xref: /linux/net/ipv4/fib_semantics.c (revision 95db3b255fde4e830e5f8cc011eb404023f669d4)
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 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
483 			return -EINVAL;
484 
485 		nexthop_nh->nh_flags =
486 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
487 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
488 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
489 
490 		attrlen = rtnh_attrlen(rtnh);
491 		if (attrlen > 0) {
492 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
493 
494 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
495 			nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
496 #ifdef CONFIG_IP_ROUTE_CLASSID
497 			nla = nla_find(attrs, attrlen, RTA_FLOW);
498 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
499 			if (nexthop_nh->nh_tclassid)
500 				fi->fib_net->ipv4.fib_num_tclassid_users++;
501 #endif
502 			nla = nla_find(attrs, attrlen, RTA_ENCAP);
503 			if (nla) {
504 				struct lwtunnel_state *lwtstate;
505 				struct net_device *dev = NULL;
506 				struct nlattr *nla_entype;
507 
508 				nla_entype = nla_find(attrs, attrlen,
509 						      RTA_ENCAP_TYPE);
510 				if (!nla_entype)
511 					goto err_inval;
512 				if (cfg->fc_oif)
513 					dev = __dev_get_by_index(net, cfg->fc_oif);
514 				ret = lwtunnel_build_state(dev, nla_get_u16(
515 							   nla_entype),
516 							   nla,  AF_INET, cfg,
517 							   &lwtstate);
518 				if (ret)
519 					goto errout;
520 				nexthop_nh->nh_lwtstate =
521 					lwtstate_get(lwtstate);
522 			}
523 		}
524 
525 		rtnh = rtnh_next(rtnh, &remaining);
526 	} endfor_nexthops(fi);
527 
528 	return 0;
529 
530 err_inval:
531 	ret = -EINVAL;
532 
533 errout:
534 	return ret;
535 }
536 
537 static void fib_rebalance(struct fib_info *fi)
538 {
539 	int total;
540 	int w;
541 	struct in_device *in_dev;
542 
543 	if (fi->fib_nhs < 2)
544 		return;
545 
546 	total = 0;
547 	for_nexthops(fi) {
548 		if (nh->nh_flags & RTNH_F_DEAD)
549 			continue;
550 
551 		in_dev = __in_dev_get_rtnl(nh->nh_dev);
552 
553 		if (in_dev &&
554 		    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
555 		    nh->nh_flags & RTNH_F_LINKDOWN)
556 			continue;
557 
558 		total += nh->nh_weight;
559 	} endfor_nexthops(fi);
560 
561 	w = 0;
562 	change_nexthops(fi) {
563 		int upper_bound;
564 
565 		in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
566 
567 		if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
568 			upper_bound = -1;
569 		} else if (in_dev &&
570 			   IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
571 			   nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
572 			upper_bound = -1;
573 		} else {
574 			w += nexthop_nh->nh_weight;
575 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
576 							    total) - 1;
577 		}
578 
579 		atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
580 	} endfor_nexthops(fi);
581 
582 	net_get_random_once(&fib_multipath_secret,
583 			    sizeof(fib_multipath_secret));
584 }
585 
586 static inline void fib_add_weight(struct fib_info *fi,
587 				  const struct fib_nh *nh)
588 {
589 	fi->fib_weight += nh->nh_weight;
590 }
591 
592 #else /* CONFIG_IP_ROUTE_MULTIPATH */
593 
594 #define fib_rebalance(fi) do { } while (0)
595 #define fib_add_weight(fi, nh) do { } while (0)
596 
597 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
598 
599 static int fib_encap_match(struct net *net, u16 encap_type,
600 			   struct nlattr *encap,
601 			   int oif, const struct fib_nh *nh,
602 			   const struct fib_config *cfg)
603 {
604 	struct lwtunnel_state *lwtstate;
605 	struct net_device *dev = NULL;
606 	int ret, result = 0;
607 
608 	if (encap_type == LWTUNNEL_ENCAP_NONE)
609 		return 0;
610 
611 	if (oif)
612 		dev = __dev_get_by_index(net, oif);
613 	ret = lwtunnel_build_state(dev, encap_type, encap,
614 				   AF_INET, cfg, &lwtstate);
615 	if (!ret) {
616 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
617 		lwtstate_free(lwtstate);
618 	}
619 
620 	return result;
621 }
622 
623 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
624 {
625 	struct net *net = cfg->fc_nlinfo.nl_net;
