xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 4f6cce39)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		IPv4 Forwarding Information Base: semantics.
7  *
8  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9  *
10  *		This program is free software; you can redistribute it and/or
11  *		modify it under the terms of the GNU General Public License
12  *		as published by the Free Software Foundation; either version
13  *		2 of the License, or (at your option) any later version.
14  */
15 
16 #include <linux/uaccess.h>
17 #include <linux/bitops.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/jiffies.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/errno.h>
26 #include <linux/in.h>
27 #include <linux/inet.h>
28 #include <linux/inetdevice.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_arp.h>
31 #include <linux/proc_fs.h>
32 #include <linux/skbuff.h>
33 #include <linux/init.h>
34 #include <linux/slab.h>
35 
36 #include <net/arp.h>
37 #include <net/ip.h>
38 #include <net/protocol.h>
39 #include <net/route.h>
40 #include <net/tcp.h>
41 #include <net/sock.h>
42 #include <net/ip_fib.h>
43 #include <net/netlink.h>
44 #include <net/nexthop.h>
45 #include <net/lwtunnel.h>
46 
47 #include "fib_lookup.h"
48 
49 static DEFINE_SPINLOCK(fib_info_lock);
50 static struct hlist_head *fib_info_hash;
51 static struct hlist_head *fib_info_laddrhash;
52 static unsigned int fib_info_hash_size;
53 static unsigned int fib_info_cnt;
54 
55 #define DEVINDEX_HASHBITS 8
56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
58 
59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
60 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 EXPORT_SYMBOL_GPL(free_fib_info);
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 	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 nlattr *nla_entype;
506 
507 				nla_entype = nla_find(attrs, attrlen,
508 						      RTA_ENCAP_TYPE);
509 				if (!nla_entype)
510 					goto err_inval;
511 
512 				ret = lwtunnel_build_state(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 static void fib_rebalance(struct fib_info *fi)
536 {
537 	int total;
538 	int w;
539 	struct in_device *in_dev;
540 
541 	if (fi->fib_nhs < 2)
542 		return;
543 
544 	total = 0;
545 	for_nexthops(fi) {
546 		if (nh->nh_flags & RTNH_F_DEAD)
547 			continue;
548 
549 		in_dev = __in_dev_get_rtnl(nh->nh_dev);
550 
551 		if (in_dev &&
552 		    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
553 		    nh->nh_flags & RTNH_F_LINKDOWN)
554 			continue;
555 
556 		total += nh->nh_weight;
557 	} endfor_nexthops(fi);
558 
559 	w = 0;
560 	change_nexthops(fi) {
561 		int upper_bound;
562 
563 		in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
564 
565 		if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
566 			upper_bound = -1;
567 		} else if (in_dev &&
568 			   IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
569 			   nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
570 			upper_bound = -1;
571 		} else {
572 			w += nexthop_nh->nh_weight;
573 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
574 							    total) - 1;
575 		}
576 
577 		atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
578 	} endfor_nexthops(fi);
579 
580 	net_get_random_once(&fib_multipath_secret,
581 			    sizeof(fib_multipath_secret));
582 }
583 
584 static inline void fib_add_weight(struct fib_info *fi,
585 				  const struct fib_nh *nh)
586 {
587 	fi->fib_weight += nh->nh_weight;
588 }
589 
590 #else /* CONFIG_IP_ROUTE_MULTIPATH */
591 
592 #define fib_rebalance(fi) do { } while (0)
593 #define fib_add_weight(fi, nh) do { } while (0)
594 
595 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
596 
597 static int fib_encap_match(u16 encap_type,
598 			   struct nlattr *encap,
599 			   const struct fib_nh *nh,
600 			   const struct fib_config *cfg)
601 {
602 	struct lwtunnel_state *lwtstate;
603 	int ret, result = 0;
604 
605 	if (encap_type == LWTUNNEL_ENCAP_NONE)
606 		return 0;
607 
608 	ret = lwtunnel_build_state(encap_type, encap,
609 				   AF_INET, cfg, &lwtstate);
610 	if (!ret) {
611 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
612 		lwtstate_free(lwtstate);
613 	}
614 
615 	return result;
616 }
617 
618 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
619 {
620 #ifdef CONFIG_IP_ROUTE_MULTIPATH
621 	struct rtnexthop *rtnh;
622 	int remaining;
623 #endif
624 
625 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
626 		return 1;
627 
628 	if (cfg->fc_oif || cfg->fc_gw) {
629 		if (cfg->fc_encap) {
630 			if (fib_encap_match(cfg->fc_encap_type,
631 					    cfg->fc_encap, fi->fib_nh, cfg))
632 			    return 1;
633 		}
634 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
635 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
636 			return 0;
637 		return 1;
638 	}
639 
640 #ifdef CONFIG_IP_ROUTE_MULTIPATH
641 	if (!cfg->fc_mp)
642 		return 0;
643 
644 	rtnh = cfg->fc_mp;
645 	remaining = cfg->fc_mp_len;
646 
647 	for_nexthops(fi) {
648 		int attrlen;
649 
650 		if (!rtnh_ok(rtnh, remaining))
651 			return -EINVAL;
652 
653 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
654 			return 1;
655 
656 		attrlen = rtnh_attrlen(rtnh);
657 		if (attrlen > 0) {
658 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
659 
660 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
661 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
662 				return 1;
663 #ifdef CONFIG_IP_ROUTE_CLASSID
664 			nla = nla_find(attrs, attrlen, RTA_FLOW);
665 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
666 				return 1;
667 #endif
668 		}
669 
670 		rtnh = rtnh_next(rtnh, &remaining);
671 	} endfor_nexthops(fi);
672 #endif
673 	return 0;
674 }
675 
676 
677 /*
678  * Picture
679  * -------
680  *
681  * Semantics of nexthop is very messy by historical reasons.
