0;
291 
292 	for_nexthops(fi) {
293 		onh = fib_info_nh(ofi, nhsel);
294 
295 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
296 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
297 		    nh->fib_nh_scope != onh->fib_nh_scope ||
298 #ifdef CONFIG_IP_ROUTE_MULTIPATH
299 		    nh->fib_nh_weight != onh->fib_nh_weight ||
300 #endif
301 #ifdef CONFIG_IP_ROUTE_CLASSID
302 		    nh->nh_tclassid != onh->nh_tclassid ||
303 #endif
304 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
305 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
306 			return -1;
307 
308 		if (nh->fib_nh_gw_family == AF_INET &&
309 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
310 			return -1;
311 
312 		if (nh->fib_nh_gw_family == AF_INET6 &&
313 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
314 			return -1;
315 	} endfor_nexthops(fi);
316 	return 0;
317 }
318 
319 static inline unsigned int fib_devindex_hashfn(unsigned int val)
320 {
321 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
322 
323 	return (val ^
324 		(val >> DEVINDEX_HASHBITS) ^
325 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
326 }
327 
328 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
329 {
330 	unsigned int mask = (fib_info_hash_size - 1);
331 	unsigned int val = fi->fib_nhs;
332 
333 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
334 	val ^= (__force u32)fi->fib_prefsrc;
335 	val ^= fi->fib_priority;
336 
337 	if (fi->nh) {
338 		val ^= fib_devindex_hashfn(fi->nh->id);
339 	} else {
340 		for_nexthops(fi) {
341 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
342 		} endfor_nexthops(fi)
343 	}
344 
345 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
346 }
347 
348 static struct fib_info *fib_find_info(struct fib_info *nfi)
349 {
350 	struct hlist_head *head;
351 	struct fib_info *fi;
352 	unsigned int hash;
353 
354 	hash = fib_info_hashfn(nfi);
355 	head = &fib_info_hash[hash];
356 
357 	hlist_for_each_entry(fi, head, fib_hash) {
358 		if (!net_eq(fi->fib_net, nfi->fib_net))
359 			continue;
360 		if (fi->fib_nhs != nfi->fib_nhs)
361 			continue;
362 		if (nfi->fib_protocol == fi->fib_protocol &&
363 		    nfi->fib_scope == fi->fib_scope &&
364 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
365 		    nfi->fib_priority == fi->fib_priority &&
366 		    nfi->fib_type == fi->fib_type &&
367 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
368 			   sizeof(u32) * RTAX_MAX) == 0 &&
369 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
370 		    nh_comp(fi, nfi) == 0)
371 			return fi;
372 	}
373 
374 	return NULL;
375 }
376 
377 /* Check, that the gateway is already configured.
378  * Used only by redirect accept routine.
379  */
380 int ip_fib_check_default(__be32 gw, struct net_device *dev)
381 {
382 	struct hlist_head *head;
383 	struct fib_nh *nh;
384 	unsigned int hash;
385 
386 	spin_lock(&fib_info_lock);
387 
388 	hash = fib_devindex_hashfn(dev->ifindex);
389 	head = &fib_info_devhash[hash];
390 	hlist_for_each_entry(nh, head, nh_hash) {
391 		if (nh->fib_nh_dev == dev &&
392 		    nh->fib_nh_gw4 == gw &&
393 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
394 			spin_unlock(&fib_info_lock);
395 			return 0;
396 		}
397 	}
398 
399 	spin_unlock(&fib_info_lock);
400 
401 	return -1;
402 }
403 
404 static inline size_t fib_nlmsg_size(struct fib_info *fi)
405 {
406 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
407 			 + nla_total_size(4) /* RTA_TABLE */
408 			 + nla_total_size(4) /* RTA_DST */
409 			 + nla_total_size(4) /* RTA_PRIORITY */
410 			 + nla_total_size(4) /* RTA_PREFSRC */
411 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
412 	unsigned int nhs = fib_info_num_path(fi);
413 
414 	/* space for nested metrics */
415 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
416 
417 	if (fi->nh)
418 		payload += nla_total_size(4); /* RTA_NH_ID */
419 
420 	if (nhs) {
421 		size_t nh_encapsize = 0;
422 		/* Also handles the special case nhs == 1 */
423 
424 		/* each nexthop is packed in an attribute */
425 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
426 		unsigned int i;
427 
428 		/* may contain flow and gateway attribute */
429 		nhsize += 2 * nla_total_size(4);
430 
431 		/* grab encap info */
432 		for (i = 0; i < fib_info_num_path(fi); i++) {
433 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
434 
435 			if (nhc->nhc_lwtstate) {
436 				/* RTA_ENCAP_TYPE */
437 				nh_encapsize += lwtunnel_get_encap_size(
438 						nhc->nhc_lwtstate);
439 				/* RTA_ENCAP */
440 				nh_encapsize +=  nla_total_size(2);
441 			}
442 		}
443 
444 		/* all nexthops are packed in a nested attribute */
445 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
446 
447 	}
448 
449 	return payload;
450 }
451 
452 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
453 	       int dst_len, u32 tb_id, const struct nl_info *info,
454 	       unsigned int nlm_flags)
455 {
456 	struct sk_buff *skb;
457 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
458 	int err = -ENOBUFS;
459 
460 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
461 	if (!skb)
462 		goto errout;
463 
464 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
465 			    fa->fa_type, key, dst_len,
466 			    fa->fa_tos, fa->fa_info, nlm_flags);
467 	if (err < 0) {
468 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
469 		WARN_ON(err == -EMSGSIZE);
470 		kfree_skb(skb);
471 		goto errout;
472 	}
473 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
474 		    info->nlh, GFP_KERNEL);
475 	return;
476 errout:
477 	if (err < 0)
478 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
479 }
480 
481 static int fib_detect_death(struct fib_info *fi, int order,
482 			    struct fib_info **last_resort, int *last_idx,
483 			    int dflt)
484 {
485 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
486 	struct neighbour *n;
487 	int state = NUD_NONE;
488 
489 	if (likely(nhc->nhc_gw_family == AF_INET))
490 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
491 	else if (nhc->nhc_gw_family == AF_INET6)
492 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
493 				 nhc->nhc_dev);
494 	else
495 		n = NULL;
496 
497 	if (n) {
498 		state = n->nud_state;
499 		neigh_release(n);
500 	} else {
501 		return 0;
502 	}
503 	if (state == NUD_REACHABLE)
504 		return 0;
505 	if ((state & NUD_VALID) && order != dflt)
506 		return 0;
507 	if ((state & NUD_VALID) ||
508 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
509 		*last_resort = fi;
510 		*last_idx = order;
511 	}
512 	return 1;
513 }
514 
515 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
516 		       u16 encap_type, void *cfg, gfp_t gfp_flags,
517 		       struct netlink_ext_ack *extack)
