xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision e1e0a9e6)
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 
46 #include "fib_lookup.h"
47 
48 static DEFINE_SPINLOCK(fib_info_lock);
49 static struct hlist_head *fib_info_hash;
50 static struct hlist_head *fib_info_laddrhash;
51 static unsigned int fib_info_hash_size;
52 static unsigned int fib_info_cnt;
53 
54 #define DEVINDEX_HASHBITS 8
55 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
56 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
57 
58 #ifdef CONFIG_IP_ROUTE_MULTIPATH
59 
60 static DEFINE_SPINLOCK(fib_multipath_lock);
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 = nh->nh_exceptions;
161 	int i;
162 
163 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
164 		struct fib_nh_exception *fnhe;
165 
166 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
167 		while (fnhe) {
168 			struct fib_nh_exception *next;
169 
170 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
171 
172 			rt_fibinfo_free(&fnhe->fnhe_rth);
173 
174 			kfree(fnhe);
175 
176 			fnhe = next;
177 		}
178 	}
179 	kfree(hash);
180 }
181 
182 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
183 {
184 	int cpu;
185 
186 	if (!rtp)
187 		return;
188 
189 	for_each_possible_cpu(cpu) {
190 		struct rtable *rt;
191 
192 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
193 		if (rt)
194 			dst_free(&rt->dst);
195 	}
196 	free_percpu(rtp);
197 }
198 
199 /* Release a nexthop info record */
200 static void free_fib_info_rcu(struct rcu_head *head)
201 {
202 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
203 
204 	change_nexthops(fi) {
205 		if (nexthop_nh->nh_dev)
206 			dev_put(nexthop_nh->nh_dev);
207 		if (nexthop_nh->nh_exceptions)
208 			free_nh_exceptions(nexthop_nh);
209 		rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
210 		rt_fibinfo_free(&nexthop_nh->nh_rth_input);
211 	} endfor_nexthops(fi);
212 
213 	release_net(fi->fib_net);
214 	if (fi->fib_metrics != (u32 *) dst_default_metrics)
215 		kfree(fi->fib_metrics);
216 	kfree(fi);
217 }
218 
219 void free_fib_info(struct fib_info *fi)
220 {
221 	if (fi->fib_dead == 0) {
222 		pr_warn("Freeing alive fib_info %p\n", fi);
223 		return;
224 	}
225 	fib_info_cnt--;
226 #ifdef CONFIG_IP_ROUTE_CLASSID
227 	change_nexthops(fi) {
228 		if (nexthop_nh->nh_tclassid)
229 			fi->fib_net->ipv4.fib_num_tclassid_users--;
230 	} endfor_nexthops(fi);
231 #endif
232 	call_rcu(&fi->rcu, free_fib_info_rcu);
233 }
234 
235 void fib_release_info(struct fib_info *fi)
236 {
237 	spin_lock_bh(&fib_info_lock);
238 	if (fi && --fi->fib_treeref == 0) {
239 		hlist_del(&fi->fib_hash);
240 		if (fi->fib_prefsrc)
241 			hlist_del(&fi->fib_lhash);
242 		change_nexthops(fi) {
243 			if (!nexthop_nh->nh_dev)
244 				continue;
245 			hlist_del(&nexthop_nh->nh_hash);
246 		} endfor_nexthops(fi)
247 		fi->fib_dead = 1;
248 		fib_info_put(fi);
249 	}
250 	spin_unlock_bh(&fib_info_lock);
251 }
252 
253 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
254 {
255 	const struct fib_nh *onh = ofi->fib_nh;
256 
257 	for_nexthops(fi) {
258 		if (nh->nh_oif != onh->nh_oif ||
259 		    nh->nh_gw  != onh->nh_gw ||
260 		    nh->nh_scope != onh->nh_scope ||
261 #ifdef CONFIG_IP_ROUTE_MULTIPATH
262 		    nh->nh_weight != onh->nh_weight ||
263 #endif
264 #ifdef CONFIG_IP_ROUTE_CLASSID
265 		    nh->nh_tclassid != onh->nh_tclassid ||
266 #endif
267 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
268 			return -1;
269 		onh++;
270 	} endfor_nexthops(fi);
271 	return 0;
272 }
273 
274 static inline unsigned int fib_devindex_hashfn(unsigned int val)
275 {
276 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
277 
278 	return (val ^
279 		(val >> DEVINDEX_HASHBITS) ^
280 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
281 }
282 
283 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
284 {
