xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision d236d361)
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 
36 #include <net/arp.h>
37 #include <net/ip.h>
38 #include <net/protocol.h>
39 #include <net/route.h>
40 #include <net/tcp.h>
41 #include <net/sock.h>
42 #include <net/ip_fib.h>
43 #include <net/netlink.h>
44 #include <net/nexthop.h>
45 #include <net/lwtunnel.h>
46 
47 #include "fib_lookup.h"
48 
49 static DEFINE_SPINLOCK(fib_info_lock);
50 static struct hlist_head *fib_info_hash;
51 static struct hlist_head *fib_info_laddrhash;
52 static unsigned int fib_info_hash_size;
53 static unsigned int fib_info_cnt;
54 
55 #define DEVINDEX_HASHBITS 8
56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
58 
59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
60 
61 #define for_nexthops(fi) {						\
62 	int nhsel; const struct fib_nh *nh;				\
63 	for (nhsel = 0, nh = (fi)->fib_nh;				\
64 	     nhsel < (fi)->fib_nhs;					\
65 	     nh++, nhsel++)
66 
67 #define change_nexthops(fi) {						\
68 	int nhsel; struct fib_nh *nexthop_nh;				\
69 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
70 	     nhsel < (fi)->fib_nhs;					\
71 	     nexthop_nh++, nhsel++)
72 
73 #else /* CONFIG_IP_ROUTE_MULTIPATH */
74 
75 /* Hope, that gcc will optimize it to get rid of dummy loop */
76 
77 #define for_nexthops(fi) {						\
78 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
79 	for (nhsel = 0; nhsel < 1; nhsel++)
80 
81 #define change_nexthops(fi) {						\
82 	int nhsel;							\
83 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
84 	for (nhsel = 0; nhsel < 1; nhsel++)
85 
86 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
87 
88 #define endfor_nexthops(fi) }
89 
90 
91 const struct fib_prop fib_props[RTN_MAX + 1] = {
92 	[RTN_UNSPEC] = {
93 		.error	= 0,
94 		.scope	= RT_SCOPE_NOWHERE,
95 	},
96 	[RTN_UNICAST] = {
97 		.error	= 0,
98 		.scope	= RT_SCOPE_UNIVERSE,
99 	},
100 	[RTN_LOCAL] = {
101 		.error	= 0,
102 		.scope	= RT_SCOPE_HOST,
103 	},
104 	[RTN_BROADCAST] = {
105 		.error	= 0,
106 		.scope	= RT_SCOPE_LINK,
107 	},
108 	[RTN_ANYCAST] = {
109 		.error	= 0,
110 		.scope	= RT_SCOPE_LINK,
111 	},
112 	[RTN_MULTICAST] = {
113 		.error	= 0,
114 		.scope	= RT_SCOPE_UNIVERSE,
115 	},
116 	[RTN_BLACKHOLE] = {
117 		.error	= -EINVAL,
118 		.scope	= RT_SCOPE_UNIVERSE,
119 	},
120 	[RTN_UNREACHABLE] = {
121 		.error	= -EHOSTUNREACH,
122 		.scope	= RT_SCOPE_UNIVERSE,
123 	},
124 	[RTN_PROHIBIT] = {
125 		.error	= -EACCES,
126 		.scope	= RT_SCOPE_UNIVERSE,
127 	},
128 	[RTN_THROW] = {
129 		.error	= -EAGAIN,
130 		.scope	= RT_SCOPE_UNIVERSE,
131 	},
132 	[RTN_NAT] = {
133 		.error	= -EINVAL,
134 		.scope	= RT_SCOPE_NOWHERE,
135 	},
136 	[RTN_XRESOLVE] = {
137 		.error	= -EINVAL,
138 		.scope	= RT_SCOPE_NOWHERE,
139 	},
140 };
141 
142 static void rt_fibinfo_free(struct rtable __rcu **rtp)
143 {
144 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
145 
146 	if (!rt)
147 		return;
148 
149 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
150 	 * because we waited an RCU grace period before calling
151 	 * free_fib_info_rcu()
152 	 */
153 
154 	dst_free(&rt->dst);
155 }
156 
157 static void free_nh_exceptions(struct fib_nh *nh)
158 {
159 	struct fnhe_hash_bucket *hash;
160 	int i;
161 
162 	hash = rcu_dereference_protected(nh->nh_exceptions, 1);
163 	if (!hash)
164 		return;
165 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
166 		struct fib_nh_exception *fnhe;
167 
168 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
169 		while (fnhe) {
170 			struct fib_nh_exception *next;
171 
172 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
173 
174 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
175 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
176 
177 			kfree(fnhe);
178 
179 			fnhe = next;
180 		}
181 	}
182 	kfree(hash);
183 }
184 
185 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
186 {
187 	int cpu;
188 
189 	if (!rtp)
190 		return;
191 
192 	for_each_possible_cpu(cpu) {
193 		struct rtable *rt;
194 
195 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
196 		if (rt)
197 			dst_free(&rt->dst);
198 	}
199 	free_percpu(rtp);
200 }
201 
202 /* Release a nexthop info record */
203 static void free_fib_info_rcu(struct rcu_head *head)
204 {
205 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
206 
207 	change_nexthops(fi) {
208 		if (nexthop_nh->nh_dev)
209 			dev_put(nexthop_nh->nh_dev);
210 		lwtstate_put(nexthop_nh->nh_lwtstate);
211 		free_nh_exceptions(nexthop_nh);
212 		rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
213 		rt_fibinfo_free(&nexthop_nh->nh_rth_input);
214 	} endfor_nexthops(fi);
215 
216 	if (fi->fib_metrics != (u32 *) dst_default_metrics)
217 		kfree(fi->fib_metrics);
218 	kfree(fi);
219 }
220 
221 void free_fib_info(struct fib_info *fi)
222 {
223 	if (fi->fib_dead == 0) {
224 		pr_warn("Freeing alive fib_info %p\n", fi);
225 		return;
226 	}
227 	fib_info_cnt--;
228 #ifdef CONFIG_IP_ROUTE_CLASSID
229 	change_nexthops(fi) {
230 		if (nexthop_nh->nh_tclassid)
231 			fi->fib_net->ipv4.fib_num_tclassid_users--;
232 	} endfor_nexthops(fi);
233 #endif
234 	call_rcu(&fi->rcu, free_fib_info_rcu);
235 }
236 EXPORT_SYMBOL_GPL(free_fib_info);
237 
238 void fib_release_info(struct fib_info *fi)
239 {
240 	spin_lock_bh(&fib_info_lock);
241 	if (fi && --fi->fib_treeref == 0) {
242 		hlist_del(&fi->fib_hash);
243 		if (fi->fib_prefsrc)
244 			hlist_del(&fi->fib_lhash);
245 		change_nexthops(fi) {
246 			if (!nexthop_nh->nh_dev)
247 				continue;
248 			hlist_del(&nexthop_nh->nh_hash);
249 		} endfor_nexthops(fi)
250 		fi->fib_dead = 1;
251 		fib_info_put(fi);
252 	}
253 	spin_unlock_bh(&fib_info_lock);
254 }
255 
256 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