682  * We have to take into account, that:
683  * a) gateway can be actually local interface address,
684  *    so that gatewayed route is direct.
685  * b) gateway must be on-link address, possibly
686  *    described not by an ifaddr, but also by a direct route.
687  * c) If both gateway and interface are specified, they should not
688  *    contradict.
689  * d) If we use tunnel routes, gateway could be not on-link.
690  *
691  * Attempt to reconcile all of these (alas, self-contradictory) conditions
692  * results in pretty ugly and hairy code with obscure logic.
693  *
694  * I chose to generalized it instead, so that the size
695  * of code does not increase practically, but it becomes
696  * much more general.
697  * Every prefix is assigned a "scope" value: "host" is local address,
698  * "link" is direct route,
699  * [ ... "site" ... "interior" ... ]
700  * and "universe" is true gateway route with global meaning.
701  *
702  * Every prefix refers to a set of "nexthop"s (gw, oif),
703  * where gw must have narrower scope. This recursion stops
704  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
705  * which means that gw is forced to be on link.
706  *
707  * Code is still hairy, but now it is apparently logically
708  * consistent and very flexible. F.e. as by-product it allows
709  * to co-exists in peace independent exterior and interior
710  * routing processes.
711  *
712  * Normally it looks as following.
713  *
714  * {universe prefix}  -> (gw, oif) [scope link]
715  *		  |
716  *		  |-> {link prefix} -> (gw, oif) [scope local]
717  *					|
718  *					|-> {local prefix} (terminal node)
719  */
720 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
721 			struct fib_nh *nh)
722 {
723 	int err = 0;
724 	struct net *net;
725 	struct net_device *dev;
726 
727 	net = cfg->fc_nlinfo.nl_net;
728 	if (nh->nh_gw) {
729 		struct fib_result res;
730 
731 		if (nh->nh_flags & RTNH_F_ONLINK) {
732 			unsigned int addr_type;
733 
734 			if (cfg->fc_scope >= RT_SCOPE_LINK)
735 				return -EINVAL;
736 			dev = __dev_get_by_index(net, nh->nh_oif);
737 			if (!dev)
738 				return -ENODEV;
739 			if (!(dev->flags & IFF_UP))
740 				return -ENETDOWN;
741 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
742 			if (addr_type != RTN_UNICAST)
743 				return -EINVAL;
744 			if (!netif_carrier_ok(dev))
745 				nh->nh_flags |= RTNH_F_LINKDOWN;
746 			nh->nh_dev = dev;
747 			dev_hold(dev);
748 			nh->nh_scope = RT_SCOPE_LINK;
749 			return 0;
750 		}
751 		rcu_read_lock();
752 		{
753 			struct fib_table *tbl = NULL;
754 			struct flowi4 fl4 = {
755 				.daddr = nh->nh_gw,
756 				.flowi4_scope = cfg->fc_scope + 1,
757 				.flowi4_oif = nh->nh_oif,
758 				.flowi4_iif = LOOPBACK_IFINDEX,
759 			};
760 
761 			/* It is not necessary, but requires a bit of thinking */
762 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
763 				fl4.flowi4_scope = RT_SCOPE_LINK;
764 
765 			if (cfg->fc_table)
766 				tbl = fib_get_table(net, cfg->fc_table);
767 
768 			if (tbl)
769 				err = fib_table_lookup(tbl, &fl4, &res,
770 						       FIB_LOOKUP_IGNORE_LINKSTATE |
771 						       FIB_LOOKUP_NOREF);
772 
773 			/* on error or if no table given do full lookup. This
774 			 * is needed for example when nexthops are in the local
775 			 * table rather than the given table
776 			 */
777 			if (!tbl || err) {
778 				err = fib_lookup(net, &fl4, &res,
779 						 FIB_LOOKUP_IGNORE_LINKSTATE);
780 			}
781 
782 			if (err) {
783 				rcu_read_unlock();
784 				return err;
785 			}
786 		}
787 		err = -EINVAL;
788 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
789 			goto out;
790 		nh->nh_scope = res.scope;
791 		nh->nh_oif = FIB_RES_OIF(res);
792 		nh->nh_dev = dev = FIB_RES_DEV(res);
793 		if (!dev)
794 			goto out;
795 		dev_hold(dev);
796 		if (!netif_carrier_ok(dev))
797 			nh->nh_flags |= RTNH_F_LINKDOWN;
798 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