518 {
519 	int err;
520 
521 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
522 						    gfp_flags);
523 	if (!nhc->nhc_pcpu_rth_output)
524 		return -ENOMEM;
525 
526 	if (encap) {
527 		struct lwtunnel_state *lwtstate;
528 
529 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
530 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
531 			err = -EINVAL;
532 			goto lwt_failure;
533 		}
534 		err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family,
535 					   cfg, &lwtstate, extack);
536 		if (err)
537 			goto lwt_failure;
538 
539 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
540 	}
541 
542 	return 0;
543 
544 lwt_failure:
545 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
546 	nhc->nhc_pcpu_rth_output = NULL;
547 	return err;
548 }
549 EXPORT_SYMBOL_GPL(fib_nh_common_init);
550 
551 int fib_nh_init(struct net *net, struct fib_nh *nh,
552 		struct fib_config *cfg, int nh_weight,
553 		struct netlink_ext_ack *extack)
554 {
555 	int err;
556 
557 	nh->fib_nh_family = AF_INET;
558 
559 	err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
560 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
561 	if (err)
562 		return err;
563 
564 	nh->fib_nh_oif = cfg->fc_oif;
565 	nh->fib_nh_gw_family = cfg->fc_gw_family;
566 	if (cfg->fc_gw_family == AF_INET)
567 		nh->fib_nh_gw4 = cfg->fc_gw4;
568 	else if (cfg->fc_gw_family == AF_INET6)
569 		nh->fib_nh_gw6 = cfg->fc_gw6;
570 
571 	nh->fib_nh_flags = cfg->fc_flags;
572 
573 #ifdef CONFIG_IP_ROUTE_CLASSID
574 	nh->nh_tclassid = cfg->fc_flow;
575 	if (nh->nh_tclassid)
576 		net->ipv4.fib_num_tclassid_users++;
577 #endif
578 #ifdef CONFIG_IP_ROUTE_MULTIPATH
579 	nh->fib_nh_weight = nh_weight;
580 #endif
581 	return 0;
582 }
583 
584 #ifdef CONFIG_IP_ROUTE_MULTIPATH
585 
586 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
587 			      struct netlink_ext_ack *extack)
588 {
589 	int nhs = 0;
590 
591 	while (rtnh_ok(rtnh, remaining)) {
592 		nhs++;
593 		rtnh = rtnh_next(rtnh, &remaining);
594 	}
595 
596 	/* leftover implies invalid nexthop configuration, discard it */
597 	if (remaining > 0) {
598 		NL_SET_ERR_MSG(extack,
599 			       "Invalid nexthop configuration - extra data after nexthops");
600 		nhs = 0;
601 	}
602 
603 	return nhs;
604 }
605 
606 /* only called when fib_nh is integrated into fib_info */
607 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
608 		       int remaining, struct fib_config *cfg,
609 		       struct netlink_ext_ack *extack)
610 {
611 	struct net *net = fi->fib_net;
612 	struct fib_config fib_cfg;
613 	struct fib_nh *nh;
614 	int ret;
615 
616 	change_nexthops(fi) {
617 		int attrlen;
618 
619 		memset(&fib_cfg, 0, sizeof(fib_cfg));
620 
621 		if (!rtnh_ok(rtnh, remaining)) {
622 			NL_SET_ERR_MSG(extack,
623 				       "Invalid nexthop configuration - extra data after nexthop");
624 			return -EINVAL;
625 		}
626 
627 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
628 			NL_SET_ERR_MSG(extack,
629 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
630 			return -EINVAL;
631 		}
632 
633 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
634 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
635 
636 		attrlen = rtnh_attrlen(rtnh);
637 		if (attrlen > 0) {
638 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
639 
640 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
641 			nlav = nla_find(attrs, attrlen, RTA_VIA);
642 			if (nla && nlav) {
643 				NL_SET_ERR_MSG(extack,
644 					       "Nexthop configuration can not contain both GATEWAY and VIA");
645 				return -EINVAL;
646 			}
647 			if (nla) {
648 				fib_cfg.fc_gw4 = nla_get_in_addr(nla);
649 				if (fib_cfg.fc_gw4)
650 					fib_cfg.fc_gw_family = AF_INET;
651 			} else if (nlav) {
652 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
653 				if (ret)
654 					goto errout;
655 			}
656 
657 			nla = nla_find(attrs, attrlen, RTA_FLOW);
658 			if (nla)
659 				fib_cfg.fc_flow = nla_get_u32(nla);
660 
661 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
662 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
663 			if (nla)
664 				fib_cfg.fc_encap_type = nla_get_u16(nla);
665 		}
666 
667 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
668 				  rtnh->rtnh_hops + 1, extack);
669 		if (ret)
670 			goto errout;
671 
672 		rtnh = rtnh_next(rtnh, &remaining);
673 	} endfor_nexthops(fi);
674 
675 	ret = -EINVAL;
676 	nh = fib_info_nh(fi, 0);
677 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
678 		NL_SET_ERR_MSG(extack,
679 			       "Nexthop device index does not match RTA_OIF");
680 		goto errout;
681 	}
682 	if (cfg->fc_gw_family) {
683 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
684 		    (cfg->fc_gw_family == AF_INET &&
685 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
686 		    (cfg->fc_gw_family == AF_INET6 &&
687 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
688 			NL_SET_ERR_MSG(extack,
689 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
690 			goto errout;
691 		}
692 	}
693 #ifdef CONFIG_IP_ROUTE_CLASSID
694 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
695 		NL_SET_ERR_MSG(extack,
696 			       "Nexthop class id does not match RTA_FLOW");
697 		goto errout;
698 	}
699 #endif
700 	ret = 0;
701 errout:
702 	return ret;
703 }
704 
705 /* only called when fib_nh is integrated into fib_info */
706 static void fib_rebalance(struct fib_info *fi)
707 {
708 	int total;
709 	int w;
710 
711 	if (fib_info_num_path(fi) < 2)
712 		return;
713 
714 	total = 0;
715 	for_nexthops(fi) {
716 		if (nh->fib_nh_flags & RTNH_F_DEAD)
717 			continue;
718 
719 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
720 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
721 			continue;
722 
723 		total += nh->fib_nh_weight;
724 	} endfor_nexthops(fi);
725 
726 	w = 0;
727 	change_nexthops(fi) {
728 		int upper_bound;
729 
730 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
731 			upper_bound = -1;
732 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
733 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
734 			upper_bound = -1;
735 		} else {
736 			w += nexthop_nh->fib_nh_weight;
737 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
738 							    total) - 1;
739 		}
740 
741 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
742 	} endfor_nexthops(fi);
743 }
744 #else /* CONFIG_IP_ROUTE_MULTIPATH */
745 
746 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
747 		       int remaining, struct fib_config *cfg,
748 		       struct netlink_ext_ack *extack)
749 {
750 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
751 
752 	return -EINVAL;
753 }
754 