257 {
258 	const struct fib_nh *onh = ofi->fib_nh;
259 
260 	for_nexthops(fi) {
261 		if (nh->nh_oif != onh->nh_oif ||
262 		    nh->nh_gw  != onh->nh_gw ||
263 		    nh->nh_scope != onh->nh_scope ||
264 #ifdef CONFIG_IP_ROUTE_MULTIPATH
265 		    nh->nh_weight != onh->nh_weight ||
266 #endif
267 #ifdef CONFIG_IP_ROUTE_CLASSID
268 		    nh->nh_tclassid != onh->nh_tclassid ||
269 #endif
270 		    lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
271 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
272 			return -1;
273 		onh++;
274 	} endfor_nexthops(fi);
275 	return 0;
276 }
277 
278 static inline unsigned int fib_devindex_hashfn(unsigned int val)
279 {
280 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
281 
282 	return (val ^
283 		(val >> DEVINDEX_HASHBITS) ^
284 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
285 }
286 
287 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
288 {
289 	unsigned int mask = (fib_info_hash_size - 1);
290 	unsigned int val = fi->fib_nhs;
291 
292 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
293 	val ^= (__force u32)fi->fib_prefsrc;
294 	val ^= fi->fib_priority;
295 	for_nexthops(fi) {
296 		val ^= fib_devindex_hashfn(nh->nh_oif);
297 	} endfor_nexthops(fi)
298 
299 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
300 }
301 
302 static struct fib_info *fib_find_info(const struct fib_info *nfi)
303 {
304 	struct hlist_head *head;
305 	struct fib_info *fi;
306 	unsigned int hash;
307 
308 	hash = fib_info_hashfn(nfi);
309 	head = &fib_info_hash[hash];
310 
311 	hlist_for_each_entry(fi, head, fib_hash) {
312 		if (!net_eq(fi->fib_net, nfi->fib_net))
313 			continue;
314 		if (fi->fib_nhs != nfi->fib_nhs)
315 			continue;
316 		if (nfi->fib_protocol == fi->fib_protocol &&
317 		    nfi->fib_scope == fi->fib_scope &&
318 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
319 		    nfi->fib_priority == fi->fib_priority &&
320 		    nfi->fib_type == fi->fib_type &&
321 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
322 			   sizeof(u32) * RTAX_MAX) == 0 &&
323 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
324 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
325 			return fi;
326 	}
327 
328 	return NULL;
329 }
330 
331 /* Check, that the gateway is already configured.
332  * Used only by redirect accept routine.
333  */
334 int ip_fib_check_default(__be32 gw, struct net_device *dev)
335 {
336 	struct hlist_head *head;
337 	struct fib_nh *nh;
338 	unsigned int hash;
339 
340 	spin_lock(&fib_info_lock);
341 
342 	hash = fib_devindex_hashfn(dev->ifindex);
343 	head = &fib_info_devhash[hash];
344 	hlist_for_each_entry(nh, head, nh_hash) {
345 		if (nh->nh_dev == dev &&
346 		    nh->nh_gw == gw &&
347 		    !(nh->nh_flags & RTNH_F_DEAD)) {
348 			spin_unlock(&fib_info_lock);
349 			return 0;
350 		}
351 	}
352 
353 	spin_unlock(&fib_info_lock);
354 
355 	return -1;
356 }
357 
358 static inline size_t fib_nlmsg_size(struct fib_info *fi)
359 {
360 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
361 			 + nla_total_size(4) /* RTA_TABLE */
362 			 + nla_total_size(4) /* RTA_DST */
363 			 + nla_total_size(4) /* RTA_PRIORITY */
364 			 + nla_total_size(4) /* RTA_PREFSRC */
365 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
366 
367 	/* space for nested metrics */
368 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
369 
370 	if (fi->fib_nhs) {
371 		size_t nh_encapsize = 0;
372 		/* Also handles the special case fib_nhs == 1 */
373 
374 		/* each nexthop is packed in an attribute */
375 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
376 
377 		/* may contain flow and gateway attribute */
378 		nhsize += 2 * nla_total_size(4);
379 
380 		/* grab encap info */
381 		for_nexthops(fi) {
382 			if (nh->nh_lwtstate) {
383 				/* RTA_ENCAP_TYPE */
384 				nh_encapsize += lwtunnel_get_encap_size(
385 						nh->nh_lwtstate);
386 				/* RTA_ENCAP */
387 				nh_encapsize +=  nla_total_size(2);
388 			}
389 		} endfor_nexthops(fi);
390 
391 		/* all nexthops are packed in a nested attribute */
392 		payload += nla_total_size((fi->fib_nhs * nhsize) +
393 					  nh_encapsize);
394 
395 	}
396 
397 	return payload;
398 }
399 
400 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
401 	       int dst_len, u32 tb_id, const struct nl_info *info,
402 	       unsigned int nlm_flags)
403 {
404 	struct sk_buff *skb;
405 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
406 	int err = -ENOBUFS;
407 
408 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
409 	if (!skb)
410 		goto errout;
411 
412 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
413 			    fa->fa_type, key, dst_len,
414 			    fa->fa_tos, fa->fa_info, nlm_flags);
415 	if (err < 0) {
416 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
417 		WARN_ON(err == -EMSGSIZE);
418 		kfree_skb(skb);
419 		goto errout;
420 	}
421 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
422 		    info->nlh, GFP_KERNEL);
423 	return;
424 errout:
425 	if (err < 0)
426 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
427 }
428 
429 static int fib_detect_death(struct fib_info *fi, int order,
430 			    struct fib_info **last_resort, int *last_idx,
431 			    int dflt)
432 {
433 	struct neighbour *n;
434 	int state = NUD_NONE;
435 
436 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
437 	if (n) {
438 		state = n->nud_state;
439 		neigh_release(n);
440 	} else {
441 		return 0;
442 	}
443 	if (state == NUD_REACHABLE)
444 		return 0;
445 	if ((state & NUD_VALID) && order != dflt)
446 		return 0;
447 	if ((state & NUD_VALID) ||
448 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
449 		*last_resort = fi;
450 		*last_idx = order;
451 	}
452 	return 1;
453 }
454 
