xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 3557b3fd)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		IPv4 Forwarding Information Base: semantics.
7  *
8  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9  *
10  *		This program is free software; you can redistribute it and/or
11  *		modify it under the terms of the GNU General Public License
12  *		as published by the Free Software Foundation; either version
13  *		2 of the License, or (at your option) any later version.
14  */
15 
16 #include <linux/uaccess.h>
17 #include <linux/bitops.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/jiffies.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/errno.h>
26 #include <linux/in.h>
27 #include <linux/inet.h>
28 #include <linux/inetdevice.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_arp.h>
31 #include <linux/proc_fs.h>
32 #include <linux/skbuff.h>
33 #include <linux/init.h>
34 #include <linux/slab.h>
35 #include <linux/netlink.h>
36 
37 #include <net/arp.h>
38 #include <net/ip.h>
39 #include <net/protocol.h>
40 #include <net/route.h>
41 #include <net/tcp.h>
42 #include <net/sock.h>
43 #include <net/ip_fib.h>
44 #include <net/ip6_fib.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 static DEFINE_SPINLOCK(fib_info_lock);
54 static struct hlist_head *fib_info_hash;
55 static struct hlist_head *fib_info_laddrhash;
56 static unsigned int fib_info_hash_size;
57 static unsigned int fib_info_cnt;
58 
59 #define DEVINDEX_HASHBITS 8
60 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
61 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
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 < (fi)->fib_nhs;					\
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 < (fi)->fib_nhs;					\
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 *nh)
163 {
164 	struct fnhe_hash_bucket *hash;
165 	int i;
166 
167 	hash = rcu_dereference_protected(nh->nh_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 }
216 EXPORT_SYMBOL_GPL(fib_nh_common_release);
217 
218 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
219 {
220 #ifdef CONFIG_IP_ROUTE_CLASSID
221 	if (fib_nh->nh_tclassid)
222 		net->ipv4.fib_num_tclassid_users--;
223 #endif
224 	fib_nh_common_release(&fib_nh->nh_common);
225 	free_nh_exceptions(fib_nh);
226 	rt_fibinfo_free_cpus(fib_nh->nh_pcpu_rth_output);
227 	rt_fibinfo_free(&fib_nh->nh_rth_input);
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 	change_nexthops(fi) {
236 		fib_nh_release(fi->fib_net, nexthop_nh);
237 	} endfor_nexthops(fi);
238 
239 	ip_fib_metrics_put(fi->fib_metrics);
240 
241 	kfree(fi);
242 }
243 
244 void free_fib_info(struct fib_info *fi)
245 {
246 	if (fi->fib_dead == 0) {
247 		pr_warn("Freeing alive fib_info %p\n", fi);
248 		return;
249 	}
250 	fib_info_cnt--;
251 
252 	call_rcu(&fi->rcu, free_fib_info_rcu);
253 }
254 EXPORT_SYMBOL_GPL(free_fib_info);
255 
256 void fib_release_info(struct fib_info *fi)
257 {
258 	spin_lock_bh(&fib_info_lock);
259 	if (fi && --fi->fib_treeref == 0) {
260 		hlist_del(&fi->fib_hash);
261 		if (fi->fib_prefsrc)
262 			hlist_del(&fi->fib_lhash);
263 		change_nexthops(fi) {
264 			if (!nexthop_nh->fib_nh_dev)
265 				continue;
266 			hlist_del(&nexthop_nh->nh_hash);
267 		} endfor_nexthops(fi)
268 		fi->fib_dead = 1;
269 		fib_info_put(fi);
270 	}
271 	spin_unlock_bh(&fib_info_lock);
272 }
273 
274 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
275 {
276 	const struct fib_nh *onh = ofi->fib_nh;
277 
278 	for_nexthops(fi) {
279 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
280 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
281 		    nh->fib_nh_scope != onh->fib_nh_scope ||
282 #ifdef CONFIG_IP_ROUTE_MULTIPATH
283 		    nh->fib_nh_weight != onh->fib_nh_weight ||
284 #endif
285 #ifdef CONFIG_IP_ROUTE_CLASSID
286 		    nh->nh_tclassid != onh->nh_tclassid ||
287 #endif
288 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
289 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
290 			return -1;
291 
292 		if (nh->fib_nh_gw_family == AF_INET &&
293 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
294 			return -1;
295 
296 		if (nh->fib_nh_gw_family == AF_INET6 &&
297 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
298 			return -1;
299 
300 		onh++;
301 	} endfor_nexthops(fi);
302 	return 0;
303 }
304 
305 static inline unsigned int fib_devindex_hashfn(unsigned int val)
306 {
307 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
308 
309 	return (val ^
310 		(val >> DEVINDEX_HASHBITS) ^
311 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
312 }
313 
314 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
315 {
316 	unsigned int mask = (fib_info_hash_size - 1);
317 	unsigned int val = fi->fib_nhs;
318 
319 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
320 	val ^= (__force u32)fi->fib_prefsrc;
321 	val ^= fi->fib_priority;
322 	for_nexthops(fi) {
323 		val ^= fib_devindex_hashfn(nh->fib_nh_oif);
324 	} endfor_nexthops(fi)
325 
326 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
327 }
328 
329 static struct fib_info *fib_find_info(const struct fib_info *nfi)
330 {
331 	struct hlist_head *head;
332 	struct fib_info *fi;
333 	unsigned int hash;
334 
335 	hash = fib_info_hashfn(nfi);
336 	head = &fib_info_hash[hash];
337 
338 	hlist_for_each_entry(fi, head, fib_hash) {
339 		if (!net_eq(fi->fib_net, nfi->fib_net))
340 			continue;
341 		if (fi->fib_nhs != nfi->fib_nhs)
342 			continue;
343 		if (nfi->fib_protocol == fi->fib_protocol &&
344 		    nfi->fib_scope == fi->fib_scope &&
345 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
346 		    nfi->fib_priority == fi->fib_priority &&
347 		    nfi->fib_type == fi->fib_type &&
348 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
349 			   sizeof(u32) * RTAX_MAX) == 0 &&
350 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
351 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
352 			return fi;
353 	}
354 
355 	return NULL;
356 }
357 
358 /* Check, that the gateway is already configured.
359  * Used only by redirect accept routine.