799 	} else {
800 		struct in_device *in_dev;
801 
802 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
803 			return -EINVAL;
804 
805 		rcu_read_lock();
806 		err = -ENODEV;
807 		in_dev = inetdev_by_index(net, nh->nh_oif);
808 		if (!in_dev)
809 			goto out;
810 		err = -ENETDOWN;
811 		if (!(in_dev->dev->flags & IFF_UP))
812 			goto out;
813 		nh->nh_dev = in_dev->dev;
814 		dev_hold(nh->nh_dev);
815 		nh->nh_scope = RT_SCOPE_HOST;
816 		if (!netif_carrier_ok(nh->nh_dev))
817 			nh->nh_flags |= RTNH_F_LINKDOWN;
818 		err = 0;
819 	}
820 out:
821 	rcu_read_unlock();
822 	return err;
823 }
824 
825 static inline unsigned int fib_laddr_hashfn(__be32 val)
826 {
827 	unsigned int mask = (fib_info_hash_size - 1);
828 
829 	return ((__force u32)val ^
830 		((__force u32)val >> 7) ^
831 		((__force u32)val >> 14)) & mask;
832 }
833 
834 static struct hlist_head *fib_info_hash_alloc(int bytes)
835 {
836 	if (bytes <= PAGE_SIZE)
837 		return kzalloc(bytes, GFP_KERNEL);
838 	else
839 		return (struct hlist_head *)
840 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
841 					 get_order(bytes));
842 }
843 
844 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
845 {
846 	if (!hash)
847 		return;
848 
849 	if (bytes <= PAGE_SIZE)
850 		kfree(hash);
851 	else
852 		free_pages((unsigned long) hash, get_order(bytes));
853 }
854 
855 static void fib_info_hash_move(struct hlist_head *new_info_hash,
856 			       struct hlist_head *new_laddrhash,
857 			       unsigned int new_size)
858 {
859 	struct hlist_head *old_info_hash, *old_laddrhash;
860 	unsigned int old_size = fib_info_hash_size;
861 	unsigned int i, bytes;
862 
863 	spin_lock_bh(&fib_info_lock);
864 	old_info_hash = fib_info_hash;
865 	old_laddrhash = fib_info_laddrhash;
866 	fib_info_hash_size = new_size;
867 
868 	for (i = 0; i < old_size; i++) {
869 		struct hlist_head *head = &fib_info_hash[i];
870 		struct hlist_node *n;
871 		struct fib_info *fi;
872 
873 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
874 			struct hlist_head *dest;
875 			unsigned int new_hash;
876 
877 			new_hash = fib_info_hashfn(fi);
878 			dest = &new_info_hash[new_hash];
879 			hlist_add_head(&fi->fib_hash, dest);
880 		}
881 	}
882 	fib_info_hash = new_info_hash;
883 
884 	for (i = 0; i < old_size; i++) {
885 		struct hlist_head *lhead = &fib_info_laddrhash[i];
886 		struct hlist_node *n;
887 		struct fib_info *fi;
888 
889 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
890 			struct hlist_head *ldest;
891 			unsigned int new_hash;
892 
893 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
894 			ldest = &new_laddrhash[new_hash];
895 			hlist_add_head(&fi->fib_lhash, ldest);
896 		}
897 	}
898 	fib_info_laddrhash = new_laddrhash;
899 
900 	spin_unlock_bh(&fib_info_lock);
901 
902 	bytes = old_size * sizeof(struct hlist_head *);
903 	fib_info_hash_free(old_info_hash, bytes);
904 	fib_info_hash_free(old_laddrhash, bytes);
905 }
906 
907 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
908 {
909 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
910 					nh->nh_gw,
911 					nh->nh_parent->fib_scope);
912 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
913 
914 	return nh->nh_saddr;
915 }
916 
917 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
918 {
919 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
920 	    fib_prefsrc != cfg->fc_dst) {
921 		u32 tb_id = cfg->fc_table;
922 		int rc;
923 
924 		if (tb_id == RT_TABLE_MAIN)
925 			tb_id = RT_TABLE_LOCAL;
926 
927 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
928 					  fib_prefsrc, tb_id);
929 
930 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
931 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
932 						  fib_prefsrc, RT_TABLE_LOCAL);
933 		}
934 
935 		if (rc != RTN_LOCAL)
936 			return false;
937 	}
938 	return true;
939 }
940 
941 static int
942 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
