xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 7587eb18)
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 <asm/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 u32 fib_multipath_secret __read_mostly;
61 
62 #define for_nexthops(fi) {						\
63 	int nhsel; const struct fib_nh *nh;				\
64 	for (nhsel = 0, nh = (fi)->fib_nh;				\
65 	     nhsel < (fi)->fib_nhs;					\
66 	     nh++, nhsel++)
67 
68 #define change_nexthops(fi) {						\
69 	int nhsel; struct fib_nh *nexthop_nh;				\
70 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
71 	     nhsel < (fi)->fib_nhs;					\
72 	     nexthop_nh++, nhsel++)
73 
74 #else /* CONFIG_IP_ROUTE_MULTIPATH */
75 
76 /* Hope, that gcc will optimize it to get rid of dummy loop */
77 
78 #define for_nexthops(fi) {						\
79 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
80 	for (nhsel = 0; nhsel < 1; nhsel++)
81 
82 #define change_nexthops(fi) {						\
83 	int nhsel;							\
84 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
85 	for (nhsel = 0; nhsel < 1; nhsel++)
86 
87 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
88 
89 #define endfor_nexthops(fi) }
90 
91 
92 const struct fib_prop fib_props[RTN_MAX + 1] = {
93 	[RTN_UNSPEC] = {
94 		.error	= 0,
95 		.scope	= RT_SCOPE_NOWHERE,
96 	},
97 	[RTN_UNICAST] = {
98 		.error	= 0,
99 		.scope	= RT_SCOPE_UNIVERSE,
100 	},
101 	[RTN_LOCAL] = {
102 		.error	= 0,
103 		.scope	= RT_SCOPE_HOST,
104 	},
105 	[RTN_BROADCAST] = {
106 		.error	= 0,
107 		.scope	= RT_SCOPE_LINK,
108 	},
109 	[RTN_ANYCAST] = {
110 		.error	= 0,
111 		.scope	= RT_SCOPE_LINK,
112 	},
113 	[RTN_MULTICAST] = {
114 		.error	= 0,
115 		.scope	= RT_SCOPE_UNIVERSE,
116 	},
117 	[RTN_BLACKHOLE] = {
118 		.error	= -EINVAL,
119 		.scope	= RT_SCOPE_UNIVERSE,
120 	},
121 	[RTN_UNREACHABLE] = {
122 		.error	= -EHOSTUNREACH,
123 		.scope	= RT_SCOPE_UNIVERSE,
124 	},
125 	[RTN_PROHIBIT] = {
126 		.error	= -EACCES,
127 		.scope	= RT_SCOPE_UNIVERSE,
128 	},
129 	[RTN_THROW] = {
130 		.error	= -EAGAIN,
131 		.scope	= RT_SCOPE_UNIVERSE,
132 	},
133 	[RTN_NAT] = {
134 		.error	= -EINVAL,
135 		.scope	= RT_SCOPE_NOWHERE,
136 	},
137 	[RTN_XRESOLVE] = {
138 		.error	= -EINVAL,
139 		.scope	= RT_SCOPE_NOWHERE,
140 	},
141 };
142 
143 static void rt_fibinfo_free(struct rtable __rcu **rtp)
144 {
145 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
146 
147 	if (!rt)
148 		return;
149 
150 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
151 	 * because we waited an RCU grace period before calling
152 	 * free_fib_info_rcu()
153 	 */
154 
155 	dst_free(&rt->dst);
156 }
157 
158 static void free_nh_exceptions(struct fib_nh *nh)
159 {
160 	struct fnhe_hash_bucket *hash;
161 	int i;
162 
163 	hash = rcu_dereference_protected(nh->nh_exceptions, 1);
164 	if (!hash)
165 		return;
166 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
167 		struct fib_nh_exception *fnhe;
168 
169 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
170 		while (fnhe) {
171 			struct fib_nh_exception *next;
172 
173 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
174 
175 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
176 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
177 
178 			kfree(fnhe);
179 
180 			fnhe = next;
181 		}
182 	}
183 	kfree(hash);
184 }
185 
186 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
187 {
188 	int cpu;
189 
190 	if (!rtp)
191 		return;
192 
193 	for_each_possible_cpu(cpu) {
194 		struct rtable *rt;
195 
196 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
197 		if (rt)
198 			dst_free(&rt->dst);
199 	}
200 	free_percpu(rtp);
201 }
202 
203 /* Release a nexthop info record */
204 static void free_fib_info_rcu(struct rcu_head *head)
205 {
206 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
207 
208 	change_nexthops(fi) {
209 		if (nexthop_nh->nh_dev)
210 			dev_put(nexthop_nh->nh_dev);
211 		lwtstate_put(nexthop_nh->nh_lwtstate);
212 		free_nh_exceptions(nexthop_nh);
213 		rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
214 		rt_fibinfo_free(&nexthop_nh->nh_rth_input);
215 	} endfor_nexthops(fi);
216 
217 	if (fi->fib_metrics != (u32 *) dst_default_metrics)
218 		kfree(fi->fib_metrics);
219 	kfree(fi);
220 }
221 
222 void free_fib_info(struct fib_info *fi)
223 {
224 	if (fi->fib_dead == 0) {
225 		pr_warn("Freeing alive fib_info %p\n", fi);
226 		return;
227 	}
228 	fib_info_cnt--;
229 #ifdef CONFIG_IP_ROUTE_CLASSID
230 	change_nexthops(fi) {
231 		if (nexthop_nh->nh_tclassid)