755 #define fib_rebalance(fi) do { } while (0)
756 
757 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
758 
759 static int fib_encap_match(u16 encap_type,
760 			   struct nlattr *encap,
761 			   const struct fib_nh *nh,
762 			   const struct fib_config *cfg,
763 			   struct netlink_ext_ack *extack)
764 {
765 	struct lwtunnel_state *lwtstate;
766 	int ret, result = 0;
767 
768 	if (encap_type == LWTUNNEL_ENCAP_NONE)
769 		return 0;
770 
771 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
772 				   cfg, &lwtstate, extack);
773 	if (!ret) {
774 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
775 		lwtstate_free(lwtstate);
776 	}
777 
778 	return result;
779 }
780 
781 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
782 		 struct netlink_ext_ack *extack)
783 {
784 #ifdef CONFIG_IP_ROUTE_MULTIPATH
785 	struct rtnexthop *rtnh;
786 	int remaining;
787 #endif
788 
789 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
790 		return 1;
791 
792 	if (cfg->fc_oif || cfg->fc_gw_family) {
793 		struct fib_nh *nh = fib_info_nh(fi, 0);
794 
795 		if (cfg->fc_encap) {
796 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
797 					    nh, cfg, extack))
798 				return 1;
799 		}
800 #ifdef CONFIG_IP_ROUTE_CLASSID
801 		if (cfg->fc_flow &&
802 		    cfg->fc_flow != nh->nh_tclassid)
803 			return 1;
804 #endif
805 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
806 		    (cfg->fc_gw_family &&
807 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
808 			return 1;
809 
810 		if (cfg->fc_gw_family == AF_INET &&
811 		    cfg->fc_gw4 != nh->fib_nh_gw4)
812 			return 1;
813 
814 		if (cfg->fc_gw_family == AF_INET6 &&
815 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
816 			return 1;
817 
818 		return 0;
819 	}
820 
821 #ifdef CONFIG_IP_ROUTE_MULTIPATH
822 	if (!cfg->fc_mp)
823 		return 0;
824 
825 	rtnh = cfg->fc_mp;
826 	remaining = cfg->fc_mp_len;
827 
828 	for_nexthops(fi) {
829 		int attrlen;
830 
831 		if (!rtnh_ok(rtnh, remaining))
832 			return -EINVAL;
833 
834 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
835 			return 1;
836 
837 		attrlen = rtnh_attrlen(rtnh);
838 		if (attrlen > 0) {
839 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
840 
841 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
842 			nlav = nla_find(attrs, attrlen, RTA_VIA);
843 			if (nla && nlav) {
844 				NL_SET_ERR_MSG(extack,
845 					       "Nexthop configuration can not contain both GATEWAY and VIA");
846 				return -EINVAL;
847 			}
848 
849 			if (nla) {
850 				if (nh->fib_nh_gw_family != AF_INET ||
851 				    nla_get_in_addr(nla) != nh->fib_nh_gw4)
852 					return 1;
853 			} else if (nlav) {
854 				struct fib_config cfg2;
855 				int err;
856 
857 				err = fib_gw_from_via(&cfg2, nlav, extack);
858 				if (err)
859 					return err;
860 
861 				switch (nh->fib_nh_gw_family) {
862 				case AF_INET:
863 					if (cfg2.fc_gw_family != AF_INET ||
864 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
865 						return 1;
866 					break;
867 				case AF_INET6:
868 					if (cfg2.fc_gw_family != AF_INET6 ||
869 					    ipv6_addr_cmp(&cfg2.fc_gw6,
870 							  &nh->fib_nh_gw6))
871 						return 1;
872 					break;
873 				}
874 			}
875 
876 #ifdef CONFIG_IP_ROUTE_CLASSID
877 			nla = nla_find(attrs, attrlen, RTA_FLOW);
878 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
879 				return 1;
880 #endif
881 		}
882 
883 		rtnh = rtnh_next(rtnh, &remaining);
884 	} endfor_nexthops(fi);
885 #endif
886 	return 0;
887 }
888 
889 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
890 {
891 	struct nlattr *nla;
892 	int remaining;
893 
894 	if (!cfg->fc_mx)
895 		return true;
896 
897 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
898 		int type = nla_type(nla);
899 		u32 fi_val, val;
900 
901 		if (!type)
902 			continue;
903 		if (type > RTAX_MAX)
904 			return false;
905 
906 		if (type == RTAX_CC_ALGO) {
907 			char tmp[TCP_CA_NAME_MAX];
908 			bool ecn_ca = false;
909 
910 			nla_strlcpy(tmp, nla, sizeof(tmp));
911 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
912 		} else {
913 			if (nla_len(nla) != sizeof(u32))
914 				return false;
915 			val = nla_get_u32(nla);
916 		}
917 
918 		fi_val = fi->fib_metrics->metrics[type - 1];
919 		if (type == RTAX_FEATURES)
920 			fi_val &= ~DST_FEATURE_ECN_CA;
921 
922 		if (fi_val != val)
923 			return false;
924 	}
925 
926 	return true;
927 }
928 
929 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
930 			      u32 table, struct netlink_ext_ack *extack)
931 {
932 	struct fib6_config cfg = {
933 		.fc_table = table,
934 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
935 		.fc_ifindex = nh->fib_nh_oif,
936 		.fc_gateway = nh->fib_nh_gw6,
937 	};
938 	struct fib6_nh fib6_nh = {};
939 	int err;
940 
941 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
942 	if (!err) {
943 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
944 		dev_hold(nh->fib_nh_dev);
945 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
946 		nh->fib_nh_scope = RT_SCOPE_LINK;
947 
948 		ipv6_stub->fib6_nh_release(&fib6_nh);
949 	}
950 
951 	return err;
952 }
953 
954 /*
955  * Picture
956  * -------
957  *
958  * Semantics of nexthop is very messy by historical reasons.
959  * We have to take into account, that:
960  * a) gateway can be actually local interface address,
961  *    so that gatewayed route is direct.
962  * b) gateway must be on-link address, possibly
963  *    described not by an ifaddr, but also by a direct route.
964  * c) If both gateway and interface are specified, they should not
965  *    contradict.
966  * d) If we use tunnel routes, gateway could be not on-link.
967  *
968  * Attempt to reconcile all of these (alas, self-contradictory) conditions
969  * results in pretty ugly and hairy code with obscure logic.
970  *
971  * I chose to generalized it instead, so that the size
972  * of code does not increase practically, but it becomes
973  * much more general.
974  * Every prefix is assigned a "scope" value: "host" is local address,
975  * "link" is direct route,
976  * [ ... "site" ... "interior" ... ]
977  * and "universe" is true gateway route with global meaning.
978  *
979  * Every prefix refers to a set of "nexthop"s (gw, oif),
980  * where gw must have narrower scope. This recursion stops
981  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
982  * which means that gw is forced to be on link.
983  *
984  * Code is still hairy, but now it is apparently logically
985  * consistent and very flexible. F.e. as by-product it allows
986  * to co-exists in peace independent exterior and interior
987  * routing processes.
988  *
989  * Normally it looks as following.