626 #ifdef CONFIG_IP_ROUTE_MULTIPATH
627 	struct rtnexthop *rtnh;
628 	int remaining;
629 #endif
630 
631 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
632 		return 1;
633 
634 	if (cfg->fc_oif || cfg->fc_gw) {
635 		if (cfg->fc_encap) {
636 			if (fib_encap_match(net, cfg->fc_encap_type,
637 					    cfg->fc_encap, cfg->fc_oif,
638 					    fi->fib_nh, cfg))
639 			    return 1;
640 		}
641 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
642 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
643 			return 0;
644 		return 1;
645 	}
646 
647 #ifdef CONFIG_IP_ROUTE_MULTIPATH
648 	if (!cfg->fc_mp)
649 		return 0;
650 
651 	rtnh = cfg->fc_mp;
652 	remaining = cfg->fc_mp_len;
653 
654 	for_nexthops(fi) {
655 		int attrlen;
656 
657 		if (!rtnh_ok(rtnh, remaining))
658 			return -EINVAL;
659 
660 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
661 			return 1;
662 
663 		attrlen = rtnh_attrlen(rtnh);
664 		if (attrlen > 0) {
665 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
666 
667 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
668 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
669 				return 1;
670 #ifdef CONFIG_IP_ROUTE_CLASSID
671 			nla = nla_find(attrs, attrlen, RTA_FLOW);
672 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
673 				return 1;
674 #endif
675 		}
676 
677 		rtnh = rtnh_next(rtnh, &remaining);
678 	} endfor_nexthops(fi);
679 #endif
680 	return 0;
681 }
682 
683 
684 /*
685  * Picture
686  * -------
687  *
688  * Semantics of nexthop is very messy by historical reasons.
689  * We have to take into account, that:
690  * a) gateway can be actually local interface address,
691  *    so that gatewayed route is direct.
692  * b) gateway must be on-link address, possibly
693  *    described not by an ifaddr, but also by a direct route.
694  * c) If both gateway and interface are specified, they should not
695  *    contradict.
696  * d) If we use tunnel routes, gateway could be not on-link.
697  *
698  * Attempt to reconcile all of these (alas, self-contradictory) conditions
699  * results in pretty ugly and hairy code with obscure logic.
700  *
701  * I chose to generalized it instead, so that the size
702  * of code does not increase practically, but it becomes
703  * much more general.
704  * Every prefix is assigned a "scope" value: "host" is local address,
705  * "link" is direct route,
706  * [ ... "site" ... "interior" ... ]
707  * and "universe" is true gateway route with global meaning.
708  *
709  * Every prefix refers to a set of "nexthop"s (gw, oif),
710  * where gw must have narrower scope. This recursion stops
711  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
712  * which means that gw is forced to be on link.
713  *
714  * Code is still hairy, but now it is apparently logically
715  * consistent and very flexible. F.e. as by-product it allows
716  * to co-exists in peace independent exterior and interior
717  * routing processes.
718  *
719  * Normally it looks as following.
720  *
721  * {universe prefix}  -> (gw, oif) [scope link]
722  *		  |
723  *		  |-> {link prefix} -> (gw, oif) [scope local]
724  *					|
725  *					|-> {local prefix} (terminal node)
726  */
727 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
728 			struct fib_nh *nh)
729 {
730 	int err = 0;
731 	struct net *net;
732 	struct net_device *dev;
733 
734 	net = cfg->fc_nlinfo.nl_net;
735 	if (nh->nh_gw) {
736 		struct fib_result res;
737 
738 		if (nh->nh_flags & RTNH_F_ONLINK) {
739 			unsigned int addr_type;
740 
741 			if (cfg->fc_scope >= RT_SCOPE_LINK)
742 				return -EINVAL;
743 			dev = __dev_get_by_index(net, nh->nh_oif);
744 			if (!dev)
745 				return -ENODEV;
746 			if (!(dev->flags & IFF_UP))
747 				return -ENETDOWN;
748 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
749 			if (addr_type != RTN_UNICAST)
750 				return -EINVAL;
751 			if (!netif_carrier_ok(dev))
752 				nh->nh_flags |= RTNH_F_LINKDOWN;
753 			nh->nh_dev = dev;
754 			dev_hold(dev);
755 			nh->nh_scope = RT_SCOPE_LINK;
756 			return 0;
757 		}
758 		rcu_read_lock();
759 		{
760 			struct fib_table *tbl = NULL;
761 			struct flowi4 fl4 = {