360  */
361 int ip_fib_check_default(__be32 gw, struct net_device *dev)
362 {
363 	struct hlist_head *head;
364 	struct fib_nh *nh;
365 	unsigned int hash;
366 
367 	spin_lock(&fib_info_lock);
368 
369 	hash = fib_devindex_hashfn(dev->ifindex);
370 	head = &fib_info_devhash[hash];
371 	hlist_for_each_entry(nh, head, nh_hash) {
372 		if (nh->fib_nh_dev == dev &&
373 		    nh->fib_nh_gw4 == gw &&
374 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
375 			spin_unlock(&fib_info_lock);
376 			return 0;
377 		}
378 	}
379 
380 	spin_unlock(&fib_info_lock);
381 
382 	return -1;
383 }
384 
385 static inline size_t fib_nlmsg_size(struct fib_info *fi)
386 {
387 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
388 			 + nla_total_size(4) /* RTA_TABLE */
389 			 + nla_total_size(4) /* RTA_DST */
390 			 + nla_total_size(4) /* RTA_PRIORITY */
391 			 + nla_total_size(4) /* RTA_PREFSRC */
392 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
393 
394 	/* space for nested metrics */
395 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
396 
397 	if (fi->fib_nhs) {
398 		size_t nh_encapsize = 0;
399 		/* Also handles the special case fib_nhs == 1 */
400 
401 		/* each nexthop is packed in an attribute */
402 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
403 
404 		/* may contain flow and gateway attribute */
405 		nhsize += 2 * nla_total_size(4);
406 
407 		/* grab encap info */
408 		for_nexthops(fi) {
409 			if (nh->fib_nh_lws) {
410 				/* RTA_ENCAP_TYPE */
411 				nh_encapsize += lwtunnel_get_encap_size(
412 						nh->fib_nh_lws);
413 				/* RTA_ENCAP */
414 				nh_encapsize +=  nla_total_size(2);
415 			}
416 		} endfor_nexthops(fi);
417 
418 		/* all nexthops are packed in a nested attribute */
419 		payload += nla_total_size((fi->fib_nhs * nhsize) +
420 					  nh_encapsize);
421 
422 	}
423 
424 	return payload;
425 }
426 
427 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
428 	       int dst_len, u32 tb_id, const struct nl_info *info,
429 	       unsigned int nlm_flags)
430 {
431 	struct sk_buff *skb;
432 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
433 	int err = -ENOBUFS;
434 
435 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
436 	if (!skb)
437 		goto errout;
438 
439 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
440 			    fa->fa_type, key, dst_len,
441 			    fa->fa_tos, fa->fa_info, nlm_flags);
442 	if (err < 0) {
443 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
444 		WARN_ON(err == -EMSGSIZE);
445 		kfree_skb(skb);
446 		goto errout;
447 	}
448 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
449 		    info->nlh, GFP_KERNEL);
450 	return;
451 errout:
452 	if (err < 0)
453 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
454 }
455 
456 static int fib_detect_death(struct fib_info *fi, int order,
457 			    struct fib_info **last_resort, int *last_idx,
458 			    int dflt)
459 {
460 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
461 	struct neighbour *n;
462 	int state = NUD_NONE;
463 
464 	if (likely(nhc->nhc_gw_family == AF_INET))
465 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
466 	else if (nhc->nhc_gw_family == AF_INET6)
467 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
468 				 nhc->nhc_dev);
469 	else
470 		n = NULL;
471 
472 	if (n) {
473 		state = n->nud_state;
474 		neigh_release(n);
475 	} else {
476 		return 0;
477 	}
478 	if (state == NUD_REACHABLE)
479 		return 0;
480 	if ((state & NUD_VALID) && order != dflt)
481 		return 0;
482 	if ((state & NUD_VALID) ||
483 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
484 		*last_resort = fi;
485 		*last_idx = order;
486 	}
487 	return 1;
488 }
489 
490 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
491 		       u16 encap_type, void *cfg, gfp_t gfp_flags,
492 		       struct netlink_ext_ack *extack)
493 {
494 	if (encap) {
495 		struct lwtunnel_state *lwtstate;
496 		int err;
497 
498 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
499 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
500 			return -EINVAL;
501 		}
502 		err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family,
503 					   cfg, &lwtstate, extack);
504 		if (err)
505 			return err;
506 
507 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
508 	}
509 
510 	return 0;
511 }
512 EXPORT_SYMBOL_GPL(fib_nh_common_init);
513 
514 int fib_nh_init(struct net *net, struct fib_nh *nh,
515 		struct fib_config *cfg, int nh_weight,
516 		struct netlink_ext_ack *extack)
517 {
518 	int err = -ENOMEM;
519 
520 	nh->fib_nh_family = AF_INET;
521 
522 	nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
523 	if (!nh->nh_pcpu_rth_output)
524 		goto err_out;
525 
526 	err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
527 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
528 	if (err)
529 		goto init_failure;
530 
531 	nh->fib_nh_oif = cfg->fc_oif;
532 	nh->fib_nh_gw_family = cfg->fc_gw_family;
533 	if (cfg->fc_gw_family == AF_INET)
534 		nh->fib_nh_gw4 = cfg->fc_gw4;
535 	else if (cfg->fc_gw_family == AF_INET6)
536 		nh->fib_nh_gw6 = cfg->fc_gw6;
537 
538 	nh->fib_nh_flags = cfg->fc_flags;
539 
540 #ifdef CONFIG_IP_ROUTE_CLASSID
541 	nh->nh_tclassid = cfg->fc_flow;
542 	if (nh->nh_tclassid)
543 		net->ipv4.fib_num_tclassid_users++;
544 #endif
545 #ifdef CONFIG_IP_ROUTE_MULTIPATH
546 	nh->fib_nh_weight = nh_weight;
547 #endif
548 	return 0;
549 
550 init_failure:
551 	rt_fibinfo_free_cpus(nh->nh_pcpu_rth_output);
552 	nh->nh_pcpu_rth_output = NULL;
553 err_out:
554 	return err;
555 }
556 
557 #ifdef CONFIG_IP_ROUTE_MULTIPATH
558 
559 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
560 			      struct netlink_ext_ack *extack)
561 {
562 	int nhs = 0;
563 
564 	while (rtnh_ok(rtnh, remaining)) {
565 		nhs++;
566 		rtnh = rtnh_next(rtnh, &remaining);
567 	}
568 
569 	/* leftover implies invalid nexthop configuration, discard it */
570 	if (remaining > 0) {
571 		NL_SET_ERR_MSG(extack,
572 			       "Invalid nexthop configuration - extra data after nexthops");
573 		nhs = 0;
574 	}
575 
576 	return nhs;
577 }
578 
579 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
580 		       int remaining, struct fib_config *cfg,
581 		       struct netlink_ext_ack *extack)
582 {
583 	struct net *net = fi->fib_net;
584 	struct fib_config fib_cfg;
585 	int ret;
586 
587 	change_nexthops(fi) {
588 		int attrlen;
589 
590 		memset(&fib_cfg, 0, sizeof(fib_cfg));
591 
592 		if (!rtnh_ok(rtnh, remaining)) {
593 			NL_SET_ERR_MSG(extack,
594 				       "Invalid nexthop configuration - extra data after nexthop");
595 			return -EINVAL;
596 		}
597 
598 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
599 			NL_SET_ERR_MSG(extack,
600 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
601 			return -EINVAL;
602 		}
603 
604 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
605 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
606 
607 		attrlen = rtnh_attrlen(rtnh);
608 		if (attrlen > 0) {
609 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
610 
611 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
612 			nlav = nla_find(attrs, attrlen, RTA_VIA);
613 			if (nla && nlav) {
614 				NL_SET_ERR_MSG(extack,
615 					       "Nexthop configuration can not contain both GATEWAY and VIA");
616 				return -EINVAL;
617 			}
618 			if (nla) {
619 				fib_cfg.fc_gw4 = nla_get_in_addr(nla);
620 				if (fib_cfg.fc_gw4)
621 					fib_cfg.fc_gw_family = AF_INET;
622 			} else if (nlav) {
623 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
624 				if (ret)
625 					goto errout;
626 			}
627 
628 			nla = nla_find(attrs, attrlen, RTA_FLOW);
629 			if (nla)
630 				fib_cfg.fc_flow = nla_get_u32(nla);
631 
632 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