285 	unsigned int mask = (fib_info_hash_size - 1);
286 	unsigned int val = fi->fib_nhs;
287 
288 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
289 	val ^= (__force u32)fi->fib_prefsrc;
290 	val ^= fi->fib_priority;
291 	for_nexthops(fi) {
292 		val ^= fib_devindex_hashfn(nh->nh_oif);
293 	} endfor_nexthops(fi)
294 
295 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
296 }
297 
298 static struct fib_info *fib_find_info(const struct fib_info *nfi)
299 {
300 	struct hlist_head *head;
301 	struct hlist_node *node;
302 	struct fib_info *fi;
303 	unsigned int hash;
304 
305 	hash = fib_info_hashfn(nfi);
306 	head = &fib_info_hash[hash];
307 
308 	hlist_for_each_entry(fi, node, head, fib_hash) {
309 		if (!net_eq(fi->fib_net, nfi->fib_net))
310 			continue;
311 		if (fi->fib_nhs != nfi->fib_nhs)
312 			continue;
313 		if (nfi->fib_protocol == fi->fib_protocol &&
314 		    nfi->fib_scope == fi->fib_scope &&
315 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
316 		    nfi->fib_priority == fi->fib_priority &&
317 		    nfi->fib_type == fi->fib_type &&
318 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
319 			   sizeof(u32) * RTAX_MAX) == 0 &&
320 		    ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
321 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
322 			return fi;
323 	}
324 
325 	return NULL;
326 }
327 
328 /* Check, that the gateway is already configured.
329  * Used only by redirect accept routine.
330  */
331 int ip_fib_check_default(__be32 gw, struct net_device *dev)
332 {
333 	struct hlist_head *head;
334 	struct hlist_node *node;
335 	struct fib_nh *nh;
336 	unsigned int hash;
337 
338 	spin_lock(&fib_info_lock);
339 
340 	hash = fib_devindex_hashfn(dev->ifindex);
341 	head = &fib_info_devhash[hash];
342 	hlist_for_each_entry(nh, node, head, nh_hash) {
343 		if (nh->nh_dev == dev &&
344 		    nh->nh_gw == gw &&
345 		    !(nh->nh_flags & RTNH_F_DEAD)) {
346 			spin_unlock(&fib_info_lock);
347 			return 0;
348 		}
349 	}
350 
351 	spin_unlock(&fib_info_lock);
352 
353 	return -1;
354 }
355 
356 static inline size_t fib_nlmsg_size(struct fib_info *fi)
357 {
358 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
359 			 + nla_total_size(4) /* RTA_TABLE */
360 			 + nla_total_size(4) /* RTA_DST */
361 			 + nla_total_size(4) /* RTA_PRIORITY */
362 			 + nla_total_size(4); /* RTA_PREFSRC */
363 
364 	/* space for nested metrics */
365 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
366 
367 	if (fi->fib_nhs) {
368 		/* Also handles the special case fib_nhs == 1 */
369 
370 		/* each nexthop is packed in an attribute */
371 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
372 
373 		/* may contain flow and gateway attribute */
374 		nhsize += 2 * nla_total_size(4);
375 
376 		/* all nexthops are packed in a nested attribute */
377 		payload += nla_total_size(fi->fib_nhs * nhsize);
378 	}
379 
380 	return payload;
381 }
382 
383 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
384 	       int dst_len, u32 tb_id, struct nl_info *info,
385 	       unsigned int nlm_flags)
386 {
387 	struct sk_buff *skb;
388 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
389 	int err = -ENOBUFS;
390 
391 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
392 	if (skb == NULL)
393 		goto errout;
394 
395 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
396 			    fa->fa_type, key, dst_len,
397 			    fa->fa_tos, fa->fa_info, nlm_flags);
398 	if (err < 0) {
399 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
400 		WARN_ON(err == -EMSGSIZE);
401 		kfree_skb(skb);
402 		goto errout;
403 	}
404 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
405 		    info->nlh, GFP_KERNEL);
406 	return;
407 errout:
408 	if (err < 0)
409 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
410 }
411 
412 /* Return the first fib alias matching TOS with
413  * priority less than or equal to PRIO.