990  *
991  * {universe prefix}  -> (gw, oif) [scope link]
992  *		  |
993  *		  |-> {link prefix} -> (gw, oif) [scope local]
994  *					|
995  *					|-> {local prefix} (terminal node)
996  */
997 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
998 			      u8 scope, struct netlink_ext_ack *extack)
999 {
1000 	struct net_device *dev;
1001 	struct fib_result res;
1002 	int err;
1003 
1004 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1005 		unsigned int addr_type;
1006 
1007 		if (scope >= RT_SCOPE_LINK) {
1008 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1009 			return -EINVAL;
1010 		}
1011 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1012 		if (!dev) {
1013 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1014 			return -ENODEV;
1015 		}
1016 		if (!(dev->flags & IFF_UP)) {
1017 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1018 			return -ENETDOWN;
1019 		}
1020 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1021 		if (addr_type != RTN_UNICAST) {
1022 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1023 			return -EINVAL;
1024 		}
1025 		if (!netif_carrier_ok(dev))
1026 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1027 		nh->fib_nh_dev = dev;
1028 		dev_hold(dev);
1029 		nh->fib_nh_scope = RT_SCOPE_LINK;
1030 		return 0;
1031 	}
1032 	rcu_read_lock();
1033 	{
1034 		struct fib_table *tbl = NULL;
1035 		struct flowi4 fl4 = {
1036 			.daddr = nh->fib_nh_gw4,
1037 			.flowi4_scope = scope + 1,
1038 			.flowi4_oif = nh->fib_nh_oif,
1039 			.flowi4_iif = LOOPBACK_IFINDEX,
1040 		};
1041 
1042 		/* It is not necessary, but requires a bit of thinking */
1043 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1044 			fl4.flowi4_scope = RT_SCOPE_LINK;
1045 
1046 		if (table)
1047 			tbl = fib_get_table(net, table);
1048 
1049 		if (tbl)
1050 			err = fib_table_lookup(tbl, &fl4, &res,
1051 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1052 					       FIB_LOOKUP_NOREF);
1053 
1054 		/* on error or if no table given do full lookup. This
1055 		 * is needed for example when nexthops are in the local
1056 		 * table rather than the given table
1057 		 */
1058 		if (!tbl || err) {
1059 			err = fib_lookup(net, &fl4, &res,
1060 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1061 		}
1062 
1063 		if (err) {
1064 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1065 			goto out;
1066 		}
1067 	}
1068 
1069 	err = -EINVAL;
1070 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1071 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1072 		goto out;
1073 	}
1074 	nh->fib_nh_scope = res.scope;
1075 	nh->fib_nh_oif = FIB_RES_OIF(res);
1076 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1077 	if (!dev) {
1078 		NL_SET_ERR_MSG(extack,
1079 			       "No egress device for nexthop gateway");
1080 		goto out;
1081 	}
1082 	dev_hold(dev);
1083 	if (!netif_carrier_ok(dev))
1084 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1085 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1086 out:
1087 	rcu_read_unlock();
1088 	return err;
1089 }
1090 
1091 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1092 			      struct netlink_ext_ack *extack)
1093 {
1094 	struct in_device *in_dev;
1095 	int err;
1096 
1097 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1098 		NL_SET_ERR_MSG(extack,
1099 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1100 		return -EINVAL;
1101 	}
1102 
1103 	rcu_read_lock();
1104 
1105 	err = -ENODEV;
1106 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1107 	if (!in_dev)
1108 		goto out;
1109 	err = -ENETDOWN;
1110 	if (!(in_dev->dev->flags & IFF_UP)) {
1111 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1112 		goto out;
1113 	}
1114 
1115 	nh->fib_nh_dev = in_dev->dev;
1116 	dev_hold(nh->fib_nh_dev);
1117 	nh->fib_nh_scope = RT_SCOPE_HOST;
1118 	if (!netif_carrier_ok(nh->fib_nh_dev))
1119 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1120 	err = 0;
1121 out:
1122 	rcu_read_unlock();
1123 	return err;
1124 }
1125 
1126 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1127 		 struct netlink_ext_ack *extack)
1128 {
1129 	int err;
1130 
1131 	if (nh->fib_nh_gw_family == AF_INET)
1132 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1133 	else if (nh->fib_nh_gw_family == AF_INET6)
1134 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1135 	else
1136 		err = fib_check_nh_nongw(net, nh, extack);
1137 
1138 	return err;
1139 }
1140 
1141 static inline unsigned int fib_laddr_hashfn(__be32 val)
1142 {
1143 	unsigned int mask = (fib_info_hash_size - 1);
1144 
1145 	return ((__force u32)val ^
1146 		((__force u32)val >> 7) ^
1147 		((__force u32)val >> 14)) & mask;
1148 }
1149 
1150 static struct hlist_head *fib_info_hash_alloc(int bytes)
1151 {
1152 	if (bytes <= PAGE_SIZE)
1153 		return kzalloc(bytes, GFP_KERNEL);
1154 	else
1155 		return (struct hlist_head *)
1156 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1157 					 get_order(bytes));
1158 }
1159 
1160 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1161 {
1162 	if (!hash)
1163 		return;
1164 
1165 	if (bytes <= PAGE_SIZE)
1166 		kfree(hash);
1167 	else
1168 		free_pages((unsigned long) hash, get_order(bytes));
1169 }
1170 
1171 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1172 			       struct hlist_head *new_laddrhash,
1173 			       unsigned int new_size)
1174 {
1175 	struct hlist_head *old_info_hash, *old_laddrhash;
1176 	unsigned int old_size = fib_info_hash_size;
1177 	unsigned int i, bytes;
1178 
1179 	spin_lock_bh(&fib_info_lock);
1180 	old_info_hash = fib_info_hash;
1181 	old_laddrhash = fib_info_laddrhash;
1182 	fib_info_hash_size = new_size;
1183 
1184 	for (i = 0; i < old_size; i++) {
1185 		struct hlist_head *head = &fib_info_hash[i];
1186 		struct hlist_node *n;
1187 		struct fib_info *fi;
1188 
1189 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1190 			struct hlist_head *dest;
1191 			unsigned int new_hash;
1192 
1193 			new_hash = fib_info_hashfn(fi);
1194 			dest = &new_info_hash[new_hash];
1195 			hlist_add_head(&fi->fib_hash, dest);
1196 		}
1197 	}
1198 	fib_info_hash = new_info_hash;
1199 
1200 	for (i = 0; i < old_size; i++) {
1201 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1202 		struct hlist_node *n;
1203 		struct fib_info *fi;
1204 
1205 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1206 			struct hlist_head *ldest;
1207 			unsigned int new_hash;
1208 
1209 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1210 			ldest = &new_laddrhash[new_hash];
1211 			hlist_add_head(&fi->fib_lhash, ldest);
1212 		}
1213 	}
1214 	fib_info_laddrhash = new_laddrhash;
1215 
1216 	spin_unlock_bh(&fib_info_lock);
1217 
1218 	bytes = old_size * sizeof(struct hlist_head *);
1219 	fib_info_hash_free(old_info_hash, bytes);
1220 	fib_info_hash_free(old_laddrhash, bytes);
1221 }
1222 
1223 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1224 				 unsigned char scope)
1225 {
1226 	struct fib_nh *nh;
1227 
1228 	if (nhc->nhc_family != AF_INET)
1229 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1230 
1231 	nh = container_of(nhc, struct fib_nh, nh_common);