633 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
634 			if (nla)
635 				fib_cfg.fc_encap_type = nla_get_u16(nla);
636 		}
637 
638 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
639 				  rtnh->rtnh_hops + 1, extack);
640 		if (ret)
641 			goto errout;
642 
643 		rtnh = rtnh_next(rtnh, &remaining);
644 	} endfor_nexthops(fi);
645 
646 	ret = -EINVAL;
647 	if (cfg->fc_oif && fi->fib_nh->fib_nh_oif != cfg->fc_oif) {
648 		NL_SET_ERR_MSG(extack,
649 			       "Nexthop device index does not match RTA_OIF");
650 		goto errout;
651 	}
652 	if (cfg->fc_gw_family) {
653 		if (cfg->fc_gw_family != fi->fib_nh->fib_nh_gw_family ||
654 		    (cfg->fc_gw_family == AF_INET &&
655 		     fi->fib_nh->fib_nh_gw4 != cfg->fc_gw4) ||
656 		    (cfg->fc_gw_family == AF_INET6 &&
657 		     ipv6_addr_cmp(&fi->fib_nh->fib_nh_gw6, &cfg->fc_gw6))) {
658 			NL_SET_ERR_MSG(extack,
659 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
660 			goto errout;
661 		}
662 	}
663 #ifdef CONFIG_IP_ROUTE_CLASSID
664 	if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
665 		NL_SET_ERR_MSG(extack,
666 			       "Nexthop class id does not match RTA_FLOW");
667 		goto errout;
668 	}
669 #endif
670 	ret = 0;
671 errout:
672 	return ret;
673 }
674 
675 static void fib_rebalance(struct fib_info *fi)
676 {
677 	int total;
678 	int w;
679 
680 	if (fi->fib_nhs < 2)
681 		return;
682 
683 	total = 0;
684 	for_nexthops(fi) {
685 		if (nh->fib_nh_flags & RTNH_F_DEAD)
686 			continue;
687 
688 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
689 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
690 			continue;
691 
692 		total += nh->fib_nh_weight;
693 	} endfor_nexthops(fi);
694 
695 	w = 0;
696 	change_nexthops(fi) {
697 		int upper_bound;
698 
699 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
700 			upper_bound = -1;
701 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
702 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
703 			upper_bound = -1;
704 		} else {
705 			w += nexthop_nh->fib_nh_weight;
706 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
707 							    total) - 1;
708 		}
709 
710 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
711 	} endfor_nexthops(fi);
712 }
713 #else /* CONFIG_IP_ROUTE_MULTIPATH */
714 
715 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
716 		       int remaining, struct fib_config *cfg,
717 		       struct netlink_ext_ack *extack)
718 {
719 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
720 
721 	return -EINVAL;
722 }
723 
724 #define fib_rebalance(fi) do { } while (0)
725 
726 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
727 
728 static int fib_encap_match(u16 encap_type,
729 			   struct nlattr *encap,
730 			   const struct fib_nh *nh,
731 			   const struct fib_config *cfg,
732 			   struct netlink_ext_ack *extack)
733 {
734 	struct lwtunnel_state *lwtstate;
735 	int ret, result = 0;
736 
737 	if (encap_type == LWTUNNEL_ENCAP_NONE)
738 		return 0;
739 
740 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
741 				   cfg, &lwtstate, extack);
742 	if (!ret) {
743 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
744 		lwtstate_free(lwtstate);
745 	}
746 
747 	return result;
748 }
749 
750 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
751 		 struct netlink_ext_ack *extack)
752 {
753 #ifdef CONFIG_IP_ROUTE_MULTIPATH
754 	struct rtnexthop *rtnh;
755 	int remaining;
756 #endif
757 
758 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
759 		return 1;
760 
761 	if (cfg->fc_oif || cfg->fc_gw_family) {
762 		if (cfg->fc_encap) {
763 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
764 					    fi->fib_nh, cfg, extack))
765 				return 1;
766 		}
767 #ifdef CONFIG_IP_ROUTE_CLASSID
768 		if (cfg->fc_flow &&
769 		    cfg->fc_flow != fi->fib_nh->nh_tclassid)
770 			return 1;
771 #endif
772 		if ((cfg->fc_oif && cfg->fc_oif != fi->fib_nh->fib_nh_oif) ||
773 		    (cfg->fc_gw_family &&
774 		     cfg->fc_gw_family != fi->fib_nh->fib_nh_gw_family))
775 			return 1;
776 
777 		if (cfg->fc_gw_family == AF_INET &&
778 		    cfg->fc_gw4 != fi->fib_nh->fib_nh_gw4)
779 			return 1;
780 
781 		if (cfg->fc_gw_family == AF_INET6 &&
782 		    ipv6_addr_cmp(&cfg->fc_gw6, &fi->fib_nh->fib_nh_gw6))
783 			return 1;
784 
785 		return 0;
786 	}
787 
788 #ifdef CONFIG_IP_ROUTE_MULTIPATH
789 	if (!cfg->fc_mp)
790 		return 0;
791 
792 	rtnh = cfg->fc_mp;
793 	remaining = cfg->fc_mp_len;
794 
795 	for_nexthops(fi) {
796 		int attrlen;
797 
798 		if (!rtnh_ok(rtnh, remaining))
799 			return -EINVAL;
800 
801 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
802 			return 1;
803 
804 		attrlen = rtnh_attrlen(rtnh);
805 		if (attrlen > 0) {
806 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
807 
808 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
809 			nlav = nla_find(attrs, attrlen, RTA_VIA);
810 			if (nla && nlav) {
811 				NL_SET_ERR_MSG(extack,
812 					       "Nexthop configuration can not contain both GATEWAY and VIA");
813 				return -EINVAL;
814 			}
815 
816 			if (nla) {
817 				if (nh->fib_nh_gw_family != AF_INET ||
818 				    nla_get_in_addr(nla) != nh->fib_nh_gw4)
819 					return 1;
820 			} else if (nlav) {
821 				struct fib_config cfg2;
822 				int err;
823 
824 				err = fib_gw_from_via(&cfg2, nlav, extack);
825 				if (err)
826 					return err;
827 
828 				switch (nh->fib_nh_gw_family) {
829 				case AF_INET:
830 					if (cfg2.fc_gw_family != AF_INET ||
831 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
832 						return 1;
833 					break;
834 				case AF_INET6:
835 					if (cfg2.fc_gw_family != AF_INET6 ||
836 					    ipv6_addr_cmp(&cfg2.fc_gw6,
837 							  &nh->fib_nh_gw6))
838 						return 1;
839 					break;
840 				}
841 			}
842 
843 #ifdef CONFIG_IP_ROUTE_CLASSID
844 			nla = nla_find(attrs, attrlen, RTA_FLOW);
845 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
846 				return 1;
847 #endif
848 		}
849 
850 		rtnh = rtnh_next(rtnh, &remaining);
851 	} endfor_nexthops(fi);
852 #endif
853 	return 0;
854 }
855 
856 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
857 {
858 	struct nlattr *nla;
859 	int remaining;
860 
861 	if (!cfg->fc_mx)
862 		return true;
863 
864 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
865 		int type = nla_type(nla);
866 		u32 fi_val, val;
867 
868 		if (!type)
869 			continue;
870 		if (type > RTAX_MAX)
871 			return false;
872 
873 		if (type == RTAX_CC_ALGO) {
874 			char tmp[TCP_CA_NAME_MAX];
875 			bool ecn_ca = false;
876 
877 			nla_strlcpy(tmp, nla, sizeof(tmp));
878 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
879 		} else {
880 			if (nla_len(nla) != sizeof(u32))
881 				return false;
882 			val = nla_get_u32(nla);
883 		}
884 
885 		fi_val = fi->fib_metrics->metrics[type - 1];
886 		if (type == RTAX_FEATURES)
887 			fi_val &= ~DST_FEATURE_ECN_CA;
888 
889 		if (fi_val != val)
890 			return false;
891 	}
892 
893 	return true;
894 }
895 
896 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
897 			      u32 table, struct netlink_ext_ack *extack)
898 {
899 	struct fib6_config cfg = {
900 		.fc_table = table,
901 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
902 		.fc_ifindex = nh->fib_nh_oif,
903 		.fc_gateway = nh->fib_nh_gw6,
904 	};
905 	struct fib6_nh fib6_nh = {};
906 	int err;
907 
908 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
909 	if (!err) {
910 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
911 		dev_hold(nh->fib_nh_dev);
912 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
913 		nh->fib_nh_scope = RT_SCOPE_LINK;
914 
915 		ipv6_stub->fib6_nh_release(&fib6_nh);
916 	}
917 
918 	return err;
919 }
920 
921 /*
922  * Picture
923  * -------
924  *
925  * Semantics of nexthop is very messy by historical reasons.