455 #ifdef CONFIG_IP_ROUTE_MULTIPATH
456 
457 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
458 {
459 	int nhs = 0;
460 
461 	while (rtnh_ok(rtnh, remaining)) {
462 		nhs++;
463 		rtnh = rtnh_next(rtnh, &remaining);
464 	}
465 
466 	/* leftover implies invalid nexthop configuration, discard it */
467 	return remaining > 0 ? 0 : nhs;
468 }
469 
470 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
471 		       int remaining, struct fib_config *cfg)
472 {
473 	int ret;
474 
475 	change_nexthops(fi) {
476 		int attrlen;
477 
478 		if (!rtnh_ok(rtnh, remaining))
479 			return -EINVAL;
480 
481 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
482 			return -EINVAL;
483 
484 		nexthop_nh->nh_flags =
485 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
486 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
487 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
488 
489 		attrlen = rtnh_attrlen(rtnh);
490 		if (attrlen > 0) {
491 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
492 
493 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
494 			nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
495 #ifdef CONFIG_IP_ROUTE_CLASSID
496 			nla = nla_find(attrs, attrlen, RTA_FLOW);
497 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
498 			if (nexthop_nh->nh_tclassid)
499 				fi->fib_net->ipv4.fib_num_tclassid_users++;
500 #endif
501 			nla = nla_find(attrs, attrlen, RTA_ENCAP);
502 			if (nla) {
503 				struct lwtunnel_state *lwtstate;
504 				struct nlattr *nla_entype;
505 
506 				nla_entype = nla_find(attrs, attrlen,
507 						      RTA_ENCAP_TYPE);
508 				if (!nla_entype)
509 					goto err_inval;
510 
511 				ret = lwtunnel_build_state(nla_get_u16(
512 							   nla_entype),
513 							   nla,  AF_INET, cfg,
514 							   &lwtstate);
515 				if (ret)
516 					goto errout;
517 				nexthop_nh->nh_lwtstate =
518 					lwtstate_get(lwtstate);
519 			}
520 		}
521 
522 		rtnh = rtnh_next(rtnh, &remaining);
523 	} endfor_nexthops(fi);
524 
525 	return 0;
526 
527 err_inval:
528 	ret = -EINVAL;
529 
530 errout:
531 	return ret;
532 }
533 
534 static void fib_rebalance(struct fib_info *fi)
535 {
536 	int total;
537 	int w;
538 	struct in_device *in_dev;
539 
540 	if (fi->fib_nhs < 2)
541 		return;
542 
543 	total = 0;
544 	for_nexthops(fi) {
545 		if (nh->nh_flags & RTNH_F_DEAD)
546 			continue;
547 
548 		in_dev = __in_dev_get_rtnl(nh->nh_dev);
549 
550 		if (in_dev &&
551 		    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
552 		    nh->nh_flags & RTNH_F_LINKDOWN)
553 			continue;
554 
555 		total += nh->nh_weight;
556 	} endfor_nexthops(fi);
557 
558 	w = 0;
559 	change_nexthops(fi) {
560 		int upper_bound;
561 
562 		in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
563 
564 		if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
565 			upper_bound = -1;
566 		} else if (in_dev &&
567 			   IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
568 			   nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
569 			upper_bound = -1;
570 		} else {
571 			w += nexthop_nh->nh_weight;
572 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
573 							    total) - 1;
574 		}
575 
576 		atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
577 	} endfor_nexthops(fi);
578 }
579 
580 static inline void fib_add_weight(struct fib_info *fi,
581 				  const struct fib_nh *nh)
582 {
583 	fi->fib_weight += nh->nh_weight;
584 }
585 
586 #else /* CONFIG_IP_ROUTE_MULTIPATH */
587 
588 #define fib_rebalance(fi) do { } while (0)
589 #define fib_add_weight(fi, nh) do { } while (0)
590 
591 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
592 
593 static int fib_encap_match(u16 encap_type,
594 			   struct nlattr *encap,
595 			   const struct fib_nh *nh,
596 			   const struct fib_config *cfg)
597 {
598 	struct lwtunnel_state *lwtstate;
599 	int ret, result = 0;
600 
601 	if (encap_type == LWTUNNEL_ENCAP_NONE)
602 		return 0;
603 
604 	ret = lwtunnel_build_state(encap_type, encap,
605 				   AF_INET, cfg, &lwtstate);
606 	if (!ret) {
607 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
608 		lwtstate_free(lwtstate);
609 	}
610 
611 	return result;
612 }
613 
614 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
615 {
616 #ifdef CONFIG_IP_ROUTE_MULTIPATH
617 	struct rtnexthop *rtnh;
618 	int remaining;
619 #endif
620 
621 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
622 		return 1;
623 
624 	if (cfg->fc_oif || cfg->fc_gw) {
625 		if (cfg->fc_encap) {
626 			if (fib_encap_match(cfg->fc_encap_type,
627 					    cfg->fc_encap, fi->fib_nh, cfg))
628 			    return 1;
629 		}
630 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
631 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
632 			return 0;
633 		return 1;
634 	}
635 
636 #ifdef CONFIG_IP_ROUTE_MULTIPATH
637 	if (!cfg->fc_mp)
638 		return 0;
639 
640 	rtnh = cfg->fc_mp;
641 	remaining = cfg->fc_mp_len;
642 
643 	for_nexthops(fi) {
644 		int attrlen;
645 
646 		if (!rtnh_ok(rtnh, remaining))
647 			return -EINVAL;
648 
649 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
650 			return 1;
651 
652 		attrlen = rtnh_attrlen(rtnh);
653 		if (attrlen > 0) {
654 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
655 
656 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
657 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
658 				return 1;
659 #ifdef CONFIG_IP_ROUTE_CLASSID
660 			nla = nla_find(attrs, attrlen, RTA_FLOW);
661 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
662 				return 1;
663 #endif
664 		}
665 
666 		rtnh = rtnh_next(rtnh, &remaining);
667 	} endfor_nexthops(fi);
668 #endif
669 	return 0;
670 }
671 
672 
673 /*
674  * Picture
675  * -------
676  *
677  * Semantics of nexthop is very messy by historical reasons.