762 				.daddr = nh->nh_gw,
763 				.flowi4_scope = cfg->fc_scope + 1,
764 				.flowi4_oif = nh->nh_oif,
765 				.flowi4_iif = LOOPBACK_IFINDEX,
766 			};
767 
768 			/* It is not necessary, but requires a bit of thinking */
769 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
770 				fl4.flowi4_scope = RT_SCOPE_LINK;
771 
772 			if (cfg->fc_table)
773 				tbl = fib_get_table(net, cfg->fc_table);
774 
775 			if (tbl)
776 				err = fib_table_lookup(tbl, &fl4, &res,
777 						       FIB_LOOKUP_IGNORE_LINKSTATE |
778 						       FIB_LOOKUP_NOREF);
779 
780 			/* on error or if no table given do full lookup. This
781 			 * is needed for example when nexthops are in the local
782 			 * table rather than the given table
783 			 */
784 			if (!tbl || err) {
785 				err = fib_lookup(net, &fl4, &res,
786 						 FIB_LOOKUP_IGNORE_LINKSTATE);
787 			}
788 
789 			if (err) {
790 				rcu_read_unlock();
791 				return err;
792 			}
793 		}
794 		err = -EINVAL;
795 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
796 			goto out;
797 		nh->nh_scope = res.scope;
798 		nh->nh_oif = FIB_RES_OIF(res);
799 		nh->nh_dev = dev = FIB_RES_DEV(res);
800 		if (!dev)
801 			goto out;
802 		dev_hold(dev);
803 		if (!netif_carrier_ok(dev))
804 			nh->nh_flags |= RTNH_F_LINKDOWN;
805 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
806 	} else {
807 		struct in_device *in_dev;
808 
809 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
810 			return -EINVAL;
811 
812 		rcu_read_lock();
813 		err = -ENODEV;
814 		in_dev = inetdev_by_index(net, nh->nh_oif);
815 		if (!in_dev)
816 			goto out;
817 		err = -ENETDOWN;
818 		if (!(in_dev->dev->flags & IFF_UP))
819 			goto out;
820 		nh->nh_dev = in_dev->dev;
821 		dev_hold(nh->nh_dev);
822 		nh->nh_scope = RT_SCOPE_HOST;
823 		if (!netif_carrier_ok(nh->nh_dev))
824 			nh->nh_flags |= RTNH_F_LINKDOWN;
825 		err = 0;
826 	}
827 out:
828 	rcu_read_unlock();
829 	return err;
830 }
831 
832 static inline unsigned int fib_laddr_hashfn(__be32 val)
833 {
834 	unsigned int mask = (fib_info_hash_size - 1);
835 
836 	return ((__force u32)val ^
837 		((__force u32)val >> 7) ^
838 		((__force u32)val >> 14)) & mask;
839 }
840 
841 static struct hlist_head *fib_info_hash_alloc(int bytes)
842 {
843 	if (bytes <= PAGE_SIZE)
844 		return kzalloc(bytes, GFP_KERNEL);
845 	else
846 		return (struct hlist_head *)
847 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
848 					 get_order(bytes));
849 }
850 
851 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
852 {
853 	if (!hash)
854 		return;
855 
856 	if (bytes <= PAGE_SIZE)
857 		kfree(hash);
858 	else
859 		free_pages((unsigned long) hash, get_order(bytes));
860 }
861 
862 static void fib_info_hash_move(struct hlist_head *new_info_hash,
863 			       struct hlist_head *new_laddrhash,
864 			       unsigned int new_size)
865 {
866 	struct hlist_head *old_info_hash, *old_laddrhash;
867 	unsigned int old_size = fib_info_hash_size;
868 	unsigned int i, bytes;
869 
870 	spin_lock_bh(&fib_info_lock);
871 	old_info_hash = fib_info_hash;
872 	old_laddrhash = fib_info_laddrhash;
873 	fib_info_hash_size = new_size;
874 
875 	for (i = 0; i < old_size; i++) {
876 		struct hlist_head *head = &fib_info_hash[i];
877 		struct hlist_node *n;
878 		struct fib_info *fi;
879 
880 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
881 			struct hlist_head *dest;
882 			unsigned int new_hash;
883 
884 			new_hash = fib_info_hashfn(fi);
885 			dest = &new_info_hash[new_hash];
886 			hlist_add_head(&fi->fib_hash, dest);
887 		}
888 	}
889 	fib_info_hash = new_info_hash;
890 
891 	for (i = 0; i < old_size; i++) {
892 		struct hlist_head *lhead = &fib_info_laddrhash[i];
893 		struct hlist_node *n;
894 		struct fib_info *fi;
895 
896 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
897 			struct hlist_head *ldest;
898 			unsigned int new_hash;
899 
900 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
901 			ldest = &new_laddrhash[new_hash];
902 			hlist_add_head(&fi->fib_lhash, ldest);
903 		}
904 	}
905 	fib_info_laddrhash = new_laddrhash;
906 
907 	spin_unlock_bh(&fib_info_lock);
908 