926  * We have to take into account, that:
927  * a) gateway can be actually local interface address,
928  *    so that gatewayed route is direct.
929  * b) gateway must be on-link address, possibly
930  *    described not by an ifaddr, but also by a direct route.
931  * c) If both gateway and interface are specified, they should not
932  *    contradict.
933  * d) If we use tunnel routes, gateway could be not on-link.
934  *
935  * Attempt to reconcile all of these (alas, self-contradictory) conditions
936  * results in pretty ugly and hairy code with obscure logic.
937  *
938  * I chose to generalized it instead, so that the size
939  * of code does not increase practically, but it becomes
940  * much more general.
941  * Every prefix is assigned a "scope" value: "host" is local address,
942  * "link" is direct route,
943  * [ ... "site" ... "interior" ... ]
944  * and "universe" is true gateway route with global meaning.
945  *
946  * Every prefix refers to a set of "nexthop"s (gw, oif),
947  * where gw must have narrower scope. This recursion stops
948  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
949  * which means that gw is forced to be on link.
950  *
951  * Code is still hairy, but now it is apparently logically
952  * consistent and very flexible. F.e. as by-product it allows
953  * to co-exists in peace independent exterior and interior
954  * routing processes.
955  *
956  * Normally it looks as following.
957  *
958  * {universe prefix}  -> (gw, oif) [scope link]
959  *		  |
960  *		  |-> {link prefix} -> (gw, oif) [scope local]
961  *					|
962  *					|-> {local prefix} (terminal node)
963  */
964 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
965 			      u8 scope, struct netlink_ext_ack *extack)
966 {
967 	struct net_device *dev;
968 	struct fib_result res;
969 	int err;
970 
971 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
972 		unsigned int addr_type;
973 
974 		if (scope >= RT_SCOPE_LINK) {
975 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
976 			return -EINVAL;
977 		}
978 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
979 		if (!dev) {
980 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
981 			return -ENODEV;
982 		}
983 		if (!(dev->flags & IFF_UP)) {
984 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
985 			return -ENETDOWN;
986 		}
987 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
988 		if (addr_type != RTN_UNICAST) {
989 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
990 			return -EINVAL;
991 		}
992 		if (!netif_carrier_ok(dev))
993 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
994 		nh->fib_nh_dev = dev;
995 		dev_hold(dev);
996 		nh->fib_nh_scope = RT_SCOPE_LINK;
997 		return 0;
998 	}
999 	rcu_read_lock();
1000 	{
1001 		struct fib_table *tbl = NULL;
1002 		struct flowi4 fl4 = {
1003 			.daddr = nh->fib_nh_gw4,
1004 			.flowi4_scope = scope + 1,
1005 			.flowi4_oif = nh->fib_nh_oif,
1006 			.flowi4_iif = LOOPBACK_IFINDEX,
1007 		};
1008 
1009 		/* It is not necessary, but requires a bit of thinking */
1010 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1011 			fl4.flowi4_scope = RT_SCOPE_LINK;
1012 
1013 		if (table)
1014 			tbl = fib_get_table(net, table);
1015 
1016 		if (tbl)
1017 			err = fib_table_lookup(tbl, &fl4, &res,
1018 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1019 					       FIB_LOOKUP_NOREF);
1020 
1021 		/* on error or if no table given do full lookup. This
1022 		 * is needed for example when nexthops are in the local
1023 		 * table rather than the given table
1024 		 */
1025 		if (!tbl || err) {
1026 			err = fib_lookup(net, &fl4, &res,
1027 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1028 		}
1029 
1030 		if (err) {
1031 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1032 			goto out;
1033 		}
1034 	}
1035 
1036 	err = -EINVAL;
1037 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1038 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1039 		goto out;
1040 	}
1041 	nh->fib_nh_scope = res.scope;
1042 	nh->fib_nh_oif = FIB_RES_OIF(res);
1043 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1044 	if (!dev) {
1045 		NL_SET_ERR_MSG(extack,
1046 			       "No egress device for nexthop gateway");
1047 		goto out;
1048 	}
1049 	dev_hold(dev);
1050 	if (!netif_carrier_ok(dev))
1051 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1052 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1053 out:
1054 	rcu_read_unlock();
1055 	return err;
1056 }
1057 
1058 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1059 			      struct netlink_ext_ack *extack)
1060 {
1061 	struct in_device *in_dev;
1062 	int err;
1063 
1064 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1065 		NL_SET_ERR_MSG(extack,
1066 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1067 		return -EINVAL;
1068 	}
1069 
1070 	rcu_read_lock();
1071 
1072 	err = -ENODEV;
1073 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1074 	if (!in_dev)
1075 		goto out;
1076 	err = -ENETDOWN;
1077 	if (!(in_dev->dev->flags & IFF_UP)) {
1078 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1079 		goto out;
1080 	}
1081 
1082 	nh->fib_nh_dev = in_dev->dev;
1083 	dev_hold(nh->fib_nh_dev);
1084 	nh->fib_nh_scope = RT_SCOPE_HOST;
1085 	if (!netif_carrier_ok(nh->fib_nh_dev))
1086 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1087 	err = 0;
1088 out:
1089 	rcu_read_unlock();
1090 	return err;
1091 }
1092 
1093 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
1094 			struct netlink_ext_ack *extack)
1095 {
1096 	struct net *net = cfg->fc_nlinfo.nl_net;
1097 	u32 table = cfg->fc_table;
1098 	int err;
1099 
1100 	if (nh->fib_nh_gw_family == AF_INET)
1101 		err = fib_check_nh_v4_gw(net, nh, table, cfg->fc_scope, extack);
1102 	else if (nh->fib_nh_gw_family == AF_INET6)
1103 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1104 	else
1105 		err = fib_check_nh_nongw(net, nh, extack);
1106 
1107 	return err;
1108 }
1109 
1110 static inline unsigned int fib_laddr_hashfn(__be32 val)
1111 {
1112 	unsigned int mask = (fib_info_hash_size - 1);
1113 
1114 	return ((__force u32)val ^
1115 		((__force u32)val >> 7) ^
1116 		((__force u32)val >> 14)) & mask;
1117 }
1118 
1119 static struct hlist_head *fib_info_hash_alloc(int bytes)
1120 {
1121 	if (bytes <= PAGE_SIZE)
1122 		return kzalloc(bytes, GFP_KERNEL);
1123 	else
1124 		return (struct hlist_head *)
1125 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1126 					 get_order(bytes));
1127 }
1128 
1129 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1130 {
1131 	if (!hash)
1132 		return;
1133 
1134 	if (bytes <= PAGE_SIZE)
1135 		kfree(hash);
1136 	else
1137 		free_pages((unsigned long) hash, get_order(bytes));
1138 }
1139 
1140 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1141 			       struct hlist_head *new_laddrhash,
1142 			       unsigned int new_size)
1143 {
1144 	struct hlist_head *old_info_hash, *old_laddrhash;
1145 	unsigned int old_size = fib_info_hash_size;
1146 	unsigned int i, bytes;
1147 
1148 	spin_lock_bh(&fib_info_lock);
1149 	old_info_hash = fib_info_hash;
1150 	old_laddrhash = fib_info_laddrhash;