678  * We have to take into account, that:
679  * a) gateway can be actually local interface address,
680  *    so that gatewayed route is direct.
681  * b) gateway must be on-link address, possibly
682  *    described not by an ifaddr, but also by a direct route.
683  * c) If both gateway and interface are specified, they should not
684  *    contradict.
685  * d) If we use tunnel routes, gateway could be not on-link.
686  *
687  * Attempt to reconcile all of these (alas, self-contradictory) conditions
688  * results in pretty ugly and hairy code with obscure logic.
689  *
690  * I chose to generalized it instead, so that the size
691  * of code does not increase practically, but it becomes
692  * much more general.
693  * Every prefix is assigned a "scope" value: "host" is local address,
694  * "link" is direct route,
695  * [ ... "site" ... "interior" ... ]
696  * and "universe" is true gateway route with global meaning.
697  *
698  * Every prefix refers to a set of "nexthop"s (gw, oif),
699  * where gw must have narrower scope. This recursion stops
700  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
701  * which means that gw is forced to be on link.
702  *
703  * Code is still hairy, but now it is apparently logically
704  * consistent and very flexible. F.e. as by-product it allows
705  * to co-exists in peace independent exterior and interior
706  * routing processes.
707  *
708  * Normally it looks as following.
709  *
710  * {universe prefix}  -> (gw, oif) [scope link]
711  *		  |
712  *		  |-> {link prefix} -> (gw, oif) [scope local]
713  *					|
714  *					|-> {local prefix} (terminal node)
715  */
716 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
717 			struct fib_nh *nh)
718 {
719 	int err = 0;
720 	struct net *net;
721 	struct net_device *dev;
722 
723 	net = cfg->fc_nlinfo.nl_net;
724 	if (nh->nh_gw) {
725 		struct fib_result res;
726 
727 		if (nh->nh_flags & RTNH_F_ONLINK) {
728 			unsigned int addr_type;
729 
730 			if (cfg->fc_scope >= RT_SCOPE_LINK)
731 				return -EINVAL;
732 			dev = __dev_get_by_index(net, nh->nh_oif);
733 			if (!dev)
734 				return -ENODEV;
735 			if (!(dev->flags & IFF_UP))
736 				return -ENETDOWN;
737 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
738 			if (addr_type != RTN_UNICAST)
739 				return -EINVAL;
740 			if (!netif_carrier_ok(dev))
741 				nh->nh_flags |= RTNH_F_LINKDOWN;
742 			nh->nh_dev = dev;
743 			dev_hold(dev);
744 			nh->nh_scope = RT_SCOPE_LINK;
745 			return 0;
746 		}
747 		rcu_read_lock();
748 		{
749 			struct fib_table *tbl = NULL;
750 			struct flowi4 fl4 = {
751 				.daddr = nh->nh_gw,
752 				.flowi4_scope = cfg->fc_scope + 1,
753 				.flowi4_oif = nh->nh_oif,
754 				.flowi4_iif = LOOPBACK_IFINDEX,
755 			};
756 
757 			/* It is not necessary, but requires a bit of thinking */
758 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
759 				fl4.flowi4_scope = RT_SCOPE_LINK;
760 
761 			if (cfg->fc_table)
762 				tbl = fib_get_table(net, cfg->fc_table);
763 
764 			if (tbl)
765 				err = fib_table_lookup(tbl, &fl4, &res,
766 						       FIB_LOOKUP_IGNORE_LINKSTATE |
767 						       FIB_LOOKUP_NOREF);
768 
769 			/* on error or if no table given do full lookup. This
770 			 * is needed for example when nexthops are in the local
771 			 * table rather than the given table
772 			 */
773 			if (!tbl || err) {
774 				err = fib_lookup(net, &fl4, &res,
775 						 FIB_LOOKUP_IGNORE_LINKSTATE);
776 			}
777 
778 			if (err) {
779 				rcu_read_unlock();
780 				return err;
781 			}
782 		}
783 		err = -EINVAL;
784 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
785 			goto out;
786 		nh->nh_scope = res.scope;
787 		nh->nh_oif = FIB_RES_OIF(res);
788 		nh->nh_dev = dev = FIB_RES_DEV(res);
789 		if (!dev)
790 			goto out;
791 		dev_hold(dev);
792 		if (!netif_carrier_ok(dev))
793 			nh->nh_flags |= RTNH_F_LINKDOWN;
794 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
795 	} else {
796 		struct in_device *in_dev;
797 
798 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
799 			return -EINVAL;
800 
801 		rcu_read_lock();
802 		err = -ENODEV;
803 		in_dev = inetdev_by_index(net, nh->nh_oif);
804 		if (!in_dev)
805 			goto out;
806 		err = -ENETDOWN;
807 		if (!(in_dev->dev->flags & IFF_UP))
808 			goto out;
809 		nh->nh_dev = in_dev->dev;
810 		dev_hold(nh->nh_dev);
811 		nh->nh_scope = RT_SCOPE_HOST;
812 		if (!netif_carrier_ok(nh->nh_dev))
813 			nh->nh_flags |= RTNH_F_LINKDOWN;
814 		err = 0;
815 	}
816 out:
817 	rcu_read_unlock();
818 	return err;
819 }
820 
821 static inline unsigned int fib_laddr_hashfn(__be32 val)
822 {
823 	unsigned int mask = (fib_info_hash_size - 1);
824 
825 	return ((__force u32)val ^
826 		((__force u32)val >> 7) ^
827 		((__force u32)val >> 14)) & mask;
828 }
829 
830 static struct hlist_head *fib_info_hash_alloc(int bytes)
831 {
832 	if (bytes <= PAGE_SIZE)
833 		return kzalloc(bytes, GFP_KERNEL);
834 	else
835 		return (struct hlist_head *)
836 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
837 					 get_order(bytes));
838 }
839 
840 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
841 {
842 	if (!hash)
843 		return;
844 
845 	if (bytes <= PAGE_SIZE)
846 		kfree(hash);
847 	else
848 		free_pages((unsigned long) hash, get_order(bytes));
849 }
850 
851 static void fib_info_hash_move(struct hlist_head *new_info_hash,
852 			       struct hlist_head *new_laddrhash,
853 			       unsigned int new_size)
854 {
855 	struct hlist_head *old_info_hash, *old_laddrhash;
856 	unsigned int old_size = fib_info_hash_size;
857 	unsigned int i, bytes;
858 
859 	spin_lock_bh(&fib_info_lock);
860 	old_info_hash = fib_info_hash;
861 	old_laddrhash = fib_info_laddrhash;