232 			fi->fib_net->ipv4.fib_num_tclassid_users--;
233 	} endfor_nexthops(fi);
234 #endif
235 	call_rcu(&fi->rcu, free_fib_info_rcu);
236 }
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 	struct net *net = cfg->fc_nlinfo.nl_net;
474 	int ret;
475 
476 	change_nexthops(fi) {
477 		int attrlen;
478 
479 		if (!rtnh_ok(rtnh, remaining))
480 			return -EINVAL;
481 
482 		nexthop_nh->nh_flags =
483 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
484 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
485 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
486 
487 		attrlen = rtnh_attrlen(rtnh);
488 		if (attrlen > 0) {
489 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
490 
491 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
492 			nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
493 #ifdef CONFIG_IP_ROUTE_CLASSID
494 			nla = nla_find(attrs, attrlen, RTA_FLOW);
495 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
496 			if (nexthop_nh->nh_tclassid)
497 				fi->fib_net->ipv4.fib_num_tclassid_users++;
498 #endif
499 			nla = nla_find(attrs, attrlen, RTA_ENCAP);
500 			if (nla) {
501 				struct lwtunnel_state *lwtstate;
502 				struct net_device *dev = NULL;
503 				struct nlattr *nla_entype;
504 
505 				nla_entype = nla_find(attrs, attrlen,
506 						      RTA_ENCAP_TYPE);
507 				if (!nla_entype)
508 					goto err_inval;
509 				if (cfg->fc_oif)
510 					dev = __dev_get_by_index(net, cfg->fc_oif);
511 				ret = lwtunnel_build_state(dev, 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 	net_get_random_once(&fib_multipath_secret,
580 			    sizeof(fib_multipath_secret));
581 }
582 
583 static inline void fib_add_weight(struct fib_info *fi,
584 				  const struct fib_nh *nh)
585 {
586 	fi->fib_weight += nh->nh_weight;
587 }
588 
589 #else /* CONFIG_IP_ROUTE_MULTIPATH */
590 
591 #define fib_rebalance(fi) do { } while (0)
592 #define fib_add_weight(fi, nh) do { } while (0)
593 
594 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
595 
596 static int fib_encap_match(struct net *net, u16 encap_type,
597 			   struct nlattr *encap,
598 			   int oif, const struct fib_nh *nh,
599 			   const struct fib_config *cfg)
600 {
601 	struct lwtunnel_state *lwtstate;
602 	struct net_device *dev = NULL;
603 	int ret, result = 0;
604 
605 	if (encap_type == LWTUNNEL_ENCAP_NONE)
606 		return 0;
607 
608 	if (oif)
609 		dev = __dev_get_by_index(net, oif);
610 	ret = lwtunnel_build_state(dev, encap_type, encap,
611 				   AF_INET, cfg, &lwtstate);
612 	if (!ret) {
613 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
614 		lwtstate_free(lwtstate);
615 	}
616 
617 	return result;
618 }
619 
620 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
621 {
622 	struct net *net = cfg->fc_nlinfo.nl_net;
623 #ifdef CONFIG_IP_ROUTE_MULTIPATH
624 	struct rtnexthop *rtnh;
625 	int remaining;
626 #endif
627 
628 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
629 		return 1;
630 
631 	if (cfg->fc_oif || cfg->fc_gw) {
632 		if (cfg->fc_encap) {
633 			if (fib_encap_match(net, cfg->fc_encap_type,
634 					    cfg->fc_encap, cfg->fc_oif,
635 					    fi->fib_nh, cfg))
636 			    return 1;
637 		}
638 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
639 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
640 			return 0;
641 		return 1;
642 	}
643 
644 #ifdef CONFIG_IP_ROUTE_MULTIPATH
645 	if (!cfg->fc_mp)
646 		return 0;
647 
648 	rtnh = cfg->fc_mp;
649 	remaining = cfg->fc_mp_len;
650 
651 	for_nexthops(fi) {
652 		int attrlen;
653 
654 		if (!rtnh_ok(rtnh, remaining))
655 			return -EINVAL;
656 
657 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
658 			return 1;
659 
660 		attrlen = rtnh_attrlen(rtnh);
661 		if (attrlen > 0) {
662 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
663 
664 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
665 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
666 				return 1;
667 #ifdef CONFIG_IP_ROUTE_CLASSID
668 			nla = nla_find(attrs, attrlen, RTA_FLOW);
669 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
670 				return 1;
671 #endif
672 		}
673 
674 		rtnh = rtnh_next(rtnh, &remaining);
675 	} endfor_nexthops(fi);
676 #endif
677 	return 0;
678 }
679 
680 
681 /*
682  * Picture
683  * -------
684  *
685  * Semantics of nexthop is very messy by historical reasons.
686  * We have to take into account, that:
687  * a) gateway can be actually local interface address,
688  *    so that gatewayed route is direct.