1151 	fib_info_hash_size = new_size;
1152 
1153 	for (i = 0; i < old_size; i++) {
1154 		struct hlist_head *head = &fib_info_hash[i];
1155 		struct hlist_node *n;
1156 		struct fib_info *fi;
1157 
1158 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1159 			struct hlist_head *dest;
1160 			unsigned int new_hash;
1161 
1162 			new_hash = fib_info_hashfn(fi);
1163 			dest = &new_info_hash[new_hash];
1164 			hlist_add_head(&fi->fib_hash, dest);
1165 		}
1166 	}
1167 	fib_info_hash = new_info_hash;
1168 
1169 	for (i = 0; i < old_size; i++) {
1170 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1171 		struct hlist_node *n;
1172 		struct fib_info *fi;
1173 
1174 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1175 			struct hlist_head *ldest;
1176 			unsigned int new_hash;
1177 
1178 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1179 			ldest = &new_laddrhash[new_hash];
1180 			hlist_add_head(&fi->fib_lhash, ldest);
1181 		}
1182 	}
1183 	fib_info_laddrhash = new_laddrhash;
1184 
1185 	spin_unlock_bh(&fib_info_lock);
1186 
1187 	bytes = old_size * sizeof(struct hlist_head *);
1188 	fib_info_hash_free(old_info_hash, bytes);
1189 	fib_info_hash_free(old_laddrhash, bytes);
1190 }
1191 
1192 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
1193 {
1194 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev,
1195 					nh->fib_nh_gw4,
1196 					nh->nh_parent->fib_scope);
1197 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1198 
1199 	return nh->nh_saddr;
1200 }
1201 
1202 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1203 {
1204 	struct fib_nh_common *nhc = res->nhc;
1205 	struct fib_nh *nh;
1206 
1207 	if (res->fi->fib_prefsrc)
1208 		return res->fi->fib_prefsrc;
1209 
1210 	nh = container_of(nhc, struct fib_nh, nh_common);
1211 	if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1212 		return nh->nh_saddr;
1213 
1214 	return fib_info_update_nh_saddr(net, nh);
1215 }
1216 
1217 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1218 {
1219 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1220 	    fib_prefsrc != cfg->fc_dst) {
1221 		u32 tb_id = cfg->fc_table;
1222 		int rc;
1223 
1224 		if (tb_id == RT_TABLE_MAIN)
1225 			tb_id = RT_TABLE_LOCAL;
1226 
1227 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1228 					  fib_prefsrc, tb_id);
1229 
1230 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1231 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1232 						  fib_prefsrc, RT_TABLE_LOCAL);
1233 		}
1234 
1235 		if (rc != RTN_LOCAL)
1236 			return false;
1237 	}
1238 	return true;
1239 }
1240 
1241 struct fib_info *fib_create_info(struct fib_config *cfg,
1242 				 struct netlink_ext_ack *extack)
1243 {
1244 	int err;
1245 	struct fib_info *fi = NULL;
1246 	struct fib_info *ofi;
1247 	int nhs = 1;
1248 	struct net *net = cfg->fc_nlinfo.nl_net;
1249 
1250 	if (cfg->fc_type > RTN_MAX)
1251 		goto err_inval;
1252 
1253 	/* Fast check to catch the most weird cases */
1254 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1255 		NL_SET_ERR_MSG(extack, "Invalid scope");
1256 		goto err_inval;
1257 	}
1258 
1259 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1260 		NL_SET_ERR_MSG(extack,
1261 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1262 		goto err_inval;
1263 	}
1264 
1265 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1266 	if (cfg->fc_mp) {
1267 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1268 		if (nhs == 0)
1269 			goto err_inval;
1270 	}
1271 #endif
1272 
1273 	err = -ENOBUFS;
1274 	if (fib_info_cnt >= fib_info_hash_size) {
1275 		unsigned int new_size = fib_info_hash_size << 1;
1276 		struct hlist_head *new_info_hash;
1277 		struct hlist_head *new_laddrhash;
1278 		unsigned int bytes;
1279 
1280 		if (!new_size)
1281 			new_size = 16;
1282 		bytes = new_size * sizeof(struct hlist_head *);
1283 		new_info_hash = fib_info_hash_alloc(bytes);
1284 		new_laddrhash = fib_info_hash_alloc(bytes);
1285 		if (!new_info_hash || !new_laddrhash) {
1286 			fib_info_hash_free(new_info_hash, bytes);
1287 			fib_info_hash_free(new_laddrhash, bytes);
1288 		} else
1289 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1290 
1291 		if (!fib_info_hash_size)
1292 			goto failure;
1293 	}
1294 
1295 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1296 	if (!fi)
1297 		goto failure;
1298 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1299 					      cfg->fc_mx_len, extack);
1300 	if (unlikely(IS_ERR(fi->fib_metrics))) {
1301 		err = PTR_ERR(fi->fib_metrics);
1302 		kfree(fi);
1303 		return ERR_PTR(err);
1304 	}
1305 
1306 	fib_info_cnt++;
1307 	fi->fib_net = net;
1308 	fi->fib_protocol = cfg->fc_protocol;
1309 	fi->fib_scope = cfg->fc_scope;
1310 	fi->fib_flags = cfg->fc_flags;
1311 	fi->fib_priority = cfg->fc_priority;
1312 	fi->fib_prefsrc = cfg->fc_prefsrc;
1313 	fi->fib_type = cfg->fc_type;
1314 	fi->fib_tb_id = cfg->fc_table;
1315 
1316 	fi->fib_nhs = nhs;
1317 	change_nexthops(fi) {
1318 		nexthop_nh->nh_parent = fi;
1319 	} endfor_nexthops(fi)
1320 
1321 	if (cfg->fc_mp)
1322 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1323 	else
1324 		err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1325 
1326 	if (err != 0)
1327 		goto failure;
1328 
1329 	if (fib_props[cfg->fc_type].error) {
1330 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1331 			NL_SET_ERR_MSG(extack,
1332 				       "Gateway, device and multipath can not be specified for this route type");
1333 			goto err_inval;
1334 		}
1335 		goto link_it;
1336 	} else {
1337 		switch (cfg->fc_type) {
1338 		case RTN_UNICAST:
1339 		case RTN_LOCAL:
1340 		case RTN_BROADCAST:
1341 		case RTN_ANYCAST:
1342 		case RTN_MULTICAST:
1343 			break;
1344 		default:
1345 			NL_SET_ERR_MSG(extack, "Invalid route type");
1346 			goto err_inval;
1347 		}
1348 	}
1349 
1350 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1351 		NL_SET_ERR_MSG(extack, "Invalid scope");
1352 		goto err_inval;
1353 	}
1354 
1355 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1356 		struct fib_nh *nh = fi->fib_nh;
1357 
1358 		/* Local address is added. */
1359 		if (nhs != 1) {
1360 			NL_SET_ERR_MSG(extack,
1361 				       "Route with host scope can not have multiple nexthops");
1362 			goto err_inval;
1363 		}
1364 		if (nh->fib_nh_gw_family) {
1365 			NL_SET_ERR_MSG(extack,
1366 				       "Route with host scope can not have a gateway");
1367 			goto err_inval;
1368 		}
1369 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1370 		nh->fib_nh_dev = dev_get_by_index(net, fi->fib_nh->fib_nh_oif);
1371 		err = -ENODEV;
1372 		if (!nh->fib_nh_dev)
1373 			goto failure;
1374 	} else {
1375 		int linkdown = 0;
1376 
1377 		change_nexthops(fi) {
1378 			err = fib_check_nh(cfg, nexthop_nh, extack);
1379 			if (err != 0)
1380 				goto failure;
1381 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1382 				linkdown++;
1383 		} endfor_nexthops(fi)
1384 		if (linkdown == fi->fib_nhs)
1385 			fi->fib_flags |= RTNH_F_LINKDOWN;
1386 	}
1387 
1388 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1389 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1390 		goto err_inval;