909 	bytes = old_size * sizeof(struct hlist_head *);
910 	fib_info_hash_free(old_info_hash, bytes);
911 	fib_info_hash_free(old_laddrhash, bytes);
912 }
913 
914 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
915 {
916 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
917 					nh->nh_gw,
918 					nh->nh_parent->fib_scope);
919 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
920 
921 	return nh->nh_saddr;
922 }
923 
924 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
925 {
926 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
927 	    fib_prefsrc != cfg->fc_dst) {
928 		u32 tb_id = cfg->fc_table;
929 		int rc;
930 
931 		if (tb_id == RT_TABLE_MAIN)
932 			tb_id = RT_TABLE_LOCAL;
933 
934 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
935 					  fib_prefsrc, tb_id);
936 
937 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
938 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
939 						  fib_prefsrc, RT_TABLE_LOCAL);
940 		}
941 
942 		if (rc != RTN_LOCAL)
943 			return false;
944 	}
945 	return true;
946 }
947 
948 static int
949 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
950 {
951 	bool ecn_ca = false;
952 	struct nlattr *nla;
953 	int remaining;
954 
955 	if (!cfg->fc_mx)
956 		return 0;
957 
958 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
959 		int type = nla_type(nla);
960 		u32 val;
961 
962 		if (!type)
963 			continue;
964 		if (type > RTAX_MAX)
965 			return -EINVAL;
966 
967 		if (type == RTAX_CC_ALGO) {
968 			char tmp[TCP_CA_NAME_MAX];
969 
970 			nla_strlcpy(tmp, nla, sizeof(tmp));
971 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
972 			if (val == TCP_CA_UNSPEC)
973 				return -EINVAL;
974 		} else {
975 			val = nla_get_u32(nla);
976 		}
977 		if (type == RTAX_ADVMSS && val > 65535 - 40)
978 			val = 65535 - 40;
979 		if (type == RTAX_MTU && val > 65535 - 15)
980 			val = 65535 - 15;
981 		if (type == RTAX_HOPLIMIT && val > 255)
982 			val = 255;
983 		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
984 			return -EINVAL;
985 		fi->fib_metrics[type - 1] = val;
986 	}
987 
988 	if (ecn_ca)
989 		fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
990 
991 	return 0;
992 }
993 
994 struct fib_info *fib_create_info(struct fib_config *cfg)
995 {
996 	int err;
997 	struct fib_info *fi = NULL;
998 	struct fib_info *ofi;
999 	int nhs = 1;
1000 	struct net *net = cfg->fc_nlinfo.nl_net;
1001 
1002 	if (cfg->fc_type > RTN_MAX)
1003 		goto err_inval;
1004 
1005 	/* Fast check to catch the most weird cases */
1006 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
1007 		goto err_inval;
1008 
1009 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
1010 		goto err_inval;
1011 
1012 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1013 	if (cfg->fc_mp) {
1014 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1015 		if (nhs == 0)
1016 			goto err_inval;
1017 	}
1018 #endif
1019 
1020 	err = -ENOBUFS;
1021 	if (fib_info_cnt >= fib_info_hash_size) {
1022 		unsigned int new_size = fib_info_hash_size << 1;
1023 		struct hlist_head *new_info_hash;
1024 		struct hlist_head *new_laddrhash;
1025 		unsigned int bytes;
1026 
1027 		if (!new_size)
1028 			new_size = 16;
1029 		bytes = new_size * sizeof(struct hlist_head *);
1030 		new_info_hash = fib_info_hash_alloc(bytes);
1031 		new_laddrhash = fib_info_hash_alloc(bytes);
1032 		if (!new_info_hash || !new_laddrhash) {
1033 			fib_info_hash_free(new_info_hash, bytes);
1034 			fib_info_hash_free(new_laddrhash, bytes);
1035 		} else
1036 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1037 
1038 		if (!fib_info_hash_size)
1039 			goto failure;
1040 	}
1041 
1042 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1043 	if (!fi)
1044 		goto failure;
1045 	fib_info_cnt++;
1046 	if (cfg->fc_mx) {
1047 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1048 		if (!fi->fib_metrics)
1049 			goto failure;
1050 	} else
1051 		fi->fib_metrics = (u32 *) dst_default_metrics;
1052 
1053 	fi->fib_net = net;
1054 	fi->fib_protocol = cfg->fc_protocol;
1055 	fi->fib_scope = cfg->fc_scope;
1056 	fi->fib_flags = cfg->fc_flags;
1057 	fi->fib_priority = cfg->fc_priority;