689  * b) gateway must be on-link address, possibly
690  *    described not by an ifaddr, but also by a direct route.
691  * c) If both gateway and interface are specified, they should not
692  *    contradict.
693  * d) If we use tunnel routes, gateway could be not on-link.
694  *
695  * Attempt to reconcile all of these (alas, self-contradictory) conditions
696  * results in pretty ugly and hairy code with obscure logic.
697  *
698  * I chose to generalized it instead, so that the size
699  * of code does not increase practically, but it becomes
700  * much more general.
701  * Every prefix is assigned a "scope" value: "host" is local address,
702  * "link" is direct route,
703  * [ ... "site" ... "interior" ... ]
704  * and "universe" is true gateway route with global meaning.
705  *
706  * Every prefix refers to a set of "nexthop"s (gw, oif),
707  * where gw must have narrower scope. This recursion stops
708  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
709  * which means that gw is forced to be on link.
710  *
711  * Code is still hairy, but now it is apparently logically
712  * consistent and very flexible. F.e. as by-product it allows
713  * to co-exists in peace independent exterior and interior
714  * routing processes.
715  *
716  * Normally it looks as following.
717  *
718  * {universe prefix}  -> (gw, oif) [scope link]
719  *		  |
720  *		  |-> {link prefix} -> (gw, oif) [scope local]
721  *					|
722  *					|-> {local prefix} (terminal node)
723  */
724 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
725 			struct fib_nh *nh)
726 {
727 	int err = 0;
728 	struct net *net;
729 	struct net_device *dev;
730 
731 	net = cfg->fc_nlinfo.nl_net;
732 	if (nh->nh_gw) {
733 		struct fib_result res;
734 
735 		if (nh->nh_flags & RTNH_F_ONLINK) {
736 			unsigned int addr_type;
737 
738 			if (cfg->fc_scope >= RT_SCOPE_LINK)
739 				return -EINVAL;
740 			dev = __dev_get_by_index(net, nh->nh_oif);
741 			if (!dev)
742 				return -ENODEV;
743 			if (!(dev->flags & IFF_UP))
744 				return -ENETDOWN;
745 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
746 			if (addr_type != RTN_UNICAST)
747 				return -EINVAL;
748 			if (!netif_carrier_ok(dev))
749 				nh->nh_flags |= RTNH_F_LINKDOWN;
750 			nh->nh_dev = dev;
751 			dev_hold(dev);
752 			nh->nh_scope = RT_SCOPE_LINK;
753 			return 0;
754 		}
755 		rcu_read_lock();
756 		{
757 			struct fib_table *tbl = NULL;
758 			struct flowi4 fl4 = {
759 				.daddr = nh->nh_gw,
760 				.flowi4_scope = cfg->fc_scope + 1,
761 				.flowi4_oif = nh->nh_oif,
762 				.flowi4_iif = LOOPBACK_IFINDEX,
763 			};
764 
765 			/* It is not necessary, but requires a bit of thinking */
766 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
767 				fl4.flowi4_scope = RT_SCOPE_LINK;
768 
769 			if (cfg->fc_table)
770 				tbl = fib_get_table(net, cfg->fc_table);
771 
772 			if (tbl)
773 				err = fib_table_lookup(tbl, &fl4, &res,
774 						       FIB_LOOKUP_IGNORE_LINKSTATE |
775 						       FIB_LOOKUP_NOREF);
776 
777 			/* on error or if no table given do full lookup. This
778 			 * is needed for example when nexthops are in the local
779 			 * table rather than the given table
780 			 */
781 			if (!tbl || err) {
782 				err = fib_lookup(net, &fl4, &res,
783 						 FIB_LOOKUP_IGNORE_LINKSTATE);
784 			}
785 
786 			if (err) {
787 				rcu_read_unlock();
788 				return err;
789 			}
790 		}
791 		err = -EINVAL;
792 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
793 			goto out;
794 		nh->nh_scope = res.scope;
795 		nh->nh_oif = FIB_RES_OIF(res);
796 		nh->nh_dev = dev = FIB_RES_DEV(res);
797 		if (!dev)
798 			goto out;
799 		dev_hold(dev);
800 		if (!netif_carrier_ok(dev))
801 			nh->nh_flags |= RTNH_F_LINKDOWN;
802 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
803 	} else {
804 		struct in_device *in_dev;
805 
806 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
807 			return -EINVAL;
808 
809 		rcu_read_lock();
810 		err = -ENODEV;
811 		in_dev = inetdev_by_index(net, nh->nh_oif);
812 		if (!in_dev)
813 			goto out;
814 		err = -ENETDOWN;
815 		if (!(in_dev->dev->flags & IFF_UP))
816 			goto out;
817 		nh->nh_dev = in_dev->dev;
818 		dev_hold(nh->nh_dev);
819 		nh->nh_scope = RT_SCOPE_HOST;
820 		if (!netif_carrier_ok(nh->nh_dev))
821 			nh->nh_flags |= RTNH_F_LINKDOWN;
822 		err = 0;
823 	}
824 out:
825 	rcu_read_unlock();
826 	return err;
827 }
828 
829 static inline unsigned int fib_laddr_hashfn(__be32 val)
830 {
831 	unsigned int mask = (fib_info_hash_size - 1);
832 
833 	return ((__force u32)val ^
834 		((__force u32)val >> 7) ^
835 		((__force u32)val >> 14)) & mask;
836 }
837 
838 static struct hlist_head *fib_info_hash_alloc(int bytes)
839 {
840 	if (bytes <= PAGE_SIZE)
