xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 4fc4dca8)
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_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 	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 	int err;
495 
496 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
497 						    gfp_flags);
498 	if (!nhc->nhc_pcpu_rth_output)
499 		return -ENOMEM;
500 
501 	if (encap) {
502 		struct lwtunnel_state *lwtstate;
503 
504 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
505 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
506 			err = -EINVAL;
507 			goto lwt_failure;
508 		}
509 		err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family,
510 					   cfg, &lwtstate, extack);
511 		if (err)
512 			goto lwt_failure;
513 
514 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
515 	}
516 
517 	return 0;
518 
519 lwt_failure:
520 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
521 	nhc->nhc_pcpu_rth_output = NULL;
522 	return err;
523 }
524 EXPORT_SYMBOL_GPL(fib_nh_common_init);
525 
526 int fib_nh_init(struct net *net, struct fib_nh *nh,
527 		struct fib_config *cfg, int nh_weight,
528 		struct netlink_ext_ack *extack)
529 {
530 	int err;
531 
532 	nh->fib_nh_family = AF_INET;
533 
534 	err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
535 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
536 	if (err)
537 		return err;
538 
539 	nh->fib_nh_oif = cfg->fc_oif;
540 	nh->fib_nh_gw_family = cfg->fc_gw_family;
541 	if (cfg->fc_gw_family == AF_INET)
542 		nh->fib_nh_gw4 = cfg->fc_gw4;
543 	else if (cfg->fc_gw_family == AF_INET6)
544 		nh->fib_nh_gw6 = cfg->fc_gw6;
545 
546 	nh->fib_nh_flags = cfg->fc_flags;
547 
548 #ifdef CONFIG_IP_ROUTE_CLASSID
549 	nh->nh_tclassid = cfg->fc_flow;
550 	if (nh->nh_tclassid)
551 		net->ipv4.fib_num_tclassid_users++;
552 #endif
553 #ifdef CONFIG_IP_ROUTE_MULTIPATH
554 	nh->fib_nh_weight = nh_weight;
555 #endif
556 	return 0;
557 }
558 
559 #ifdef CONFIG_IP_ROUTE_MULTIPATH
560 
561 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
562 			      struct netlink_ext_ack *extack)
563 {
564 	int nhs = 0;
565 
566 	while (rtnh_ok(rtnh, remaining)) {
567 		nhs++;
568 		rtnh = rtnh_next(rtnh, &remaining);
569 	}
570 
571 	/* leftover implies invalid nexthop configuration, discard it */
572 	if (remaining > 0) {
573 		NL_SET_ERR_MSG(extack,
574 			       "Invalid nexthop configuration - extra data after nexthops");
575 		nhs = 0;
576 	}
577 
578 	return nhs;
579 }
580 
581 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
582 		       int remaining, struct fib_config *cfg,
583 		       struct netlink_ext_ack *extack)
584 {
585 	struct net *net = fi->fib_net;
586 	struct fib_config fib_cfg;
587 	int ret;
588 
589 	change_nexthops(fi) {
590 		int attrlen;
591 
592 		memset(&fib_cfg, 0, sizeof(fib_cfg));
593 
594 		if (!rtnh_ok(rtnh, remaining)) {
595 			NL_SET_ERR_MSG(extack,
596 				       "Invalid nexthop configuration - extra data after nexthop");
597 			return -EINVAL;
598 		}
599 
600 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
601 			NL_SET_ERR_MSG(extack,
602 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
603 			return -EINVAL;
604 		}
605 
606 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
607 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
608 
609 		attrlen = rtnh_attrlen(rtnh);
610 		if (attrlen > 0) {
611 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
612 
613 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
614 			nlav = nla_find(attrs, attrlen, RTA_VIA);
615 			if (nla && nlav) {
616 				NL_SET_ERR_MSG(extack,
617 					       "Nexthop configuration can not contain both GATEWAY and VIA");
618 				return -EINVAL;
619 			}
620 			if (nla) {
621 				fib_cfg.fc_gw4 = nla_get_in_addr(nla);
622 				if (fib_cfg.fc_gw4)
623 					fib_cfg.fc_gw_family = AF_INET;
624 			} else if (nlav) {
625 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
626 				if (ret)
627 					goto errout;
628 			}
629 
630 			nla = nla_find(attrs, attrlen, RTA_FLOW);
631 			if (nla)
632 				fib_cfg.fc_flow = nla_get_u32(nla);
633 
634 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
635 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
636 			if (nla)
637 				fib_cfg.fc_encap_type = nla_get_u16(nla);
638 		}
639 
640 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
641 				  rtnh->rtnh_hops + 1, extack);
642 		if (ret)
643 			goto errout;
644 
645 		rtnh = rtnh_next(rtnh, &remaining);
646 	} endfor_nexthops(fi);
647 
648 	ret = -EINVAL;
649 	if (cfg->fc_oif && fi->fib_nh->fib_nh_oif != cfg->fc_oif) {
650 		NL_SET_ERR_MSG(extack,
651 			       "Nexthop device index does not match RTA_OIF");
652 		goto errout;
653 	}
654 	if (cfg->fc_gw_family) {
655 		if (cfg->fc_gw_family != fi->fib_nh->fib_nh_gw_family ||
656 		    (cfg->fc_gw_family == AF_INET &&
657 		     fi->fib_nh->fib_nh_gw4 != cfg->fc_gw4) ||
658 		    (cfg->fc_gw_family == AF_INET6 &&
659 		     ipv6_addr_cmp(&fi->fib_nh->fib_nh_gw6, &cfg->fc_gw6))) {
660 			NL_SET_ERR_MSG(extack,
661 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
662 			goto errout;
663 		}
664 	}
665 #ifdef CONFIG_IP_ROUTE_CLASSID
666 	if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
667 		NL_SET_ERR_MSG(extack,
668 			       "Nexthop class id does not match RTA_FLOW");
669 		goto errout;
670 	}
671 #endif
672 	ret = 0;
673 errout:
674 	return ret;
675 }
676 
677 static void fib_rebalance(struct fib_info *fi)
678 {
679 	int total;
680 	int w;
681 
682 	if (fi->fib_nhs < 2)
683 		return;
684 
685 	total = 0;
686 	for_nexthops(fi) {
687 		if (nh->fib_nh_flags & RTNH_F_DEAD)
688 			continue;
689 
690 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
691 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
692 			continue;
693 
694 		total += nh->fib_nh_weight;
695 	} endfor_nexthops(fi);
696 
697 	w = 0;
698 	change_nexthops(fi) {
699 		int upper_bound;
700 
701 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