1232 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1233 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1234 
1235 	return nh->nh_saddr;
1236 }
1237 
1238 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1239 {
1240 	struct fib_nh_common *nhc = res->nhc;
1241 
1242 	if (res->fi->fib_prefsrc)
1243 		return res->fi->fib_prefsrc;
1244 
1245 	if (nhc->nhc_family == AF_INET) {
1246 		struct fib_nh *nh;
1247 
1248 		nh = container_of(nhc, struct fib_nh, nh_common);
1249 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1250 			return nh->nh_saddr;
1251 	}
1252 
1253 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1254 }
1255 
1256 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1257 {
1258 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1259 	    fib_prefsrc != cfg->fc_dst) {
1260 		u32 tb_id = cfg->fc_table;
1261 		int rc;
1262 
1263 		if (tb_id == RT_TABLE_MAIN)
1264 			tb_id = RT_TABLE_LOCAL;
1265 
1266 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1267 					  fib_prefsrc, tb_id);
1268 
1269 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1270 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1271 						  fib_prefsrc, RT_TABLE_LOCAL);
1272 		}
1273 
1274 		if (rc != RTN_LOCAL)
1275 			return false;
1276 	}
1277 	return true;
1278 }
1279 
1280 struct fib_info *fib_create_info(struct fib_config *cfg,
1281 				 struct netlink_ext_ack *extack)
1282 {
1283 	int err;
1284 	struct fib_info *fi = NULL;
1285 	struct nexthop *nh = NULL;
1286 	struct fib_info *ofi;
1287 	int nhs = 1;
1288 	struct net *net = cfg->fc_nlinfo.nl_net;
1289 
1290 	if (cfg->fc_type > RTN_MAX)
1291 		goto err_inval;
1292 
1293 	/* Fast check to catch the most weird cases */
1294 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1295 		NL_SET_ERR_MSG(extack, "Invalid scope");
1296 		goto err_inval;
1297 	}
1298 
1299 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1300 		NL_SET_ERR_MSG(extack,
1301 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1302 		goto err_inval;
1303 	}
1304 
1305 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1306 	if (cfg->fc_mp) {
1307 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1308 		if (nhs == 0)
1309 			goto err_inval;
1310 	}
1311 #endif
1312 
1313 	err = -ENOBUFS;
1314 	if (fib_info_cnt >= fib_info_hash_size) {
1315 		unsigned int new_size = fib_info_hash_size << 1;
1316 		struct hlist_head *new_info_hash;
1317 		struct hlist_head *new_laddrhash;
1318 		unsigned int bytes;
1319 
1320 		if (!new_size)
1321 			new_size = 16;
1322 		bytes = new_size * sizeof(struct hlist_head *);
1323 		new_info_hash = fib_info_hash_alloc(bytes);
1324 		new_laddrhash = fib_info_hash_alloc(bytes);
1325 		if (!new_info_hash || !new_laddrhash) {
1326 			fib_info_hash_free(new_info_hash, bytes);
1327 			fib_info_hash_free(new_laddrhash, bytes);
1328 		} else
1329 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1330 
1331 		if (!fib_info_hash_size)
1332 			goto failure;
1333 	}
1334 
1335 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1336 	if (!fi)
1337 		goto failure;
1338 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1339 					      cfg->fc_mx_len, extack);
1340 	if (IS_ERR(fi->fib_metrics)) {
1341 		err = PTR_ERR(fi->fib_metrics);
1342 		kfree(fi);
1343 		return ERR_PTR(err);
1344 	}
1345 
1346 	fib_info_cnt++;
1347 	fi->fib_net = net;
1348 	fi->fib_protocol = cfg->fc_protocol;
1349 	fi->fib_scope = cfg->fc_scope;
1350 	fi->fib_flags = cfg->fc_flags;
1351 	fi->fib_priority = cfg->fc_priority;
1352 	fi->fib_prefsrc = cfg->fc_prefsrc;
1353 	fi->fib_type = cfg->fc_type;
1354 	fi->fib_tb_id = cfg->fc_table;
1355 
1356 	fi->fib_nhs = nhs;
1357 	if (nh) {
1358 		if (!nexthop_get(nh)) {
1359 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1360 			err = -EINVAL;
1361 		} else {
1362 			err = 0;
1363 			fi->nh = nh;
1364 		}
1365 	} else {
1366 		change_nexthops(fi) {
1367 			nexthop_nh->nh_parent = fi;
1368 		} endfor_nexthops(fi)
1369 
1370 		if (cfg->fc_mp)
1371 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1372 					  extack);
1373 		else
1374 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1375 	}
1376 
1377 	if (err != 0)
1378 		goto failure;
1379 
1380 	if (fib_props[cfg->fc_type].error) {
1381 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1382 			NL_SET_ERR_MSG(extack,
1383 				       "Gateway, device and multipath can not be specified for this route type");
1384 			goto err_inval;
1385 		}
1386 		goto link_it;
1387 	} else {
1388 		switch (cfg->fc_type) {
1389 		case RTN_UNICAST:
1390 		case RTN_LOCAL:
1391 		case RTN_BROADCAST:
1392 		case RTN_ANYCAST:
1393 		case RTN_MULTICAST:
1394 			break;
1395 		default:
1396 			NL_SET_ERR_MSG(extack, "Invalid route type");
1397 			goto err_inval;
1398 		}
1399 	}
1400 
1401 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1402 		NL_SET_ERR_MSG(extack, "Invalid scope");
1403 		goto err_inval;
1404 	}
1405 
1406 	if (fi->nh) {
1407 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1408 		if (err)
1409 			goto failure;
1410 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1411 		struct fib_nh *nh = fi->fib_nh;
1412 
1413 		/* Local address is added. */
1414 		if (nhs != 1) {
1415 			NL_SET_ERR_MSG(extack,
1416 				       "Route with host scope can not have multiple nexthops");
1417 			goto err_inval;
1418 		}
1419 		if (nh->fib_nh_gw_family) {
1420 			NL_SET_ERR_MSG(extack,
1421 				       "Route with host scope can not have a gateway");
1422 			goto err_inval;
1423 		}
1424 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1425 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1426 		err = -ENODEV;
1427 		if (!nh->fib_nh_dev)
1428 			goto failure;
1429 	} else {
1430 		int linkdown = 0;
1431 
1432 		change_nexthops(fi) {
1433 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1434 					   cfg->fc_table, cfg->fc_scope,
1435 					   extack);
1436 			if (err != 0)
1437 				goto failure;
1438 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1439 				linkdown++;
1440 		} endfor_nexthops(fi)
1441 		if (linkdown == fi->fib_nhs)
1442 			fi->fib_flags |= RTNH_F_LINKDOWN;
1443 	}
1444 
1445 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1446 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1447 		goto err_inval;
1448 	}
1449 
1450 	if (!fi->nh) {
1451 		change_nexthops(fi) {
1452 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1453 						  fi->fib_scope);
1454 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1455 				fi->fib_nh_is_v6 = true;
1456 		} endfor_nexthops(fi)
1457 
1458 		fib_rebalance(fi);
1459 	}
1460 
1461 link_it:
1462 	ofi = fib_find_info(fi);
1463 	if (ofi) {
1464 		fi->fib_dead = 1;
1465 		free_fib_info(fi);
1466 		ofi->fib_treeref++;
1467 		return ofi;
1468 	}
1469 
1470 	fi->fib_treeref++;
1471 	refcount_set(&fi->fib_clntref, 1);
1472 	spin_lock_bh(&fib_info_lock);
1473 	hlist_add_head(&fi->fib_hash,
1474 		       &fib_info_hash[fib_info_hashfn(fi)]);
1475 	if (fi->fib_prefsrc) {
1476 		struct hlist_head *head;
1477 
1478 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1479 		hlist_add_head(&fi->fib_lhash, head);
