xref: /linux/net/ipv4/fib_semantics.c (revision f02e58f91a121ec909efad06b0a7aa806e1f7a84)
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, &lwtstate);
515 				if (ret)
516 					goto errout;
517 				nexthop_nh->nh_lwtstate =
518 					lwtstate_get(lwtstate);
519 			}
520 		}
521 
522 		rtnh = rtnh_next(rtnh, &remaining);
523 	} endfor_nexthops(fi);
524 
525 	return 0;
526 
527 err_inval:
528 	ret = -EINVAL;
529 
530 errout:
531 	return ret;
532 }
533 
534 #endif
535 
536 int fib_encap_match(struct net *net, u16 encap_type,
537 		    struct nlattr *encap,
538 		    int oif, const struct fib_nh *nh)
539 {
540 	struct lwtunnel_state *lwtstate;
541 	struct net_device *dev = NULL;
542 	int ret;
543 
544 	if (encap_type == LWTUNNEL_ENCAP_NONE)
545 		return 0;
546 
547 	if (oif)
548 		dev = __dev_get_by_index(net, oif);
549 	ret = lwtunnel_build_state(dev, encap_type,
550 				   encap, &lwtstate);
551 	if (!ret)
552 		return lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
553 
554 	return 0;
555 }
556 
557 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
558 {
559 	struct net *net = cfg->fc_nlinfo.nl_net;
560 #ifdef CONFIG_IP_ROUTE_MULTIPATH
561 	struct rtnexthop *rtnh;
562 	int remaining;
563 #endif
564 
565 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
566 		return 1;
567 
568 	if (cfg->fc_oif || cfg->fc_gw) {
569 		if (cfg->fc_encap) {
570 			if (fib_encap_match(net, cfg->fc_encap_type,
571 					    cfg->fc_encap, cfg->fc_oif,
572 					    fi->fib_nh))
573 			    return 1;
574 		}
575 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
576 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
577 			return 0;
578 		return 1;
579 	}
580 
581 #ifdef CONFIG_IP_ROUTE_MULTIPATH
582 	if (!cfg->fc_mp)
583 		return 0;
584 
585 	rtnh = cfg->fc_mp;
586 	remaining = cfg->fc_mp_len;
587 
588 	for_nexthops(fi) {
589 		int attrlen;
590 
591 		if (!rtnh_ok(rtnh, remaining))
592 			return -EINVAL;
593 
594 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
595 			return 1;
596 
597 		attrlen = rtnh_attrlen(rtnh);
598 		if (attrlen > 0) {
599 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
600 
601 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
602 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
603 				return 1;
604 #ifdef CONFIG_IP_ROUTE_CLASSID
605 			nla = nla_find(attrs, attrlen, RTA_FLOW);
606 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
607 				return 1;
608 #endif
609 		}
610 
611 		rtnh = rtnh_next(rtnh, &remaining);
612 	} endfor_nexthops(fi);
613 #endif
614 	return 0;
615 }
616 
617 
618 /*
619  * Picture
620  * -------
621  *
622  * Semantics of nexthop is very messy by historical reasons.
623  * We have to take into account, that:
624  * a) gateway can be actually local interface address,
625  *    so that gatewayed route is direct.
626  * b) gateway must be on-link address, possibly
627  *    described not by an ifaddr, but also by a direct route.
628  * c) If both gateway and interface are specified, they should not
629  *    contradict.
630  * d) If we use tunnel routes, gateway could be not on-link.
631  *
632  * Attempt to reconcile all of these (alas, self-contradictory) conditions
633  * results in pretty ugly and hairy code with obscure logic.
634  *
635  * I chose to generalized it instead, so that the size
636  * of code does not increase practically, but it becomes
637  * much more general.
638  * Every prefix is assigned a "scope" value: "host" is local address,
639  * "link" is direct route,
640  * [ ... "site" ... "interior" ... ]
641  * and "universe" is true gateway route with global meaning.
642  *
643  * Every prefix refers to a set of "nexthop"s (gw, oif),
644  * where gw must have narrower scope. This recursion stops
645  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
646  * which means that gw is forced to be on link.