862 	fib_info_hash_size = new_size;
863 
864 	for (i = 0; i < old_size; i++) {
865 		struct hlist_head *head = &fib_info_hash[i];
866 		struct hlist_node *n;
867 		struct fib_info *fi;
868 
869 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
870 			struct hlist_head *dest;
871 			unsigned int new_hash;
872 
873 			new_hash = fib_info_hashfn(fi);
874 			dest = &new_info_hash[new_hash];
875 			hlist_add_head(&fi->fib_hash, dest);
876 		}
877 	}
878 	fib_info_hash = new_info_hash;
879 
880 	for (i = 0; i < old_size; i++) {
881 		struct hlist_head *lhead = &fib_info_laddrhash[i];
882 		struct hlist_node *n;
883 		struct fib_info *fi;
884 
885 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
886 			struct hlist_head *ldest;
887 			unsigned int new_hash;
888 
889 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
890 			ldest = &new_laddrhash[new_hash];
891 			hlist_add_head(&fi->fib_lhash, ldest);
892 		}
893 	}
894 	fib_info_laddrhash = new_laddrhash;
895 
896 	spin_unlock_bh(&fib_info_lock);
897 
898 	bytes = old_size * sizeof(struct hlist_head *);
899 	fib_info_hash_free(old_info_hash, bytes);
900 	fib_info_hash_free(old_laddrhash, bytes);
901 }
902 
903 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
904 {
905 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
906 					nh->nh_gw,
907 					nh->nh_parent->fib_scope);
908 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
909 
910 	return nh->nh_saddr;
911 }
912 
913 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
914 {
915 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
916 	    fib_prefsrc != cfg->fc_dst) {
917 		u32 tb_id = cfg->fc_table;
918 		int rc;
919 
920 		if (tb_id == RT_TABLE_MAIN)
921 			tb_id = RT_TABLE_LOCAL;
922 
923 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
924 					  fib_prefsrc, tb_id);
925 
926 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
927 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
928 						  fib_prefsrc, RT_TABLE_LOCAL);
929 		}
930 
931 		if (rc != RTN_LOCAL)
932 			return false;
933 	}
934 	return true;
935 }
936 
937 static int
938 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
939 {
940 	bool ecn_ca = false;
941 	struct nlattr *nla;
942 	int remaining;
943 
944 	if (!cfg->fc_mx)
945 		return 0;
946 
947 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
948 		int type = nla_type(nla);
949 		u32 val;
950 
951 		if (!type)
952 			continue;
953 		if (type > RTAX_MAX)
954 			return -EINVAL;
955 
956 		if (type == RTAX_CC_ALGO) {
957 			char tmp[TCP_CA_NAME_MAX];
958 
959 			nla_strlcpy(tmp, nla, sizeof(tmp));
960 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
961 			if (val == TCP_CA_UNSPEC)
962 				return -EINVAL;
963 		} else {
964 			val = nla_get_u32(nla);
965 		}
966 		if (type == RTAX_ADVMSS && val > 65535 - 40)
967 			val = 65535 - 40;
968 		if (type == RTAX_MTU && val > 65535 - 15)
969 			val = 65535 - 15;
970 		if (type == RTAX_HOPLIMIT && val > 255)
971 			val = 255;
972 		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
973 			return -EINVAL;
974 		fi->fib_metrics[type - 1] = val;
975 	}
976 
977 	if (ecn_ca)
978 		fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
979 
980 	return 0;
981 }
982 
983 struct fib_info *fib_create_info(struct fib_config *cfg)
984 {
985 	int err;
986 	struct fib_info *fi = NULL;
987 	struct fib_info *ofi;
988 	int nhs = 1;
989 	struct net *net = cfg->fc_nlinfo.nl_net;
990 
991 	if (cfg->fc_type > RTN_MAX)
992 		goto err_inval;
993 
994 	/* Fast check to catch the most weird cases */
995 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
996 		goto err_inval;
997 
998 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
999 		goto err_inval;
1000 
1001 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1002 	if (cfg->fc_mp) {
1003 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1004 		if (nhs == 0)
1005 			goto err_inval;
1006 	}
1007 #endif
1008 
1009 	err = -ENOBUFS;
1010 	if (fib_info_cnt >= fib_info_hash_size) {
1011 		unsigned int new_size = fib_info_hash_size << 1;
1012 		struct hlist_head *new_info_hash;
1013 		struct hlist_head *new_laddrhash;
1014 		unsigned int bytes;
1015 
1016 		if (!new_size)
1017 			new_size = 16;
1018 		bytes = new_size * sizeof(struct hlist_head *);
1019 		new_info_hash = fib_info_hash_alloc(bytes);
1020 		new_laddrhash = fib_info_hash_alloc(bytes);
1021 		if (!new_info_hash || !new_laddrhash) {
1022 			fib_info_hash_free(new_info_hash, bytes);
1023 			fib_info_hash_free(new_laddrhash, bytes);
1024 		} else
1025 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1026 
1027 		if (!fib_info_hash_size)
1028 			goto failure;
1029 	}
1030 
1031 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1032 	if (!fi)
1033 		goto failure;
1034 	fib_info_cnt++;
1035 	if (cfg->fc_mx) {
1036 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1037 		if (!fi->fib_metrics)
1038 			goto failure;
1039 	} else
1040 		fi->fib_metrics = (u32 *) dst_default_metrics;
1041 
1042 	fi->fib_net = net;
1043 	fi->fib_protocol = cfg->fc_protocol;
1044 	fi->fib_scope = cfg->fc_scope;
1045 	fi->fib_flags = cfg->fc_flags;
1046 	fi->fib_priority = cfg->fc_priority;
1047 	fi->fib_prefsrc = cfg->fc_prefsrc;
1048 	fi->fib_type = cfg->fc_type;
1049 	fi->fib_tb_id = cfg->fc_table;
1050 
1051 	fi->fib_nhs = nhs;
1052 	change_nexthops(fi) {
1053 		nexthop_nh->nh_parent = fi;
1054 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1055 		if (!nexthop_nh->nh_pcpu_rth_output)
1056 			goto failure;
1057 	} endfor_nexthops(fi)
1058 
1059 	err = fib_convert_metrics(fi, cfg);
1060 	if (err)
1061 		goto failure;
1062 
1063 	if (cfg->fc_mp) {