1480 	}
1481 	if (fi->nh) {
1482 		list_add(&fi->nh_list, &nh->fi_list);
1483 	} else {
1484 		change_nexthops(fi) {
1485 			struct hlist_head *head;
1486 			unsigned int hash;
1487 
1488 			if (!nexthop_nh->fib_nh_dev)
1489 				continue;
1490 			hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1491 			head = &fib_info_devhash[hash];
1492 			hlist_add_head(&nexthop_nh->nh_hash, head);
1493 		} endfor_nexthops(fi)
1494 	}
1495 	spin_unlock_bh(&fib_info_lock);
1496 	return fi;
1497 
1498 err_inval:
1499 	err = -EINVAL;
1500 
1501 failure:
1502 	if (fi) {
1503 		fi->fib_dead = 1;
1504 		free_fib_info(fi);
1505 	}
1506 
1507 	return ERR_PTR(err);
1508 }
1509 
1510 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1511 		     unsigned char *flags, bool skip_oif)
1512 {
1513 	if (nhc->nhc_flags & RTNH_F_DEAD)
1514 		*flags |= RTNH_F_DEAD;
1515 
1516 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1517 		*flags |= RTNH_F_LINKDOWN;
1518 
1519 		rcu_read_lock();
1520 		switch (nhc->nhc_family) {
1521 		case AF_INET:
1522 			if (ip_ignore_linkdown(nhc->nhc_dev))
1523 				*flags |= RTNH_F_DEAD;
1524 			break;
1525 		case AF_INET6:
1526 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1527 				*flags |= RTNH_F_DEAD;
1528 			break;
1529 		}
1530 		rcu_read_unlock();
1531 	}
1532 
1533 	switch (nhc->nhc_gw_family) {
1534 	case AF_INET:
1535 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1536 			goto nla_put_failure;
1537 		break;
1538 	case AF_INET6:
1539 		/* if gateway family does not match nexthop family
1540 		 * gateway is encoded as RTA_VIA
1541 		 */
1542 		if (nhc->nhc_gw_family != nhc->nhc_family) {
1543 			int alen = sizeof(struct in6_addr);
1544 			struct nlattr *nla;
1545 			struct rtvia *via;
1546 
1547 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1548 			if (!nla)
1549 				goto nla_put_failure;
1550 
1551 			via = nla_data(nla);
1552 			via->rtvia_family = AF_INET6;
1553 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1554 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1555 					    &nhc->nhc_gw.ipv6) < 0) {
1556 			goto nla_put_failure;
1557 		}
1558 		break;
1559 	}
1560 
1561 	*flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1562 	if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1563 		*flags |= RTNH_F_OFFLOAD;
1564 
1565 	if (!skip_oif && nhc->nhc_dev &&
1566 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1567 		goto nla_put_failure;
1568 
1569 	if (nhc->nhc_lwtstate &&
1570 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1571 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1572 		goto nla_put_failure;
1573 
1574 	return 0;
1575 
1576 nla_put_failure:
1577 	return -EMSGSIZE;
1578 }
1579 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1580 
1581 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1582 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1583 		    int nh_weight)
1584 {
1585 	const struct net_device *dev = nhc->nhc_dev;
1586 	struct rtnexthop *rtnh;
1587 	unsigned char flags = 0;
1588 
1589 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1590 	if (!rtnh)
1591 		goto nla_put_failure;
1592 
1593 	rtnh->rtnh_hops = nh_weight - 1;
1594 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1595 
1596 	if (fib_nexthop_info(skb, nhc, &flags, true) < 0)
1597 		goto nla_put_failure;
1598 
1599 	rtnh->rtnh_flags = flags;
1600 
1601 	/* length of rtnetlink header + attributes */
1602 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1603 
1604 	return 0;
1605 
1606 nla_put_failure:
1607 	return -EMSGSIZE;
1608 }
1609 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1610 #endif
1611 
1612 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1613 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1614 {
1615 	struct nlattr *mp;
1616 
1617 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1618 	if (!mp)
1619 		goto nla_put_failure;
1620 
1621 	if (unlikely(fi->nh)) {
1622 		if (nexthop_mpath_fill_node(skb, fi->nh) < 0)
1623 			goto nla_put_failure;
1624 		goto mp_end;
1625 	}
1626 
1627 	for_nexthops(fi) {
1628 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight) < 0)
1629 			goto nla_put_failure;
1630 #ifdef CONFIG_IP_ROUTE_CLASSID
1631 		if (nh->nh_tclassid &&
1632 		    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1633 			goto nla_put_failure;
1634 #endif
1635 	} endfor_nexthops(fi);
1636 
1637 mp_end:
1638 	nla_nest_end(skb, mp);
1639 
1640 	return 0;
1641 
1642 nla_put_failure:
1643 	return -EMSGSIZE;
1644 }
1645 #else
1646 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1647 {
1648 	return 0;
1649 }
1650 #endif
1651 
1652 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1653 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1654 		  struct fib_info *fi, unsigned int flags)
1655 {
1656 	unsigned int nhs = fib_info_num_path(fi);
1657 	struct nlmsghdr *nlh;
1658 	struct rtmsg *rtm;
1659 
1660 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1661 	if (!nlh)
1662 		return -EMSGSIZE;
1663 
1664 	rtm = nlmsg_data(nlh);
1665 	rtm->rtm_family = AF_INET;
1666 	rtm->rtm_dst_len = dst_len;
1667 	rtm->rtm_src_len = 0;
1668 	rtm->rtm_tos = tos;
1669 	if (tb_id < 256)
1670 		rtm->rtm_table = tb_id;
1671 	else
1672 		rtm->rtm_table = RT_TABLE_COMPAT;
1673 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1674 		goto nla_put_failure;
1675 	rtm->rtm_type = type;
1676 	rtm->rtm_flags = fi->fib_flags;
1677 	rtm->rtm_scope = fi->fib_scope;
1678 	rtm->rtm_protocol = fi->fib_protocol;
1679 
1680 	if (rtm->rtm_dst_len &&
1681 	    nla_put_in_addr(skb, RTA_DST, dst))
1682 		goto nla_put_failure;
1683 	if (fi->fib_priority &&
1684 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1685 		goto nla_put_failure;
1686 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1687 		goto nla_put_failure;
1688 
1689 	if (fi->fib_prefsrc &&
1690 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1691 		goto nla_put_failure;
1692 
1693 	if (fi->nh) {
1694 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1695 			goto nla_put_failure;
1696 		if (nexthop_is_blackhole(fi->nh))
1697 			rtm->rtm_type = RTN_BLACKHOLE;
1698 	}
1699 
1700 	if (nhs == 1) {
1701 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1702 		unsigned char flags = 0;
1703 
1704 		if (fib_nexthop_info(skb, nhc, &flags, false) < 0)
1705 			goto nla_put_failure;
1706 
1707 		rtm->rtm_flags = flags;
1708 #ifdef CONFIG_IP_ROUTE_CLASSID
1709 		if (nhc->nhc_family == AF_INET) {
1710 			struct fib_nh *nh;
1711 
1712 			nh = container_of(nhc, struct fib_nh, nh_common);
1713 			if (nh->nh_tclassid &&
1714 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1715 				goto nla_put_failure;
1716 		}
1717 #endif
1718 	} else {
1719 		if (fib_add_multipath(skb, fi) < 0)
1720 			goto nla_put_failure;
1721 	}
1722 
1723 	nlmsg_end(skb, nlh);
1724 	return 0;
1725 
1726 nla_put_failure:
1727 	nlmsg_cancel(skb, nlh);
1728 	return -EMSGSIZE;
1729 }
1730 
1731 /*
1732  * Update FIB if:
1733  * - local address disappeared -> we must delete all the entries
1734  *   referring to it.