1058 	fi->fib_prefsrc = cfg->fc_prefsrc;
1059 	fi->fib_type = cfg->fc_type;
1060 
1061 	fi->fib_nhs = nhs;
1062 	change_nexthops(fi) {
1063 		nexthop_nh->nh_parent = fi;
1064 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1065 		if (!nexthop_nh->nh_pcpu_rth_output)
1066 			goto failure;
1067 	} endfor_nexthops(fi)
1068 
1069 	err = fib_convert_metrics(fi, cfg);
1070 	if (err)
1071 		goto failure;
1072 
1073 	if (cfg->fc_mp) {
1074 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1075 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1076 		if (err != 0)
1077 			goto failure;
1078 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1079 			goto err_inval;
1080 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1081 			goto err_inval;
1082 #ifdef CONFIG_IP_ROUTE_CLASSID
1083 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1084 			goto err_inval;
1085 #endif
1086 #else
1087 		goto err_inval;
1088 #endif
1089 	} else {
1090 		struct fib_nh *nh = fi->fib_nh;
1091 
1092 		if (cfg->fc_encap) {
1093 			struct lwtunnel_state *lwtstate;
1094 			struct net_device *dev = NULL;
1095 
1096 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1097 				goto err_inval;
1098 			if (cfg->fc_oif)
1099 				dev = __dev_get_by_index(net, cfg->fc_oif);
1100 			err = lwtunnel_build_state(dev, cfg->fc_encap_type,
1101 						   cfg->fc_encap, AF_INET, cfg,
1102 						   &lwtstate);
1103 			if (err)
1104 				goto failure;
1105 
1106 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1107 		}
1108 		nh->nh_oif = cfg->fc_oif;
1109 		nh->nh_gw = cfg->fc_gw;
1110 		nh->nh_flags = cfg->fc_flags;
1111 #ifdef CONFIG_IP_ROUTE_CLASSID
1112 		nh->nh_tclassid = cfg->fc_flow;
1113 		if (nh->nh_tclassid)
1114 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1115 #endif
1116 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1117 		nh->nh_weight = 1;
1118 #endif
1119 	}
1120 
1121 	if (fib_props[cfg->fc_type].error) {
1122 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1123 			goto err_inval;
1124 		goto link_it;
1125 	} else {
1126 		switch (cfg->fc_type) {
1127 		case RTN_UNICAST:
1128 		case RTN_LOCAL:
1129 		case RTN_BROADCAST:
1130 		case RTN_ANYCAST:
1131 		case RTN_MULTICAST:
1132 			break;
1133 		default:
1134 			goto err_inval;
1135 		}
1136 	}
1137 
1138 	if (cfg->fc_scope > RT_SCOPE_HOST)
1139 		goto err_inval;
1140 
1141 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1142 		struct fib_nh *nh = fi->fib_nh;
1143 
1144 		/* Local address is added. */
1145 		if (nhs != 1 || nh->nh_gw)
1146 			goto err_inval;
1147 		nh->nh_scope = RT_SCOPE_NOWHERE;
1148 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1149 		err = -ENODEV;
1150 		if (!nh->nh_dev)
1151 			goto failure;
1152 	} else {
1153 		int linkdown = 0;
1154 
1155 		change_nexthops(fi) {
1156 			err = fib_check_nh(cfg, fi, nexthop_nh);
1157 			if (err != 0)
1158 				goto failure;
1159 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1160 				linkdown++;
1161 		} endfor_nexthops(fi)
1162 		if (linkdown == fi->fib_nhs)
1163 			fi->fib_flags |= RTNH_F_LINKDOWN;
1164 	}
1165 
1166 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1167 		goto err_inval;
1168 
1169 	change_nexthops(fi) {
1170 		fib_info_update_nh_saddr(net, nexthop_nh);
1171 		fib_add_weight(fi, nexthop_nh);
1172 	} endfor_nexthops(fi)
1173 
1174 	fib_rebalance(fi);
1175 
1176 link_it:
1177 	ofi = fib_find_info(fi);
1178 	if (ofi) {
1179 		fi->fib_dead = 1;
1180 		free_fib_info(fi);
1181 		ofi->fib_treeref++;
1182 		return ofi;
1183 	}
1184 
1185 	fi->fib_treeref++;
1186 	atomic_inc(&fi->fib_clntref);
1187 	spin_lock_bh(&fib_info_lock);
1188 	hlist_add_head(&fi->fib_hash,
1189 		       &fib_info_hash[fib_info_hashfn(fi)]);
1190 	if (fi->fib_prefsrc) {
1191 		struct hlist_head *head;
1192 
1193 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1194 		hlist_add_head(&fi->fib_lhash, head);
1195 	}
1196 	change_nexthops(fi) {
1197 		struct hlist_head *head;
1198 		unsigned int hash;
1199 
1200 		if (!nexthop_nh->nh_dev)
1201 			continue;
1202 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1203 		head = &fib_info_devhash[hash];
1204 		hlist_add_head(&nexthop_nh->nh_hash, head);
