xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision eb3fcf00)
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 
61 static DEFINE_SPINLOCK(fib_multipath_lock);
62 
63 #define for_nexthops(fi) {						\
64 	int nhsel; const struct fib_nh *nh;				\
65 	for (nhsel = 0, nh = (fi)->fib_nh;				\
66 	     nhsel < (fi)->fib_nhs;					\
67 	     nh++, nhsel++)
68 
69 #define change_nexthops(fi) {						\
70 	int nhsel; struct fib_nh *nexthop_nh;				\
71 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
72 	     nhsel < (fi)->fib_nhs;					\
73 	     nexthop_nh++, nhsel++)
74 
75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
76 
77 /* Hope, that gcc will optimize it to get rid of dummy loop */
78 
79 #define for_nexthops(fi) {						\
80 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
81 	for (nhsel = 0; nhsel < 1; nhsel++)
82 
83 #define change_nexthops(fi) {						\
84 	int nhsel;							\
85 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
86 	for (nhsel = 0; nhsel < 1; nhsel++)
87 
88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
89 
90 #define endfor_nexthops(fi) }
91 
92 
93 const struct fib_prop fib_props[RTN_MAX + 1] = {
94 	[RTN_UNSPEC] = {
95 		.error	= 0,
96 		.scope	= RT_SCOPE_NOWHERE,
97 	},
98 	[RTN_UNICAST] = {
99 		.error	= 0,
100 		.scope	= RT_SCOPE_UNIVERSE,
101 	},
102 	[RTN_LOCAL] = {
103 		.error	= 0,
104 		.scope	= RT_SCOPE_HOST,
105 	},
106 	[RTN_BROADCAST] = {
107 		.error	= 0,
108 		.scope	= RT_SCOPE_LINK,
109 	},
110 	[RTN_ANYCAST] = {
111 		.error	= 0,
112 		.scope	= RT_SCOPE_LINK,
113 	},
114 	[RTN_MULTICAST] = {
115 		.error	= 0,
116 		.scope	= RT_SCOPE_UNIVERSE,
117 	},
118 	[RTN_BLACKHOLE] = {
119 		.error	= -EINVAL,
120 		.scope	= RT_SCOPE_UNIVERSE,
121 	},
122 	[RTN_UNREACHABLE] = {
123 		.error	= -EHOSTUNREACH,
124 		.scope	= RT_SCOPE_UNIVERSE,
125 	},
126 	[RTN_PROHIBIT] = {
127 		.error	= -EACCES,
128 		.scope	= RT_SCOPE_UNIVERSE,
129 	},
130 	[RTN_THROW] = {
131 		.error	= -EAGAIN,
132 		.scope	= RT_SCOPE_UNIVERSE,
133 	},
134 	[RTN_NAT] = {
135 		.error	= -EINVAL,
136 		.scope	= RT_SCOPE_NOWHERE,
137 	},
138 	[RTN_XRESOLVE] = {
139 		.error	= -EINVAL,
140 		.scope	= RT_SCOPE_NOWHERE,
141 	},
142 };
143 
144 static void rt_fibinfo_free(struct rtable __rcu **rtp)
145 {
146 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
147 
148 	if (!rt)
149 		return;
150 
151 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
152 	 * because we waited an RCU grace period before calling
153 	 * free_fib_info_rcu()
154 	 */
155 
156 	dst_free(&rt->dst);
157 }
158 
159 static void free_nh_exceptions(struct fib_nh *nh)
160 {
161 	struct fnhe_hash_bucket *hash;
162 	int i;
163 
164 	hash = rcu_dereference_protected(nh->nh_exceptions, 1);
165 	if (!hash)
166 		return;
167 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
168 		struct fib_nh_exception *fnhe;
169 
170 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
171 		while (fnhe) {
172 			struct fib_nh_exception *next;
173 
174 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
175 
176 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
177 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
178 
179 			kfree(fnhe);
180 
181 			fnhe = next;
182 		}
183 	}
184 	kfree(hash);
185 }
186 
187 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
188 {
189 	int cpu;
190 
191 	if (!rtp)
192 		return;
193 
194 	for_each_possible_cpu(cpu) {
195 		struct rtable *rt;
196 
197 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
198 		if (rt)
199 			dst_free(&rt->dst);
200 	}
201 	free_percpu(rtp);
202 }
203 
204 /* Release a nexthop info record */
205 static void free_fib_info_rcu(struct rcu_head *head)
206 {
207 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
208 
209 	change_nexthops(fi) {
210 		if (nexthop_nh->nh_dev)
211 			dev_put(nexthop_nh->nh_dev);
212 		lwtstate_put(nexthop_nh->nh_lwtstate);
213 		free_nh_exceptions(nexthop_nh);
214 		rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
215 		rt_fibinfo_free(&nexthop_nh->nh_rth_input);
216 	} endfor_nexthops(fi);
217 
218 	if (fi->fib_metrics != (u32 *) dst_default_metrics)
219 		kfree(fi->fib_metrics);
220 	kfree(fi);
221 }
222 
223 void free_fib_info(struct fib_info *fi)
224 {
225 	if (fi->fib_dead == 0) {
226 		pr_warn("Freeing alive fib_info %p\n", fi);
227 		return;
228 	}
229 	fib_info_cnt--;
230 #ifdef CONFIG_IP_ROUTE_CLASSID
231 	change_nexthops(fi) {
232 		if (nexthop_nh->nh_tclassid)
233 			fi->fib_net->ipv4.fib_num_tclassid_users--;
234 	} endfor_nexthops(fi);
235 #endif
236 	call_rcu(&fi->rcu, free_fib_info_rcu);
237 }
238 
239 void fib_release_info(struct fib_info *fi)
240 {
241 	spin_lock_bh(&fib_info_lock);
242 	if (fi && --fi->fib_treeref == 0) {
243 		hlist_del(&fi->fib_hash);
244 		if (fi->fib_prefsrc)
245 			hlist_del(&fi->fib_lhash);
246 		change_nexthops(fi) {
247 			if (!nexthop_nh->nh_dev)
248 				continue;
249 			hlist_del(&nexthop_nh->nh_hash);
250 		} endfor_nexthops(fi)
251 		fi->fib_dead = 1;
252 		fib_info_put(fi);
253 	}
254 	spin_unlock_bh(&fib_info_lock);
255 }
256 
257 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
258 {
259 	const struct fib_nh *onh = ofi->fib_nh;
260 
261 	for_nexthops(fi) {
262 		if (nh->nh_oif != onh->nh_oif ||
263 		    nh->nh_gw  != onh->nh_gw ||
264 		    nh->nh_scope != onh->nh_scope ||