702 			upper_bound = -1;
703 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
704 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
705 			upper_bound = -1;
706 		} else {
707 			w += nexthop_nh->fib_nh_weight;
708 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
709 							    total) - 1;
710 		}
711 
712 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
713 	} endfor_nexthops(fi);
714 }
715 #else /* CONFIG_IP_ROUTE_MULTIPATH */
716 
717 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
718 		       int remaining, struct fib_config *cfg,
719 		       struct netlink_ext_ack *extack)
720 {
721 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
722 
723 	return -EINVAL;
724 }
725 
726 #define fib_rebalance(fi) do { } while (0)
727 
728 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
729 
730 static int fib_encap_match(u16 encap_type,
731 			   struct nlattr *encap,
732 			   const struct fib_nh *nh,
733 			   const struct fib_config *cfg,
734 			   struct netlink_ext_ack *extack)
735 {
736 	struct lwtunnel_state *lwtstate;
737 	int ret, result = 0;
738 
739 	if (encap_type == LWTUNNEL_ENCAP_NONE)
740 		return 0;
741 
742 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
743 				   cfg, &lwtstate, extack);
744 	if (!ret) {
745 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
746 		lwtstate_free(lwtstate);
747 	}
748 
749 	return result;
750 }
751 
752 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
753 		 struct netlink_ext_ack *extack)
754 {
755 #ifdef CONFIG_IP_ROUTE_MULTIPATH
756 	struct rtnexthop *rtnh;
757 	int remaining;
758 #endif
759 
760 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
761 		return 1;
762 
763 	if (cfg->fc_oif || cfg->fc_gw_family) {
764 		if (cfg->fc_encap) {
765 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
766 					    fi->fib_nh, cfg, extack))
767 				return 1;
768 		}
769 #ifdef CONFIG_IP_ROUTE_CLASSID
770 		if (cfg->fc_flow &&
771 		    cfg->fc_flow != fi->fib_nh->nh_tclassid)
772 			return 1;
773 #endif
774 		if ((cfg->fc_oif && cfg->fc_oif != fi->fib_nh->fib_nh_oif) ||
775 		    (cfg->fc_gw_family &&
776 		     cfg->fc_gw_family != fi->fib_nh->fib_nh_gw_family))
777 			return 1;
778 
779 		if (cfg->fc_gw_family == AF_INET &&
780 		    cfg->fc_gw4 != fi->fib_nh->fib_nh_gw4)
781 			return 1;
782 
783 		if (cfg->fc_gw_family == AF_INET6 &&
784 		    ipv6_addr_cmp(&cfg->fc_gw6, &fi->fib_nh->fib_nh_gw6))
785 			return 1;
786 
787 		return 0;
788 	}
789 
790 #ifdef CONFIG_IP_ROUTE_MULTIPATH
791 	if (!cfg->fc_mp)
792 		return 0;
793 
794 	rtnh = cfg->fc_mp;
795 	remaining = cfg->fc_mp_len;
796 
797 	for_nexthops(fi) {
798 		int attrlen;
799 
800 		if (!rtnh_ok(rtnh, remaining))
801 			return -EINVAL;
802 
803 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
804 			return 1;
805 
806 		attrlen = rtnh_attrlen(rtnh);
807 		if (attrlen > 0) {
808 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
809 
810 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
811 			nlav = nla_find(attrs, attrlen, RTA_VIA);
812 			if (nla && nlav) {
813 				NL_SET_ERR_MSG(extack,
814 					       "Nexthop configuration can not contain both GATEWAY and VIA");
815 				return -EINVAL;
816 			}
817 
818 			if (nla) {
819 				if (nh->fib_nh_gw_family != AF_INET ||
820 				    nla_get_in_addr(nla) != nh->fib_nh_gw4)
821 					return 1;
822 			} else if (nlav) {
823 				struct fib_config cfg2;
824 				int err;
825 
826 				err = fib_gw_from_via(&cfg2, nlav, extack);
827 				if (err)
828 					return err;
829 
830 				switch (nh->fib_nh_gw_family) {
831 				case AF_INET:
832 					if (cfg2.fc_gw_family != AF_INET ||
833 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
834 						return 1;
835 					break;
836 				case AF_INET6:
837 					if (cfg2.fc_gw_family != AF_INET6 ||
838 					    ipv6_addr_cmp(&cfg2.fc_gw6,
839 							  &nh->fib_nh_gw6))
840 						return 1;
841 					break;
842 				}
843 			}
844 
845 #ifdef CONFIG_IP_ROUTE_CLASSID
846 			nla = nla_find(attrs, attrlen, RTA_FLOW);
847 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
848 				return 1;
849 #endif
850 		}
851 
852 		rtnh = rtnh_next(rtnh, &remaining);
853 	} endfor_nexthops(fi);
854 #endif
855 	return 0;
856 }
857 
858 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
859 {
860 	struct nlattr *nla;
861 	int remaining;
862 
863 	if (!cfg->fc_mx)
864 		return true;
865 
866 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
867 		int type = nla_type(nla);
868 		u32 fi_val, val;
869 
870 		if (!type)
871 			continue;
872 		if (type > RTAX_MAX)
873 			return false;
874 
875 		if (type == RTAX_CC_ALGO) {
876 			char tmp[TCP_CA_NAME_MAX];
877 			bool ecn_ca = false;
878 
879 			nla_strlcpy(tmp, nla, sizeof(tmp));
880 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
881 		} else {
882 			if (nla_len(nla) != sizeof(u32))
883 				return false;
884 			val = nla_get_u32(nla);
885 		}
886 
887 		fi_val = fi->fib_metrics->metrics[type - 1];
888 		if (type == RTAX_FEATURES)
889 			fi_val &= ~DST_FEATURE_ECN_CA;
890 
891 		if (fi_val != val)
892 			return false;
893 	}
894 
895 	return true;
896 }
897 
898 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
899 			      u32 table, struct netlink_ext_ack *extack)
900 {
901 	struct fib6_config cfg = {
902 		.fc_table = table,
903 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
904 		.fc_ifindex = nh->fib_nh_oif,
905 		.fc_gateway = nh->fib_nh_gw6,
906 	};
907 	struct fib6_nh fib6_nh = {};
908 	int err;
909 
910 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
911 	if (!err) {
912 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
913 		dev_hold(nh->fib_nh_dev);
914 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
915 		nh->fib_nh_scope = RT_SCOPE_LINK;
916 
917 		ipv6_stub->fib6_nh_release(&fib6_nh);
918 	}
919 
920 	return err;
921 }
922 
923 /*
924  * Picture
925  * -------
926  *
927  * Semantics of nexthop is very messy by historical reasons.