1391 	}
1392 
1393 	change_nexthops(fi) {
1394 		fib_info_update_nh_saddr(net, nexthop_nh);
1395 		if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1396 			fi->fib_nh_is_v6 = true;
1397 	} endfor_nexthops(fi)
1398 
1399 	fib_rebalance(fi);
1400 
1401 link_it:
1402 	ofi = fib_find_info(fi);
1403 	if (ofi) {
1404 		fi->fib_dead = 1;
1405 		free_fib_info(fi);
1406 		ofi->fib_treeref++;
1407 		return ofi;
1408 	}
1409 
1410 	fi->fib_treeref++;
1411 	refcount_set(&fi->fib_clntref, 1);
1412 	spin_lock_bh(&fib_info_lock);
1413 	hlist_add_head(&fi->fib_hash,
1414 		       &fib_info_hash[fib_info_hashfn(fi)]);
1415 	if (fi->fib_prefsrc) {
1416 		struct hlist_head *head;
1417 
1418 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1419 		hlist_add_head(&fi->fib_lhash, head);
1420 	}
1421 	change_nexthops(fi) {
1422 		struct hlist_head *head;
1423 		unsigned int hash;
1424 
1425 		if (!nexthop_nh->fib_nh_dev)
1426 			continue;
1427 		hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1428 		head = &fib_info_devhash[hash];
1429 		hlist_add_head(&nexthop_nh->nh_hash, head);
1430 	} endfor_nexthops(fi)
1431 	spin_unlock_bh(&fib_info_lock);
1432 	return fi;
1433 
1434 err_inval:
1435 	err = -EINVAL;
1436 
1437 failure:
1438 	if (fi) {
1439 		fi->fib_dead = 1;
1440 		free_fib_info(fi);
1441 	}
1442 
1443 	return ERR_PTR(err);
1444 }
1445 
1446 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1447 		     unsigned int *flags, bool skip_oif)
1448 {
1449 	if (nhc->nhc_flags & RTNH_F_DEAD)
1450 		*flags |= RTNH_F_DEAD;
1451 
1452 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1453 		*flags |= RTNH_F_LINKDOWN;
1454 
1455 		rcu_read_lock();
1456 		switch (nhc->nhc_family) {
1457 		case AF_INET:
1458 			if (ip_ignore_linkdown(nhc->nhc_dev))
1459 				*flags |= RTNH_F_DEAD;
1460 			break;
1461 		case AF_INET6:
1462 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1463 				*flags |= RTNH_F_DEAD;
1464 			break;
1465 		}
1466 		rcu_read_unlock();
1467 	}
1468 
1469 	switch (nhc->nhc_gw_family) {
1470 	case AF_INET:
1471 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1472 			goto nla_put_failure;
1473 		break;
1474 	case AF_INET6:
1475 		/* if gateway family does not match nexthop family
1476 		 * gateway is encoded as RTA_VIA
1477 		 */
1478 		if (nhc->nhc_gw_family != nhc->nhc_family) {
1479 			int alen = sizeof(struct in6_addr);
1480 			struct nlattr *nla;
1481 			struct rtvia *via;
1482 
1483 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1484 			if (!nla)
1485 				goto nla_put_failure;
1486 
1487 			via = nla_data(nla);
1488 			via->rtvia_family = AF_INET6;
1489 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1490 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1491 					    &nhc->nhc_gw.ipv6) < 0) {
1492 			goto nla_put_failure;
1493 		}
1494 		break;
1495 	}
1496 
1497 	*flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1498 	if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1499 		*flags |= RTNH_F_OFFLOAD;
1500 
1501 	if (!skip_oif && nhc->nhc_dev &&
1502 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1503 		goto nla_put_failure;
1504 
1505 	if (nhc->nhc_lwtstate &&
1506 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate) < 0)
1507 		goto nla_put_failure;
1508 
1509 	return 0;
1510 
1511 nla_put_failure:
1512 	return -EMSGSIZE;
1513 }
1514 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1515 
1516 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1517 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1518 		    int nh_weight)
1519 {
1520 	const struct net_device *dev = nhc->nhc_dev;
1521 	struct rtnexthop *rtnh;
1522 	unsigned int flags = 0;
1523 
1524 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1525 	if (!rtnh)
1526 		goto nla_put_failure;
1527 
1528 	rtnh->rtnh_hops = nh_weight - 1;
1529 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1530 
1531 	if (fib_nexthop_info(skb, nhc, &flags, true) < 0)
1532 		goto nla_put_failure;
1533 
1534 	rtnh->rtnh_flags = flags;
1535 
1536 	/* length of rtnetlink header + attributes */
1537 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1538 
1539 	return 0;
1540 
1541 nla_put_failure:
1542 	return -EMSGSIZE;
1543 }
1544 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1545 #endif
1546 
1547 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1548 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1549 {
1550 	struct nlattr *mp;
1551 
1552 	mp = nla_nest_start(skb, RTA_MULTIPATH);
1553 	if (!mp)
1554 		goto nla_put_failure;
1555 
1556 	for_nexthops(fi) {
1557 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight) < 0)
1558 			goto nla_put_failure;
1559 #ifdef CONFIG_IP_ROUTE_CLASSID
1560 		if (nh->nh_tclassid &&
1561 		    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1562 			goto nla_put_failure;
1563 #endif
1564 	} endfor_nexthops(fi);
1565 
1566 	nla_nest_end(skb, mp);
1567 
1568 	return 0;
1569 
1570 nla_put_failure:
1571 	return -EMSGSIZE;
1572 }
1573 #else
1574 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1575 {
1576 	return 0;
1577 }
1578 #endif
1579 
1580 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1581 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1582 		  struct fib_info *fi, unsigned int flags)
1583 {
1584 	struct nlmsghdr *nlh;
1585 	struct rtmsg *rtm;
1586 
1587 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1588 	if (!nlh)
1589 		return -EMSGSIZE;
1590 
1591 	rtm = nlmsg_data(nlh);
1592 	rtm->rtm_family = AF_INET;
1593 	rtm->rtm_dst_len = dst_len;
1594 	rtm->rtm_src_len = 0;
1595 	rtm->rtm_tos = tos;
1596 	if (tb_id < 256)
1597 		rtm->rtm_table = tb_id;
1598 	else
1599 		rtm->rtm_table = RT_TABLE_COMPAT;
1600 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1601 		goto nla_put_failure;
1602 	rtm->rtm_type = type;
1603 	rtm->rtm_flags = fi->fib_flags;
1604 	rtm->rtm_scope = fi->fib_scope;
1605 	rtm->rtm_protocol = fi->fib_protocol;
1606 
1607 	if (rtm->rtm_dst_len &&
1608 	    nla_put_in_addr(skb, RTA_DST, dst))
1609 		goto nla_put_failure;
1610 	if (fi->fib_priority &&
1611 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1612 		goto nla_put_failure;
1613 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1614 		goto nla_put_failure;
1615 
1616 	if (fi->fib_prefsrc &&
1617 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1618 		goto nla_put_failure;
1619 	if (fi->fib_nhs == 1) {
1620 		struct fib_nh *nh = &fi->fib_nh[0];
1621 		unsigned int flags = 0;
1622 
1623 		if (fib_nexthop_info(skb, &nh->nh_common, &flags, false) < 0)
1624 			goto nla_put_failure;
1625 
1626 		rtm->rtm_flags = flags;
1627 #ifdef CONFIG_IP_ROUTE_CLASSID
1628 		if (nh->nh_tclassid &&
1629 		    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1630 			goto nla_put_failure;
1631 #endif
1632 	} else {
1633 		if (fib_add_multipath(skb, fi) < 0)
1634 			goto nla_put_failure;
1635 	}
1636 
1637 	nlmsg_end(skb, nlh);
1638 	return 0;
1639 
1640 nla_put_failure:
1641 	nlmsg_cancel(skb, nlh);
1642 	return -EMSGSIZE;
1643 }
1644 
1645 /*
1646  * Update FIB if:
1647  * - local address disappeared -> we must delete all the entries
1648  *   referring to it.