265 #ifdef CONFIG_IP_ROUTE_MULTIPATH
266 		    nh->nh_weight != onh->nh_weight ||
267 #endif
268 #ifdef CONFIG_IP_ROUTE_CLASSID
269 		    nh->nh_tclassid != onh->nh_tclassid ||
270 #endif
271 		    lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
272 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
273 			return -1;
274 		onh++;
275 	} endfor_nexthops(fi);
276 	return 0;
277 }
278 
279 static inline unsigned int fib_devindex_hashfn(unsigned int val)
280 {
281 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
282 
283 	return (val ^
284 		(val >> DEVINDEX_HASHBITS) ^
285 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
286 }
287 
288 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
289 {
290 	unsigned int mask = (fib_info_hash_size - 1);
291 	unsigned int val = fi->fib_nhs;
292 
293 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
294 	val ^= (__force u32)fi->fib_prefsrc;
295 	val ^= fi->fib_priority;
296 	for_nexthops(fi) {
297 		val ^= fib_devindex_hashfn(nh->nh_oif);
298 	} endfor_nexthops(fi)
299 
300 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
301 }
302 
303 static struct fib_info *fib_find_info(const struct fib_info *nfi)
304 {
305 	struct hlist_head *head;
306 	struct fib_info *fi;
307 	unsigned int hash;
308 
309 	hash = fib_info_hashfn(nfi);
310 	head = &fib_info_hash[hash];
311 
312 	hlist_for_each_entry(fi, head, fib_hash) {
313 		if (!net_eq(fi->fib_net, nfi->fib_net))
314 			continue;
315 		if (fi->fib_nhs != nfi->fib_nhs)
316 			continue;
317 		if (nfi->fib_protocol == fi->fib_protocol &&
318 		    nfi->fib_scope == fi->fib_scope &&
319 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
320 		    nfi->fib_priority == fi->fib_priority &&
321 		    nfi->fib_type == fi->fib_type &&
322 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
323 			   sizeof(u32) * RTAX_MAX) == 0 &&
324 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
325 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
326 			return fi;
327 	}
328 
329 	return NULL;
330 }
331 
332 /* Check, that the gateway is already configured.
333  * Used only by redirect accept routine.
334  */
335 int ip_fib_check_default(__be32 gw, struct net_device *dev)
336 {
337 	struct hlist_head *head;
338 	struct fib_nh *nh;
339 	unsigned int hash;
340 
341 	spin_lock(&fib_info_lock);
342 
343 	hash = fib_devindex_hashfn(dev->ifindex);
344 	head = &fib_info_devhash[hash];
345 	hlist_for_each_entry(nh, head, nh_hash) {
346 		if (nh->nh_dev == dev &&
347 		    nh->nh_gw == gw &&
348 		    !(nh->nh_flags & RTNH_F_DEAD)) {
349 			spin_unlock(&fib_info_lock);
350 			return 0;
351 		}
352 	}
353 
354 	spin_unlock(&fib_info_lock);
355 
356 	return -1;
357 }
358 
359 static inline size_t fib_nlmsg_size(struct fib_info *fi)
360 {
361 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
362 			 + nla_total_size(4) /* RTA_TABLE */
363 			 + nla_total_size(4) /* RTA_DST */
364 			 + nla_total_size(4) /* RTA_PRIORITY */
365 			 + nla_total_size(4) /* RTA_PREFSRC */
366 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
367 
368 	/* space for nested metrics */
369 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
370 
371 	if (fi->fib_nhs) {
372 		size_t nh_encapsize = 0;
373 		/* Also handles the special case fib_nhs == 1 */
374 
375 		/* each nexthop is packed in an attribute */
376 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
377 
378 		/* may contain flow and gateway attribute */
379 		nhsize += 2 * nla_total_size(4);
380 
381 		/* grab encap info */
382 		for_nexthops(fi) {
383 			if (nh->nh_lwtstate) {
384 				/* RTA_ENCAP_TYPE */
385 				nh_encapsize += lwtunnel_get_encap_size(
386 						nh->nh_lwtstate);
387 				/* RTA_ENCAP */
388 				nh_encapsize +=  nla_total_size(2);
389 			}
390 		} endfor_nexthops(fi);
391 
392 		/* all nexthops are packed in a nested attribute */
393 		payload += nla_total_size((fi->fib_nhs * nhsize) +
394 					  nh_encapsize);
395 
396 	}
397 
398 	return payload;
399 }
400 
401 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
402 	       int dst_len, u32 tb_id, const struct nl_info *info,
403 	       unsigned int nlm_flags)
404 {
405 	struct sk_buff *skb;
406 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
407 	int err = -ENOBUFS;
408 
409 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
410 	if (!skb)
411 		goto errout;
412 
413 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
414 			    fa->fa_type, key, dst_len,
415 			    fa->fa_tos, fa->fa_info, nlm_flags);
416 	if (err < 0) {
417 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
418 		WARN_ON(err == -EMSGSIZE);
419 		kfree_skb(skb);
420 		goto errout;
421 	}
422 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
423 		    info->nlh, GFP_KERNEL);
424 	return;
425 errout:
426 	if (err < 0)
427 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
428 }
429 
430 static int fib_detect_death(struct fib_info *fi, int order,
431 			    struct fib_info **last_resort, int *last_idx,
432 			    int dflt)
433 {
434 	struct neighbour *n;
435 	int state = NUD_NONE;
436 
437 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
438 	if (n) {
439 		state = n->nud_state;
440 		neigh_release(n);
441 	} else {
442 		return 0;
443 	}
444 	if (state == NUD_REACHABLE)
445 		return 0;
446 	if ((state & NUD_VALID) && order != dflt)