928  * We have to take into account, that:
929  * a) gateway can be actually local interface address,
930  *    so that gatewayed route is direct.
931  * b) gateway must be on-link address, possibly
932  *    described not by an ifaddr, but also by a direct route.
933  * c) If both gateway and interface are specified, they should not
934  *    contradict.
935  * d) If we use tunnel routes, gateway could be not on-link.
936  *
937  * Attempt to reconcile all of these (alas, self-contradictory) conditions
938  * results in pretty ugly and hairy code with obscure logic.
939  *
940  * I chose to generalized it instead, so that the size
941  * of code does not increase practically, but it becomes
942  * much more general.
943  * Every prefix is assigned a "scope" value: "host" is local address,
944  * "link" is direct route,
945  * [ ... "site" ... "interior" ... ]
946  * and "universe" is true gateway route with global meaning.
947  *
948  * Every prefix refers to a set of "nexthop"s (gw, oif),
949  * where gw must have narrower scope. This recursion stops
950  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
951  * which means that gw is forced to be on link.
952  *
953  * Code is still hairy, but now it is apparently logically
954  * consistent and very flexible. F.e. as by-product it allows
955  * to co-exists in peace independent exterior and interior
956  * routing processes.
957  *
958  * Normally it looks as following.
959  *
960  * {universe prefix}  -> (gw, oif) [scope link]
961  *		  |
962  *		  |-> {link prefix} -> (gw, oif) [scope local]
963  *					|
964  *					|-> {local prefix} (terminal node)
965  */
966 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
967 			      u8 scope, struct netlink_ext_ack *extack)
968 {
969 	struct net_device *dev;
970 	struct fib_result res;
971 	int err;
972 
973 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
974 		unsigned int addr_type;
975 
976 		if (scope >= RT_SCOPE_LINK) {
977 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
978 			return -EINVAL;
979 		}
980 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
981 		if (!dev) {
982 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
983 			return -ENODEV;
984 		}
985 		if (!(dev->flags & IFF_UP)) {
986 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
987 			return -ENETDOWN;
988 		}
989 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
990 		if (addr_type != RTN_UNICAST) {
991 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
992 			return -EINVAL;
993 		}
994 		if (!netif_carrier_ok(dev))
995 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
996 		nh->fib_nh_dev = dev;
997 		dev_hold(dev);
998 		nh->fib_nh_scope = RT_SCOPE_LINK;
999 		return 0;
1000 	}
1001 	rcu_read_lock();
1002 	{
1003 		struct fib_table *tbl = NULL;
1004 		struct flowi4 fl4 = {
1005 			.daddr = nh->fib_nh_gw4,
1006 			.flowi4_scope = scope + 1,
1007 			.flowi4_oif = nh->fib_nh_oif,
1008 			.flowi4_iif = LOOPBACK_IFINDEX,
1009 		};
1010 
1011 		/* It is not necessary, but requires a bit of thinking */
1012 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1013 			fl4.flowi4_scope = RT_SCOPE_LINK;
1014 
1015 		if (table)
1016 			tbl = fib_get_table(net, table);
1017 
1018 		if (tbl)
1019 			err = fib_table_lookup(tbl, &fl4, &res,
1020 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1021 					       FIB_LOOKUP_NOREF);
1022 
1023 		/* on error or if no table given do full lookup. This
1024 		 * is needed for example when nexthops are in the local
1025 		 * table rather than the given table
1026 		 */
1027 		if (!tbl || err) {
1028 			err = fib_lookup(net, &fl4, &res,
1029 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1030 		}
1031 
1032 		if (err) {
1033 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1034 			goto out;
1035 		}
1036 	}
1037 
1038 	err = -EINVAL;
1039 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1040 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1041 		goto out;
1042 	}
1043 	nh->fib_nh_scope = res.scope;
1044 	nh->fib_nh_oif = FIB_RES_OIF(res);
1045 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1046 	if (!dev) {
1047 		NL_SET_ERR_MSG(extack,
1048 			       "No egress device for nexthop gateway");
1049 		goto out;
1050 	}
1051 	dev_hold(dev);
1052 	if (!netif_carrier_ok(dev))
1053 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1054 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1055 out:
1056 	rcu_read_unlock();
1057 	return err;
1058 }
1059 
1060 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1061 			      struct netlink_ext_ack *extack)
1062 {
1063 	struct in_device *in_dev;
1064 	int err;
1065 
1066 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1067 		NL_SET_ERR_MSG(extack,
1068 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1069 		return -EINVAL;
1070 	}
1071 
1072 	rcu_read_lock();
1073 
1074 	err = -ENODEV;
1075 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1076 	if (!in_dev)
1077 		goto out;
1078 	err = -ENETDOWN;
1079 	if (!(in_dev->dev->flags & IFF_UP)) {
1080 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1081 		goto out;
1082 	}
1083 
1084 	nh->fib_nh_dev = in_dev->dev;
1085 	dev_hold(nh->fib_nh_dev);
1086 	nh->fib_nh_scope = RT_SCOPE_HOST;
1087 	if (!netif_carrier_ok(nh->fib_nh_dev))
1088 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1089 	err = 0;