1649  * - device went down -> we must shutdown all nexthops going via it.
1650  */
1651 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1652 {
1653 	int ret = 0;
1654 	unsigned int hash = fib_laddr_hashfn(local);
1655 	struct hlist_head *head = &fib_info_laddrhash[hash];
1656 	struct net *net = dev_net(dev);
1657 	int tb_id = l3mdev_fib_table(dev);
1658 	struct fib_info *fi;
1659 
1660 	if (!fib_info_laddrhash || local == 0)
1661 		return 0;
1662 
1663 	hlist_for_each_entry(fi, head, fib_lhash) {
1664 		if (!net_eq(fi->fib_net, net) ||
1665 		    fi->fib_tb_id != tb_id)
1666 			continue;
1667 		if (fi->fib_prefsrc == local) {
1668 			fi->fib_flags |= RTNH_F_DEAD;
1669 			ret++;
1670 		}
1671 	}
1672 	return ret;
1673 }
1674 
1675 static int call_fib_nh_notifiers(struct fib_nh *nh,
1676 				 enum fib_event_type event_type)
1677 {
1678 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1679 	struct fib_nh_notifier_info info = {
1680 		.fib_nh = nh,
1681 	};
1682 
1683 	switch (event_type) {
1684 	case FIB_EVENT_NH_ADD:
1685 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1686 			break;
1687 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1688 			break;
1689 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1690 					   &info.info);
1691 	case FIB_EVENT_NH_DEL:
1692 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1693 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1694 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1695 						   event_type, &info.info);
1696 	default:
1697 		break;
1698 	}
1699 
1700 	return NOTIFY_DONE;
1701 }
1702 
1703 /* Update the PMTU of exceptions when:
1704  * - the new MTU of the first hop becomes smaller than the PMTU
1705  * - the old MTU was the same as the PMTU, and it limited discovery of
1706  *   larger MTUs on the path. With that limit raised, we can now
1707  *   discover larger MTUs
1708  * A special case is locked exceptions, for which the PMTU is smaller
1709  * than the minimal accepted PMTU:
1710  * - if the new MTU is greater than the PMTU, don't make any change
1711  * - otherwise, unlock and set PMTU
1712  */
1713 static void nh_update_mtu(struct fib_nh *nh, u32 new, u32 orig)
1714 {
1715 	struct fnhe_hash_bucket *bucket;
1716 	int i;
1717 
1718 	bucket = rcu_dereference_protected(nh->nh_exceptions, 1);
1719 	if (!bucket)
1720 		return;
1721 
1722 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1723 		struct fib_nh_exception *fnhe;
1724 
1725 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1726 		     fnhe;
1727 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1728 			if (fnhe->fnhe_mtu_locked) {
1729 				if (new <= fnhe->fnhe_pmtu) {
1730 					fnhe->fnhe_pmtu = new;
1731 					fnhe->fnhe_mtu_locked = false;
1732 				}
1733 			} else if (new < fnhe->fnhe_pmtu ||
1734 				   orig == fnhe->fnhe_pmtu) {
1735 				fnhe->fnhe_pmtu = new;
1736 			}
1737 		}
1738 	}
1739 }
1740 
1741 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1742 {
1743 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1744 	struct hlist_head *head = &fib_info_devhash[hash];
1745 	struct fib_nh *nh;
1746 
1747 	hlist_for_each_entry(nh, head, nh_hash) {
1748 		if (nh->fib_nh_dev == dev)
1749 			nh_update_mtu(nh, dev->mtu, orig_mtu);
1750 	}
1751 }
1752 
1753 /* Event              force Flags           Description
1754  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1755  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1756  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1757  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1758  */
1759 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1760 {
1761 	int ret = 0;
1762 	int scope = RT_SCOPE_NOWHERE;
1763 	struct fib_info *prev_fi = NULL;
1764 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1765 	struct hlist_head *head = &fib_info_devhash[hash];
1766 	struct fib_nh *nh;
1767 
1768 	if (force)
1769 		scope = -1;
1770 
1771 	hlist_for_each_entry(nh, head, nh_hash) {
1772 		struct fib_info *fi = nh->nh_parent;
1773 		int dead;
1774 
1775 		BUG_ON(!fi->fib_nhs);
1776 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1777 			continue;
1778 		prev_fi = fi;
1779 		dead = 0;
1780 		change_nexthops(fi) {
1781 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1782 				dead++;
1783 			else if (nexthop_nh->fib_nh_dev == dev &&
1784 				 nexthop_nh->fib_nh_scope != scope) {
1785 				switch (event) {
1786 				case NETDEV_DOWN:
1787 				case NETDEV_UNREGISTER:
1788 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1789 					/* fall through */
1790 				case NETDEV_CHANGE:
1791 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1792 					break;
1793 				}
1794 				call_fib_nh_notifiers(nexthop_nh,
1795 						      FIB_EVENT_NH_DEL);
1796 				dead++;
1797 			}
1798 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1799 			if (event == NETDEV_UNREGISTER &&
1800 			    nexthop_nh->fib_nh_dev == dev) {
1801 				dead = fi->fib_nhs;
1802 				break;
1803 			}
1804 #endif
1805 		} endfor_nexthops(fi)
1806 		if (dead == fi->fib_nhs) {
1807 			switch (event) {
1808 			case NETDEV_DOWN:
1809 			case NETDEV_UNREGISTER:
1810 				fi->fib_flags |= RTNH_F_DEAD;
1811 				/* fall through */
1812 			case NETDEV_CHANGE:
1813 				fi->fib_flags |= RTNH_F_LINKDOWN;
1814 				break;
1815 			}
1816 			ret++;
1817 		}
1818 
1819 		fib_rebalance(fi);
1820 	}
1821 
1822 	return ret;
1823 }
1824 
1825 /* Must be invoked inside of an RCU protected region.  */
1826 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1827 {
1828 	struct fib_info *fi = NULL, *last_resort = NULL;
1829 	struct hlist_head *fa_head = res->fa_head;