447 		return 0;
448 	if ((state & NUD_VALID) ||
449 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
450 		*last_resort = fi;
451 		*last_idx = order;
452 	}
453 	return 1;
454 }
455 
456 #ifdef CONFIG_IP_ROUTE_MULTIPATH
457 
458 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
459 {
460 	int nhs = 0;
461 
462 	while (rtnh_ok(rtnh, remaining)) {
463 		nhs++;
464 		rtnh = rtnh_next(rtnh, &remaining);
465 	}
466 
467 	/* leftover implies invalid nexthop configuration, discard it */
468 	return remaining > 0 ? 0 : nhs;
469 }
470 
471 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
472 		       int remaining, struct fib_config *cfg)
473 {
474 	struct net *net = cfg->fc_nlinfo.nl_net;
475 	int ret;
476 
477 	change_nexthops(fi) {
478 		int attrlen;
479 
480 		if (!rtnh_ok(rtnh, remaining))
481 			return -EINVAL;
482 
483 		nexthop_nh->nh_flags =
484 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
485 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
486 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
487 
488 		attrlen = rtnh_attrlen(rtnh);
489 		if (attrlen > 0) {
490 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
491 
492 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
493 			nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
494 #ifdef CONFIG_IP_ROUTE_CLASSID
495 			nla = nla_find(attrs, attrlen, RTA_FLOW);
496 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
497 			if (nexthop_nh->nh_tclassid)
498 				fi->fib_net->ipv4.fib_num_tclassid_users++;
499 #endif
500 			nla = nla_find(attrs, attrlen, RTA_ENCAP);
501 			if (nla) {
502 				struct lwtunnel_state *lwtstate;
503 				struct net_device *dev = NULL;
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 				if (cfg->fc_oif)
511 					dev = __dev_get_by_index(net, cfg->fc_oif);
512 				ret = lwtunnel_build_state(dev, nla_get_u16(
513 							   nla_entype),
514 							   nla,  AF_INET, cfg,
515 							   &lwtstate);
516 				if (ret)
517 					goto errout;
518 				nexthop_nh->nh_lwtstate =
519 					lwtstate_get(lwtstate);
520 			}
521 		}
522 
523 		rtnh = rtnh_next(rtnh, &remaining);
524 	} endfor_nexthops(fi);
525 
526 	return 0;
527 
528 err_inval:
529 	ret = -EINVAL;
530 
531 errout:
532 	return ret;
533 }
534 
535 #endif
536 
537 static int fib_encap_match(struct net *net, u16 encap_type,
538 			   struct nlattr *encap,
539 			   int oif, const struct fib_nh *nh,
540 			   const struct fib_config *cfg)
541 {
542 	struct lwtunnel_state *lwtstate;
543 	struct net_device *dev = NULL;
544 	int ret, result = 0;
545 
546 	if (encap_type == LWTUNNEL_ENCAP_NONE)
547 		return 0;
548 
549 	if (oif)
550 		dev = __dev_get_by_index(net, oif);
551 	ret = lwtunnel_build_state(dev, encap_type, encap,
552 				   AF_INET, cfg, &lwtstate);
553 	if (!ret) {
554 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
555 		lwtstate_free(lwtstate);
556 	}
557 
558 	return result;
559 }
560 
561 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
562 {
563 	struct net *net = cfg->fc_nlinfo.nl_net;
564 #ifdef CONFIG_IP_ROUTE_MULTIPATH
565 	struct rtnexthop *rtnh;
566 	int remaining;
567 #endif
568 
569 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
570 		return 1;
571 
572 	if (cfg->fc_oif || cfg->fc_gw) {
573 		if (cfg->fc_encap) {
574 			if (fib_encap_match(net, cfg->fc_encap_type,
575 					    cfg->fc_encap, cfg->fc_oif,
576 					    fi->fib_nh, cfg))
577 			    return 1;
578 		}
579 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
580 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
581 			return 0;
582 		return 1;
583 	}
584 
585 #ifdef CONFIG_IP_ROUTE_MULTIPATH
586 	if (!cfg->fc_mp)
587 		return 0;
588 
589 	rtnh = cfg->fc_mp;
590 	remaining = cfg->fc_mp_len;
591 
592 	for_nexthops(fi) {
593 		int attrlen;
594 
595 		if (!rtnh_ok(rtnh, remaining))
596 			return -EINVAL;
597 
598 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
599 			return 1;
600 
601 		attrlen = rtnh_attrlen(rtnh);
602 		if (attrlen > 0) {
603 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
604 
605 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
606 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
607 				return 1;
608 #ifdef CONFIG_IP_ROUTE_CLASSID
609 			nla = nla_find(attrs, attrlen, RTA_FLOW);
610 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
611 				return 1;
612 #endif
613 		}
614 
615 		rtnh = rtnh_next(rtnh, &remaining);
616 	} endfor_nexthops(fi);
617 #endif
618 	return 0;
619 }
620 
621 
622 /*
623  * Picture
624  * -------
625  *
626  * Semantics of nexthop is very messy by historical reasons.
627  * We have to take into account, that:
628  * a) gateway can be actually local interface address,
629  *    so that gatewayed route is direct.
630  * b) gateway must be on-link address, possibly
631  *    described not by an ifaddr, but also by a direct route.
632  * c) If both gateway and interface are specified, they should not
633  *    contradict.
634  * d) If we use tunnel routes, gateway could be not on-link.
635  *
636  * Attempt to reconcile all of these (alas, self-contradictory) conditions
637  * results in pretty ugly and hairy code with obscure logic.
638  *
639  * I chose to generalized it instead, so that the size
640  * of code does not increase practically, but it becomes
641  * much more general.