1090 out:
1091 	rcu_read_unlock();
1092 	return err;
1093 }
1094 
1095 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
1096 			struct netlink_ext_ack *extack)
1097 {
1098 	struct net *net = cfg->fc_nlinfo.nl_net;
1099 	u32 table = cfg->fc_table;
1100 	int err;
1101 
1102 	if (nh->fib_nh_gw_family == AF_INET)
1103 		err = fib_check_nh_v4_gw(net, nh, table, cfg->fc_scope, extack);
1104 	else if (nh->fib_nh_gw_family == AF_INET6)
1105 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1106 	else
1107 		err = fib_check_nh_nongw(net, nh, extack);
1108 
1109 	return err;
1110 }
1111 
1112 static inline unsigned int fib_laddr_hashfn(__be32 val)
1113 {
1114 	unsigned int mask = (fib_info_hash_size - 1);
1115 
1116 	return ((__force u32)val ^
1117 		((__force u32)val >> 7) ^
1118 		((__force u32)val >> 14)) & mask;
1119 }
1120 
1121 static struct hlist_head *fib_info_hash_alloc(int bytes)
1122 {
1123 	if (bytes <= PAGE_SIZE)
1124 		return kzalloc(bytes, GFP_KERNEL);
1125 	else
1126 		return (struct hlist_head *)
1127 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1128 					 get_order(bytes));
1129 }
1130 
1131 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1132 {
1133 	if (!hash)
1134 		return;
1135 
1136 	if (bytes <= PAGE_SIZE)
1137 		kfree(hash);
1138 	else
1139 		free_pages((unsigned long) hash, get_order(bytes));
1140 }
1141 
1142 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1143 			       struct hlist_head *new_laddrhash,
1144 			       unsigned int new_size)
1145 {
1146 	struct hlist_head *old_info_hash, *old_laddrhash;
1147 	unsigned int old_size = fib_info_hash_size;
1148 	unsigned int i, bytes;
1149 
1150 	spin_lock_bh(&fib_info_lock);
1151 	old_info_hash = fib_info_hash;
1152 	old_laddrhash = fib_info_laddrhash;
1153 	fib_info_hash_size = new_size;
1154 
1155 	for (i = 0; i < old_size; i++) {
1156 		struct hlist_head *head = &fib_info_hash[i];
1157 		struct hlist_node *n;
1158 		struct fib_info *fi;
1159 
1160 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1161 			struct hlist_head *dest;
1162 			unsigned int new_hash;
1163 
1164 			new_hash = fib_info_hashfn(fi);
1165 			dest = &new_info_hash[new_hash];
1166 			hlist_add_head(&fi->fib_hash, dest);
1167 		}
1168 	}
1169 	fib_info_hash = new_info_hash;
1170 
1171 	for (i = 0; i < old_size; i++) {
1172 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1173 		struct hlist_node *n;
1174 		struct fib_info *fi;
1175 
1176 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1177 			struct hlist_head *ldest;
1178 			unsigned int new_hash;
1179 
1180 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1181 			ldest = &new_laddrhash[new_hash];
1182 			hlist_add_head(&fi->fib_lhash, ldest);
1183 		}
1184 	}
1185 	fib_info_laddrhash = new_laddrhash;
1186 
1187 	spin_unlock_bh(&fib_info_lock);
1188 
1189 	bytes = old_size * sizeof(struct hlist_head *);
1190 	fib_info_hash_free(old_info_hash, bytes);
1191 	fib_info_hash_free(old_laddrhash, bytes);
1192 }
1193 
1194 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
1195 {
1196 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev,
1197 					nh->fib_nh_gw4,
1198 					nh->nh_parent->fib_scope);
1199 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1200 
1201 	return nh->nh_saddr;
1202 }
1203 
1204 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1205 {
1206 	struct fib_nh_common *nhc = res->nhc;
1207 	struct fib_nh *nh;
1208 
1209 	if (res->fi->fib_prefsrc)
1210 		return res->fi->fib_prefsrc;
1211 
1212 	nh = container_of(nhc, struct fib_nh, nh_common);
1213 	if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1214 		return nh->nh_saddr;
1215 
1216 	return fib_info_update_nh_saddr(net, nh);
1217 }
1218 
1219 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1220 {
1221 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1222 	    fib_prefsrc != cfg->fc_dst) {
1223 		u32 tb_id = cfg->fc_table;
1224 		int rc;
1225 
1226 		if (tb_id == RT_TABLE_MAIN)
1227 			tb_id = RT_TABLE_LOCAL;
1228 
1229 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1230 					  fib_prefsrc, tb_id);
1231 
1232 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1233 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1234 						  fib_prefsrc, RT_TABLE_LOCAL);
1235 		}
1236 
1237 		if (rc != RTN_LOCAL)
1238 			return false;
1239 	}
1240 	return true;
1241 }
1242 
1243 struct fib_info *fib_create_info(struct fib_config *cfg,
1244 				 struct netlink_ext_ack *extack)
1245 {
1246 	int err;
1247 	struct fib_info *fi = NULL;
1248 	struct fib_info *ofi;
1249 	int nhs = 1;
1250 	struct net *net = cfg->fc_nlinfo.nl_net;
1251 
1252 	if (cfg->fc_type > RTN_MAX)
1253 		goto err_inval;
1254 
1255 	/* Fast check to catch the most weird cases */
1256 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1257 		NL_SET_ERR_MSG(extack, "Invalid scope");
1258 		goto err_inval;
1259 	}
1260 
1261 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1262 		NL_SET_ERR_MSG(extack,
1263 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1264 		goto err_inval;
1265 	}
1266 
1267 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1268 	if (cfg->fc_mp) {
1269 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1270 		if (nhs == 0)
1271 			goto err_inval;
1272 	}
1273 #endif
1274 
1275 	err = -ENOBUFS;
1276 	if (fib_info_cnt >= fib_info_hash_size) {