647  *
648  * Code is still hairy, but now it is apparently logically
649  * consistent and very flexible. F.e. as by-product it allows
650  * to co-exists in peace independent exterior and interior
651  * routing processes.
652  *
653  * Normally it looks as following.
654  *
655  * {universe prefix}  -> (gw, oif) [scope link]
656  *		  |
657  *		  |-> {link prefix} -> (gw, oif) [scope local]
658  *					|
659  *					|-> {local prefix} (terminal node)
660  */
661 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
662 			struct fib_nh *nh)
663 {
664 	int err;
665 	struct net *net;
666 	struct net_device *dev;
667 
668 	net = cfg->fc_nlinfo.nl_net;
669 	if (nh->nh_gw) {
670 		struct fib_result res;
671 
672 		if (nh->nh_flags & RTNH_F_ONLINK) {
673 
674 			if (cfg->fc_scope >= RT_SCOPE_LINK)
675 				return -EINVAL;
676 			if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
677 				return -EINVAL;
678 			dev = __dev_get_by_index(net, nh->nh_oif);
679 			if (!dev)
680 				return -ENODEV;
681 			if (!(dev->flags & IFF_UP))
682 				return -ENETDOWN;
683 			if (!netif_carrier_ok(dev))
684 				nh->nh_flags |= RTNH_F_LINKDOWN;
685 			nh->nh_dev = dev;
686 			dev_hold(dev);
687 			nh->nh_scope = RT_SCOPE_LINK;
688 			return 0;
689 		}
690 		rcu_read_lock();
691 		{
692 			struct flowi4 fl4 = {
693 				.daddr = nh->nh_gw,
694 				.flowi4_scope = cfg->fc_scope + 1,
695 				.flowi4_oif = nh->nh_oif,
696 				.flowi4_iif = LOOPBACK_IFINDEX,
697 			};
698 
699 			/* It is not necessary, but requires a bit of thinking */
700 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
701 				fl4.flowi4_scope = RT_SCOPE_LINK;
702 			err = fib_lookup(net, &fl4, &res,
703 					 FIB_LOOKUP_IGNORE_LINKSTATE);
704 			if (err) {
705 				rcu_read_unlock();
706 				return err;
707 			}
708 		}
709 		err = -EINVAL;
710 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
711 			goto out;
712 		nh->nh_scope = res.scope;
713 		nh->nh_oif = FIB_RES_OIF(res);
714 		nh->nh_dev = dev = FIB_RES_DEV(res);
715 		if (!dev)
716 			goto out;
717 		dev_hold(dev);
718 		if (!netif_carrier_ok(dev))
719 			nh->nh_flags |= RTNH_F_LINKDOWN;
720 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
721 	} else {
722 		struct in_device *in_dev;
723 
724 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
725 			return -EINVAL;
726 
727 		rcu_read_lock();
728 		err = -ENODEV;
729 		in_dev = inetdev_by_index(net, nh->nh_oif);
730 		if (!in_dev)
731 			goto out;
732 		err = -ENETDOWN;
733 		if (!(in_dev->dev->flags & IFF_UP))
734 			goto out;
735 		nh->nh_dev = in_dev->dev;
736 		dev_hold(nh->nh_dev);
737 		nh->nh_scope = RT_SCOPE_HOST;
738 		if (!netif_carrier_ok(nh->nh_dev))
739 			nh->nh_flags |= RTNH_F_LINKDOWN;
740 		err = 0;
741 	}
742 out:
743 	rcu_read_unlock();
744 	return err;
745 }
746 
747 static inline unsigned int fib_laddr_hashfn(__be32 val)
748 {
749 	unsigned int mask = (fib_info_hash_size - 1);
750 
751 	return ((__force u32)val ^
752 		((__force u32)val >> 7) ^
753 		((__force u32)val >> 14)) & mask;
754 }
755 
756 static struct hlist_head *fib_info_hash_alloc(int bytes)
757 {
758 	if (bytes <= PAGE_SIZE)
759 		return kzalloc(bytes, GFP_KERNEL);
760 	else
761 		return (struct hlist_head *)
762 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
763 					 get_order(bytes));
764 }
765 
766 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
767 {
768 	if (!hash)
769 		return;
770 
771 	if (bytes <= PAGE_SIZE)
772 		kfree(hash);
773 	else