1205 	} endfor_nexthops(fi)
1206 	spin_unlock_bh(&fib_info_lock);
1207 	return fi;
1208 
1209 err_inval:
1210 	err = -EINVAL;
1211 
1212 failure:
1213 	if (fi) {
1214 		fi->fib_dead = 1;
1215 		free_fib_info(fi);
1216 	}
1217 
1218 	return ERR_PTR(err);
1219 }
1220 
1221 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1222 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1223 		  struct fib_info *fi, unsigned int flags)
1224 {
1225 	struct nlmsghdr *nlh;
1226 	struct rtmsg *rtm;
1227 
1228 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1229 	if (!nlh)
1230 		return -EMSGSIZE;
1231 
1232 	rtm = nlmsg_data(nlh);
1233 	rtm->rtm_family = AF_INET;
1234 	rtm->rtm_dst_len = dst_len;
1235 	rtm->rtm_src_len = 0;
1236 	rtm->rtm_tos = tos;
1237 	if (tb_id < 256)
1238 		rtm->rtm_table = tb_id;
1239 	else
1240 		rtm->rtm_table = RT_TABLE_COMPAT;
1241 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1242 		goto nla_put_failure;
1243 	rtm->rtm_type = type;
1244 	rtm->rtm_flags = fi->fib_flags;
1245 	rtm->rtm_scope = fi->fib_scope;
1246 	rtm->rtm_protocol = fi->fib_protocol;
1247 
1248 	if (rtm->rtm_dst_len &&
1249 	    nla_put_in_addr(skb, RTA_DST, dst))
1250 		goto nla_put_failure;
1251 	if (fi->fib_priority &&
1252 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1253 		goto nla_put_failure;
1254 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1255 		goto nla_put_failure;
1256 
1257 	if (fi->fib_prefsrc &&
1258 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1259 		goto nla_put_failure;
1260 	if (fi->fib_nhs == 1) {
1261 		struct in_device *in_dev;
1262 
1263 		if (fi->fib_nh->nh_gw &&
1264 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1265 			goto nla_put_failure;
1266 		if (fi->fib_nh->nh_oif &&
1267 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1268 			goto nla_put_failure;
1269 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1270 			in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1271 			if (in_dev &&
1272 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1273 				rtm->rtm_flags |= RTNH_F_DEAD;
1274 		}
1275 #ifdef CONFIG_IP_ROUTE_CLASSID
1276 		if (fi->fib_nh[0].nh_tclassid &&
1277 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1278 			goto nla_put_failure;
1279 #endif
1280 		if (fi->fib_nh->nh_lwtstate)
1281 			lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1282 	}
1283 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1284 	if (fi->fib_nhs > 1) {
1285 		struct rtnexthop *rtnh;
1286 		struct nlattr *mp;
1287 
1288 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1289 		if (!mp)
1290 			goto nla_put_failure;
1291 
1292 		for_nexthops(fi) {
1293 			struct in_device *in_dev;
1294 
1295 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1296 			if (!rtnh)
1297 				goto nla_put_failure;
1298 
1299 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1300 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1301 				in_dev = __in_dev_get_rtnl(nh->nh_dev);
1302 				if (in_dev &&
1303 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1304 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1305 			}
1306 			rtnh->rtnh_hops = nh->nh_weight - 1;
1307 			rtnh->rtnh_ifindex = nh->nh_oif;
1308 
1309 			if (nh->nh_gw &&
1310 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1311 				goto nla_put_failure;
1312 #ifdef CONFIG_IP_ROUTE_CLASSID
1313 			if (nh->nh_tclassid &&
1314 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1315 				goto nla_put_failure;
1316 #endif
1317 			if (nh->nh_lwtstate)
1318 				lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1319 			/* length of rtnetlink header + attributes */
1320 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1321 		} endfor_nexthops(fi);
1322 
1323 		nla_nest_end(skb, mp);
1324 	}
1325 #endif
1326 	nlmsg_end(skb, nlh);
1327 	return 0;
1328 
1329 nla_put_failure:
1330 	nlmsg_cancel(skb, nlh);
1331 	return -EMSGSIZE;
1332 }
1333 
1334 /*
1335  * Update FIB if:
1336  * - local address disappeared -> we must delete all the entries
1337  *   referring to it.
1338  * - device went down -> we must shutdown all nexthops going via it.