642  * Every prefix is assigned a "scope" value: "host" is local address,
643  * "link" is direct route,
644  * [ ... "site" ... "interior" ... ]
645  * and "universe" is true gateway route with global meaning.
646  *
647  * Every prefix refers to a set of "nexthop"s (gw, oif),
648  * where gw must have narrower scope. This recursion stops
649  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
650  * which means that gw is forced to be on link.
651  *
652  * Code is still hairy, but now it is apparently logically
653  * consistent and very flexible. F.e. as by-product it allows
654  * to co-exists in peace independent exterior and interior
655  * routing processes.
656  *
657  * Normally it looks as following.
658  *
659  * {universe prefix}  -> (gw, oif) [scope link]
660  *		  |
661  *		  |-> {link prefix} -> (gw, oif) [scope local]
662  *					|
663  *					|-> {local prefix} (terminal node)
664  */
665 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
666 			struct fib_nh *nh)
667 {
668 	int err = 0;
669 	struct net *net;
670 	struct net_device *dev;
671 
672 	net = cfg->fc_nlinfo.nl_net;
673 	if (nh->nh_gw) {
674 		struct fib_result res;
675 
676 		if (nh->nh_flags & RTNH_F_ONLINK) {
677 			unsigned int addr_type;
678 
679 			if (cfg->fc_scope >= RT_SCOPE_LINK)
680 				return -EINVAL;
681 			dev = __dev_get_by_index(net, nh->nh_oif);
682 			if (!dev)
683 				return -ENODEV;
684 			if (!(dev->flags & IFF_UP))
685 				return -ENETDOWN;
686 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
687 			if (addr_type != RTN_UNICAST)
688 				return -EINVAL;
689 			if (!netif_carrier_ok(dev))
690 				nh->nh_flags |= RTNH_F_LINKDOWN;
691 			nh->nh_dev = dev;
692 			dev_hold(dev);
693 			nh->nh_scope = RT_SCOPE_LINK;
694 			return 0;
695 		}
696 		rcu_read_lock();
697 		{
698 			struct fib_table *tbl = NULL;
699 			struct flowi4 fl4 = {
700 				.daddr = nh->nh_gw,
701 				.flowi4_scope = cfg->fc_scope + 1,
702 				.flowi4_oif = nh->nh_oif,
703 				.flowi4_iif = LOOPBACK_IFINDEX,
704 			};
705 
706 			/* It is not necessary, but requires a bit of thinking */
707 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
708 				fl4.flowi4_scope = RT_SCOPE_LINK;
709 
710 			if (cfg->fc_table)
711 				tbl = fib_get_table(net, cfg->fc_table);
712 
713 			if (tbl)
714 				err = fib_table_lookup(tbl, &fl4, &res,
715 						       FIB_LOOKUP_IGNORE_LINKSTATE |
716 						       FIB_LOOKUP_NOREF);
717 
718 			/* on error or if no table given do full lookup. This
719 			 * is needed for example when nexthops are in the local
720 			 * table rather than the given table
721 			 */
722 			if (!tbl || err) {
723 				err = fib_lookup(net, &fl4, &res,
724 						 FIB_LOOKUP_IGNORE_LINKSTATE);
725 			}
726 
727 			if (err) {
728 				rcu_read_unlock();
729 				return err;
730 			}
731 		}
732 		err = -EINVAL;
733 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
734 			goto out;
735 		nh->nh_scope = res.scope;
736 		nh->nh_oif = FIB_RES_OIF(res);
737 		nh->nh_dev = dev = FIB_RES_DEV(res);
738 		if (!dev)
739 			goto out;
740 		dev_hold(dev);
741 		if (!netif_carrier_ok(dev))
742 			nh->nh_flags |= RTNH_F_LINKDOWN;
743 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
744 	} else {
745 		struct in_device *in_dev;
746 
747 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
748 			return -EINVAL;
749 
750 		rcu_read_lock();
751 		err = -ENODEV;
752 		in_dev = inetdev_by_index(net, nh->nh_oif);
753 		if (!in_dev)
754 			goto out;
755 		err = -ENETDOWN;
756 		if (!(in_dev->dev->flags & IFF_UP))
757 			goto out;
758 		nh->nh_dev = in_dev->dev;
759 		dev_hold(nh->nh_dev);
760 		nh->nh_scope = RT_SCOPE_HOST;
761 		if (!netif_carrier_ok(nh->nh_dev))
762 			nh->nh_flags |= RTNH_F_LINKDOWN;
763 		err = 0;
764 	}
765 out:
766 	rcu_read_unlock();
767 	return err;
768 }
769 
770 static inline unsigned int fib_laddr_hashfn(__be32 val)
771 {
772 	unsigned int mask = (fib_info_hash_size - 1);
773 
774 	return ((__force u32)val ^
775 		((__force u32)val >> 7) ^
776 		((__force u32)val >> 14)) & mask;
777 }
778 
779 static struct hlist_head *fib_info_hash_alloc(int bytes)
780 {
781 	if (bytes <= PAGE_SIZE)
782 		return kzalloc(bytes, GFP_KERNEL);
783 	else
784 		return (struct hlist_head *)
785 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
786 					 get_order(bytes));
787 }
788 
789 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
790 {
791 	if (!hash)
792 		return;
793 
794 	if (bytes <= PAGE_SIZE)
795 		kfree(hash);
796 	else
797 		free_pages((unsigned long) hash, get_order(bytes));
798 }
799 
800 static void fib_info_hash_move(struct hlist_head *new_info_hash,
801 			       struct hlist_head *new_laddrhash,
802 			       unsigned int new_size)
803 {
804 	struct hlist_head *old_info_hash, *old_laddrhash;
805 	unsigned int old_size = fib_info_hash_size;
806 	unsigned int i, bytes;
807 
808 	spin_lock_bh(&fib_info_lock);
809 	old_info_hash = fib_info_hash;
810 	old_laddrhash = fib_info_laddrhash;
811 	fib_info_hash_size = new_size;
812 
813 	for (i = 0; i < old_size; i++) {
814 		struct hlist_head *head = &fib_info_hash[i];
815 		struct hlist_node *n;
816 		struct fib_info *fi;
817 
818 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