774 		free_pages((unsigned long) hash, get_order(bytes));
775 }
776 
777 static void fib_info_hash_move(struct hlist_head *new_info_hash,
778 			       struct hlist_head *new_laddrhash,
779 			       unsigned int new_size)
780 {
781 	struct hlist_head *old_info_hash, *old_laddrhash;
782 	unsigned int old_size = fib_info_hash_size;
783 	unsigned int i, bytes;
784 
785 	spin_lock_bh(&fib_info_lock);
786 	old_info_hash = fib_info_hash;
787 	old_laddrhash = fib_info_laddrhash;
788 	fib_info_hash_size = new_size;
789 
790 	for (i = 0; i < old_size; i++) {
791 		struct hlist_head *head = &fib_info_hash[i];
792 		struct hlist_node *n;
793 		struct fib_info *fi;
794 
795 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
796 			struct hlist_head *dest;
797 			unsigned int new_hash;
798 
799 			new_hash = fib_info_hashfn(fi);
800 			dest = &new_info_hash[new_hash];
801 			hlist_add_head(&fi->fib_hash, dest);
802 		}
803 	}
804 	fib_info_hash = new_info_hash;
805 
806 	for (i = 0; i < old_size; i++) {
807 		struct hlist_head *lhead = &fib_info_laddrhash[i];
808 		struct hlist_node *n;
809 		struct fib_info *fi;
810 
811 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
812 			struct hlist_head *ldest;
813 			unsigned int new_hash;
814 
815 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
816 			ldest = &new_laddrhash[new_hash];
817 			hlist_add_head(&fi->fib_lhash, ldest);
818 		}
819 	}
820 	fib_info_laddrhash = new_laddrhash;
821 
822 	spin_unlock_bh(&fib_info_lock);
823 
824 	bytes = old_size * sizeof(struct hlist_head *);
825 	fib_info_hash_free(old_info_hash, bytes);
826 	fib_info_hash_free(old_laddrhash, bytes);
827 }
828 
829 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
830 {
831 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
832 					nh->nh_gw,
833 					nh->nh_parent->fib_scope);
834 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
835 
836 	return nh->nh_saddr;
837 }
838 
839 struct fib_info *fib_create_info(struct fib_config *cfg)
840 {
841 	int err;
842 	struct fib_info *fi = NULL;
843 	struct fib_info *ofi;
844 	int nhs = 1;
845 	struct net *net = cfg->fc_nlinfo.nl_net;
846 
847 	if (cfg->fc_type > RTN_MAX)
848 		goto err_inval;
849 
850 	/* Fast check to catch the most weird cases */
851 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
852 		goto err_inval;
853 
854 #ifdef CONFIG_IP_ROUTE_MULTIPATH
855 	if (cfg->fc_mp) {
856 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
857 		if (nhs == 0)
858 			goto err_inval;
859 	}
860 #endif
861 
862 	err = -ENOBUFS;
863 	if (fib_info_cnt >= fib_info_hash_size) {
864 		unsigned int new_size = fib_info_hash_size << 1;
865 		struct hlist_head *new_info_hash;
866 		struct hlist_head *new_laddrhash;
867 		unsigned int bytes;
868 
869 		if (!new_size)
870 			new_size = 16;
871 		bytes = new_size * sizeof(struct hlist_head *);
872 		new_info_hash = fib_info_hash_alloc(bytes);
873 		new_laddrhash = fib_info_hash_alloc(bytes);
874 		if (!new_info_hash || !new_laddrhash) {
875 			fib_info_hash_free(new_info_hash, bytes);
876 			fib_info_hash_free(new_laddrhash, bytes);
877 		} else
878 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
879 
880 		if (!fib_info_hash_size)
881 			goto failure;
882 	}
883 
884 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
885 	if (!fi)
886 		goto failure;
887 	fib_info_cnt++;
888 	if (cfg->fc_mx) {
889 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
890 		if (!fi->fib_metrics)
891 			goto failure;