1339  */
1340 int fib_sync_down_addr(struct net *net, __be32 local)
1341 {
1342 	int ret = 0;
1343 	unsigned int hash = fib_laddr_hashfn(local);
1344 	struct hlist_head *head = &fib_info_laddrhash[hash];
1345 	struct fib_info *fi;
1346 
1347 	if (!fib_info_laddrhash || local == 0)
1348 		return 0;
1349 
1350 	hlist_for_each_entry(fi, head, fib_lhash) {
1351 		if (!net_eq(fi->fib_net, net))
1352 			continue;
1353 		if (fi->fib_prefsrc == local) {
1354 			fi->fib_flags |= RTNH_F_DEAD;
1355 			ret++;
1356 		}
1357 	}
1358 	return ret;
1359 }
1360 
1361 /* Event              force Flags           Description
1362  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1363  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1364  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1365  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1366  */
1367 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1368 {
1369 	int ret = 0;
1370 	int scope = RT_SCOPE_NOWHERE;
1371 	struct fib_info *prev_fi = NULL;
1372 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1373 	struct hlist_head *head = &fib_info_devhash[hash];
1374 	struct fib_nh *nh;
1375 
1376 	if (force)
1377 		scope = -1;
1378 
1379 	hlist_for_each_entry(nh, head, nh_hash) {
1380 		struct fib_info *fi = nh->nh_parent;
1381 		int dead;
1382 
1383 		BUG_ON(!fi->fib_nhs);
1384 		if (nh->nh_dev != dev || fi == prev_fi)
1385 			continue;
1386 		prev_fi = fi;
1387 		dead = 0;
1388 		change_nexthops(fi) {
1389 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1390 				dead++;
1391 			else if (nexthop_nh->nh_dev == dev &&
1392 				 nexthop_nh->nh_scope != scope) {
1393 				switch (event) {
1394 				case NETDEV_DOWN:
1395 				case NETDEV_UNREGISTER:
1396 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1397 					/* fall through */
1398 				case NETDEV_CHANGE:
1399 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1400 					break;
1401 				}
1402 				dead++;
1403 			}
1404 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1405 			if (event == NETDEV_UNREGISTER &&
1406 			    nexthop_nh->nh_dev == dev) {
1407 				dead = fi->fib_nhs;
1408 				break;
1409 			}
1410 #endif
1411 		} endfor_nexthops(fi)
1412 		if (dead == fi->fib_nhs) {
1413 			switch (event) {
1414 			case NETDEV_DOWN:
1415 			case NETDEV_UNREGISTER:
1416 				fi->fib_flags |= RTNH_F_DEAD;
1417 				/* fall through */
1418 			case NETDEV_CHANGE:
1419 				fi->fib_flags |= RTNH_F_LINKDOWN;
1420 				break;
1421 			}
1422 			ret++;
1423 		}
1424 
1425 		fib_rebalance(fi);
1426 	}
1427 
1428 	return ret;
1429 }
1430 
1431 /* Must be invoked inside of an RCU protected region.  */
1432 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1433 {
1434 	struct fib_info *fi = NULL, *last_resort = NULL;
1435 	struct hlist_head *fa_head = res->fa_head;
1436 	struct fib_table *tb = res->table;
1437 	u8 slen = 32 - res->prefixlen;
1438 	int order = -1, last_idx = -1;
1439 	struct fib_alias *fa, *fa1 = NULL;
1440 	u32 last_prio = res->fi->fib_priority;
1441 	u8 last_tos = 0;
1442 
1443 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1444 		struct fib_info *next_fi = fa->fa_info;
1445 
1446 		if (fa->fa_slen != slen)
1447 			continue;
1448 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1449 			continue;
1450 		if (fa->tb_id != tb->tb_id)
1451 			continue;
1452 		if (next_fi->fib_priority > last_prio &&
1453 		    fa->fa_tos == last_tos) {
1454 			if (last_tos)
1455 				continue;
1456 			break;
1457 		}
1458 		if (next_fi->fib_flags & RTNH_F_DEAD)
1459 			continue;
1460 		last_tos = fa->fa_tos;
1461 		last_prio = next_fi->fib_priority;
1462 
1463 		if (next_fi->fib_scope != res->scope ||
1464 		    fa->fa_type != RTN_UNICAST)
1465 			continue;
1466 		if (!next_fi->fib_nh[0].nh_gw ||
1467 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1468 			continue;
1469 
1470 		fib_alias_accessed(fa);
1471 
1472 		if (!fi) {
1473 			if (next_fi != res->fi)
1474 				break;
1475 			fa1 = fa;
1476 		} else if (!fib_detect_death(fi, order, &last_resort,
1477 					     &last_idx, fa1->fa_default)) {
1478 			fib_result_assign(res, fi);
1479 			fa1->fa_default = order;
1480 			goto out;
1481 		}
1482 		fi = next_fi;
1483 		order++;
1484 	}
1485 
1486 	if (order <= 0 || !fi) {
1487 		if (fa1)
1488 			fa1->fa_default = -1;
1489 		goto out;
1490 	}
1491 
1492 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1493 			      fa1->fa_default)) {
1494 		fib_result_assign(res, fi);
1495 		fa1->fa_default = order;
1496 		goto out;
1497 	}
1498 
1499 	if (last_idx >= 0)
1500 		fib_result_assign(res, last_resort);
1501 	fa1->fa_default = last_idx;
1502 out:
1503 	return;
1504 }
1505 
1506 /*
1507  * Dead device goes up. We wake up dead nexthops.