1064 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1065 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1066 		if (err != 0)
1067 			goto failure;
1068 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1069 			goto err_inval;
1070 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1071 			goto err_inval;
1072 #ifdef CONFIG_IP_ROUTE_CLASSID
1073 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1074 			goto err_inval;
1075 #endif
1076 #else
1077 		goto err_inval;
1078 #endif
1079 	} else {
1080 		struct fib_nh *nh = fi->fib_nh;
1081 
1082 		if (cfg->fc_encap) {
1083 			struct lwtunnel_state *lwtstate;
1084 
1085 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1086 				goto err_inval;
1087 			err = lwtunnel_build_state(cfg->fc_encap_type,
1088 						   cfg->fc_encap, AF_INET, cfg,
1089 						   &lwtstate);
1090 			if (err)
1091 				goto failure;
1092 
1093 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1094 		}
1095 		nh->nh_oif = cfg->fc_oif;
1096 		nh->nh_gw = cfg->fc_gw;
1097 		nh->nh_flags = cfg->fc_flags;
1098 #ifdef CONFIG_IP_ROUTE_CLASSID
1099 		nh->nh_tclassid = cfg->fc_flow;
1100 		if (nh->nh_tclassid)
1101 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1102 #endif
1103 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1104 		nh->nh_weight = 1;
1105 #endif
1106 	}
1107 
1108 	if (fib_props[cfg->fc_type].error) {
1109 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1110 			goto err_inval;
1111 		goto link_it;
1112 	} else {
1113 		switch (cfg->fc_type) {
1114 		case RTN_UNICAST:
1115 		case RTN_LOCAL:
1116 		case RTN_BROADCAST:
1117 		case RTN_ANYCAST:
1118 		case RTN_MULTICAST:
1119 			break;
1120 		default:
1121 			goto err_inval;
1122 		}
1123 	}
1124 
1125 	if (cfg->fc_scope > RT_SCOPE_HOST)
1126 		goto err_inval;
1127 
1128 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1129 		struct fib_nh *nh = fi->fib_nh;
1130 
1131 		/* Local address is added. */
1132 		if (nhs != 1 || nh->nh_gw)
1133 			goto err_inval;
1134 		nh->nh_scope = RT_SCOPE_NOWHERE;
1135 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1136 		err = -ENODEV;
1137 		if (!nh->nh_dev)
1138 			goto failure;
1139 	} else {
1140 		int linkdown = 0;
1141 
1142 		change_nexthops(fi) {
1143 			err = fib_check_nh(cfg, fi, nexthop_nh);
1144 			if (err != 0)
1145 				goto failure;
1146 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1147 				linkdown++;
1148 		} endfor_nexthops(fi)
1149 		if (linkdown == fi->fib_nhs)
1150 			fi->fib_flags |= RTNH_F_LINKDOWN;
1151 	}
1152 
1153 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1154 		goto err_inval;
1155 
1156 	change_nexthops(fi) {
1157 		fib_info_update_nh_saddr(net, nexthop_nh);
1158 		fib_add_weight(fi, nexthop_nh);
1159 	} endfor_nexthops(fi)
1160 
1161 	fib_rebalance(fi);
1162 
1163 link_it:
1164 	ofi = fib_find_info(fi);
1165 	if (ofi) {
1166 		fi->fib_dead = 1;
1167 		free_fib_info(fi);
1168 		ofi->fib_treeref++;
1169 		return ofi;
1170 	}
1171 
1172 	fi->fib_treeref++;
1173 	atomic_inc(&fi->fib_clntref);
1174 	spin_lock_bh(&fib_info_lock);
1175 	hlist_add_head(&fi->fib_hash,
1176 		       &fib_info_hash[fib_info_hashfn(fi)]);
1177 	if (fi->fib_prefsrc) {
1178 		struct hlist_head *head;
1179 
1180 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1181 		hlist_add_head(&fi->fib_lhash, head);
1182 	}
1183 	change_nexthops(fi) {
1184 		struct hlist_head *head;
1185 		unsigned int hash;
1186 
1187 		if (!nexthop_nh->nh_dev)
1188 			continue;
1189 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1190 		head = &fib_info_devhash[hash];
1191 		hlist_add_head(&nexthop_nh->nh_hash, head);
1192 	} endfor_nexthops(fi)
1193 	spin_unlock_bh(&fib_info_lock);
1194 	return fi;
1195 
1196 err_inval:
1197 	err = -EINVAL;
1198 
1199 failure:
1200 	if (fi) {
1201 		fi->fib_dead = 1;
1202 		free_fib_info(fi);
1203 	}
1204 
1205 	return ERR_PTR(err);
1206 }
1207 
1208 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1209 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1210 		  struct fib_info *fi, unsigned int flags)
1211 {
1212 	struct nlmsghdr *nlh;
1213 	struct rtmsg *rtm;
1214 
1215 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1216 	if (!nlh)
1217 		return -EMSGSIZE;
1218 
1219 	rtm = nlmsg_data(nlh);
1220 	rtm->rtm_family = AF_INET;
1221 	rtm->rtm_dst_len = dst_len;
1222 	rtm->rtm_src_len = 0;
1223 	rtm->rtm_tos = tos;
1224 	if (tb_id < 256)
1225 		rtm->rtm_table = tb_id;
1226 	else
1227 		rtm->rtm_table = RT_TABLE_COMPAT;
1228 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1229 		goto nla_put_failure;
1230 	rtm->rtm_type = type;
1231 	rtm->rtm_flags = fi->fib_flags;
1232 	rtm->rtm_scope = fi->fib_scope;
1233 	rtm->rtm_protocol = fi->fib_protocol;
1234 
1235 	if (rtm->rtm_dst_len &&
1236 	    nla_put_in_addr(skb, RTA_DST, dst))
1237 		goto nla_put_failure;
1238 	if (fi->fib_priority &&
1239 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1240 		goto nla_put_failure;
1241 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1242 		goto nla_put_failure;
1243 
1244 	if (fi->fib_prefsrc &&
1245 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1246 		goto nla_put_failure;
1247 	if (fi->fib_nhs == 1) {
1248 		struct in_device *in_dev;
1249 
1250 		if (fi->fib_nh->nh_gw &&
1251 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1252 			goto nla_put_failure;
1253 		if (fi->fib_nh->nh_oif &&
1254 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1255 			goto nla_put_failure;
1256 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1257 			in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1258 			if (in_dev &&
1259 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1260 				rtm->rtm_flags |= RTNH_F_DEAD;
1261 		}
1262 #ifdef CONFIG_IP_ROUTE_CLASSID