1277 		unsigned int new_size = fib_info_hash_size << 1;
1278 		struct hlist_head *new_info_hash;
1279 		struct hlist_head *new_laddrhash;
1280 		unsigned int bytes;
1281 
1282 		if (!new_size)
1283 			new_size = 16;
1284 		bytes = new_size * sizeof(struct hlist_head *);
1285 		new_info_hash = fib_info_hash_alloc(bytes);
1286 		new_laddrhash = fib_info_hash_alloc(bytes);
1287 		if (!new_info_hash || !new_laddrhash) {
1288 			fib_info_hash_free(new_info_hash, bytes);
1289 			fib_info_hash_free(new_laddrhash, bytes);
1290 		} else
1291 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1292 
1293 		if (!fib_info_hash_size)
1294 			goto failure;
1295 	}
1296 
1297 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1298 	if (!fi)
1299 		goto failure;
1300 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1301 					      cfg->fc_mx_len, extack);
1302 	if (unlikely(IS_ERR(fi->fib_metrics))) {
1303 		err = PTR_ERR(fi->fib_metrics);
1304 		kfree(fi);
1305 		return ERR_PTR(err);
1306 	}
1307 
1308 	fib_info_cnt++;
1309 	fi->fib_net = net;
1310 	fi->fib_protocol = cfg->fc_protocol;
1311 	fi->fib_scope = cfg->fc_scope;
1312 	fi->fib_flags = cfg->fc_flags;
1313 	fi->fib_priority = cfg->fc_priority;
1314 	fi->fib_prefsrc = cfg->fc_prefsrc;
1315 	fi->fib_type = cfg->fc_type;
1316 	fi->fib_tb_id = cfg->fc_table;
1317 
1318 	fi->fib_nhs = nhs;
1319 	change_nexthops(fi) {
1320 		nexthop_nh->nh_parent = fi;
1321 	} endfor_nexthops(fi)
1322 
1323 	if (cfg->fc_mp)
1324 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1325 	else
1326 		err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1327 
1328 	if (err != 0)
1329 		goto failure;
1330 
1331 	if (fib_props[cfg->fc_type].error) {
1332 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1333 			NL_SET_ERR_MSG(extack,
1334 				       "Gateway, device and multipath can not be specified for this route type");
1335 			goto err_inval;
1336 		}
1337 		goto link_it;
1338 	} else {
1339 		switch (cfg->fc_type) {
1340 		case RTN_UNICAST:
1341 		case RTN_LOCAL:
1342 		case RTN_BROADCAST:
1343 		case RTN_ANYCAST:
1344 		case RTN_MULTICAST:
1345 			break;
1346 		default:
1347 			NL_SET_ERR_MSG(extack, "Invalid route type");
1348 			goto err_inval;
1349 		}
1350 	}
1351 
1352 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1353 		NL_SET_ERR_MSG(extack, "Invalid scope");
1354 		goto err_inval;
1355 	}
1356 
1357 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1358 		struct fib_nh *nh = fi->fib_nh;
1359 
1360 		/* Local address is added. */
1361 		if (nhs != 1) {
1362 			NL_SET_ERR_MSG(extack,
1363 				       "Route with host scope can not have multiple nexthops");
1364 			goto err_inval;
1365 		}
1366 		if (nh->fib_nh_gw_family) {
1367 			NL_SET_ERR_MSG(extack,
1368 				       "Route with host scope can not have a gateway");
1369 			goto err_inval;
1370 		}
1371 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1372 		nh->fib_nh_dev = dev_get_by_index(net, fi->fib_nh->fib_nh_oif);
1373 		err = -ENODEV;
1374 		if (!nh->fib_nh_dev)
1375 			goto failure;
1376 	} else {
1377 		int linkdown = 0;
1378 
1379 		change_nexthops(fi) {
1380 			err = fib_check_nh(cfg, nexthop_nh, extack);
1381 			if (err != 0)
1382 				goto failure;
1383 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1384 				linkdown++;
1385 		} endfor_nexthops(fi)
1386 		if (linkdown == fi->fib_nhs)
1387 			fi->fib_flags |= RTNH_F_LINKDOWN;
1388 	}
1389 
1390 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1391 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1392 		goto err_inval;
1393 	}
1394 
1395 	change_nexthops(fi) {
1396 		fib_info_update_nh_saddr(net, nexthop_nh);
1397 		if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1398 			fi->fib_nh_is_v6 = true;
1399 	} endfor_nexthops(fi)
1400 
1401 	fib_rebalance(fi);
1402 
1403 link_it:
1404 	ofi = fib_find_info(fi);
1405 	if (ofi) {
1406 		fi->fib_dead = 1;
1407 		free_fib_info(fi);
1408 		ofi->fib_treeref++;
1409 		return ofi;
1410 	}
1411 
1412 	fi->fib_treeref++;
1413 	refcount_set(&fi->fib_clntref, 1);
1414 	spin_lock_bh(&fib_info_lock);
1415 	hlist_add_head(&fi->fib_hash,
1416 		       &fib_info_hash[fib_info_hashfn(fi)]);
1417 	if (fi->fib_prefsrc) {
1418 		struct hlist_head *head;
1419 
1420 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1421 		hlist_add_head(&fi->fib_lhash, head);
1422 	}
1423 	change_nexthops(fi) {
1424 		struct hlist_head *head;
1425 		unsigned int hash;
1426 
1427 		if (!nexthop_nh->fib_nh_dev)
1428 			continue;
1429 		hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1430 		head = &fib_info_devhash[hash];
1431 		hlist_add_head(&nexthop_nh->nh_hash, head);
1432 	} endfor_nexthops(fi)
1433 	spin_unlock_bh(&fib_info_lock);
1434 	return fi;
1435 
1436 err_inval:
1437 	err = -EINVAL;
1438 
1439 failure:
1440 	if (fi) {
1441 		fi->fib_dead = 1;
1442 		free_fib_info(fi);
1443 	}
1444 
1445 	return ERR_PTR(err);
1446 }
1447 
1448 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1449 		     unsigned char *flags, bool skip_oif)
1450 {
1451 	if (nhc->nhc_flags & RTNH_F_DEAD)
1452 		*flags |= RTNH_F_DEAD;
1453 
1454 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1455 		*flags |= RTNH_F_LINKDOWN;
1456 
1457 		rcu_read_lock();
1458 		switch (nhc->nhc_family) {
1459 		case AF_INET:
1460 			if (ip_ignore_linkdown(nhc->nhc_dev))
1461 				*flags |= RTNH_F_DEAD;