1735  * - device went down -> we must shutdown all nexthops going via it.
1736  */
1737 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1738 {
1739 	int ret = 0;
1740 	unsigned int hash = fib_laddr_hashfn(local);
1741 	struct hlist_head *head = &fib_info_laddrhash[hash];
1742 	struct net *net = dev_net(dev);
1743 	int tb_id = l3mdev_fib_table(dev);
1744 	struct fib_info *fi;
1745 
1746 	if (!fib_info_laddrhash || local == 0)
1747 		return 0;
1748 
1749 	hlist_for_each_entry(fi, head, fib_lhash) {
1750 		if (!net_eq(fi->fib_net, net) ||
1751 		    fi->fib_tb_id != tb_id)
1752 			continue;
1753 		if (fi->fib_prefsrc == local) {
1754 			fi->fib_flags |= RTNH_F_DEAD;
1755 			ret++;
1756 		}
1757 	}
1758 	return ret;
1759 }
1760 
1761 static int call_fib_nh_notifiers(struct fib_nh *nh,
1762 				 enum fib_event_type event_type)
1763 {
1764 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1765 	struct fib_nh_notifier_info info = {
1766 		.fib_nh = nh,
1767 	};
1768 
1769 	switch (event_type) {
1770 	case FIB_EVENT_NH_ADD:
1771 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1772 			break;
1773 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1774 			break;
1775 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1776 					   &info.info);
1777 	case FIB_EVENT_NH_DEL:
1778 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1779 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1780 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1781 						   event_type, &info.info);
1782 	default:
1783 		break;
1784 	}
1785 
1786 	return NOTIFY_DONE;
1787 }
1788 
1789 /* Update the PMTU of exceptions when:
1790  * - the new MTU of the first hop becomes smaller than the PMTU
1791  * - the old MTU was the same as the PMTU, and it limited discovery of
1792  *   larger MTUs on the path. With that limit raised, we can now
1793  *   discover larger MTUs
1794  * A special case is locked exceptions, for which the PMTU is smaller
1795  * than the minimal accepted PMTU:
1796  * - if the new MTU is greater than the PMTU, don't make any change
1797  * - otherwise, unlock and set PMTU
1798  */
1799 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1800 {
1801 	struct fnhe_hash_bucket *bucket;
1802 	int i;
1803 
1804 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1805 	if (!bucket)
1806 		return;
1807 
1808 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1809 		struct fib_nh_exception *fnhe;
1810 
1811 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1812 		     fnhe;
1813 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1814 			if (fnhe->fnhe_mtu_locked) {
1815 				if (new <= fnhe->fnhe_pmtu) {
1816 					fnhe->fnhe_pmtu = new;
1817 					fnhe->fnhe_mtu_locked = false;
1818 				}
1819 			} else if (new < fnhe->fnhe_pmtu ||
1820 				   orig == fnhe->fnhe_pmtu) {
1821 				fnhe->fnhe_pmtu = new;
1822 			}
1823 		}
1824 	}
1825 }
1826 
1827 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1828 {
1829 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1830 	struct hlist_head *head = &fib_info_devhash[hash];
1831 	struct fib_nh *nh;
1832 
1833 	hlist_for_each_entry(nh, head, nh_hash) {
1834 		if (nh->fib_nh_dev == dev)
1835 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1836 	}
1837 }
1838 
1839 /* Event              force Flags           Description
1840  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1841  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1842  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1843  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1844  *
1845  * only used when fib_nh is built into fib_info
1846  */
1847 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1848 {
1849 	int ret = 0;
1850 	int scope = RT_SCOPE_NOWHERE;
1851 	struct fib_info *prev_fi = NULL;
1852 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1853 	struct hlist_head *head = &fib_info_devhash[hash];
1854 	struct fib_nh *nh;
1855 
1856 	if (force)
1857 		scope = -1;
1858 
1859 	hlist_for_each_entry(nh, head, nh_hash) {
1860 		struct fib_info *fi = nh->nh_parent;
1861 		int dead;
1862 
1863 		BUG_ON(!fi->fib_nhs);
1864 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1865 			continue;
1866 		prev_fi = fi;
1867 		dead = 0;
1868 		change_nexthops(fi) {
1869 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1870 				dead++;
1871 			else if (nexthop_nh->fib_nh_dev == dev &&
1872 				 nexthop_nh->fib_nh_scope != scope) {
1873 				switch (event) {
1874 				case NETDEV_DOWN:
1875 				case NETDEV_UNREGISTER:
1876 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1877 					/* fall through */
1878 				case NETDEV_CHANGE:
1879 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1880 					break;
1881 				}
1882 				call_fib_nh_notifiers(nexthop_nh,
1883 						      FIB_EVENT_NH_DEL);
1884 				dead++;
1885 			}
1886 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1887 			if (event == NETDEV_UNREGISTER &&
1888 			    nexthop_nh->fib_nh_dev == dev) {
1889 				dead = fi->fib_nhs;
1890 				break;
1891 			}
1892 #endif
1893 		} endfor_nexthops(fi)
1894 		if (dead == fi->fib_nhs) {
1895 			switch (event) {
1896 			case NETDEV_DOWN:
1897 			case NETDEV_UNREGISTER:
1898 				fi->fib_flags |= RTNH_F_DEAD;
1899 				/* fall through */
1900 			case NETDEV_CHANGE:
1901 				fi->fib_flags |= RTNH_F_LINKDOWN;
1902 				break;
1903 			}
1904 			ret++;
1905 		}
1906 
1907 		fib_rebalance(fi);
1908 	}
1909 
1910 	return ret;
1911 }
1912 
1913 /* Must be invoked inside of an RCU protected region.  */
1914 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1915 {
1916 	struct fib_info *fi = NULL, *last_resort = NULL;
1917 	struct hlist_head *fa_head = res->fa_head;
1918 	struct fib_table *tb = res->table;
1919 	u8 slen = 32 - res->prefixlen;