1508  * It takes sense only on multipath routes.
1509  */
1510 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1511 {
1512 	struct fib_info *prev_fi;
1513 	unsigned int hash;
1514 	struct hlist_head *head;
1515 	struct fib_nh *nh;
1516 	int ret;
1517 
1518 	if (!(dev->flags & IFF_UP))
1519 		return 0;
1520 
1521 	if (nh_flags & RTNH_F_DEAD) {
1522 		unsigned int flags = dev_get_flags(dev);
1523 
1524 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1525 			nh_flags |= RTNH_F_LINKDOWN;
1526 	}
1527 
1528 	prev_fi = NULL;
1529 	hash = fib_devindex_hashfn(dev->ifindex);
1530 	head = &fib_info_devhash[hash];
1531 	ret = 0;
1532 
1533 	hlist_for_each_entry(nh, head, nh_hash) {
1534 		struct fib_info *fi = nh->nh_parent;
1535 		int alive;
1536 
1537 		BUG_ON(!fi->fib_nhs);
1538 		if (nh->nh_dev != dev || fi == prev_fi)
1539 			continue;
1540 
1541 		prev_fi = fi;
1542 		alive = 0;
1543 		change_nexthops(fi) {
1544 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1545 				alive++;
1546 				continue;
1547 			}
1548 			if (!nexthop_nh->nh_dev ||
1549 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1550 				continue;
1551 			if (nexthop_nh->nh_dev != dev ||
1552 			    !__in_dev_get_rtnl(dev))
1553 				continue;
1554 			alive++;
1555 			nexthop_nh->nh_flags &= ~nh_flags;
1556 		} endfor_nexthops(fi)
1557 
1558 		if (alive > 0) {
1559 			fi->fib_flags &= ~nh_flags;
1560 			ret++;
1561 		}
1562 
1563 		fib_rebalance(fi);
1564 	}
1565 
1566 	return ret;
1567 }
1568 
1569 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1570 static bool fib_good_nh(const struct fib_nh *nh)
1571 {
1572 	int state = NUD_REACHABLE;
1573 
1574 	if (nh->nh_scope == RT_SCOPE_LINK) {
1575 		struct neighbour *n;
1576 
1577 		rcu_read_lock_bh();
1578 
1579 		n = __ipv4_neigh_lookup_noref(nh->nh_dev, nh->nh_gw);
1580 		if (n)
1581 			state = n->nud_state;
1582 
1583 		rcu_read_unlock_bh();
1584 	}
1585 
1586 	return !!(state & NUD_VALID);
1587 }
1588 
1589 void fib_select_multipath(struct fib_result *res, int hash)
1590 {
1591 	struct fib_info *fi = res->fi;
1592 	struct net *net = fi->fib_net;
1593 	bool first = false;
1594 
1595 	for_nexthops(fi) {
1596 		if (hash > atomic_read(&nh->nh_upper_bound))
1597 			continue;
1598 
1599 		if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1600 		    fib_good_nh(nh)) {
1601 			res->nh_sel = nhsel;
1602 			return;
1603 		}
1604 		if (!first) {
1605 			res->nh_sel = nhsel;
1606 			first = true;
1607 		}
1608 	} endfor_nexthops(fi);
1609 }
1610 #endif
1611 
1612 void fib_select_path(struct net *net, struct fib_result *res,
1613 		     struct flowi4 *fl4, int mp_hash)
1614 {
1615 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1616 	if (res->fi->fib_nhs > 1 && fl4->flowi4_oif == 0) {
1617 		if (mp_hash < 0)
1618 			mp_hash = get_hash_from_flowi4(fl4) >> 1;
1619 
1620 		fib_select_multipath(res, mp_hash);
1621 	}
1622 	else
1623 #endif
1624 	if (!res->prefixlen &&
1625 	    res->table->tb_num_default > 1 &&
1626 	    res->type == RTN_UNICAST && !fl4->flowi4_oif)
1627 		fib_select_default(fl4, res);
1628 
1629 	if (!fl4->saddr)
1630 		fl4->saddr = FIB_RES_PREFSRC(net, *res);
1631 }
1632 EXPORT_SYMBOL_GPL(fib_select_path);
1633