1462 			break;
1463 		case AF_INET6:
1464 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1465 				*flags |= RTNH_F_DEAD;
1466 			break;
1467 		}
1468 		rcu_read_unlock();
1469 	}
1470 
1471 	switch (nhc->nhc_gw_family) {
1472 	case AF_INET:
1473 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1474 			goto nla_put_failure;
1475 		break;
1476 	case AF_INET6:
1477 		/* if gateway family does not match nexthop family
1478 		 * gateway is encoded as RTA_VIA
1479 		 */
1480 		if (nhc->nhc_gw_family != nhc->nhc_family) {
1481 			int alen = sizeof(struct in6_addr);
1482 			struct nlattr *nla;
1483 			struct rtvia *via;
1484 
1485 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1486 			if (!nla)
1487 				goto nla_put_failure;
1488 
1489 			via = nla_data(nla);
1490 			via->rtvia_family = AF_INET6;
1491 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1492 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1493 					    &nhc->nhc_gw.ipv6) < 0) {
1494 			goto nla_put_failure;
1495 		}
1496 		break;
1497 	}
1498 
1499 	*flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1500 	if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1501 		*flags |= RTNH_F_OFFLOAD;
1502 
1503 	if (!skip_oif && nhc->nhc_dev &&
1504 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1505 		goto nla_put_failure;
1506 
1507 	if (nhc->nhc_lwtstate &&
1508 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1509 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1510 		goto nla_put_failure;
1511 
1512 	return 0;
1513 
1514 nla_put_failure:
1515 	return -EMSGSIZE;
1516 }
1517 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1518 
1519 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1520 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1521 		    int nh_weight)
1522 {
1523 	const struct net_device *dev = nhc->nhc_dev;
1524 	struct rtnexthop *rtnh;
1525 	unsigned char flags = 0;
1526 
1527 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1528 	if (!rtnh)
1529 		goto nla_put_failure;
1530 
1531 	rtnh->rtnh_hops = nh_weight - 1;
1532 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1533 
1534 	if (fib_nexthop_info(skb, nhc, &flags, true) < 0)
1535 		goto nla_put_failure;
1536 
1537 	rtnh->rtnh_flags = flags;
1538 
1539 	/* length of rtnetlink header + attributes */
1540 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1541 
1542 	return 0;
1543 
1544 nla_put_failure:
1545 	return -EMSGSIZE;
1546 }
1547 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1548 #endif
1549 
1550 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1551 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1552 {
1553 	struct nlattr *mp;
1554 
1555 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1556 	if (!mp)
1557 		goto nla_put_failure;
1558 
1559 	for_nexthops(fi) {
1560 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight) < 0)
1561 			goto nla_put_failure;
1562 #ifdef CONFIG_IP_ROUTE_CLASSID
1563 		if (nh->nh_tclassid &&
1564 		    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1565 			goto nla_put_failure;
1566 #endif
1567 	} endfor_nexthops(fi);
1568 
1569 	nla_nest_end(skb, mp);
1570 
1571 	return 0;
1572 
1573 nla_put_failure:
1574 	return -EMSGSIZE;
1575 }
1576 #else
1577 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1578 {
1579 	return 0;
1580 }
1581 #endif
1582 
1583 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1584 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1585 		  struct fib_info *fi, unsigned int flags)
1586 {
1587 	struct nlmsghdr *nlh;
1588 	struct rtmsg *rtm;
1589 
1590 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1591 	if (!nlh)
1592 		return -EMSGSIZE;
1593 
1594 	rtm = nlmsg_data(nlh);
1595 	rtm->rtm_family = AF_INET;
1596 	rtm->rtm_dst_len = dst_len;
1597 	rtm->rtm_src_len = 0;
1598 	rtm->rtm_tos = tos;
1599 	if (tb_id < 256)
1600 		rtm->rtm_table = tb_id;
1601 	else
1602 		rtm->rtm_table = RT_TABLE_COMPAT;
1603 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1604 		goto nla_put_failure;
1605 	rtm->rtm_type = type;
1606 	rtm->rtm_flags = fi->fib_flags;
1607 	rtm->rtm_scope = fi->fib_scope;
1608 	rtm->rtm_protocol = fi->fib_protocol;
1609 
1610 	if (rtm->rtm_dst_len &&
1611 	    nla_put_in_addr(skb, RTA_DST, dst))
1612 		goto nla_put_failure;
1613 	if (fi->fib_priority &&
1614 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1615 		goto nla_put_failure;
1616 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1617 		goto nla_put_failure;
1618 
1619 	if (fi->fib_prefsrc &&
1620 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1621 		goto nla_put_failure;
1622 	if (fi->fib_nhs == 1) {
1623 		struct fib_nh *nh = &fi->fib_nh[0];
1624 		unsigned char flags = 0;
1625 
1626 		if (fib_nexthop_info(skb, &nh->nh_common, &flags, false) < 0)
1627 			goto nla_put_failure;
1628 
1629 		rtm->rtm_flags = flags;
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 	} else {
1636 		if (fib_add_multipath(skb, fi) < 0)
1637 			goto nla_put_failure;
1638 	}
1639 
1640 	nlmsg_end(skb, nlh);
1641 	return 0;
1642 
1643 nla_put_failure:
1644 	nlmsg_cancel(skb, nlh);
1645 	return -EMSGSIZE;
1646 }
1647 
1648 /*
1649  * Update FIB if:
1650  * - local address disappeared -> we must delete all the entries
1651  *   referring to it.
1652  * - device went down -> we must shutdown all nexthops going via it.