943 {
944 	bool ecn_ca = false;
945 	struct nlattr *nla;
946 	int remaining;
947 
948 	if (!cfg->fc_mx)
949 		return 0;
950 
951 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
952 		int type = nla_type(nla);
953 		u32 val;
954 
955 		if (!type)
956 			continue;
957 		if (type > RTAX_MAX)
958 			return -EINVAL;
959 
960 		if (type == RTAX_CC_ALGO) {
961 			char tmp[TCP_CA_NAME_MAX];
962 
963 			nla_strlcpy(tmp, nla, sizeof(tmp));
964 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
965 			if (val == TCP_CA_UNSPEC)
966 				return -EINVAL;
967 		} else {
968 			val = nla_get_u32(nla);
969 		}
970 		if (type == RTAX_ADVMSS && val > 65535 - 40)
971 			val = 65535 - 40;
972 		if (type == RTAX_MTU && val > 65535 - 15)
973 			val = 65535 - 15;
974 		if (type == RTAX_HOPLIMIT && val > 255)
975 			val = 255;
976 		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
977 			return -EINVAL;
978 		fi->fib_metrics[type - 1] = val;
979 	}
980 
981 	if (ecn_ca)
982 		fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
983 
984 	return 0;
985 }
986 
987 struct fib_info *fib_create_info(struct fib_config *cfg)
988 {
989 	int err;
990 	struct fib_info *fi = NULL;
991 	struct fib_info *ofi;
992 	int nhs = 1;
993 	struct net *net = cfg->fc_nlinfo.nl_net;
994 
995 	if (cfg->fc_type > RTN_MAX)
996 		goto err_inval;
997 
998 	/* Fast check to catch the most weird cases */
999 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
1000 		goto err_inval;
1001 
1002 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
1003 		goto err_inval;
1004 
1005 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1006 	if (cfg->fc_mp) {
1007 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1008 		if (nhs == 0)
1009 			goto err_inval;
1010 	}
1011 #endif
1012 
1013 	err = -ENOBUFS;
1014 	if (fib_info_cnt >= fib_info_hash_size) {
1015 		unsigned int new_size = fib_info_hash_size << 1;
1016 		struct hlist_head *new_info_hash;
1017 		struct hlist_head *new_laddrhash;
1018 		unsigned int bytes;
1019 
1020 		if (!new_size)
1021 			new_size = 16;
1022 		bytes = new_size * sizeof(struct hlist_head *);
1023 		new_info_hash = fib_info_hash_alloc(bytes);
1024 		new_laddrhash = fib_info_hash_alloc(bytes);
1025 		if (!new_info_hash || !new_laddrhash) {
1026 			fib_info_hash_free(new_info_hash, bytes);
1027 			fib_info_hash_free(new_laddrhash, bytes);
1028 		} else
1029 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1030 
1031 		if (!fib_info_hash_size)
1032 			goto failure;
1033 	}
1034 
1035 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1036 	if (!fi)
1037 		goto failure;
1038 	fib_info_cnt++;
1039 	if (cfg->fc_mx) {
1040 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1041 		if (!fi->fib_metrics)
1042 			goto failure;
1043 	} else
1044 		fi->fib_metrics = (u32 *) dst_default_metrics;
1045 
1046 	fi->fib_net = net;
1047 	fi->fib_protocol = cfg->fc_protocol;
1048 	fi->fib_scope = cfg->fc_scope;
1049 	fi->fib_flags = cfg->fc_flags;
1050 	fi->fib_priority = cfg->fc_priority;
1051 	fi->fib_prefsrc = cfg->fc_prefsrc;
1052 	fi->fib_type = cfg->fc_type;
1053 	fi->fib_tb_id = cfg->fc_table;
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 
1089 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1090 				goto err_inval;
1091 			err = lwtunnel_build_state(cfg->fc_encap_type,
1092 						   cfg->fc_encap, AF_INET, cfg,
1093 						   &lwtstate);
1094 			if (err)
1095 				goto failure;
1096 
1097 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1098 		}
1099 		nh->nh_oif = cfg->fc_oif;
1100 		nh->nh_gw = cfg->fc_gw;
1101 		nh->nh_flags = cfg->fc_flags;
1102 #ifdef CONFIG_IP_ROUTE_CLASSID
1103 		nh->nh_tclassid = cfg->fc_flow;
1104 		if (nh->nh_tclassid)
1105 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1106 #endif
1107 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1108 		nh->nh_weight = 1;
1109 #endif
1110 	}
1111 