414  */
415 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
416 {
417 	if (fah) {
418 		struct fib_alias *fa;
419 		list_for_each_entry(fa, fah, fa_list) {
420 			if (fa->fa_tos > tos)
421 				continue;
422 			if (fa->fa_info->fib_priority >= prio ||
423 			    fa->fa_tos < tos)
424 				return fa;
425 		}
426 	}
427 	return NULL;
428 }
429 
430 int fib_detect_death(struct fib_info *fi, int order,
431 		     struct fib_info **last_resort, int *last_idx, int dflt)
432 {
433 	struct neighbour *n;
434 	int state = NUD_NONE;
435 
436 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
437 	if (n) {
438 		state = n->nud_state;
439 		neigh_release(n);
440 	}
441 	if (state == NUD_REACHABLE)
442 		return 0;
443 	if ((state & NUD_VALID) && order != dflt)
444 		return 0;
445 	if ((state & NUD_VALID) ||
446 	    (*last_idx < 0 && order > dflt)) {
447 		*last_resort = fi;
448 		*last_idx = order;
449 	}
450 	return 1;
451 }
452 
453 #ifdef CONFIG_IP_ROUTE_MULTIPATH
454 
455 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
456 {
457 	int nhs = 0;
458 
459 	while (rtnh_ok(rtnh, remaining)) {
460 		nhs++;
461 		rtnh = rtnh_next(rtnh, &remaining);
462 	}
463 
464 	/* leftover implies invalid nexthop configuration, discard it */
465 	return remaining > 0 ? 0 : nhs;
466 }
467 
468 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
469 		       int remaining, struct fib_config *cfg)
470 {
471 	change_nexthops(fi) {
472 		int attrlen;
473 
474 		if (!rtnh_ok(rtnh, remaining))
475 			return -EINVAL;
476 
477 		nexthop_nh->nh_flags =
478 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
479 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
480 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
481 
482 		attrlen = rtnh_attrlen(rtnh);
483 		if (attrlen > 0) {
484 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
485 
486 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
487 			nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
488 #ifdef CONFIG_IP_ROUTE_CLASSID
489 			nla = nla_find(attrs, attrlen, RTA_FLOW);
490 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
491 			if (nexthop_nh->nh_tclassid)
492 				fi->fib_net->ipv4.fib_num_tclassid_users++;
493 #endif
494 		}
495 
496 		rtnh = rtnh_next(rtnh, &remaining);
497 	} endfor_nexthops(fi);
498 
499 	return 0;
500 }
501 
502 #endif
503 
504 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
505 {
506 #ifdef CONFIG_IP_ROUTE_MULTIPATH
507 	struct rtnexthop *rtnh;
508 	int remaining;
509 #endif
510 
511 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
512 		return 1;
513 
514 	if (cfg->fc_oif || cfg->fc_gw) {
515 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
516 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
517 			return 0;
518 		return 1;
519 	}
520 
521 #ifdef CONFIG_IP_ROUTE_MULTIPATH
522 	if (cfg->fc_mp == NULL)
523 		return 0;
524 
525 	rtnh = cfg->fc_mp;
526 	remaining = cfg->fc_mp_len;
527 
528 	for_nexthops(fi) {
529 		int attrlen;
530 
531 		if (!rtnh_ok(rtnh, remaining))
532 			return -EINVAL;
533 
534 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
535 			return 1;
536 
537 		attrlen = rtnh_attrlen(rtnh);
538 		if (attrlen < 0) {
539 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
540 
541 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
542 			if (nla && nla_get_be32(nla) != nh->nh_gw)
543 				return 1;
544 #ifdef CONFIG_IP_ROUTE_CLASSID
545 			nla = nla_find(attrs, attrlen, RTA_FLOW);
546 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
547 				return 1;
548 #endif
549 		}
550 
551 		rtnh = rtnh_next(rtnh, &remaining);
552 	} endfor_nexthops(fi);
553 #endif
554 	return 0;
555 }
556 
557 
558 /*
559  * Picture
560  * -------
561  *
562  * Semantics of nexthop is very messy by historical reasons.