1653  */
1654 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1655 {
1656 	int ret = 0;
1657 	unsigned int hash = fib_laddr_hashfn(local);
1658 	struct hlist_head *head = &fib_info_laddrhash[hash];
1659 	struct net *net = dev_net(dev);
1660 	int tb_id = l3mdev_fib_table(dev);
1661 	struct fib_info *fi;
1662 
1663 	if (!fib_info_laddrhash || local == 0)
1664 		return 0;
1665 
1666 	hlist_for_each_entry(fi, head, fib_lhash) {
1667 		if (!net_eq(fi->fib_net, net) ||
1668 		    fi->fib_tb_id != tb_id)
1669 			continue;
1670 		if (fi->fib_prefsrc == local) {
1671 			fi->fib_flags |= RTNH_F_DEAD;
1672 			ret++;
1673 		}
1674 	}
1675 	return ret;
1676 }
1677 
1678 static int call_fib_nh_notifiers(struct fib_nh *nh,
1679 				 enum fib_event_type event_type)
1680 {
1681 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1682 	struct fib_nh_notifier_info info = {
1683 		.fib_nh = nh,
1684 	};
1685 
1686 	switch (event_type) {
1687 	case FIB_EVENT_NH_ADD:
1688 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1689 			break;
1690 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1691 			break;
1692 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1693 					   &info.info);
1694 	case FIB_EVENT_NH_DEL:
1695 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1696 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1697 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1698 						   event_type, &info.info);
1699 	default:
1700 		break;
1701 	}
1702 
1703 	return NOTIFY_DONE;
1704 }
1705 
1706 /* Update the PMTU of exceptions when:
1707  * - the new MTU of the first hop becomes smaller than the PMTU
1708  * - the old MTU was the same as the PMTU, and it limited discovery of
1709  *   larger MTUs on the path. With that limit raised, we can now
1710  *   discover larger MTUs
1711  * A special case is locked exceptions, for which the PMTU is smaller
1712  * than the minimal accepted PMTU:
1713  * - if the new MTU is greater than the PMTU, don't make any change
1714  * - otherwise, unlock and set PMTU
1715  */
1716 static void nh_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1717 {
1718 	struct fnhe_hash_bucket *bucket;
1719 	int i;
1720 
1721 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1722 	if (!bucket)
1723 		return;
1724 
1725 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1726 		struct fib_nh_exception *fnhe;
1727 
1728 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1729 		     fnhe;
1730 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1731 			if (fnhe->fnhe_mtu_locked) {
1732 				if (new <= fnhe->fnhe_pmtu) {
1733 					fnhe->fnhe_pmtu = new;
1734 					fnhe->fnhe_mtu_locked = false;
1735 				}
1736 			} else if (new < fnhe->fnhe_pmtu ||
1737 				   orig == fnhe->fnhe_pmtu) {
1738 				fnhe->fnhe_pmtu = new;
1739 			}
1740 		}
1741 	}
1742 }
1743 
1744 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1745 {
1746 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1747 	struct hlist_head *head = &fib_info_devhash[hash];
1748 	struct fib_nh *nh;
1749 
1750 	hlist_for_each_entry(nh, head, nh_hash) {
1751 		if (nh->fib_nh_dev == dev)
1752 			nh_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1753 	}
1754 }
1755 
1756 /* Event              force Flags           Description
1757  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1758  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1759  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1760  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1761  */
1762 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1763 {
1764 	int ret = 0;
1765 	int scope = RT_SCOPE_NOWHERE;
1766 	struct fib_info *prev_fi = NULL;
1767 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1768 	struct hlist_head *head = &fib_info_devhash[hash];
1769 	struct fib_nh *nh;
1770 
1771 	if (force)
1772 		scope = -1;
1773 
1774 	hlist_for_each_entry(nh, head, nh_hash) {
1775 		struct fib_info *fi = nh->nh_parent;
1776 		int dead;
1777 
1778 		BUG_ON(!fi->fib_nhs);
1779 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1780 			continue;
1781 		prev_fi = fi;
1782 		dead = 0;
1783 		change_nexthops(fi) {
1784 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1785 				dead++;
1786 			else if (nexthop_nh->fib_nh_dev == dev &&
1787 				 nexthop_nh->fib_nh_scope != scope) {
1788 				switch (event) {
1789 				case NETDEV_DOWN:
1790 				case NETDEV_UNREGISTER:
1791 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1792 					/* fall through */
1793 				case NETDEV_CHANGE:
1794 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1795 					break;
1796 				}
1797 				call_fib_nh_notifiers(nexthop_nh,
1798 						      FIB_EVENT_NH_DEL);
1799 				dead++;
1800 			}
1801 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1802 			if (event == NETDEV_UNREGISTER &&
1803 			    nexthop_nh->fib_nh_dev == dev) {
1804 				dead = fi->fib_nhs;
1805 				break;
1806 			}
1807 #endif
1808 		} endfor_nexthops(fi)
1809 		if (dead == fi->fib_nhs) {
1810 			switch (event) {
1811 			case NETDEV_DOWN:
1812 			case NETDEV_UNREGISTER:
1813 				fi->fib_flags |= RTNH_F_DEAD;
1814 				/* fall through */
1815 			case NETDEV_CHANGE:
1816 				fi->fib_flags |= RTNH_F_LINKDOWN;
1817 				break;
1818 			}
1819 			ret++;
1820 		}
1821 
1822 		fib_rebalance(fi);
1823 	}
1824 
1825 	return ret;
1826 }
1827 
1828 /* Must be invoked inside of an RCU protected region.  */
1829 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1830 {
1831 	struct fib_info *fi = NULL, *last_resort = NULL;
1832 	struct hlist_head *fa_head = res->fa_head;
1833 	struct fib_table *tb = res->table;