841 		return kzalloc(bytes, GFP_KERNEL);
842 	else
843 		return (struct hlist_head *)
844 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
845 					 get_order(bytes));
846 }
847 
848 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
849 {
850 	if (!hash)
851 		return;
852 
853 	if (bytes <= PAGE_SIZE)
854 		kfree(hash);
855 	else
856 		free_pages((unsigned long) hash, get_order(bytes));
857 }
858 
859 static void fib_info_hash_move(struct hlist_head *new_info_hash,
860 			       struct hlist_head *new_laddrhash,
861 			       unsigned int new_size)
862 {
863 	struct hlist_head *old_info_hash, *old_laddrhash;
864 	unsigned int old_size = fib_info_hash_size;
865 	unsigned int i, bytes;
866 
867 	spin_lock_bh(&fib_info_lock);
868 	old_info_hash = fib_info_hash;
869 	old_laddrhash = fib_info_laddrhash;
870 	fib_info_hash_size = new_size;
871 
872 	for (i = 0; i < old_size; i++) {
873 		struct hlist_head *head = &fib_info_hash[i];
874 		struct hlist_node *n;
875 		struct fib_info *fi;
876 
877 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
878 			struct hlist_head *dest;
879 			unsigned int new_hash;
880 
881 			new_hash = fib_info_hashfn(fi);
882 			dest = &new_info_hash[new_hash];
883 			hlist_add_head(&fi->fib_hash, dest);
884 		}
885 	}
886 	fib_info_hash = new_info_hash;
887 
888 	for (i = 0; i < old_size; i++) {
889 		struct hlist_head *lhead = &fib_info_laddrhash[i];
890 		struct hlist_node *n;
891 		struct fib_info *fi;
892 
893 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
894 			struct hlist_head *ldest;
895 			unsigned int new_hash;
896 
897 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
898 			ldest = &new_laddrhash[new_hash];
899 			hlist_add_head(&fi->fib_lhash, ldest);
900 		}
901 	}
902 	fib_info_laddrhash = new_laddrhash;
903 
904 	spin_unlock_bh(&fib_info_lock);
905 
906 	bytes = old_size * sizeof(struct hlist_head *);
907 	fib_info_hash_free(old_info_hash, bytes);
908 	fib_info_hash_free(old_laddrhash, bytes);
909 }
910 
911 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
912 {
913 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
914 					nh->nh_gw,
915 					nh->nh_parent->fib_scope);
916 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
917 
918 	return nh->nh_saddr;
919 }
920 
921 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
922 {
923 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
924 	    fib_prefsrc != cfg->fc_dst) {
925 		u32 tb_id = cfg->fc_table;
926 		int rc;
927 
928 		if (tb_id == RT_TABLE_MAIN)
929 			tb_id = RT_TABLE_LOCAL;
930 
931 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
932 					  fib_prefsrc, tb_id);
933 
934 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
935 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
936 						  fib_prefsrc, RT_TABLE_LOCAL);
937 		}
938 
939 		if (rc != RTN_LOCAL)
940 			return false;
941 	}
942 	return true;
943 }
944 
945 static int
946 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
947 {
948 	bool ecn_ca = false;
949 	struct nlattr *nla;
950 	int remaining;
951 
952 	if (!cfg->fc_mx)
953 		return 0;
954 
955 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
956 		int type = nla_type(nla);
957 		u32 val;
958 
959 		if (!type)
960 			continue;
961 		if (type > RTAX_MAX)
962 			return -EINVAL;
963 
964 		if (type == RTAX_CC_ALGO) {
965 			char tmp[TCP_CA_NAME_MAX];
966 
967 			nla_strlcpy(tmp, nla, sizeof(tmp));
968 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
969 			if (val == TCP_CA_UNSPEC)
970 				return -EINVAL;
971 		} else {
972 			val = nla_get_u32(nla);
973 		}
974 		if (type == RTAX_ADVMSS && val > 65535 - 40)
975 			val = 65535 - 40;
976 		if (type == RTAX_MTU && val > 65535 - 15)
977 			val = 65535 - 15;
978 		if (type == RTAX_HOPLIMIT && val > 255)
979 			val = 255;
980 		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
981 			return -EINVAL;
982 		fi->fib_metrics[type - 1] = val;
983 	}
984 
985 	if (ecn_ca)
986 		fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
987 
988 	return 0;
989 }
990 
991 struct fib_info *fib_create_info(struct fib_config *cfg)
992 {
993 	int err;
994 	struct fib_info *fi = NULL;
995 	struct fib_info *ofi;
996 	int nhs = 1;
997 	struct net *net = cfg->fc_nlinfo.nl_net;
998 
999 	if (cfg->fc_type > RTN_MAX)
1000 		goto err_inval;
1001 
1002 	/* Fast check to catch the most weird cases */
1003 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
1004 		goto err_inval;
1005 
1006 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1007 	if (cfg->fc_mp) {
1008 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1009 		if (nhs == 0)
1010 			goto err_inval;
1011 	}
1012 #endif
1013 
1014 	err = -ENOBUFS;