892 	} else
893 		fi->fib_metrics = (u32 *) dst_default_metrics;
894 
895 	fi->fib_net = net;
896 	fi->fib_protocol = cfg->fc_protocol;
897 	fi->fib_scope = cfg->fc_scope;
898 	fi->fib_flags = cfg->fc_flags;
899 	fi->fib_priority = cfg->fc_priority;
900 	fi->fib_prefsrc = cfg->fc_prefsrc;
901 	fi->fib_type = cfg->fc_type;
902 
903 	fi->fib_nhs = nhs;
904 	change_nexthops(fi) {
905 		nexthop_nh->nh_parent = fi;
906 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
907 		if (!nexthop_nh->nh_pcpu_rth_output)
908 			goto failure;
909 	} endfor_nexthops(fi)
910 
911 	if (cfg->fc_mx) {
912 		struct nlattr *nla;
913 		int remaining;
914 
915 		nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
916 			int type = nla_type(nla);
917 
918 			if (type) {
919 				u32 val;
920 
921 				if (type > RTAX_MAX)
922 					goto err_inval;
923 				if (type == RTAX_CC_ALGO) {
924 					char tmp[TCP_CA_NAME_MAX];
925 
926 					nla_strlcpy(tmp, nla, sizeof(tmp));
927 					val = tcp_ca_get_key_by_name(tmp);
928 					if (val == TCP_CA_UNSPEC)
929 						goto err_inval;
930 				} else {
931 					val = nla_get_u32(nla);
932 				}
933 				if (type == RTAX_ADVMSS && val > 65535 - 40)
934 					val = 65535 - 40;
935 				if (type == RTAX_MTU && val > 65535 - 15)
936 					val = 65535 - 15;
937 				fi->fib_metrics[type - 1] = val;
938 			}
939 		}
940 	}
941 
942 	if (cfg->fc_mp) {
943 #ifdef CONFIG_IP_ROUTE_MULTIPATH
944 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
945 		if (err != 0)
946 			goto failure;
947 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
948 			goto err_inval;
949 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
950 			goto err_inval;
951 #ifdef CONFIG_IP_ROUTE_CLASSID
952 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
953 			goto err_inval;
954 #endif
955 #else
956 		goto err_inval;
957 #endif
958 	} else {
959 		struct fib_nh *nh = fi->fib_nh;
960 
961 		if (cfg->fc_encap) {
962 			struct lwtunnel_state *lwtstate;
963 			struct net_device *dev = NULL;
964 
965 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
966 				goto err_inval;
967 			if (cfg->fc_oif)
968 				dev = __dev_get_by_index(net, cfg->fc_oif);
969 			err = lwtunnel_build_state(dev, cfg->fc_encap_type,
970 						   cfg->fc_encap, &lwtstate);
971 			if (err)
972 				goto failure;
973 
974 			nh->nh_lwtstate = lwtstate_get(lwtstate);
975 		}
976 		nh->nh_oif = cfg->fc_oif;
977 		nh->nh_gw = cfg->fc_gw;
978 		nh->nh_flags = cfg->fc_flags;
979 #ifdef CONFIG_IP_ROUTE_CLASSID
980 		nh->nh_tclassid = cfg->fc_flow;
981 		if (nh->nh_tclassid)
982 			fi->fib_net->ipv4.fib_num_tclassid_users++;
983 #endif
984 #ifdef CONFIG_IP_ROUTE_MULTIPATH
985 		nh->nh_weight = 1;
986 #endif
987 	}
988 
989 	if (fib_props[cfg->fc_type].error) {
990 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
991 			goto err_inval;
992 		goto link_it;
993 	} else {
994 		switch (cfg->fc_type) {
995 		case RTN_UNICAST:
996 		case RTN_LOCAL:
997 		case RTN_BROADCAST:
998 		case RTN_ANYCAST:
999 		case RTN_MULTICAST:
1000 			break;
1001 		default:
1002 			goto err_inval;
1003 		}
1004 	}
1005 
1006 	if (cfg->fc_scope > RT_SCOPE_HOST)
1007 		goto err_inval;
1008 
1009 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1010 		struct fib_nh *nh = fi->fib_nh;
1011 
1012 		/* Local address is added. */
1013 		if (nhs != 1 || nh->nh_gw)
1014 			goto err_inval;
1015 		nh->nh_scope = RT_SCOPE_NOWHERE;
1016 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1017 		err = -ENODEV;