563  * We have to take into account, that:
564  * a) gateway can be actually local interface address,
565  *    so that gatewayed route is direct.
566  * b) gateway must be on-link address, possibly
567  *    described not by an ifaddr, but also by a direct route.
568  * c) If both gateway and interface are specified, they should not
569  *    contradict.
570  * d) If we use tunnel routes, gateway could be not on-link.
571  *
572  * Attempt to reconcile all of these (alas, self-contradictory) conditions
573  * results in pretty ugly and hairy code with obscure logic.
574  *
575  * I chose to generalized it instead, so that the size
576  * of code does not increase practically, but it becomes
577  * much more general.
578  * Every prefix is assigned a "scope" value: "host" is local address,
579  * "link" is direct route,
580  * [ ... "site" ... "interior" ... ]
581  * and "universe" is true gateway route with global meaning.
582  *
583  * Every prefix refers to a set of "nexthop"s (gw, oif),
584  * where gw must have narrower scope. This recursion stops
585  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
586  * which means that gw is forced to be on link.
587  *
588  * Code is still hairy, but now it is apparently logically
589  * consistent and very flexible. F.e. as by-product it allows
590  * to co-exists in peace independent exterior and interior
591  * routing processes.
592  *
593  * Normally it looks as following.
594  *
595  * {universe prefix}  -> (gw, oif) [scope link]
596  *		  |
597  *		  |-> {link prefix} -> (gw, oif) [scope local]
598  *					|
599  *					|-> {local prefix} (terminal node)
600  */
601 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
602 			struct fib_nh *nh)
603 {
604 	int err;
605 	struct net *net;
606 	struct net_device *dev;
607 
608 	net = cfg->fc_nlinfo.nl_net;
609 	if (nh->nh_gw) {
610 		struct fib_result res;
611 
612 		if (nh->nh_flags & RTNH_F_ONLINK) {
613 
614 			if (cfg->fc_scope >= RT_SCOPE_LINK)
615 				return -EINVAL;
616 			if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
617 				return -EINVAL;
618 			dev = __dev_get_by_index(net, nh->nh_oif);
619 			if (!dev)
620 				return -ENODEV;
621 			if (!(dev->flags & IFF_UP))
622 				return -ENETDOWN;
623 			nh->nh_dev = dev;
624 			dev_hold(dev);
625 			nh->nh_scope = RT_SCOPE_LINK;
626 			return 0;
627 		}
628 		rcu_read_lock();
629 		{
630 			struct flowi4 fl4 = {
631 				.daddr = nh->nh_gw,
632 				.flowi4_scope = cfg->fc_scope + 1,
633 				.flowi4_oif = nh->nh_oif,
634 			};
635 
636 			/* It is not necessary, but requires a bit of thinking */
637 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
638 				fl4.flowi4_scope = RT_SCOPE_LINK;
639 			err = fib_lookup(net, &fl4, &res);
640 			if (err) {
641 				rcu_read_unlock();
642 				return err;
643 			}
644 		}
645 		err = -EINVAL;
646 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
647 			goto out;
648 		nh->nh_scope = res.scope;
649 		nh->nh_oif = FIB_RES_OIF(res);
650 		nh->nh_dev = dev = FIB_RES_DEV(res);
651 		if (!dev)
652 			goto out;
653 		dev_hold(dev);
654 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
655 	} else {
656 		struct in_device *in_dev;
657 
658 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
659 			return -EINVAL;
660 
661 		rcu_read_lock();
662 		err = -ENODEV;
663 		in_dev = inetdev_by_index(net, nh->nh_oif);
664 		if (in_dev == NULL)
665 			goto out;
666 		err = -ENETDOWN;
667 		if (!(in_dev->dev->flags & IFF_UP))
668 			goto out;
669 		nh->nh_dev = in_dev->dev;
670 		dev_hold(nh->nh_dev);