1112 	if (fib_props[cfg->fc_type].error) {
1113 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1114 			goto err_inval;
1115 		goto link_it;
1116 	} else {
1117 		switch (cfg->fc_type) {
1118 		case RTN_UNICAST:
1119 		case RTN_LOCAL:
1120 		case RTN_BROADCAST:
1121 		case RTN_ANYCAST:
1122 		case RTN_MULTICAST:
1123 			break;
1124 		default:
1125 			goto err_inval;
1126 		}
1127 	}
1128 
1129 	if (cfg->fc_scope > RT_SCOPE_HOST)
1130 		goto err_inval;
1131 
1132 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1133 		struct fib_nh *nh = fi->fib_nh;
1134 
1135 		/* Local address is added. */
1136 		if (nhs != 1 || nh->nh_gw)
1137 			goto err_inval;
1138 		nh->nh_scope = RT_SCOPE_NOWHERE;
1139 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1140 		err = -ENODEV;
1141 		if (!nh->nh_dev)
1142 			goto failure;
1143 	} else {
1144 		int linkdown = 0;
1145 
1146 		change_nexthops(fi) {
1147 			err = fib_check_nh(cfg, fi, nexthop_nh);
1148 			if (err != 0)
1149 				goto failure;
1150 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1151 				linkdown++;
1152 		} endfor_nexthops(fi)
1153 		if (linkdown == fi->fib_nhs)
1154 			fi->fib_flags |= RTNH_F_LINKDOWN;
1155 	}
1156 
1157 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1158 		goto err_inval;
1159 
1160 	change_nexthops(fi) {
1161 		fib_info_update_nh_saddr(net, nexthop_nh);
1162 		fib_add_weight(fi, nexthop_nh);
1163 	} endfor_nexthops(fi)
1164 
1165 	fib_rebalance(fi);
1166 
1167 link_it:
1168 	ofi = fib_find_info(fi);
1169 	if (ofi) {
1170 		fi->fib_dead = 1;
1171 		free_fib_info(fi);
1172 		ofi->fib_treeref++;
1173 		return ofi;
1174 	}
1175 
1176 	fi->fib_treeref++;
1177 	atomic_inc(&fi->fib_clntref);
1178 	spin_lock_bh(&fib_info_lock);
1179 	hlist_add_head(&fi->fib_hash,
1180 		       &fib_info_hash[fib_info_hashfn(fi)]);
1181 	if (fi->fib_prefsrc) {
1182 		struct hlist_head *head;
1183 
1184 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1185 		hlist_add_head(&fi->fib_lhash, head);
1186 	}
1187 	change_nexthops(fi) {
1188 		struct hlist_head *head;
1189 		unsigned int hash;
1190 
1191 		if (!nexthop_nh->nh_dev)
1192 			continue;
1193 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1194 		head = &fib_info_devhash[hash];
1195 		hlist_add_head(&nexthop_nh->nh_hash, head);
1196 	} endfor_nexthops(fi)
1197 	spin_unlock_bh(&fib_info_lock);
1198 	return fi;
1199 
1200 err_inval:
1201 	err = -EINVAL;
1202 
1203 failure:
1204 	if (fi) {
1205 		fi->fib_dead = 1;
1206 		free_fib_info(fi);
1207 	}
1208 
1209 	return ERR_PTR(err);
1210 }
1211 
1212 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1213 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1214 		  struct fib_info *fi, unsigned int flags)
1215 {
1216 	struct nlmsghdr *nlh;
1217 	struct rtmsg *rtm;
1218 
1219 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1220 	if (!nlh)
1221 		return -EMSGSIZE;
1222 
1223 	rtm = nlmsg_data(nlh);
1224 	rtm->rtm_family = AF_INET;
1225 	rtm->rtm_dst_len = dst_len;
1226 	rtm->rtm_src_len = 0;
1227 	rtm->rtm_tos = tos;
1228 	if (tb_id < 256)
1229 		rtm->rtm_table = tb_id;
1230 	else
1231 		rtm->rtm_table = RT_TABLE_COMPAT;
1232 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1233 		goto nla_put_failure;
1234 	rtm->rtm_type = type;
1235 	rtm->rtm_flags = fi->fib_flags;
1236 	rtm->rtm_scope = fi->fib_scope;
1237 	rtm->rtm_protocol = fi->fib_protocol;
1238 
1239 	if (rtm->rtm_dst_len &&
1240 	    nla_put_in_addr(skb, RTA_DST, dst))
1241 		goto nla_put_failure;
1242 	if (fi->fib_priority &&
1243 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1244 		goto nla_put_failure;
1245 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1246 		goto nla_put_failure;
1247 
1248 	if (fi->fib_prefsrc &&
1249 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1250 		goto nla_put_failure;
1251 	if (fi->fib_nhs == 1) {
1252 		struct in_device *in_dev;
1253 
1254 		if (fi->fib_nh->nh_gw &&