1830 	struct fib_table *tb = res->table;
1831 	u8 slen = 32 - res->prefixlen;
1832 	int order = -1, last_idx = -1;
1833 	struct fib_alias *fa, *fa1 = NULL;
1834 	u32 last_prio = res->fi->fib_priority;
1835 	u8 last_tos = 0;
1836 
1837 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1838 		struct fib_info *next_fi = fa->fa_info;
1839 
1840 		if (fa->fa_slen != slen)
1841 			continue;
1842 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1843 			continue;
1844 		if (fa->tb_id != tb->tb_id)
1845 			continue;
1846 		if (next_fi->fib_priority > last_prio &&
1847 		    fa->fa_tos == last_tos) {
1848 			if (last_tos)
1849 				continue;
1850 			break;
1851 		}
1852 		if (next_fi->fib_flags & RTNH_F_DEAD)
1853 			continue;
1854 		last_tos = fa->fa_tos;
1855 		last_prio = next_fi->fib_priority;
1856 
1857 		if (next_fi->fib_scope != res->scope ||
1858 		    fa->fa_type != RTN_UNICAST)
1859 			continue;
1860 		if (!next_fi->fib_nh[0].fib_nh_gw4 ||
1861 		    next_fi->fib_nh[0].fib_nh_scope != RT_SCOPE_LINK)
1862 			continue;
1863 
1864 		fib_alias_accessed(fa);
1865 
1866 		if (!fi) {
1867 			if (next_fi != res->fi)
1868 				break;
1869 			fa1 = fa;
1870 		} else if (!fib_detect_death(fi, order, &last_resort,
1871 					     &last_idx, fa1->fa_default)) {
1872 			fib_result_assign(res, fi);
1873 			fa1->fa_default = order;
1874 			goto out;
1875 		}
1876 		fi = next_fi;
1877 		order++;
1878 	}
1879 
1880 	if (order <= 0 || !fi) {
1881 		if (fa1)
1882 			fa1->fa_default = -1;
1883 		goto out;
1884 	}
1885 
1886 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1887 			      fa1->fa_default)) {
1888 		fib_result_assign(res, fi);
1889 		fa1->fa_default = order;
1890 		goto out;
1891 	}
1892 
1893 	if (last_idx >= 0)
1894 		fib_result_assign(res, last_resort);
1895 	fa1->fa_default = last_idx;
1896 out:
1897 	return;
1898 }
1899 
1900 /*
1901  * Dead device goes up. We wake up dead nexthops.
1902  * It takes sense only on multipath routes.
1903  */
1904 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1905 {
1906 	struct fib_info *prev_fi;
1907 	unsigned int hash;
1908 	struct hlist_head *head;
1909 	struct fib_nh *nh;
1910 	int ret;
1911 
1912 	if (!(dev->flags & IFF_UP))
1913 		return 0;
1914 
1915 	if (nh_flags & RTNH_F_DEAD) {
1916 		unsigned int flags = dev_get_flags(dev);
1917 
1918 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1919 			nh_flags |= RTNH_F_LINKDOWN;
1920 	}
1921 
1922 	prev_fi = NULL;
1923 	hash = fib_devindex_hashfn(dev->ifindex);
1924 	head = &fib_info_devhash[hash];
1925 	ret = 0;
1926 
1927 	hlist_for_each_entry(nh, head, nh_hash) {
1928 		struct fib_info *fi = nh->nh_parent;
1929 		int alive;
1930 
1931 		BUG_ON(!fi->fib_nhs);
1932 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1933 			continue;
1934 
1935 		prev_fi = fi;
1936 		alive = 0;
1937 		change_nexthops(fi) {
1938 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
1939 				alive++;
1940 				continue;
1941 			}
1942 			if (!nexthop_nh->fib_nh_dev ||
1943 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
1944 				continue;
1945 			if (nexthop_nh->fib_nh_dev != dev ||
1946 			    !__in_dev_get_rtnl(dev))
1947 				continue;
1948 			alive++;
1949 			nexthop_nh->fib_nh_flags &= ~nh_flags;
1950 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1951 		} endfor_nexthops(fi)
1952 
1953 		if (alive > 0) {
1954 			fi->fib_flags &= ~nh_flags;
1955 			ret++;
1956 		}
1957 
1958 		fib_rebalance(fi);
1959 	}
1960 
1961 	return ret;
1962 }
1963 
1964 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1965 static bool fib_good_nh(const struct fib_nh *nh)
1966 {
1967 	int state = NUD_REACHABLE;
1968 
1969 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
1970 		struct neighbour *n;
1971 
1972 		rcu_read_lock_bh();
1973 
1974 		if (likely(nh->fib_nh_gw_family == AF_INET))
1975 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
1976 						   (__force u32)nh->fib_nh_gw4);
1977 		else if (nh->fib_nh_gw_family == AF_INET6)
1978 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
1979 							   &nh->fib_nh_gw6);
1980 		else
1981 			n = NULL;
1982 		if (n)
1983 			state = n->nud_state;
1984 
1985 		rcu_read_unlock_bh();
1986 	}
1987 
1988 	return !!(state & NUD_VALID);
1989 }
1990 
1991 void fib_select_multipath(struct fib_result *res, int hash)
1992 {
1993 	struct fib_info *fi = res->fi;
1994 	struct net *net = fi->fib_net;
1995 	bool first = false;
1996 
1997 	change_nexthops(fi) {
1998 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
1999 			if (!fib_good_nh(nexthop_nh))
2000 				continue;
2001 			if (!first) {
2002 				res->nh_sel = nhsel;
2003 				res->nhc = &nexthop_nh->nh_common;
2004 				first = true;
2005 			}
2006 		}
2007 
2008 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2009 			continue;
2010 
2011 		res->nh_sel = nhsel;
2012 		res->nhc = &nexthop_nh->nh_common;
2013 		return;
2014 	} endfor_nexthops(fi);
2015 }
2016 #endif
2017 
2018 void fib_select_path(struct net *net, struct fib_result *res,
2019 		     struct flowi4 *fl4, const struct sk_buff *skb)
2020 {
2021 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2022 		goto check_saddr;
2023 
2024 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2025 	if (res->fi->fib_nhs > 1) {
2026 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2027 
2028 		fib_select_multipath(res, h);
2029 	}
2030 	else
2031 #endif
2032 	if (!res->prefixlen &&
2033 	    res->table->tb_num_default > 1 &&
2034 	    res->type == RTN_UNICAST)
2035 		fib_select_default(fl4, res);
2036 
2037 check_saddr:
2038 	if (!fl4->saddr)
2039 		fl4->saddr = fib_result_prefsrc(net, res);
2040 }
2041