1018 		if (!nh->nh_dev)
1019 			goto failure;
1020 	} else {
1021 		int linkdown = 0;
1022 
1023 		change_nexthops(fi) {
1024 			err = fib_check_nh(cfg, fi, nexthop_nh);
1025 			if (err != 0)
1026 				goto failure;
1027 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1028 				linkdown++;
1029 		} endfor_nexthops(fi)
1030 		if (linkdown == fi->fib_nhs)
1031 			fi->fib_flags |= RTNH_F_LINKDOWN;
1032 	}
1033 
1034 	if (fi->fib_prefsrc) {
1035 		if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1036 		    fi->fib_prefsrc != cfg->fc_dst)
1037 			if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
1038 				goto err_inval;
1039 	}
1040 
1041 	change_nexthops(fi) {
1042 		fib_info_update_nh_saddr(net, nexthop_nh);
1043 	} endfor_nexthops(fi)
1044 
1045 link_it:
1046 	ofi = fib_find_info(fi);
1047 	if (ofi) {
1048 		fi->fib_dead = 1;
1049 		free_fib_info(fi);
1050 		ofi->fib_treeref++;
1051 		return ofi;
1052 	}
1053 
1054 	fi->fib_treeref++;
1055 	atomic_inc(&fi->fib_clntref);
1056 	spin_lock_bh(&fib_info_lock);
1057 	hlist_add_head(&fi->fib_hash,
1058 		       &fib_info_hash[fib_info_hashfn(fi)]);
1059 	if (fi->fib_prefsrc) {
1060 		struct hlist_head *head;
1061 
1062 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1063 		hlist_add_head(&fi->fib_lhash, head);
1064 	}
1065 	change_nexthops(fi) {
1066 		struct hlist_head *head;
1067 		unsigned int hash;
1068 
1069 		if (!nexthop_nh->nh_dev)
1070 			continue;
1071 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1072 		head = &fib_info_devhash[hash];
1073 		hlist_add_head(&nexthop_nh->nh_hash, head);
1074 	} endfor_nexthops(fi)
1075 	spin_unlock_bh(&fib_info_lock);
1076 	return fi;
1077 
1078 err_inval:
1079 	err = -EINVAL;
1080 
1081 failure:
1082 	if (fi) {
1083 		fi->fib_dead = 1;
1084 		free_fib_info(fi);
1085 	}
1086 
1087 	return ERR_PTR(err);
1088 }
1089 
1090 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1091 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1092 		  struct fib_info *fi, unsigned int flags)
1093 {
1094 	struct nlmsghdr *nlh;
1095 	struct rtmsg *rtm;
1096 
1097 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1098 	if (!nlh)
1099 		return -EMSGSIZE;
1100 
1101 	rtm = nlmsg_data(nlh);
1102 	rtm->rtm_family = AF_INET;
1103 	rtm->rtm_dst_len = dst_len;
1104 	rtm->rtm_src_len = 0;
1105 	rtm->rtm_tos = tos;
1106 	if (tb_id < 256)
1107 		rtm->rtm_table = tb_id;
1108 	else
1109 		rtm->rtm_table = RT_TABLE_COMPAT;
1110 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1111 		goto nla_put_failure;
1112 	rtm->rtm_type = type;
1113 	rtm->rtm_flags = fi->fib_flags;
1114 	rtm->rtm_scope = fi->fib_scope;
1115 	rtm->rtm_protocol = fi->fib_protocol;
1116 
1117 	if (rtm->rtm_dst_len &&
1118 	    nla_put_in_addr(skb, RTA_DST, dst))
1119 		goto nla_put_failure;
1120 	if (fi->fib_priority &&
1121 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1122 		goto nla_put_failure;
1123 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1124 		goto nla_put_failure;
1125 
1126 	if (fi->fib_prefsrc &&
1127 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1128 		goto nla_put_failure;
1129 	if (fi->fib_nhs == 1) {
1130 		struct in_device *in_dev;
1131 
1132 		if (fi->fib_nh->nh_gw &&
1133 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1134 			goto nla_put_failure;
1135 		if (fi->fib_nh->nh_oif &&
1136 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1137 			goto nla_put_failure;