1255 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1256 			goto nla_put_failure;
1257 		if (fi->fib_nh->nh_oif &&
1258 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1259 			goto nla_put_failure;
1260 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1261 			in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1262 			if (in_dev &&
1263 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1264 				rtm->rtm_flags |= RTNH_F_DEAD;
1265 		}
1266 #ifdef CONFIG_IP_ROUTE_CLASSID
1267 		if (fi->fib_nh[0].nh_tclassid &&
1268 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1269 			goto nla_put_failure;
1270 #endif
1271 		if (fi->fib_nh->nh_lwtstate &&
1272 		    lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1273 			goto nla_put_failure;
1274 	}
1275 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1276 	if (fi->fib_nhs > 1) {
1277 		struct rtnexthop *rtnh;
1278 		struct nlattr *mp;
1279 
1280 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1281 		if (!mp)
1282 			goto nla_put_failure;
1283 
1284 		for_nexthops(fi) {
1285 			struct in_device *in_dev;
1286 
1287 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1288 			if (!rtnh)
1289 				goto nla_put_failure;
1290 
1291 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1292 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1293 				in_dev = __in_dev_get_rtnl(nh->nh_dev);
1294 				if (in_dev &&
1295 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1296 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1297 			}
1298 			rtnh->rtnh_hops = nh->nh_weight - 1;
1299 			rtnh->rtnh_ifindex = nh->nh_oif;
1300 
1301 			if (nh->nh_gw &&
1302 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1303 				goto nla_put_failure;
1304 #ifdef CONFIG_IP_ROUTE_CLASSID
1305 			if (nh->nh_tclassid &&
1306 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1307 				goto nla_put_failure;
1308 #endif
1309 			if (nh->nh_lwtstate &&
1310 			    lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1311 				goto nla_put_failure;
1312 
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_device *dev, __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 net *net = dev_net(dev);
1340 	int tb_id = l3mdev_fib_table(dev);
1341 	struct fib_info *fi;
1342 
1343 	if (!fib_info_laddrhash || local == 0)
1344 		return 0;
1345 
1346 	hlist_for_each_entry(fi, head, fib_lhash) {
1347 		if (!net_eq(fi->fib_net, net) ||
1348 		    fi->fib_tb_id != tb_id)
1349 			continue;
1350 		if (fi->fib_prefsrc == local) {
1351 			fi->fib_flags |= RTNH_F_DEAD;
1352 			ret++;
1353 		}
1354 	}
1355 	return ret;
1356 }
1357 
1358 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1359 				 enum fib_event_type event_type)
1360 {
1361 	struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1362 	struct fib_nh_notifier_info info = {
1363 		.fib_nh = fib_nh,
1364 	};
1365 
1366 	switch (event_type) {
1367 	case FIB_EVENT_NH_ADD:
1368 		if (fib_nh->nh_flags & RTNH_F_DEAD)
1369 			break;
1370 		if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1371 		    fib_nh->nh_flags & RTNH_F_LINKDOWN)
1372 			break;
1373 		return call_fib_notifiers(dev_net(fib_nh->nh_dev), event_type,
1374 					  &info.info);
1375 	case FIB_EVENT_NH_DEL:
1376 		if ((IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1377 		     fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1378 		    (fib_nh->nh_flags & RTNH_F_DEAD))
1379 			return call_fib_notifiers(dev_net(fib_nh->nh_dev),
1380 						  event_type, &info.info);
1381 	default:
1382 		break;
1383 	}
1384 
1385 	return NOTIFY_DONE;
1386 }
1387 
1388 /* Event              force Flags           Description
1389  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1390  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1391  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1392  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1393  */