1834 	u8 slen = 32 - res->prefixlen;
1835 	int order = -1, last_idx = -1;
1836 	struct fib_alias *fa, *fa1 = NULL;
1837 	u32 last_prio = res->fi->fib_priority;
1838 	u8 last_tos = 0;
1839 
1840 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1841 		struct fib_info *next_fi = fa->fa_info;
1842 
1843 		if (fa->fa_slen != slen)
1844 			continue;
1845 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1846 			continue;
1847 		if (fa->tb_id != tb->tb_id)
1848 			continue;
1849 		if (next_fi->fib_priority > last_prio &&
1850 		    fa->fa_tos == last_tos) {
1851 			if (last_tos)
1852 				continue;
1853 			break;
1854 		}
1855 		if (next_fi->fib_flags & RTNH_F_DEAD)
1856 			continue;
1857 		last_tos = fa->fa_tos;
1858 		last_prio = next_fi->fib_priority;
1859 
1860 		if (next_fi->fib_scope != res->scope ||
1861 		    fa->fa_type != RTN_UNICAST)
1862 			continue;
1863 		if (!next_fi->fib_nh[0].fib_nh_gw4 ||
1864 		    next_fi->fib_nh[0].fib_nh_scope != RT_SCOPE_LINK)
1865 			continue;
1866 
1867 		fib_alias_accessed(fa);
1868 
1869 		if (!fi) {
1870 			if (next_fi != res->fi)
1871 				break;
1872 			fa1 = fa;
1873 		} else if (!fib_detect_death(fi, order, &last_resort,
1874 					     &last_idx, fa1->fa_default)) {
1875 			fib_result_assign(res, fi);
1876 			fa1->fa_default = order;
1877 			goto out;
1878 		}
1879 		fi = next_fi;
1880 		order++;
1881 	}
1882 
1883 	if (order <= 0 || !fi) {
1884 		if (fa1)
1885 			fa1->fa_default = -1;
1886 		goto out;
1887 	}
1888 
1889 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1890 			      fa1->fa_default)) {
1891 		fib_result_assign(res, fi);
1892 		fa1->fa_default = order;
1893 		goto out;
1894 	}
1895 
1896 	if (last_idx >= 0)
1897 		fib_result_assign(res, last_resort);
1898 	fa1->fa_default = last_idx;
1899 out:
1900 	return;
1901 }
1902 
1903 /*
1904  * Dead device goes up. We wake up dead nexthops.
1905  * It takes sense only on multipath routes.
1906  */
1907 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
1908 {
1909 	struct fib_info *prev_fi;
1910 	unsigned int hash;
1911 	struct hlist_head *head;
1912 	struct fib_nh *nh;
1913 	int ret;
1914 
1915 	if (!(dev->flags & IFF_UP))
1916 		return 0;
1917 
1918 	if (nh_flags & RTNH_F_DEAD) {
1919 		unsigned int flags = dev_get_flags(dev);
1920 
1921 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1922 			nh_flags |= RTNH_F_LINKDOWN;
1923 	}
1924 
1925 	prev_fi = NULL;
1926 	hash = fib_devindex_hashfn(dev->ifindex);
1927 	head = &fib_info_devhash[hash];
1928 	ret = 0;
1929 
1930 	hlist_for_each_entry(nh, head, nh_hash) {
1931 		struct fib_info *fi = nh->nh_parent;
1932 		int alive;
1933 
1934 		BUG_ON(!fi->fib_nhs);
1935 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1936 			continue;
1937 
1938 		prev_fi = fi;
1939 		alive = 0;
1940 		change_nexthops(fi) {
1941 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
1942 				alive++;
1943 				continue;
1944 			}
1945 			if (!nexthop_nh->fib_nh_dev ||
1946 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
1947 				continue;
1948 			if (nexthop_nh->fib_nh_dev != dev ||
1949 			    !__in_dev_get_rtnl(dev))
1950 				continue;
1951 			alive++;
1952 			nexthop_nh->fib_nh_flags &= ~nh_flags;
1953 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1954 		} endfor_nexthops(fi)
1955 
1956 		if (alive > 0) {
1957 			fi->fib_flags &= ~nh_flags;
1958 			ret++;
1959 		}
1960 
1961 		fib_rebalance(fi);
1962 	}
1963 
1964 	return ret;
1965 }
1966 
1967 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1968 static bool fib_good_nh(const struct fib_nh *nh)
1969 {
1970 	int state = NUD_REACHABLE;
1971 
1972 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
1973 		struct neighbour *n;
1974 
1975 		rcu_read_lock_bh();
1976 
1977 		if (likely(nh->fib_nh_gw_family == AF_INET))
1978 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
1979 						   (__force u32)nh->fib_nh_gw4);
1980 		else if (nh->fib_nh_gw_family == AF_INET6)
1981 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
1982 							   &nh->fib_nh_gw6);
1983 		else
1984 			n = NULL;
1985 		if (n)
1986 			state = n->nud_state;
1987 
1988 		rcu_read_unlock_bh();
1989 	}
1990 
1991 	return !!(state & NUD_VALID);
1992 }
1993 
1994 void fib_select_multipath(struct fib_result *res, int hash)
1995 {
1996 	struct fib_info *fi = res->fi;
1997 	struct net *net = fi->fib_net;
1998 	bool first = false;
1999 
2000 	change_nexthops(fi) {
2001 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2002 			if (!fib_good_nh(nexthop_nh))
2003 				continue;
2004 			if (!first) {
2005 				res->nh_sel = nhsel;
2006 				res->nhc = &nexthop_nh->nh_common;
2007 				first = true;
2008 			}
2009 		}
2010 
2011 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2012 			continue;
2013 
2014 		res->nh_sel = nhsel;
2015 		res->nhc = &nexthop_nh->nh_common;
2016 		return;
2017 	} endfor_nexthops(fi);
2018 }
2019 #endif
2020 
2021 void fib_select_path(struct net *net, struct fib_result *res,
2022 		     struct flowi4 *fl4, const struct sk_buff *skb)
2023 {
2024 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2025 		goto check_saddr;
2026 
2027 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2028 	if (res->fi->fib_nhs > 1) {
2029 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2030 
2031 		fib_select_multipath(res, h);
2032 	}
2033 	else
2034 #endif
2035 	if (!res->prefixlen &&
2036 	    res->table->tb_num_default > 1 &&
2037 	    res->type == RTN_UNICAST)
2038 		fib_select_default(fl4, res);
2039 
2040 check_saddr:
2041 	if (!fl4->saddr)
2042 		fl4->saddr = fib_result_prefsrc(net, res);
2043 }
2044