671 		nh->nh_scope = RT_SCOPE_HOST;
672 		err = 0;
673 	}
674 out:
675 	rcu_read_unlock();
676 	return err;
677 }
678 
679 static inline unsigned int fib_laddr_hashfn(__be32 val)
680 {
681 	unsigned int mask = (fib_info_hash_size - 1);
682 
683 	return ((__force u32)val ^
684 		((__force u32)val >> 7) ^
685 		((__force u32)val >> 14)) & mask;
686 }
687 
688 static struct hlist_head *fib_info_hash_alloc(int bytes)
689 {
690 	if (bytes <= PAGE_SIZE)
691 		return kzalloc(bytes, GFP_KERNEL);
692 	else
693 		return (struct hlist_head *)
694 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
695 					 get_order(bytes));
696 }
697 
698 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
699 {
700 	if (!hash)
701 		return;
702 
703 	if (bytes <= PAGE_SIZE)
704 		kfree(hash);
705 	else
706 		free_pages((unsigned long) hash, get_order(bytes));
707 }
708 
709 static void fib_info_hash_move(struct hlist_head *new_info_hash,
710 			       struct hlist_head *new_laddrhash,
711 			       unsigned int new_size)
712 {
713 	struct hlist_head *old_info_hash, *old_laddrhash;
714 	unsigned int old_size = fib_info_hash_size;
715 	unsigned int i, bytes;
716 
717 	spin_lock_bh(&fib_info_lock);
718 	old_info_hash = fib_info_hash;
719 	old_laddrhash = fib_info_laddrhash;
720 	fib_info_hash_size = new_size;
721 
722 	for (i = 0; i < old_size; i++) {
723 		struct hlist_head *head = &fib_info_hash[i];
724 		struct hlist_node *node, *n;
725 		struct fib_info *fi;
726 
727 		hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
728 			struct hlist_head *dest;
729 			unsigned int new_hash;
730 
731 			hlist_del(&fi->fib_hash);
732 
733 			new_hash = fib_info_hashfn(fi);
734 			dest = &new_info_hash[new_hash];
735 			hlist_add_head(&fi->fib_hash, dest);
736 		}
737 	}
738 	fib_info_hash = new_info_hash;
739 
740 	for (i = 0; i < old_size; i++) {
741 		struct hlist_head *lhead = &fib_info_laddrhash[i];
742 		struct hlist_node *node, *n;
743 		struct fib_info *fi;
744 
745 		hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
746 			struct hlist_head *ldest;
747 			unsigned int new_hash;
748 
749 			hlist_del(&fi->fib_lhash);
750 
751 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
752 			ldest = &new_laddrhash[new_hash];
753 			hlist_add_head(&fi->fib_lhash, ldest);
754 		}
755 	}
756 	fib_info_laddrhash = new_laddrhash;
757 
758 	spin_unlock_bh(&fib_info_lock);
759 
760 	bytes = old_size * sizeof(struct hlist_head *);
761 	fib_info_hash_free(old_info_hash, bytes);
762 	fib_info_hash_free(old_laddrhash, bytes);
763 }
764 
765 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
766 {
767 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
768 					nh->nh_gw,
769 					nh->nh_parent->fib_scope);
770 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
771 
772 	return nh->nh_saddr;
773 }
774 
775 struct fib_info *fib_create_info(struct fib_config *cfg)
776 {
777 	int err;
778 	struct fib_info *fi = NULL;
779 	struct fib_info *ofi;
780 	int nhs = 1;
781 	struct net *net = cfg->fc_nlinfo.nl_net;
782 
783 	if (cfg->fc_type > RTN_MAX)
784 		goto err_inval;
785 
786 	/* Fast check to catch the most weird cases */
787 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
788 		goto err_inval;
789 
790 #ifdef CONFIG_IP_ROUTE_MULTIPATH
791 	if (cfg->fc_mp) {
792 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
793 		if (nhs == 0)
794 			goto err_inval;
795 	}
796 #endif
797 
798 	err = -ENOBUFS;
799 	if (fib_info_cnt >= fib_info_hash_size) {
800 		unsigned int new_size = fib_info_hash_size << 1;
801 		struct hlist_head *new_info_hash;