819 			struct hlist_head *dest;
820 			unsigned int new_hash;
821 
822 			new_hash = fib_info_hashfn(fi);
823 			dest = &new_info_hash[new_hash];
824 			hlist_add_head(&fi->fib_hash, dest);
825 		}
826 	}
827 	fib_info_hash = new_info_hash;
828 
829 	for (i = 0; i < old_size; i++) {
830 		struct hlist_head *lhead = &fib_info_laddrhash[i];
831 		struct hlist_node *n;
832 		struct fib_info *fi;
833 
834 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
835 			struct hlist_head *ldest;
836 			unsigned int new_hash;
837 
838 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
839 			ldest = &new_laddrhash[new_hash];
840 			hlist_add_head(&fi->fib_lhash, ldest);
841 		}
842 	}
843 	fib_info_laddrhash = new_laddrhash;
844 
845 	spin_unlock_bh(&fib_info_lock);
846 
847 	bytes = old_size * sizeof(struct hlist_head *);
848 	fib_info_hash_free(old_info_hash, bytes);
849 	fib_info_hash_free(old_laddrhash, bytes);
850 }
851 
852 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
853 {
854 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
855 					nh->nh_gw,
856 					nh->nh_parent->fib_scope);
857 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
858 
859 	return nh->nh_saddr;
860 }
861 
862 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
863 {
864 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
865 	    fib_prefsrc != cfg->fc_dst) {
866 		u32 tb_id = cfg->fc_table;
867 
868 		if (tb_id == RT_TABLE_MAIN)
869 			tb_id = RT_TABLE_LOCAL;
870 
871 		if (inet_addr_type_table(cfg->fc_nlinfo.nl_net,
872 					 fib_prefsrc, tb_id) != RTN_LOCAL) {
873 			return false;
874 		}
875 	}
876 	return true;
877 }
878 
879 static int
880 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
881 {
882 	bool ecn_ca = false;
883 	struct nlattr *nla;
884 	int remaining;
885 
886 	if (!cfg->fc_mx)
887 		return 0;
888 
889 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
890 		int type = nla_type(nla);
891 		u32 val;
892 
893 		if (!type)
894 			continue;
895 		if (type > RTAX_MAX)
896 			return -EINVAL;
897 
898 		if (type == RTAX_CC_ALGO) {
899 			char tmp[TCP_CA_NAME_MAX];
900 
901 			nla_strlcpy(tmp, nla, sizeof(tmp));
902 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
903 			if (val == TCP_CA_UNSPEC)
904 				return -EINVAL;
905 		} else {
906 			val = nla_get_u32(nla);
907 		}
908 		if (type == RTAX_ADVMSS && val > 65535 - 40)
909 			val = 65535 - 40;
910 		if (type == RTAX_MTU && val > 65535 - 15)
911 			val = 65535 - 15;
912 		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
913 			return -EINVAL;
914 		fi->fib_metrics[type - 1] = val;
915 	}
916 
917 	if (ecn_ca)
918 		fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
919 
920 	return 0;
921 }
922 
923 struct fib_info *fib_create_info(struct fib_config *cfg)
924 {
925 	int err;
926 	struct fib_info *fi = NULL;
927 	struct fib_info *ofi;
928 	int nhs = 1;
929 	struct net *net = cfg->fc_nlinfo.nl_net;
930 
931 	if (cfg->fc_type > RTN_MAX)
932 		goto err_inval;
933 
934 	/* Fast check to catch the most weird cases */
935 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
936 		goto err_inval;
937 
938 #ifdef CONFIG_IP_ROUTE_MULTIPATH
939 	if (cfg->fc_mp) {
940 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
941 		if (nhs == 0)
942 			goto err_inval;
943 	}
944 #endif
945 
946 	err = -ENOBUFS;
947 	if (fib_info_cnt >= fib_info_hash_size) {
948 		unsigned int new_size = fib_info_hash_size << 1;
949 		struct hlist_head *new_info_hash;
950 		struct hlist_head *new_laddrhash;
951 		unsigned int bytes;
952 
953 		if (!new_size)
954 			new_size = 16;
955 		bytes = new_size * sizeof(struct hlist_head *);
956 		new_info_hash = fib_info_hash_alloc(bytes);
957 		new_laddrhash = fib_info_hash_alloc(bytes);
958 		if (!new_info_hash || !new_laddrhash) {
959 			fib_info_hash_free(new_info_hash, bytes);
960 			fib_info_hash_free(new_laddrhash, bytes);
961 		} else
962 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
963 
964 		if (!fib_info_hash_size)
965 			goto failure;
966 	}
967 
968 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
969 	if (!fi)
970 		goto failure;
971 	fib_info_cnt++;
972 	if (cfg->fc_mx) {
973 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
974 		if (!fi->fib_metrics)
975 			goto failure;
976 	} else
977 		fi->fib_metrics = (u32 *) dst_default_metrics;
978 
979 	fi->fib_net = net;
980 	fi->fib_protocol = cfg->fc_protocol;
981 	fi->fib_scope = cfg->fc_scope;
982 	fi->fib_flags = cfg->fc_flags;
983 	fi->fib_priority = cfg->fc_priority;
984 	fi->fib_prefsrc = cfg->fc_prefsrc;
985 	fi->fib_type = cfg->fc_type;
986 
987 	fi->fib_nhs = nhs;
988 	change_nexthops(fi) {
989 		nexthop_nh->nh_parent = fi;
990 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
991 		if (!nexthop_nh->nh_pcpu_rth_output)
992 			goto failure;
993 	} endfor_nexthops(fi)
994 
995 	err = fib_convert_metrics(fi, cfg);
996 	if (err)
997 		goto failure;
998 
999 	if (cfg->fc_mp) {
1000 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1001 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1002 		if (err != 0)
1003 			goto failure;
1004 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1005 			goto err_inval;