1263 		if (fi->fib_nh[0].nh_tclassid &&
1264 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1265 			goto nla_put_failure;
1266 #endif
1267 		if (fi->fib_nh->nh_lwtstate &&
1268 		    lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1269 			goto nla_put_failure;
1270 	}
1271 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1272 	if (fi->fib_nhs > 1) {
1273 		struct rtnexthop *rtnh;
1274 		struct nlattr *mp;
1275 
1276 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1277 		if (!mp)
1278 			goto nla_put_failure;
1279 
1280 		for_nexthops(fi) {
1281 			struct in_device *in_dev;
1282 
1283 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1284 			if (!rtnh)
1285 				goto nla_put_failure;
1286 
1287 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1288 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1289 				in_dev = __in_dev_get_rtnl(nh->nh_dev);
1290 				if (in_dev &&
1291 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1292 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1293 			}
1294 			rtnh->rtnh_hops = nh->nh_weight - 1;
1295 			rtnh->rtnh_ifindex = nh->nh_oif;
1296 
1297 			if (nh->nh_gw &&
1298 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1299 				goto nla_put_failure;
1300 #ifdef CONFIG_IP_ROUTE_CLASSID
1301 			if (nh->nh_tclassid &&
1302 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1303 				goto nla_put_failure;
1304 #endif
1305 			if (nh->nh_lwtstate &&
1306 			    lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1307 				goto nla_put_failure;
1308 
1309 			/* length of rtnetlink header + attributes */
1310 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1311 		} endfor_nexthops(fi);
1312 
1313 		nla_nest_end(skb, mp);
1314 	}
1315 #endif
1316 	nlmsg_end(skb, nlh);
1317 	return 0;
1318 
1319 nla_put_failure:
1320 	nlmsg_cancel(skb, nlh);
1321 	return -EMSGSIZE;
1322 }
1323 
1324 /*
1325  * Update FIB if:
1326  * - local address disappeared -> we must delete all the entries
1327  *   referring to it.
1328  * - device went down -> we must shutdown all nexthops going via it.
1329  */
1330 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1331 {
1332 	int ret = 0;
1333 	unsigned int hash = fib_laddr_hashfn(local);
1334 	struct hlist_head *head = &fib_info_laddrhash[hash];
1335 	struct net *net = dev_net(dev);
1336 	int tb_id = l3mdev_fib_table(dev);
1337 	struct fib_info *fi;
1338 
1339 	if (!fib_info_laddrhash || local == 0)
1340 		return 0;
1341 
1342 	hlist_for_each_entry(fi, head, fib_lhash) {
1343 		if (!net_eq(fi->fib_net, net) ||
1344 		    fi->fib_tb_id != tb_id)
1345 			continue;
1346 		if (fi->fib_prefsrc == local) {
1347 			fi->fib_flags |= RTNH_F_DEAD;
1348 			ret++;
1349 		}
1350 	}
1351 	return ret;
1352 }
1353 
1354 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1355 				 enum fib_event_type event_type)
1356 {
1357 	struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1358 	struct fib_nh_notifier_info info = {
1359 		.fib_nh = fib_nh,
1360 	};
1361 
1362 	switch (event_type) {
1363 	case FIB_EVENT_NH_ADD:
1364 		if (fib_nh->nh_flags & RTNH_F_DEAD)
1365 			break;
1366 		if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1367 		    fib_nh->nh_flags & RTNH_F_LINKDOWN)
1368 			break;
1369 		return call_fib_notifiers(dev_net(fib_nh->nh_dev), event_type,
1370 					  &info.info);
1371 	case FIB_EVENT_NH_DEL:
1372 		if ((IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1373 		     fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1374 		    (fib_nh->nh_flags & RTNH_F_DEAD))
1375 			return call_fib_notifiers(dev_net(fib_nh->nh_dev),
1376 						  event_type, &info.info);
1377 	default:
1378 		break;
1379 	}
1380 
1381 	return NOTIFY_DONE;
1382 }
1383 
1384 /* Event              force Flags           Description
1385  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1386  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1387  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1388  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1389  */
1390 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1391 {
1392 	int ret = 0;
1393 	int scope = RT_SCOPE_NOWHERE;
1394 	struct fib_info *prev_fi = NULL;
1395 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1396 	struct hlist_head *head = &fib_info_devhash[hash];
1397 	struct fib_nh *nh;
1398 
1399 	if (force)
1400 		scope = -1;
1401 
1402 	hlist_for_each_entry(nh, head, nh_hash) {
1403 		struct fib_info *fi = nh->nh_parent;
1404 		int dead;
1405 
1406 		BUG_ON(!fi->fib_nhs);
1407 		if (nh->nh_dev != dev || fi == prev_fi)
1408 			continue;
1409 		prev_fi = fi;
1410 		dead = 0;
1411 		change_nexthops(fi) {
1412 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1413 				dead++;
1414 			else if (nexthop_nh->nh_dev == dev &&
1415 				 nexthop_nh->nh_scope != scope) {
1416 				switch (event) {
1417 				case NETDEV_DOWN:
1418 				case NETDEV_UNREGISTER:
1419 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1420 					/* fall through */
1421 				case NETDEV_CHANGE:
1422 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1423 					break;
1424 				}
1425 				call_fib_nh_notifiers(nexthop_nh,
1426 						      FIB_EVENT_NH_DEL);
1427 				dead++;
1428 			}
1429 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1430 			if (event == NETDEV_UNREGISTER &&
1431 			    nexthop_nh->nh_dev == dev) {
1432 				dead = fi->fib_nhs;
1433 				break;
1434 			}
1435 #endif
1436 		} endfor_nexthops(fi)
1437 		if (dead == fi->fib_nhs) {
1438 			switch (event) {
1439 			case NETDEV_DOWN:
1440 			case NETDEV_UNREGISTER:
1441 				fi->fib_flags |= RTNH_F_DEAD;
1442 				/* fall through */
1443 			case NETDEV_CHANGE:
1444 				fi->fib_flags |= RTNH_F_LINKDOWN;