1394 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1395 {
1396 	int ret = 0;
1397 	int scope = RT_SCOPE_NOWHERE;
1398 	struct fib_info *prev_fi = NULL;
1399 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1400 	struct hlist_head *head = &fib_info_devhash[hash];
1401 	struct fib_nh *nh;
1402 
1403 	if (force)
1404 		scope = -1;
1405 
1406 	hlist_for_each_entry(nh, head, nh_hash) {
1407 		struct fib_info *fi = nh->nh_parent;
1408 		int dead;
1409 
1410 		BUG_ON(!fi->fib_nhs);
1411 		if (nh->nh_dev != dev || fi == prev_fi)
1412 			continue;
1413 		prev_fi = fi;
1414 		dead = 0;
1415 		change_nexthops(fi) {
1416 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1417 				dead++;
1418 			else if (nexthop_nh->nh_dev == dev &&
1419 				 nexthop_nh->nh_scope != scope) {
1420 				switch (event) {
1421 				case NETDEV_DOWN:
1422 				case NETDEV_UNREGISTER:
1423 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1424 					/* fall through */
1425 				case NETDEV_CHANGE:
1426 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1427 					break;
1428 				}
1429 				call_fib_nh_notifiers(nexthop_nh,
1430 						      FIB_EVENT_NH_DEL);
1431 				dead++;
1432 			}
1433 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1434 			if (event == NETDEV_UNREGISTER &&
1435 			    nexthop_nh->nh_dev == dev) {
1436 				dead = fi->fib_nhs;
1437 				break;
1438 			}
1439 #endif
1440 		} endfor_nexthops(fi)
1441 		if (dead == fi->fib_nhs) {
1442 			switch (event) {
1443 			case NETDEV_DOWN:
1444 			case NETDEV_UNREGISTER:
1445 				fi->fib_flags |= RTNH_F_DEAD;
1446 				/* fall through */
1447 			case NETDEV_CHANGE:
1448 				fi->fib_flags |= RTNH_F_LINKDOWN;
1449 				break;
1450 			}
1451 			ret++;
1452 		}
1453 
1454 		fib_rebalance(fi);
1455 	}
1456 
1457 	return ret;
1458 }
1459 
1460 /* Must be invoked inside of an RCU protected region.  */
1461 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1462 {
1463 	struct fib_info *fi = NULL, *last_resort = NULL;
1464 	struct hlist_head *fa_head = res->fa_head;
1465 	struct fib_table *tb = res->table;
1466 	u8 slen = 32 - res->prefixlen;
1467 	int order = -1, last_idx = -1;
1468 	struct fib_alias *fa, *fa1 = NULL;
1469 	u32 last_prio = res->fi->fib_priority;
1470 	u8 last_tos = 0;
1471 
1472 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1473 		struct fib_info *next_fi = fa->fa_info;
1474 
1475 		if (fa->fa_slen != slen)
1476 			continue;
1477 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1478 			continue;
1479 		if (fa->tb_id != tb->tb_id)
1480 			continue;
1481 		if (next_fi->fib_priority > last_prio &&
1482 		    fa->fa_tos == last_tos) {
1483 			if (last_tos)
1484 				continue;
1485 			break;
1486 		}
1487 		if (next_fi->fib_flags & RTNH_F_DEAD)
1488 			continue;
1489 		last_tos = fa->fa_tos;
1490 		last_prio = next_fi->fib_priority;
1491 
1492 		if (next_fi->fib_scope != res->scope ||
1493 		    fa->fa_type != RTN_UNICAST)
1494 			continue;
1495 		if (!next_fi->fib_nh[0].nh_gw ||
1496 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1497 			continue;
1498 
1499 		fib_alias_accessed(fa);
1500 
1501 		if (!fi) {
1502 			if (next_fi != res->fi)
1503 				break;
1504 			fa1 = fa;
1505 		} else if (!fib_detect_death(fi, order, &last_resort,
1506 					     &last_idx, fa1->fa_default)) {
1507 			fib_result_assign(res, fi);
1508 			fa1->fa_default = order;
1509 			goto out;
1510 		}
1511 		fi = next_fi;
1512 		order++;
1513 	}
1514 
1515 	if (order <= 0 || !fi) {
1516 		if (fa1)
1517 			fa1->fa_default = -1;
1518 		goto out;
1519 	}
1520 
1521 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1522 			      fa1->fa_default)) {
1523 		fib_result_assign(res, fi);
1524 		fa1->fa_default = order;
1525 		goto out;
1526 	}
1527 
1528 	if (last_idx >= 0)
1529 		fib_result_assign(res, last_resort);
1530 	fa1->fa_default = last_idx;
1531 out:
1532 	return;
1533 }
1534 
1535 /*
1536  * Dead device goes up. We wake up dead nexthops.