1015 	if (fib_info_cnt >= fib_info_hash_size) {
1016 		unsigned int new_size = fib_info_hash_size << 1;
1017 		struct hlist_head *new_info_hash;
1018 		struct hlist_head *new_laddrhash;
1019 		unsigned int bytes;
1020 
1021 		if (!new_size)
1022 			new_size = 16;
1023 		bytes = new_size * sizeof(struct hlist_head *);
1024 		new_info_hash = fib_info_hash_alloc(bytes);
1025 		new_laddrhash = fib_info_hash_alloc(bytes);
1026 		if (!new_info_hash || !new_laddrhash) {
1027 			fib_info_hash_free(new_info_hash, bytes);
1028 			fib_info_hash_free(new_laddrhash, bytes);
1029 		} else
1030 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1031 
1032 		if (!fib_info_hash_size)
1033 			goto failure;
1034 	}
1035 
1036 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1037 	if (!fi)
1038 		goto failure;
1039 	fib_info_cnt++;
1040 	if (cfg->fc_mx) {
1041 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1042 		if (!fi->fib_metrics)
1043 			goto failure;
1044 	} else
1045 		fi->fib_metrics = (u32 *) dst_default_metrics;
1046 
1047 	fi->fib_net = net;
1048 	fi->fib_protocol = cfg->fc_protocol;
1049 	fi->fib_scope = cfg->fc_scope;
1050 	fi->fib_flags = cfg->fc_flags;
1051 	fi->fib_priority = cfg->fc_priority;
1052 	fi->fib_prefsrc = cfg->fc_prefsrc;
1053 	fi->fib_type = cfg->fc_type;
1054 
1055 	fi->fib_nhs = nhs;
1056 	change_nexthops(fi) {
1057 		nexthop_nh->nh_parent = fi;
1058 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1059 		if (!nexthop_nh->nh_pcpu_rth_output)
1060 			goto failure;
1061 	} endfor_nexthops(fi)
1062 
1063 	err = fib_convert_metrics(fi, cfg);
1064 	if (err)
1065 		goto failure;
1066 
1067 	if (cfg->fc_mp) {
1068 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1069 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1070 		if (err != 0)
1071 			goto failure;
1072 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1073 			goto err_inval;
1074 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1075 			goto err_inval;
1076 #ifdef CONFIG_IP_ROUTE_CLASSID
1077 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1078 			goto err_inval;
1079 #endif
1080 #else
1081 		goto err_inval;
1082 #endif
1083 	} else {
1084 		struct fib_nh *nh = fi->fib_nh;
1085 
1086 		if (cfg->fc_encap) {
1087 			struct lwtunnel_state *lwtstate;
1088 			struct net_device *dev = NULL;
1089 
1090 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1091 				goto err_inval;
1092 			if (cfg->fc_oif)
1093 				dev = __dev_get_by_index(net, cfg->fc_oif);
1094 			err = lwtunnel_build_state(dev, cfg->fc_encap_type,
1095 						   cfg->fc_encap, AF_INET, cfg,
1096 						   &lwtstate);
1097 			if (err)
1098 				goto failure;
1099 
1100 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1101 		}
1102 		nh->nh_oif = cfg->fc_oif;
1103 		nh->nh_gw = cfg->fc_gw;
1104 		nh->nh_flags = cfg->fc_flags;
1105 #ifdef CONFIG_IP_ROUTE_CLASSID
1106 		nh->nh_tclassid = cfg->fc_flow;
1107 		if (nh->nh_tclassid)
1108 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1109 #endif
1110 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1111 		nh->nh_weight = 1;
1112 #endif
1113 	}
1114 
1115 	if (fib_props[cfg->fc_type].error) {
1116 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1117 			goto err_inval;
1118 		goto link_it;
1119 	} else {
1120 		switch (cfg->fc_type) {
1121 		case RTN_UNICAST:
1122 		case RTN_LOCAL:
1123 		case RTN_BROADCAST:
1124 		case RTN_ANYCAST:
1125 		case RTN_MULTICAST:
1126 			break;
1127 		default:
1128 			goto err_inval;
1129 		}
1130 	}
1131 
1132 	if (cfg->fc_scope > RT_SCOPE_HOST)
1133 		goto err_inval;
1134 
1135 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1136 		struct fib_nh *nh = fi->fib_nh;
1137 
1138 		/* Local address is added. */
1139 		if (nhs != 1 || nh->nh_gw)
1140 			goto err_inval;
1141 		nh->nh_scope = RT_SCOPE_NOWHERE;
1142 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1143 		err = -ENODEV;
1144 		if (!nh->nh_dev)
1145 			goto failure;
1146 	} else {
1147 		int linkdown = 0;
1148 
1149 		change_nexthops(fi) {
1150 			err = fib_check_nh(cfg, fi, nexthop_nh);
1151 			if (err != 0)
1152 				goto failure;
1153 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1154 				linkdown++;
1155 		} endfor_nexthops(fi)
1156 		if (linkdown == fi->fib_nhs)
1157 			fi->fib_flags |= RTNH_F_LINKDOWN;
1158 	}
1159 
1160 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1161 		goto err_inval;
1162 
1163 	change_nexthops(fi) {
1164 		fib_info_update_nh_saddr(net, nexthop_nh);
1165 		fib_add_weight(fi, nexthop_nh);
1166 	} endfor_nexthops(fi)
1167 
1168 	fib_rebalance(fi);
1169 
1170 link_it:
1171 	ofi = fib_find_info(fi);
1172 	if (ofi) {
1173 		fi->fib_dead = 1;
1174 		free_fib_info(fi);
1175 		ofi->fib_treeref++;
1176 		return ofi;