1138 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1139 			in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1140 			if (in_dev &&
1141 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1142 				rtm->rtm_flags |= RTNH_F_DEAD;
1143 		}
1144 #ifdef CONFIG_IP_ROUTE_CLASSID
1145 		if (fi->fib_nh[0].nh_tclassid &&
1146 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1147 			goto nla_put_failure;
1148 #endif
1149 		if (fi->fib_nh->nh_lwtstate)
1150 			lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1151 	}
1152 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1153 	if (fi->fib_nhs > 1) {
1154 		struct rtnexthop *rtnh;
1155 		struct nlattr *mp;
1156 
1157 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1158 		if (!mp)
1159 			goto nla_put_failure;
1160 
1161 		for_nexthops(fi) {
1162 			struct in_device *in_dev;
1163 
1164 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1165 			if (!rtnh)
1166 				goto nla_put_failure;
1167 
1168 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1169 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1170 				in_dev = __in_dev_get_rtnl(nh->nh_dev);
1171 				if (in_dev &&
1172 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1173 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1174 			}
1175 			rtnh->rtnh_hops = nh->nh_weight - 1;
1176 			rtnh->rtnh_ifindex = nh->nh_oif;
1177 
1178 			if (nh->nh_gw &&
1179 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1180 				goto nla_put_failure;
1181 #ifdef CONFIG_IP_ROUTE_CLASSID
1182 			if (nh->nh_tclassid &&
1183 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1184 				goto nla_put_failure;
1185 #endif
1186 			if (nh->nh_lwtstate)
1187 				lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1188 			/* length of rtnetlink header + attributes */
1189 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1190 		} endfor_nexthops(fi);
1191 
1192 		nla_nest_end(skb, mp);
1193 	}
1194 #endif
1195 	nlmsg_end(skb, nlh);
1196 	return 0;
1197 
1198 nla_put_failure:
1199 	nlmsg_cancel(skb, nlh);
1200 	return -EMSGSIZE;
1201 }
1202 
1203 /*
1204  * Update FIB if:
1205  * - local address disappeared -> we must delete all the entries
1206  *   referring to it.
1207  * - device went down -> we must shutdown all nexthops going via it.
1208  */
1209 int fib_sync_down_addr(struct net *net, __be32 local)
1210 {
1211 	int ret = 0;
1212 	unsigned int hash = fib_laddr_hashfn(local);
1213 	struct hlist_head *head = &fib_info_laddrhash[hash];
1214 	struct fib_info *fi;
1215 
1216 	if (!fib_info_laddrhash || local == 0)
1217 		return 0;
1218 
1219 	hlist_for_each_entry(fi, head, fib_lhash) {
1220 		if (!net_eq(fi->fib_net, net))
1221 			continue;
1222 		if (fi->fib_prefsrc == local) {
1223 			fi->fib_flags |= RTNH_F_DEAD;
1224 			ret++;
1225 		}
1226 	}
1227 	return ret;
1228 }
1229 
1230 int fib_sync_down_dev(struct net_device *dev, unsigned long event)
1231 {
1232 	int ret = 0;
1233 	int scope = RT_SCOPE_NOWHERE;
1234 	struct fib_info *prev_fi = NULL;
1235 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1236 	struct hlist_head *head = &fib_info_devhash[hash];
1237 	struct fib_nh *nh;
1238 
1239 	if (event == NETDEV_UNREGISTER ||
1240 	    event == NETDEV_DOWN)
1241 		scope = -1;
1242 
1243 	hlist_for_each_entry(nh, head, nh_hash) {
1244 		struct fib_info *fi = nh->nh_parent;
1245 		int dead;
1246 
1247 		BUG_ON(!fi->fib_nhs);
1248 		if (nh->nh_dev != dev || fi == prev_fi)
1249 			continue;
1250 		prev_fi = fi;
1251 		dead = 0;