1537  * It takes sense only on multipath routes.
1538  */
1539 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1540 {
1541 	struct fib_info *prev_fi;
1542 	unsigned int hash;
1543 	struct hlist_head *head;
1544 	struct fib_nh *nh;
1545 	int ret;
1546 
1547 	if (!(dev->flags & IFF_UP))
1548 		return 0;
1549 
1550 	if (nh_flags & RTNH_F_DEAD) {
1551 		unsigned int flags = dev_get_flags(dev);
1552 
1553 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1554 			nh_flags |= RTNH_F_LINKDOWN;
1555 	}
1556 
1557 	prev_fi = NULL;
1558 	hash = fib_devindex_hashfn(dev->ifindex);
1559 	head = &fib_info_devhash[hash];
1560 	ret = 0;
1561 
1562 	hlist_for_each_entry(nh, head, nh_hash) {
1563 		struct fib_info *fi = nh->nh_parent;
1564 		int alive;
1565 
1566 		BUG_ON(!fi->fib_nhs);
1567 		if (nh->nh_dev != dev || fi == prev_fi)
1568 			continue;
1569 
1570 		prev_fi = fi;
1571 		alive = 0;
1572 		change_nexthops(fi) {
1573 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1574 				alive++;
1575 				continue;
1576 			}
1577 			if (!nexthop_nh->nh_dev ||
1578 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1579 				continue;
1580 			if (nexthop_nh->nh_dev != dev ||
1581 			    !__in_dev_get_rtnl(dev))
1582 				continue;
1583 			alive++;
1584 			nexthop_nh->nh_flags &= ~nh_flags;
1585 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1586 		} endfor_nexthops(fi)
1587 
1588 		if (alive > 0) {
1589 			fi->fib_flags &= ~nh_flags;
1590 			ret++;
1591 		}
1592 
1593 		fib_rebalance(fi);
1594 	}
1595 
1596 	return ret;
1597 }
1598 
1599 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1600 static bool fib_good_nh(const struct fib_nh *nh)
1601 {
1602 	int state = NUD_REACHABLE;
1603 
1604 	if (nh->nh_scope == RT_SCOPE_LINK) {
1605 		struct neighbour *n;
1606 
1607 		rcu_read_lock_bh();
1608 
1609 		n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1610 					      (__force u32)nh->nh_gw);
1611 		if (n)
1612 			state = n->nud_state;
1613 
1614 		rcu_read_unlock_bh();
1615 	}
1616 
1617 	return !!(state & NUD_VALID);
1618 }
1619 
1620 void fib_select_multipath(struct fib_result *res, int hash)
1621 {
1622 	struct fib_info *fi = res->fi;
1623 	struct net *net = fi->fib_net;
1624 	bool first = false;
1625 
1626 	for_nexthops(fi) {
1627 		if (hash > atomic_read(&nh->nh_upper_bound))
1628 			continue;
1629 
1630 		if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1631 		    fib_good_nh(nh)) {
1632 			res->nh_sel = nhsel;
1633 			return;
1634 		}
1635 		if (!first) {
1636 			res->nh_sel = nhsel;
1637 			first = true;
1638 		}
1639 	} endfor_nexthops(fi);
1640 }
1641 #endif
1642 
1643 void fib_select_path(struct net *net, struct fib_result *res,
1644 		     struct flowi4 *fl4, int mp_hash)
1645 {
1646 	bool oif_check;
1647 
1648 	oif_check = (fl4->flowi4_oif == 0 ||
1649 		     fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1650 
1651 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1652 	if (res->fi->fib_nhs > 1 && oif_check) {
1653 		if (mp_hash < 0)
1654 			mp_hash = get_hash_from_flowi4(fl4) >> 1;
1655 
1656 		fib_select_multipath(res, mp_hash);
1657 	}
1658 	else
1659 #endif
1660 	if (!res->prefixlen &&
1661 	    res->table->tb_num_default > 1 &&
1662 	    res->type == RTN_UNICAST && oif_check)
1663 		fib_select_default(fl4, res);
1664 
1665 	if (!fl4->saddr)
1666 		fl4->saddr = FIB_RES_PREFSRC(net, *res);
1667 }
1668 EXPORT_SYMBOL_GPL(fib_select_path);
1669