1177 	}
1178 
1179 	fi->fib_treeref++;
1180 	atomic_inc(&fi->fib_clntref);
1181 	spin_lock_bh(&fib_info_lock);
1182 	hlist_add_head(&fi->fib_hash,
1183 		       &fib_info_hash[fib_info_hashfn(fi)]);
1184 	if (fi->fib_prefsrc) {
1185 		struct hlist_head *head;
1186 
1187 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1188 		hlist_add_head(&fi->fib_lhash, head);
1189 	}
1190 	change_nexthops(fi) {
1191 		struct hlist_head *head;
1192 		unsigned int hash;
1193 
1194 		if (!nexthop_nh->nh_dev)
1195 			continue;
1196 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1197 		head = &fib_info_devhash[hash];
1198 		hlist_add_head(&nexthop_nh->nh_hash, head);
1199 	} endfor_nexthops(fi)
1200 	spin_unlock_bh(&fib_info_lock);
1201 	return fi;
1202 
1203 err_inval:
1204 	err = -EINVAL;
1205 
1206 failure:
1207 	if (fi) {
1208 		fi->fib_dead = 1;
1209 		free_fib_info(fi);
1210 	}
1211 
1212 	return ERR_PTR(err);
1213 }
1214 
1215 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1216 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1217 		  struct fib_info *fi, unsigned int flags)
1218 {
1219 	struct nlmsghdr *nlh;
1220 	struct rtmsg *rtm;
1221 
1222 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1223 	if (!nlh)
1224 		return -EMSGSIZE;
1225 
1226 	rtm = nlmsg_data(nlh);
1227 	rtm->rtm_family = AF_INET;
1228 	rtm->rtm_dst_len = dst_len;
1229 	rtm->rtm_src_len = 0;
1230 	rtm->rtm_tos = tos;
1231 	if (tb_id < 256)
1232 		rtm->rtm_table = tb_id;
1233 	else
1234 		rtm->rtm_table = RT_TABLE_COMPAT;
1235 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1236 		goto nla_put_failure;
1237 	rtm->rtm_type = type;
1238 	rtm->rtm_flags = fi->fib_flags;
1239 	rtm->rtm_scope = fi->fib_scope;
1240 	rtm->rtm_protocol = fi->fib_protocol;
1241 
1242 	if (rtm->rtm_dst_len &&
1243 	    nla_put_in_addr(skb, RTA_DST, dst))
1244 		goto nla_put_failure;
1245 	if (fi->fib_priority &&
1246 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1247 		goto nla_put_failure;
1248 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1249 		goto nla_put_failure;
1250 
1251 	if (fi->fib_prefsrc &&
1252 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1253 		goto nla_put_failure;
1254 	if (fi->fib_nhs == 1) {
1255 		struct in_device *in_dev;
1256 
1257 		if (fi->fib_nh->nh_gw &&
1258 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1259 			goto nla_put_failure;
1260 		if (fi->fib_nh->nh_oif &&
1261 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1262 			goto nla_put_failure;
1263 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1264 			in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1265 			if (in_dev &&
1266 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1267 				rtm->rtm_flags |= RTNH_F_DEAD;
1268 		}
1269 #ifdef CONFIG_IP_ROUTE_CLASSID
1270 		if (fi->fib_nh[0].nh_tclassid &&
1271 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1272 			goto nla_put_failure;
1273 #endif
1274 		if (fi->fib_nh->nh_lwtstate)
1275 			lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1276 	}
1277 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1278 	if (fi->fib_nhs > 1) {
1279 		struct rtnexthop *rtnh;
1280 		struct nlattr *mp;
1281 
1282 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1283 		if (!mp)
1284 			goto nla_put_failure;
1285 
1286 		for_nexthops(fi) {
1287 			struct in_device *in_dev;
1288 
1289 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1290 			if (!rtnh)
1291 				goto nla_put_failure;
1292 
1293 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1294 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1295 				in_dev = __in_dev_get_rtnl(nh->nh_dev);
1296 				if (in_dev &&
1297 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1298 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1299 			}
1300 			rtnh->rtnh_hops = nh->nh_weight - 1;
1301 			rtnh->rtnh_ifindex = nh->nh_oif;
1302 
1303 			if (nh->nh_gw &&
1304 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1305 				goto nla_put_failure;
1306 #ifdef CONFIG_IP_ROUTE_CLASSID
1307 			if (nh->nh_tclassid &&
1308 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1309 				goto nla_put_failure;
1310 #endif
1311 			if (nh->nh_lwtstate)
1312 				lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1313 			/* length of rtnetlink header + attributes */
1314 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1315 		} endfor_nexthops(fi);
1316 
1317 		nla_nest_end(skb, mp);
1318 	}
1319 #endif
1320 	nlmsg_end(skb, nlh);
1321 	return 0;
1322 
1323 nla_put_failure:
1324 	nlmsg_cancel(skb, nlh);
1325 	return -EMSGSIZE;
1326 }
1327 
1328 /*
1329  * Update FIB if:
1330  * - local address disappeared -> we must delete all the entries
1331  *   referring to it.