1445 				break;
1446 			}
1447 			ret++;
1448 		}
1449 
1450 		fib_rebalance(fi);
1451 	}
1452 
1453 	return ret;
1454 }
1455 
1456 /* Must be invoked inside of an RCU protected region.  */
1457 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1458 {
1459 	struct fib_info *fi = NULL, *last_resort = NULL;
1460 	struct hlist_head *fa_head = res->fa_head;
1461 	struct fib_table *tb = res->table;
1462 	u8 slen = 32 - res->prefixlen;
1463 	int order = -1, last_idx = -1;
1464 	struct fib_alias *fa, *fa1 = NULL;
1465 	u32 last_prio = res->fi->fib_priority;
1466 	u8 last_tos = 0;
1467 
1468 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1469 		struct fib_info *next_fi = fa->fa_info;
1470 
1471 		if (fa->fa_slen != slen)
1472 			continue;
1473 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1474 			continue;
1475 		if (fa->tb_id != tb->tb_id)
1476 			continue;
1477 		if (next_fi->fib_priority > last_prio &&
1478 		    fa->fa_tos == last_tos) {
1479 			if (last_tos)
1480 				continue;
1481 			break;
1482 		}
1483 		if (next_fi->fib_flags & RTNH_F_DEAD)
1484 			continue;
1485 		last_tos = fa->fa_tos;
1486 		last_prio = next_fi->fib_priority;
1487 
1488 		if (next_fi->fib_scope != res->scope ||
1489 		    fa->fa_type != RTN_UNICAST)
1490 			continue;
1491 		if (!next_fi->fib_nh[0].nh_gw ||
1492 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1493 			continue;
1494 
1495 		fib_alias_accessed(fa);
1496 
1497 		if (!fi) {
1498 			if (next_fi != res->fi)
1499 				break;
1500 			fa1 = fa;
1501 		} else if (!fib_detect_death(fi, order, &last_resort,
1502 					     &last_idx, fa1->fa_default)) {
1503 			fib_result_assign(res, fi);
1504 			fa1->fa_default = order;
1505 			goto out;
1506 		}
1507 		fi = next_fi;
1508 		order++;
1509 	}
1510 
1511 	if (order <= 0 || !fi) {
1512 		if (fa1)
1513 			fa1->fa_default = -1;
1514 		goto out;
1515 	}
1516 
1517 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1518 			      fa1->fa_default)) {
1519 		fib_result_assign(res, fi);
1520 		fa1->fa_default = order;
1521 		goto out;
1522 	}
1523 
1524 	if (last_idx >= 0)
1525 		fib_result_assign(res, last_resort);
1526 	fa1->fa_default = last_idx;
1527 out:
1528 	return;
1529 }
1530 
1531 /*
1532  * Dead device goes up. We wake up dead nexthops.
1533  * It takes sense only on multipath routes.
1534  */
1535 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1536 {
1537 	struct fib_info *prev_fi;
1538 	unsigned int hash;
1539 	struct hlist_head *head;
1540 	struct fib_nh *nh;
1541 	int ret;
1542 
1543 	if (!(dev->flags & IFF_UP))
1544 		return 0;
1545 
1546 	if (nh_flags & RTNH_F_DEAD) {
1547 		unsigned int flags = dev_get_flags(dev);
1548 
1549 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1550 			nh_flags |= RTNH_F_LINKDOWN;
1551 	}
1552 
1553 	prev_fi = NULL;
1554 	hash = fib_devindex_hashfn(dev->ifindex);
1555 	head = &fib_info_devhash[hash];
1556 	ret = 0;
1557 
1558 	hlist_for_each_entry(nh, head, nh_hash) {
1559 		struct fib_info *fi = nh->nh_parent;
1560 		int alive;
1561 
1562 		BUG_ON(!fi->fib_nhs);
1563 		if (nh->nh_dev != dev || fi == prev_fi)
1564 			continue;
1565 
1566 		prev_fi = fi;
1567 		alive = 0;
1568 		change_nexthops(fi) {
1569 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1570 				alive++;
1571 				continue;
1572 			}
1573 			if (!nexthop_nh->nh_dev ||
1574 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1575 				continue;
1576 			if (nexthop_nh->nh_dev != dev ||
1577 			    !__in_dev_get_rtnl(dev))
1578 				continue;
1579 			alive++;
1580 			nexthop_nh->nh_flags &= ~nh_flags;
1581 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1582 		} endfor_nexthops(fi)
1583 
1584 		if (alive > 0) {
1585 			fi->fib_flags &= ~nh_flags;
1586 			ret++;
1587 		}
1588 
1589 		fib_rebalance(fi);
1590 	}
1591 
1592 	return ret;
1593 }
1594 
1595 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1596 static bool fib_good_nh(const struct fib_nh *nh)
1597 {
1598 	int state = NUD_REACHABLE;
1599 
1600 	if (nh->nh_scope == RT_SCOPE_LINK) {
1601 		struct neighbour *n;
1602 
1603 		rcu_read_lock_bh();
1604 
1605 		n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1606 					      (__force u32)nh->nh_gw);
1607 		if (n)
1608 			state = n->nud_state;
1609 
1610 		rcu_read_unlock_bh();
1611 	}
1612 
1613 	return !!(state & NUD_VALID);
1614 }
1615 
1616 void fib_select_multipath(struct fib_result *res, int hash)
1617 {
1618 	struct fib_info *fi = res->fi;
1619 	struct net *net = fi->fib_net;
1620 	bool first = false;
1621 
1622 	for_nexthops(fi) {
1623 		if (hash > atomic_read(&nh->nh_upper_bound))
1624 			continue;
1625 
1626 		if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1627 		    fib_good_nh(nh)) {
1628 			res->nh_sel = nhsel;
1629 			return;
1630 		}
1631 		if (!first) {
1632 			res->nh_sel = nhsel;
1633 			first = true;
1634 		}
1635 	} endfor_nexthops(fi);
1636 }
1637 #endif
1638 
1639 void fib_select_path(struct net *net, struct fib_result *res,
1640 		     struct flowi4 *fl4, const struct sk_buff *skb)
1641 {
1642 	bool oif_check;
1643 
1644 	oif_check = (fl4->flowi4_oif == 0 ||
1645 		     fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1646 
1647 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1648 	if (res->fi->fib_nhs > 1 && oif_check) {
1649 		int h = fib_multipath_hash(res->fi, fl4, skb);
1650 
1651 		fib_select_multipath(res, h);
1652 	}
1653 	else
1654 #endif
1655 	if (!res->prefixlen &&
1656 	    res->table->tb_num_default > 1 &&
1657 	    res->type == RTN_UNICAST && oif_check)
1658 		fib_select_default(fl4, res);
1659 
1660 	if (!fl4->saddr)
1661 		fl4->saddr = FIB_RES_PREFSRC(net, *res);
1662 }
1663 EXPORT_SYMBOL_GPL(fib_select_path);
1664