1252 		change_nexthops(fi) {
1253 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1254 				dead++;
1255 			else if (nexthop_nh->nh_dev == dev &&
1256 				 nexthop_nh->nh_scope != scope) {
1257 				switch (event) {
1258 				case NETDEV_DOWN:
1259 				case NETDEV_UNREGISTER:
1260 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1261 					/* fall through */
1262 				case NETDEV_CHANGE:
1263 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1264 					break;
1265 				}
1266 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1267 				spin_lock_bh(&fib_multipath_lock);
1268 				fi->fib_power -= nexthop_nh->nh_power;
1269 				nexthop_nh->nh_power = 0;
1270 				spin_unlock_bh(&fib_multipath_lock);
1271 #endif
1272 				dead++;
1273 			}
1274 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1275 			if (event == NETDEV_UNREGISTER &&
1276 			    nexthop_nh->nh_dev == dev) {
1277 				dead = fi->fib_nhs;
1278 				break;
1279 			}
1280 #endif
1281 		} endfor_nexthops(fi)
1282 		if (dead == fi->fib_nhs) {
1283 			switch (event) {
1284 			case NETDEV_DOWN:
1285 			case NETDEV_UNREGISTER:
1286 				fi->fib_flags |= RTNH_F_DEAD;
1287 				/* fall through */
1288 			case NETDEV_CHANGE:
1289 				fi->fib_flags |= RTNH_F_LINKDOWN;
1290 				break;
1291 			}
1292 			ret++;
1293 		}
1294 	}
1295 
1296 	return ret;
1297 }
1298 
1299 /* Must be invoked inside of an RCU protected region.  */
1300 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1301 {
1302 	struct fib_info *fi = NULL, *last_resort = NULL;
1303 	struct hlist_head *fa_head = res->fa_head;
1304 	struct fib_table *tb = res->table;
1305 	u8 slen = 32 - res->prefixlen;
1306 	int order = -1, last_idx = -1;
1307 	struct fib_alias *fa, *fa1 = NULL;
1308 	u32 last_prio = res->fi->fib_priority;
1309 	u8 last_tos = 0;
1310 
1311 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1312 		struct fib_info *next_fi = fa->fa_info;
1313 
1314 		if (fa->fa_slen != slen)
1315 			continue;
1316 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1317 			continue;
1318 		if (fa->tb_id != tb->tb_id)
1319 			continue;
1320 		if (next_fi->fib_priority > last_prio &&
1321 		    fa->fa_tos == last_tos) {
1322 			if (last_tos)
1323 				continue;
1324 			break;
1325 		}
1326 		if (next_fi->fib_flags & RTNH_F_DEAD)
1327 			continue;
1328 		last_tos = fa->fa_tos;
1329 		last_prio = next_fi->fib_priority;
1330 
1331 		if (next_fi->fib_scope != res->scope ||
1332 		    fa->fa_type != RTN_UNICAST)
1333 			continue;
1334 		if (!next_fi->fib_nh[0].nh_gw ||
1335 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1336 			continue;
1337 
1338 		fib_alias_accessed(fa);
1339 
1340 		if (!fi) {
1341 			if (next_fi != res->fi)
1342 				break;
1343 			fa1 = fa;
1344 		} else if (!fib_detect_death(fi, order, &last_resort,
1345 					     &last_idx, fa1->fa_default)) {
1346 			fib_result_assign(res, fi);
1347 			fa1->fa_default = order;
1348 			goto out;
1349 		}
1350 		fi = next_fi;
1351 		order++;
1352 	}
1353 
1354 	if (order <= 0 || !fi) {
1355 		if (fa1)
1356 			fa1->fa_default = -1;
1357 		goto out;
1358 	}
1359 
1360 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1361 			      fa1->fa_default)) {
1362 		fib_result_assign(res, fi);
1363 		fa1->fa_default = order;
1364 		goto out;
1365 	}
1366 
1367 	if (last_idx >= 0)
1368 		fib_result_assign(res, last_resort);
1369 	fa1->fa_default = last_idx;
1370 out:
1371 	return;
1372 }
1373 
1374 /*
1375  * Dead device goes up. We wake up dead nexthops.