1006 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1007 			goto err_inval;
1008 #ifdef CONFIG_IP_ROUTE_CLASSID
1009 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1010 			goto err_inval;
1011 #endif
1012 #else
1013 		goto err_inval;
1014 #endif
1015 	} else {
1016 		struct fib_nh *nh = fi->fib_nh;
1017 
1018 		if (cfg->fc_encap) {
1019 			struct lwtunnel_state *lwtstate;
1020 			struct net_device *dev = NULL;
1021 
1022 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1023 				goto err_inval;
1024 			if (cfg->fc_oif)
1025 				dev = __dev_get_by_index(net, cfg->fc_oif);
1026 			err = lwtunnel_build_state(dev, cfg->fc_encap_type,
1027 						   cfg->fc_encap, AF_INET, cfg,
1028 						   &lwtstate);
1029 			if (err)
1030 				goto failure;
1031 
1032 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1033 		}
1034 		nh->nh_oif = cfg->fc_oif;
1035 		nh->nh_gw = cfg->fc_gw;
1036 		nh->nh_flags = cfg->fc_flags;
1037 #ifdef CONFIG_IP_ROUTE_CLASSID
1038 		nh->nh_tclassid = cfg->fc_flow;
1039 		if (nh->nh_tclassid)
1040 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1041 #endif
1042 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1043 		nh->nh_weight = 1;
1044 #endif
1045 	}
1046 
1047 	if (fib_props[cfg->fc_type].error) {
1048 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1049 			goto err_inval;
1050 		goto link_it;
1051 	} else {
1052 		switch (cfg->fc_type) {
1053 		case RTN_UNICAST:
1054 		case RTN_LOCAL:
1055 		case RTN_BROADCAST:
1056 		case RTN_ANYCAST:
1057 		case RTN_MULTICAST:
1058 			break;
1059 		default:
1060 			goto err_inval;
1061 		}
1062 	}
1063 
1064 	if (cfg->fc_scope > RT_SCOPE_HOST)
1065 		goto err_inval;
1066 
1067 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1068 		struct fib_nh *nh = fi->fib_nh;
1069 
1070 		/* Local address is added. */
1071 		if (nhs != 1 || nh->nh_gw)
1072 			goto err_inval;
1073 		nh->nh_scope = RT_SCOPE_NOWHERE;
1074 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1075 		err = -ENODEV;
1076 		if (!nh->nh_dev)
1077 			goto failure;
1078 	} else {
1079 		int linkdown = 0;
1080 
1081 		change_nexthops(fi) {
1082 			err = fib_check_nh(cfg, fi, nexthop_nh);
1083 			if (err != 0)
1084 				goto failure;
1085 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1086 				linkdown++;
1087 		} endfor_nexthops(fi)
1088 		if (linkdown == fi->fib_nhs)
1089 			fi->fib_flags |= RTNH_F_LINKDOWN;
1090 	}
1091 
1092 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1093 		goto err_inval;
1094 
1095 	change_nexthops(fi) {
1096 		fib_info_update_nh_saddr(net, nexthop_nh);
1097 	} endfor_nexthops(fi)
1098 
1099 link_it:
1100 	ofi = fib_find_info(fi);
1101 	if (ofi) {
1102 		fi->fib_dead = 1;
1103 		free_fib_info(fi);
1104 		ofi->fib_treeref++;
1105 		return ofi;
1106 	}
1107 
1108 	fi->fib_treeref++;
1109 	atomic_inc(&fi->fib_clntref);
1110 	spin_lock_bh(&fib_info_lock);
1111 	hlist_add_head(&fi->fib_hash,
1112 		       &fib_info_hash[fib_info_hashfn(fi)]);
1113 	if (fi->fib_prefsrc) {
1114 		struct hlist_head *head;
1115 
1116 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1117 		hlist_add_head(&fi->fib_lhash, head);
1118 	}
1119 	change_nexthops(fi) {
1120 		struct hlist_head *head;
1121 		unsigned int hash;
1122 
1123 		if (!nexthop_nh->nh_dev)
1124 			continue;
1125 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1126 		head = &fib_info_devhash[hash];
1127 		hlist_add_head(&nexthop_nh->nh_hash, head);
1128 	} endfor_nexthops(fi)
1129 	spin_unlock_bh(&fib_info_lock);
1130 	return fi;
1131 
1132 err_inval:
1133 	err = -EINVAL;
1134 
1135 failure:
1136 	if (fi) {
1137 		fi->fib_dead = 1;
1138 		free_fib_info(fi);
1139 	}
1140 
1141 	return ERR_PTR(err);
1142 }
1143 
1144 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1145 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1146 		  struct fib_info *fi, unsigned int flags)
1147 {
1148 	struct nlmsghdr *nlh;
1149 	struct rtmsg *rtm;
1150 
1151 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1152 	if (!nlh)
1153 		return -EMSGSIZE;
1154 
1155 	rtm = nlmsg_data(nlh);
1156 	rtm->rtm_family = AF_INET;
1157 	rtm->rtm_dst_len = dst_len;
1158 	rtm->rtm_src_len = 0;
1159 	rtm->rtm_tos = tos;
1160 	if (tb_id < 256)
1161 		rtm->rtm_table = tb_id;
1162 	else
1163 		rtm->rtm_table = RT_TABLE_COMPAT;
1164 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1165 		goto nla_put_failure;
1166 	rtm->rtm_type = type;
1167 	rtm->rtm_flags = fi->fib_flags;
1168 	rtm->rtm_scope = fi->fib_scope;
1169 	rtm->rtm_protocol = fi->fib_protocol;
1170 
1171 	if (rtm->rtm_dst_len &&
1172 	    nla_put_in_addr(skb, RTA_DST, dst))
1173 		goto nla_put_failure;
1174 	if (fi->fib_priority &&
1175 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1176 		goto nla_put_failure;
1177 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1178 		goto nla_put_failure;
1179 
1180 	if (fi->fib_prefsrc &&
1181 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1182 		goto nla_put_failure;
1183 	if (fi->fib_nhs == 1) {
1184 		struct in_device *in_dev;
1185 
1186 		if (fi->fib_nh->nh_gw &&
1187 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1188 			goto nla_put_failure;
1189 		if (fi->fib_nh->nh_oif &&