1920 	int order = -1, last_idx = -1;
1921 	struct fib_alias *fa, *fa1 = NULL;
1922 	u32 last_prio = res->fi->fib_priority;
1923 	u8 last_tos = 0;
1924 
1925 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1926 		struct fib_info *next_fi = fa->fa_info;
1927 		struct fib_nh *nh;
1928 
1929 		if (fa->fa_slen != slen)
1930 			continue;
1931 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1932 			continue;
1933 		if (fa->tb_id != tb->tb_id)
1934 			continue;
1935 		if (next_fi->fib_priority > last_prio &&
1936 		    fa->fa_tos == last_tos) {
1937 			if (last_tos)
1938 				continue;
1939 			break;
1940 		}
1941 		if (next_fi->fib_flags & RTNH_F_DEAD)
1942 			continue;
1943 		last_tos = fa->fa_tos;
1944 		last_prio = next_fi->fib_priority;
1945 
1946 		if (next_fi->fib_scope != res->scope ||
1947 		    fa->fa_type != RTN_UNICAST)
1948 			continue;
1949 
1950 		nh = fib_info_nh(next_fi, 0);
1951 		if (!nh->fib_nh_gw4 || nh->fib_nh_scope != RT_SCOPE_LINK)
1952 			continue;
1953 
1954 		fib_alias_accessed(fa);
1955 
1956 		if (!fi) {
1957 			if (next_fi != res->fi)
1958 				break;
1959 			fa1 = fa;
1960 		} else if (!fib_detect_death(fi, order, &last_resort,
1961 					     &last_idx, fa1->fa_default)) {
1962 			fib_result_assign(res, fi);
1963 			fa1->fa_default = order;
1964 			goto out;
1965 		}
1966 		fi = next_fi;
1967 		order++;
1968 	}
1969 
1970 	if (order <= 0 || !fi) {
1971 		if (fa1)
1972 			fa1->fa_default = -1;
1973 		goto out;
1974 	}
1975 
1976 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1977 			      fa1->fa_default)) {
1978 		fib_result_assign(res, fi);
1979 		fa1->fa_default = order;
1980 		goto out;
1981 	}
1982 
1983 	if (last_idx >= 0)
1984 		fib_result_assign(res, last_resort);
1985 	fa1->fa_default = last_idx;
1986 out:
1987 	return;
1988 }
1989 
1990 /*
1991  * Dead device goes up. We wake up dead nexthops.
1992  * It takes sense only on multipath routes.
1993  *
1994  * only used when fib_nh is built into fib_info
1995  */
1996 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
1997 {
1998 	struct fib_info *prev_fi;
1999 	unsigned int hash;
2000 	struct hlist_head *head;
2001 	struct fib_nh *nh;
2002 	int ret;
2003 
2004 	if (!(dev->flags & IFF_UP))
2005 		return 0;
2006 
2007 	if (nh_flags & RTNH_F_DEAD) {
2008 		unsigned int flags = dev_get_flags(dev);
2009 
2010 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2011 			nh_flags |= RTNH_F_LINKDOWN;
2012 	}
2013 
2014 	prev_fi = NULL;
2015 	hash = fib_devindex_hashfn(dev->ifindex);
2016 	head = &fib_info_devhash[hash];
2017 	ret = 0;
2018 
2019 	hlist_for_each_entry(nh, head, nh_hash) {
2020 		struct fib_info *fi = nh->nh_parent;
2021 		int alive;
2022 
2023 		BUG_ON(!fi->fib_nhs);
2024 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2025 			continue;
2026 
2027 		prev_fi = fi;
2028 		alive = 0;
2029 		change_nexthops(fi) {
2030 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2031 				alive++;
2032 				continue;
2033 			}
2034 			if (!nexthop_nh->fib_nh_dev ||
2035 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2036 				continue;
2037 			if (nexthop_nh->fib_nh_dev != dev ||
2038 			    !__in_dev_get_rtnl(dev))
2039 				continue;
2040 			alive++;
2041 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2042 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2043 		} endfor_nexthops(fi)
2044 
2045 		if (alive > 0) {
2046 			fi->fib_flags &= ~nh_flags;
2047 			ret++;
2048 		}
2049 
2050 		fib_rebalance(fi);
2051 	}
2052 
2053 	return ret;
2054 }
2055 
2056 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2057 static bool fib_good_nh(const struct fib_nh *nh)
2058 {
2059 	int state = NUD_REACHABLE;
2060 
2061 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2062 		struct neighbour *n;
2063 
2064 		rcu_read_lock_bh();
2065 
2066 		if (likely(nh->fib_nh_gw_family == AF_INET))
2067 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2068 						   (__force u32)nh->fib_nh_gw4);
2069 		else if (nh->fib_nh_gw_family == AF_INET6)
2070 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2071 							   &nh->fib_nh_gw6);
2072 		else
2073 			n = NULL;
2074 		if (n)
2075 			state = n->nud_state;
2076 
2077 		rcu_read_unlock_bh();
2078 	}
2079 
2080 	return !!(state & NUD_VALID);
2081 }
2082 
2083 void fib_select_multipath(struct fib_result *res, int hash)
2084 {
2085 	struct fib_info *fi = res->fi;
2086 	struct net *net = fi->fib_net;
2087 	bool first = false;
2088 
2089 	if (unlikely(res->fi->nh)) {
2090 		nexthop_path_fib_result(res, hash);
2091 		return;
2092 	}
2093 
2094 	change_nexthops(fi) {
2095 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2096 			if (!fib_good_nh(nexthop_nh))
2097 				continue;
2098 			if (!first) {
2099 				res->nh_sel = nhsel;
2100 				res->nhc = &nexthop_nh->nh_common;
2101 				first = true;
2102 			}
2103 		}
2104 
2105 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2106 			continue;
2107 
2108 		res->nh_sel = nhsel;
2109 		res->nhc = &nexthop_nh->nh_common;
2110 		return;
2111 	} endfor_nexthops(fi);
2112 }
2113 #endif
2114 
2115 void fib_select_path(struct net *net, struct fib_result *res,
2116 		     struct flowi4 *fl4, const struct sk_buff *skb)
2117 {
2118 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2119 		goto check_saddr;
2120 
2121 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2122 	if (fib_info_num_path(res->fi) > 1) {
2123 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2124 
2125 		fib_select_multipath(res, h);
2126 	}
2127 	else
2128 #endif
2129 	if (!res->prefixlen &&
2130 	    res->table->tb_num_default > 1 &&
2131 	    res->type == RTN_UNICAST)
2132 		fib_select_default(fl4, res);
2133 
2134 check_saddr:
2135 	if (!fl4->saddr)
2136 		fl4->saddr = fib_result_prefsrc(net, res);
2137 }
2138