802 		struct hlist_head *new_laddrhash;
803 		unsigned int bytes;
804 
805 		if (!new_size)
806 			new_size = 16;
807 		bytes = new_size * sizeof(struct hlist_head *);
808 		new_info_hash = fib_info_hash_alloc(bytes);
809 		new_laddrhash = fib_info_hash_alloc(bytes);
810 		if (!new_info_hash || !new_laddrhash) {
811 			fib_info_hash_free(new_info_hash, bytes);
812 			fib_info_hash_free(new_laddrhash, bytes);
813 		} else
814 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
815 
816 		if (!fib_info_hash_size)
817 			goto failure;
818 	}
819 
820 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
821 	if (fi == NULL)
822 		goto failure;
823 	if (cfg->fc_mx) {
824 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
825 		if (!fi->fib_metrics)
826 			goto failure;
827 	} else
828 		fi->fib_metrics = (u32 *) dst_default_metrics;
829 	fib_info_cnt++;
830 
831 	fi->fib_net = hold_net(net);
832 	fi->fib_protocol = cfg->fc_protocol;
833 	fi->fib_scope = cfg->fc_scope;
834 	fi->fib_flags = cfg->fc_flags;
835 	fi->fib_priority = cfg->fc_priority;
836 	fi->fib_prefsrc = cfg->fc_prefsrc;
837 	fi->fib_type = cfg->fc_type;
838 
839 	fi->fib_nhs = nhs;
840 	change_nexthops(fi) {
841 		nexthop_nh->nh_parent = fi;
842 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
843 		if (!nexthop_nh->nh_pcpu_rth_output)
844 			goto failure;
845 	} endfor_nexthops(fi)
846 
847 	if (cfg->fc_mx) {
848 		struct nlattr *nla;
849 		int remaining;
850 
851 		nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
852 			int type = nla_type(nla);
853 
854 			if (type) {
855 				u32 val;
856 
857 				if (type > RTAX_MAX)
858 					goto err_inval;
859 				val = nla_get_u32(nla);
860 				if (type == RTAX_ADVMSS && val > 65535 - 40)
861 					val = 65535 - 40;
862 				if (type == RTAX_MTU && val > 65535 - 15)
863 					val = 65535 - 15;
864 				fi->fib_metrics[type - 1] = val;
865 			}
866 		}
867 	}
868 
869 	if (cfg->fc_mp) {
870 #ifdef CONFIG_IP_ROUTE_MULTIPATH
871 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
872 		if (err != 0)
873 			goto failure;
874 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
875 			goto err_inval;
876 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
877 			goto err_inval;
878 #ifdef CONFIG_IP_ROUTE_CLASSID
879 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
880 			goto err_inval;
881 #endif
882 #else
883 		goto err_inval;
884 #endif
885 	} else {
886 		struct fib_nh *nh = fi->fib_nh;
887 
888 		nh->nh_oif = cfg->fc_oif;
889 		nh->nh_gw = cfg->fc_gw;
890 		nh->nh_flags = cfg->fc_flags;
891 #ifdef CONFIG_IP_ROUTE_CLASSID
892 		nh->nh_tclassid = cfg->fc_flow;
893 		if (nh->nh_tclassid)
894 			fi->fib_net->ipv4.fib_num_tclassid_users++;
895 #endif
896 #ifdef CONFIG_IP_ROUTE_MULTIPATH
897 		nh->nh_weight = 1;
898 #endif
899 	}
900 
901 	if (fib_props[cfg->fc_type].error) {
902 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
903 			goto err_inval;
904 		goto link_it;
905 	} else {
906 		switch (cfg->fc_type) {
907 		case RTN_UNICAST:
908 		case RTN_LOCAL:
909 		case RTN_BROADCAST:
910 		case RTN_ANYCAST:
911 		case RTN_MULTICAST:
912 			break;
913 		default:
914 			goto err_inval;
915 		}
916 	}
917 
918 	if (cfg->fc_scope > RT_SCOPE_HOST)
919 		goto err_inval;
920 
921 	if (cfg->fc_scope == RT_SCOPE_HOST) {
922 		struct fib_nh *nh = fi->fib_nh;
923 
924 		/* Local address is added. */
925 		if (nhs != 1 || nh->nh_gw)
926 			goto err_inval;
927 		nh->nh_scope = RT_SCOPE_NOWHERE;
928 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
929 		err = -ENODEV;