1190 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1191 			goto nla_put_failure;
1192 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1193 			in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1194 			if (in_dev &&
1195 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1196 				rtm->rtm_flags |= RTNH_F_DEAD;
1197 		}
1198 #ifdef CONFIG_IP_ROUTE_CLASSID
1199 		if (fi->fib_nh[0].nh_tclassid &&
1200 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1201 			goto nla_put_failure;
1202 #endif
1203 		if (fi->fib_nh->nh_lwtstate)
1204 			lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1205 	}
1206 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1207 	if (fi->fib_nhs > 1) {
1208 		struct rtnexthop *rtnh;
1209 		struct nlattr *mp;
1210 
1211 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1212 		if (!mp)
1213 			goto nla_put_failure;
1214 
1215 		for_nexthops(fi) {
1216 			struct in_device *in_dev;
1217 
1218 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1219 			if (!rtnh)
1220 				goto nla_put_failure;
1221 
1222 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1223 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1224 				in_dev = __in_dev_get_rtnl(nh->nh_dev);
1225 				if (in_dev &&
1226 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1227 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1228 			}
1229 			rtnh->rtnh_hops = nh->nh_weight - 1;
1230 			rtnh->rtnh_ifindex = nh->nh_oif;
1231 
1232 			if (nh->nh_gw &&
1233 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1234 				goto nla_put_failure;
1235 #ifdef CONFIG_IP_ROUTE_CLASSID
1236 			if (nh->nh_tclassid &&
1237 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1238 				goto nla_put_failure;
1239 #endif
1240 			if (nh->nh_lwtstate)
1241 				lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1242 			/* length of rtnetlink header + attributes */
1243 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1244 		} endfor_nexthops(fi);
1245 
1246 		nla_nest_end(skb, mp);
1247 	}
1248 #endif
1249 	nlmsg_end(skb, nlh);
1250 	return 0;
1251 
1252 nla_put_failure:
1253 	nlmsg_cancel(skb, nlh);
1254 	return -EMSGSIZE;
1255 }
1256 
1257 /*
1258  * Update FIB if:
1259  * - local address disappeared -> we must delete all the entries
1260  *   referring to it.
1261  * - device went down -> we must shutdown all nexthops going via it.
1262  */
1263 int fib_sync_down_addr(struct net *net, __be32 local)
1264 {
1265 	int ret = 0;
1266 	unsigned int hash = fib_laddr_hashfn(local);
1267 	struct hlist_head *head = &fib_info_laddrhash[hash];
1268 	struct fib_info *fi;
1269 
1270 	if (!fib_info_laddrhash || local == 0)
1271 		return 0;
1272 
1273 	hlist_for_each_entry(fi, head, fib_lhash) {
1274 		if (!net_eq(fi->fib_net, net))
1275 			continue;
1276 		if (fi->fib_prefsrc == local) {
1277 			fi->fib_flags |= RTNH_F_DEAD;
1278 			ret++;
1279 		}
1280 	}
1281 	return ret;
1282 }
1283 
1284 int fib_sync_down_dev(struct net_device *dev, unsigned long event)
1285 {
1286 	int ret = 0;
1287 	int scope = RT_SCOPE_NOWHERE;
1288 	struct fib_info *prev_fi = NULL;
1289 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1290 	struct hlist_head *head = &fib_info_devhash[hash];
1291 	struct fib_nh *nh;
1292 
1293 	if (event == NETDEV_UNREGISTER ||
1294 	    event == NETDEV_DOWN)
1295 		scope = -1;
1296 
1297 	hlist_for_each_entry(nh, head, nh_hash) {
1298 		struct fib_info *fi = nh->nh_parent;
1299 		int dead;
1300 
1301 		BUG_ON(!fi->fib_nhs);
1302 		if (nh->nh_dev != dev || fi == prev_fi)
1303 			continue;
1304 		prev_fi = fi;
1305 		dead = 0;
1306 		change_nexthops(fi) {
1307 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1308 				dead++;
1309 			else if (nexthop_nh->nh_dev == dev &&
1310 				 nexthop_nh->nh_scope != scope) {
1311 				switch (event) {
1312 				case NETDEV_DOWN:
1313 				case NETDEV_UNREGISTER:
1314 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1315 					/* fall through */
1316 				case NETDEV_CHANGE:
1317 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1318 					break;
1319 				}
1320 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1321 				spin_lock_bh(&fib_multipath_lock);
1322 				fi->fib_power -= nexthop_nh->nh_power;
1323 				nexthop_nh->nh_power = 0;
1324 				spin_unlock_bh(&fib_multipath_lock);
1325 #endif
1326 				dead++;
1327 			}
1328 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1329 			if (event == NETDEV_UNREGISTER &&
1330 			    nexthop_nh->nh_dev == dev) {
1331 				dead = fi->fib_nhs;
1332 				break;
1333 			}
1334 #endif
1335 		} endfor_nexthops(fi)
1336 		if (dead == fi->fib_nhs) {
1337 			switch (event) {
1338 			case NETDEV_DOWN:
1339 			case NETDEV_UNREGISTER:
1340 				fi->fib_flags |= RTNH_F_DEAD;
1341 				/* fall through */
1342 			case NETDEV_CHANGE:
1343 				fi->fib_flags |= RTNH_F_LINKDOWN;
1344 				break;
1345 			}
1346 			ret++;
1347 		}
1348 	}
1349 
1350 	return ret;
1351 }
1352 
1353 /* Must be invoked inside of an RCU protected region.  */
1354 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1355 {
1356 	struct fib_info *fi = NULL, *last_resort = NULL;
1357 	struct hlist_head *fa_head = res->fa_head;
1358 	struct fib_table *tb = res->table;