1332  * - device went down -> we must shutdown all nexthops going via it.
1333  */
1334 int fib_sync_down_addr(struct net *net, __be32 local)
1335 {
1336 	int ret = 0;
1337 	unsigned int hash = fib_laddr_hashfn(local);
1338 	struct hlist_head *head = &fib_info_laddrhash[hash];
1339 	struct fib_info *fi;
1340 
1341 	if (!fib_info_laddrhash || local == 0)
1342 		return 0;
1343 
1344 	hlist_for_each_entry(fi, head, fib_lhash) {
1345 		if (!net_eq(fi->fib_net, net))
1346 			continue;
1347 		if (fi->fib_prefsrc == local) {
1348 			fi->fib_flags |= RTNH_F_DEAD;
1349 			ret++;
1350 		}
1351 	}
1352 	return ret;
1353 }
1354 
1355 /* Event              force Flags           Description
1356  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1357  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1358  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1359  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1360  */
1361 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1362 {
1363 	int ret = 0;
1364 	int scope = RT_SCOPE_NOWHERE;
1365 	struct fib_info *prev_fi = NULL;
1366 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1367 	struct hlist_head *head = &fib_info_devhash[hash];
1368 	struct fib_nh *nh;
1369 
1370 	if (force)
1371 		scope = -1;
1372 
1373 	hlist_for_each_entry(nh, head, nh_hash) {
1374 		struct fib_info *fi = nh->nh_parent;
1375 		int dead;
1376 
1377 		BUG_ON(!fi->fib_nhs);
1378 		if (nh->nh_dev != dev || fi == prev_fi)
1379 			continue;
1380 		prev_fi = fi;
1381 		dead = 0;
1382 		change_nexthops(fi) {
1383 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1384 				dead++;
1385 			else if (nexthop_nh->nh_dev == dev &&
1386 				 nexthop_nh->nh_scope != scope) {
1387 				switch (event) {
1388 				case NETDEV_DOWN:
1389 				case NETDEV_UNREGISTER:
1390 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1391 					/* fall through */
1392 				case NETDEV_CHANGE:
1393 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1394 					break;
1395 				}
1396 				dead++;
1397 			}
1398 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1399 			if (event == NETDEV_UNREGISTER &&
1400 			    nexthop_nh->nh_dev == dev) {
1401 				dead = fi->fib_nhs;
1402 				break;
1403 			}
1404 #endif
1405 		} endfor_nexthops(fi)
1406 		if (dead == fi->fib_nhs) {
1407 			switch (event) {
1408 			case NETDEV_DOWN:
1409 			case NETDEV_UNREGISTER:
1410 				fi->fib_flags |= RTNH_F_DEAD;
1411 				/* fall through */
1412 			case NETDEV_CHANGE:
1413 				fi->fib_flags |= RTNH_F_LINKDOWN;
1414 				break;
1415 			}
1416 			ret++;
1417 		}
1418 
1419 		fib_rebalance(fi);
1420 	}
1421 
1422 	return ret;
1423 }
1424 
1425 /* Must be invoked inside of an RCU protected region.  */
1426 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1427 {
1428 	struct fib_info *fi = NULL, *last_resort = NULL;
1429 	struct hlist_head *fa_head = res->fa_head;
1430 	struct fib_table *tb = res->table;
1431 	u8 slen = 32 - res->prefixlen;
1432 	int order = -1, last_idx = -1;
1433 	struct fib_alias *fa, *fa1 = NULL;
1434 	u32 last_prio = res->fi->fib_priority;
1435 	u8 last_tos = 0;
1436 
1437 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1438 		struct fib_info *next_fi = fa->fa_info;
1439 
1440 		if (fa->fa_slen != slen)
1441 			continue;
1442 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1443 			continue;
1444 		if (fa->tb_id != tb->tb_id)
1445 			continue;
1446 		if (next_fi->fib_priority > last_prio &&
1447 		    fa->fa_tos == last_tos) {
1448 			if (last_tos)
1449 				continue;
1450 			break;
1451 		}
1452 		if (next_fi->fib_flags & RTNH_F_DEAD)
1453 			continue;
1454 		last_tos = fa->fa_tos;
1455 		last_prio = next_fi->fib_priority;
1456 
1457 		if (next_fi->fib_scope != res->scope ||
1458 		    fa->fa_type != RTN_UNICAST)
1459 			continue;
1460 		if (!next_fi->fib_nh[0].nh_gw ||
1461 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1462 			continue;
1463 
1464 		fib_alias_accessed(fa);
1465 
1466 		if (!fi) {
1467 			if (next_fi != res->fi)
1468 				break;
1469 			fa1 = fa;
1470 		} else if (!fib_detect_death(fi, order, &last_resort,
1471 					     &last_idx, fa1->fa_default)) {
1472 			fib_result_assign(res, fi);
1473 			fa1->fa_default = order;
1474 			goto out;
1475 		}
1476 		fi = next_fi;
1477 		order++;
1478 	}
1479 
1480 	if (order <= 0 || !fi) {
1481 		if (fa1)
1482 			fa1->fa_default = -1;
1483 		goto out;
1484 	}
1485 
1486 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1487 			      fa1->fa_default)) {
1488 		fib_result_assign(res, fi);
1489 		fa1->fa_default = order;
1490 		goto out;
1491 	}
1492 
1493 	if (last_idx >= 0)
1494 		fib_result_assign(res, last_resort);
1495 	fa1->fa_default = last_idx;
1496 out:
1497 	return;
1498 }
1499 
1500 /*
1501  * Dead device goes up. We wake up dead nexthops.