1376  * It takes sense only on multipath routes.
1377  */
1378 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1379 {
1380 	struct fib_info *prev_fi;
1381 	unsigned int hash;
1382 	struct hlist_head *head;
1383 	struct fib_nh *nh;
1384 	int ret;
1385 
1386 	if (!(dev->flags & IFF_UP))
1387 		return 0;
1388 
1389 	prev_fi = NULL;
1390 	hash = fib_devindex_hashfn(dev->ifindex);
1391 	head = &fib_info_devhash[hash];
1392 	ret = 0;
1393 
1394 	hlist_for_each_entry(nh, head, nh_hash) {
1395 		struct fib_info *fi = nh->nh_parent;
1396 		int alive;
1397 
1398 		BUG_ON(!fi->fib_nhs);
1399 		if (nh->nh_dev != dev || fi == prev_fi)
1400 			continue;
1401 
1402 		prev_fi = fi;
1403 		alive = 0;
1404 		change_nexthops(fi) {
1405 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1406 				alive++;
1407 				continue;
1408 			}
1409 			if (!nexthop_nh->nh_dev ||
1410 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1411 				continue;
1412 			if (nexthop_nh->nh_dev != dev ||
1413 			    !__in_dev_get_rtnl(dev))
1414 				continue;
1415 			alive++;
1416 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1417 			spin_lock_bh(&fib_multipath_lock);
1418 			nexthop_nh->nh_power = 0;
1419 			nexthop_nh->nh_flags &= ~nh_flags;
1420 			spin_unlock_bh(&fib_multipath_lock);
1421 #else
1422 			nexthop_nh->nh_flags &= ~nh_flags;
1423 #endif
1424 		} endfor_nexthops(fi)
1425 
1426 		if (alive > 0) {
1427 			fi->fib_flags &= ~nh_flags;
1428 			ret++;
1429 		}
1430 	}
1431 
1432 	return ret;
1433 }
1434 
1435 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1436 
1437 /*
1438  * The algorithm is suboptimal, but it provides really
1439  * fair weighted route distribution.
1440  */
1441 void fib_select_multipath(struct fib_result *res)
1442 {
1443 	struct fib_info *fi = res->fi;
1444 	struct in_device *in_dev;
1445 	int w;
1446 
1447 	spin_lock_bh(&fib_multipath_lock);
1448 	if (fi->fib_power <= 0) {
1449 		int power = 0;
1450 		change_nexthops(fi) {
1451 			in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev);
1452 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1453 				continue;
1454 			if (in_dev &&
1455 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1456 			    nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1457 				continue;
1458 			power += nexthop_nh->nh_weight;
1459 			nexthop_nh->nh_power = nexthop_nh->nh_weight;
1460 		} endfor_nexthops(fi);
1461 		fi->fib_power = power;
1462 		if (power <= 0) {
1463 			spin_unlock_bh(&fib_multipath_lock);
1464 			/* Race condition: route has just become dead. */
1465 			res->nh_sel = 0;
1466 			return;
1467 		}
1468 	}
1469 
1470 
1471 	/* w should be random number [0..fi->fib_power-1],
1472 	 * it is pretty bad approximation.
1473 	 */
1474 
1475 	w = jiffies % fi->fib_power;
1476 
1477 	change_nexthops(fi) {
1478 		if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1479 		    nexthop_nh->nh_power) {
1480 			w -= nexthop_nh->nh_power;
1481 			if (w <= 0) {
1482 				nexthop_nh->nh_power--;
1483 				fi->fib_power--;
1484 				res->nh_sel = nhsel;
1485 				spin_unlock_bh(&fib_multipath_lock);
1486 				return;
1487 			}
1488 		}
1489 	} endfor_nexthops(fi);
1490 
1491 	/* Race condition: route has just become dead. */
1492 	res->nh_sel = 0;
1493 	spin_unlock_bh(&fib_multipath_lock);
1494 }
1495 #endif
1496