1359 	u8 slen = 32 - res->prefixlen;
1360 	int order = -1, last_idx = -1;
1361 	struct fib_alias *fa, *fa1 = NULL;
1362 	u32 last_prio = res->fi->fib_priority;
1363 	u8 last_tos = 0;
1364 
1365 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1366 		struct fib_info *next_fi = fa->fa_info;
1367 
1368 		if (fa->fa_slen != slen)
1369 			continue;
1370 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1371 			continue;
1372 		if (fa->tb_id != tb->tb_id)
1373 			continue;
1374 		if (next_fi->fib_priority > last_prio &&
1375 		    fa->fa_tos == last_tos) {
1376 			if (last_tos)
1377 				continue;
1378 			break;
1379 		}
1380 		if (next_fi->fib_flags & RTNH_F_DEAD)
1381 			continue;
1382 		last_tos = fa->fa_tos;
1383 		last_prio = next_fi->fib_priority;
1384 
1385 		if (next_fi->fib_scope != res->scope ||
1386 		    fa->fa_type != RTN_UNICAST)
1387 			continue;
1388 		if (!next_fi->fib_nh[0].nh_gw ||
1389 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1390 			continue;
1391 
1392 		fib_alias_accessed(fa);
1393 
1394 		if (!fi) {
1395 			if (next_fi != res->fi)
1396 				break;
1397 			fa1 = fa;
1398 		} else if (!fib_detect_death(fi, order, &last_resort,
1399 					     &last_idx, fa1->fa_default)) {
1400 			fib_result_assign(res, fi);
1401 			fa1->fa_default = order;
1402 			goto out;
1403 		}
1404 		fi = next_fi;
1405 		order++;
1406 	}
1407 
1408 	if (order <= 0 || !fi) {
1409 		if (fa1)
1410 			fa1->fa_default = -1;
1411 		goto out;
1412 	}
1413 
1414 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1415 			      fa1->fa_default)) {
1416 		fib_result_assign(res, fi);
1417 		fa1->fa_default = order;
1418 		goto out;
1419 	}
1420 
1421 	if (last_idx >= 0)
1422 		fib_result_assign(res, last_resort);
1423 	fa1->fa_default = last_idx;
1424 out:
1425 	return;
1426 }
1427 
1428 /*
1429  * Dead device goes up. We wake up dead nexthops.
1430  * It takes sense only on multipath routes.
1431  */
1432 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1433 {
1434 	struct fib_info *prev_fi;
1435 	unsigned int hash;
1436 	struct hlist_head *head;
1437 	struct fib_nh *nh;
1438 	int ret;
1439 
1440 	if (!(dev->flags & IFF_UP))
1441 		return 0;
1442 
1443 	prev_fi = NULL;
1444 	hash = fib_devindex_hashfn(dev->ifindex);
1445 	head = &fib_info_devhash[hash];
1446 	ret = 0;
1447 
1448 	hlist_for_each_entry(nh, head, nh_hash) {
1449 		struct fib_info *fi = nh->nh_parent;
1450 		int alive;
1451 
1452 		BUG_ON(!fi->fib_nhs);
1453 		if (nh->nh_dev != dev || fi == prev_fi)
1454 			continue;
1455 
1456 		prev_fi = fi;
1457 		alive = 0;
1458 		change_nexthops(fi) {
1459 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1460 				alive++;
1461 				continue;
1462 			}
1463 			if (!nexthop_nh->nh_dev ||
1464 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1465 				continue;
1466 			if (nexthop_nh->nh_dev != dev ||
1467 			    !__in_dev_get_rtnl(dev))
1468 				continue;
1469 			alive++;
1470 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1471 			spin_lock_bh(&fib_multipath_lock);
1472 			nexthop_nh->nh_power = 0;
1473 			nexthop_nh->nh_flags &= ~nh_flags;
1474 			spin_unlock_bh(&fib_multipath_lock);
1475 #else
1476 			nexthop_nh->nh_flags &= ~nh_flags;
1477 #endif
1478 		} endfor_nexthops(fi)
1479 
1480 		if (alive > 0) {
1481 			fi->fib_flags &= ~nh_flags;
1482 			ret++;
1483 		}
1484 	}
1485 
1486 	return ret;
1487 }
1488 
1489 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1490 
1491 /*
1492  * The algorithm is suboptimal, but it provides really
1493  * fair weighted route distribution.
1494  */
1495 void fib_select_multipath(struct fib_result *res)
1496 {
1497 	struct fib_info *fi = res->fi;
1498 	struct in_device *in_dev;
1499 	int w;
1500 
1501 	spin_lock_bh(&fib_multipath_lock);
1502 	if (fi->fib_power <= 0) {
1503 		int power = 0;
1504 		change_nexthops(fi) {
1505 			in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev);
1506 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1507 				continue;
1508 			if (in_dev &&
1509 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1510 			    nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1511 				continue;
1512 			power += nexthop_nh->nh_weight;
1513 			nexthop_nh->nh_power = nexthop_nh->nh_weight;
1514 		} endfor_nexthops(fi);
1515 		fi->fib_power = power;
1516 		if (power <= 0) {
1517 			spin_unlock_bh(&fib_multipath_lock);
1518 			/* Race condition: route has just become dead. */
1519 			res->nh_sel = 0;
1520 			return;
1521 		}
1522 	}
1523 
1524 
1525 	/* w should be random number [0..fi->fib_power-1],
1526 	 * it is pretty bad approximation.
1527 	 */
1528 
1529 	w = jiffies % fi->fib_power;
1530 
1531 	change_nexthops(fi) {
1532 		if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1533 		    nexthop_nh->nh_power) {
1534 			w -= nexthop_nh->nh_power;
1535 			if (w <= 0) {
1536 				nexthop_nh->nh_power--;
1537 				fi->fib_power--;
1538 				res->nh_sel = nhsel;
1539 				spin_unlock_bh(&fib_multipath_lock);
1540 				return;
1541 			}
1542 		}
1543 	} endfor_nexthops(fi);
1544 
1545 	/* Race condition: route has just become dead. */
1546 	res->nh_sel = 0;
1547 	spin_unlock_bh(&fib_multipath_lock);
1548 }
1549 #endif
1550