1502  * It takes sense only on multipath routes.
1503  */
1504 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1505 {
1506 	struct fib_info *prev_fi;
1507 	unsigned int hash;
1508 	struct hlist_head *head;
1509 	struct fib_nh *nh;
1510 	int ret;
1511 
1512 	if (!(dev->flags & IFF_UP))
1513 		return 0;
1514 
1515 	if (nh_flags & RTNH_F_DEAD) {
1516 		unsigned int flags = dev_get_flags(dev);
1517 
1518 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1519 			nh_flags |= RTNH_F_LINKDOWN;
1520 	}
1521 
1522 	prev_fi = NULL;
1523 	hash = fib_devindex_hashfn(dev->ifindex);
1524 	head = &fib_info_devhash[hash];
1525 	ret = 0;
1526 
1527 	hlist_for_each_entry(nh, head, nh_hash) {
1528 		struct fib_info *fi = nh->nh_parent;
1529 		int alive;
1530 
1531 		BUG_ON(!fi->fib_nhs);
1532 		if (nh->nh_dev != dev || fi == prev_fi)
1533 			continue;
1534 
1535 		prev_fi = fi;
1536 		alive = 0;
1537 		change_nexthops(fi) {
1538 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1539 				alive++;
1540 				continue;
1541 			}
1542 			if (!nexthop_nh->nh_dev ||
1543 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1544 				continue;
1545 			if (nexthop_nh->nh_dev != dev ||
1546 			    !__in_dev_get_rtnl(dev))
1547 				continue;
1548 			alive++;
1549 			nexthop_nh->nh_flags &= ~nh_flags;
1550 		} endfor_nexthops(fi)
1551 
1552 		if (alive > 0) {
1553 			fi->fib_flags &= ~nh_flags;
1554 			ret++;
1555 		}
1556 
1557 		fib_rebalance(fi);
1558 	}
1559 
1560 	return ret;
1561 }
1562 
1563 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1564 static bool fib_good_nh(const struct fib_nh *nh)
1565 {
1566 	int state = NUD_REACHABLE;
1567 
1568 	if (nh->nh_scope == RT_SCOPE_LINK) {
1569 		struct neighbour *n;
1570 
1571 		rcu_read_lock_bh();
1572 
1573 		n = __ipv4_neigh_lookup_noref(nh->nh_dev, nh->nh_gw);
1574 		if (n)
1575 			state = n->nud_state;
1576 
1577 		rcu_read_unlock_bh();
1578 	}
1579 
1580 	return !!(state & NUD_VALID);
1581 }
1582 
1583 void fib_select_multipath(struct fib_result *res, int hash)
1584 {
1585 	struct fib_info *fi = res->fi;
1586 	struct net *net = fi->fib_net;
1587 	bool first = false;
1588 
1589 	for_nexthops(fi) {
1590 		if (hash > atomic_read(&nh->nh_upper_bound))
1591 			continue;
1592 
1593 		if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1594 		    fib_good_nh(nh)) {
1595 			res->nh_sel = nhsel;
1596 			return;
1597 		}
1598 		if (!first) {
1599 			res->nh_sel = nhsel;
1600 			first = true;
1601 		}
1602 	} endfor_nexthops(fi);
1603 }
1604 #endif
1605 
1606 void fib_select_path(struct net *net, struct fib_result *res,
1607 		     struct flowi4 *fl4, int mp_hash)
1608 {
1609 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1610 	if (res->fi->fib_nhs > 1 && fl4->flowi4_oif == 0) {
1611 		if (mp_hash < 0)
1612 			mp_hash = get_hash_from_flowi4(fl4) >> 1;
1613 
1614 		fib_select_multipath(res, mp_hash);
1615 	}
1616 	else
1617 #endif
1618 	if (!res->prefixlen &&
1619 	    res->table->tb_num_default > 1 &&
1620 	    res->type == RTN_UNICAST && !fl4->flowi4_oif)
1621 		fib_select_default(fl4, res);
1622 
1623 	if (!fl4->saddr)
1624 		fl4->saddr = FIB_RES_PREFSRC(net, *res);
1625 }
1626 EXPORT_SYMBOL_GPL(fib_select_path);
1627