930 		if (nh->nh_dev == NULL)
931 			goto failure;
932 	} else {
933 		change_nexthops(fi) {
934 			err = fib_check_nh(cfg, fi, nexthop_nh);
935 			if (err != 0)
936 				goto failure;
937 		} endfor_nexthops(fi)
938 	}
939 
940 	if (fi->fib_prefsrc) {
941 		if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
942 		    fi->fib_prefsrc != cfg->fc_dst)
943 			if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
944 				goto err_inval;
945 	}
946 
947 	change_nexthops(fi) {
948 		fib_info_update_nh_saddr(net, nexthop_nh);
949 	} endfor_nexthops(fi)
950 
951 link_it:
952 	ofi = fib_find_info(fi);
953 	if (ofi) {
954 		fi->fib_dead = 1;
955 		free_fib_info(fi);
956 		ofi->fib_treeref++;
957 		return ofi;
958 	}
959 
960 	fi->fib_treeref++;
961 	atomic_inc(&fi->fib_clntref);
962 	spin_lock_bh(&fib_info_lock);
963 	hlist_add_head(&fi->fib_hash,
964 		       &fib_info_hash[fib_info_hashfn(fi)]);
965 	if (fi->fib_prefsrc) {
966 		struct hlist_head *head;
967 
968 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
969 		hlist_add_head(&fi->fib_lhash, head);
970 	}
971 	change_nexthops(fi) {
972 		struct hlist_head *head;
973 		unsigned int hash;
974 
975 		if (!nexthop_nh->nh_dev)
976 			continue;
977 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
978 		head = &fib_info_devhash[hash];
979 		hlist_add_head(&nexthop_nh->nh_hash, head);
980 	} endfor_nexthops(fi)
981 	spin_unlock_bh(&fib_info_lock);
982 	return fi;
983 
984 err_inval:
985 	err = -EINVAL;
986 
987 failure:
988 	if (fi) {
989 		fi->fib_dead = 1;
990 		free_fib_info(fi);
991 	}
992 
993 	return ERR_PTR(err);
994 }
995 
996 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
997 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
998 		  struct fib_info *fi, unsigned int flags)
999 {
1000 	struct nlmsghdr *nlh;
1001 	struct rtmsg *rtm;
1002 
1003 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1004 	if (nlh == NULL)
1005 		return -EMSGSIZE;
1006 
1007 	rtm = nlmsg_data(nlh);
1008 	rtm->rtm_family = AF_INET;
1009 	rtm->rtm_dst_len = dst_len;
1010 	rtm->rtm_src_len = 0;
1011 	rtm->rtm_tos = tos;
1012 	if (tb_id < 256)
1013 		rtm->rtm_table = tb_id;
1014 	else
1015 		rtm->rtm_table = RT_TABLE_COMPAT;
1016 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1017 		goto nla_put_failure;
1018 	rtm->rtm_type = type;
1019 	rtm->rtm_flags = fi->fib_flags;
1020 	rtm->rtm_scope = fi->fib_scope;
1021 	rtm->rtm_protocol = fi->fib_protocol;
1022 
1023 	if (rtm->rtm_dst_len &&
1024 	    nla_put_be32(skb, RTA_DST, dst))
1025 		goto nla_put_failure;
1026 	if (fi->fib_priority &&
1027 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1028 		goto nla_put_failure;
1029 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1030 		goto nla_put_failure;
1031 
1032 	if (fi->fib_prefsrc &&
1033 	    nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc))
1034 		goto nla_put_failure;
1035 	if (fi->fib_nhs == 1) {
1036 		if (fi->fib_nh->nh_gw &&
1037 		    nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1038 			goto nla_put_failure;
1039 		if (fi->fib_nh->nh_oif &&
1040 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1041 			goto nla_put_failure;
1042 #ifdef CONFIG_IP_ROUTE_CLASSID
1043 		if (fi->fib_nh[0].nh_tclassid &&
1044 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1045 			goto nla_put_failure;
1046 #endif
1047 	}
1048 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1049 	if (fi->fib_nhs > 1) {
1050 		struct rtnexthop *rtnh;
1051 		struct nlattr *mp;
1052 
1053 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1054 		if (mp == NULL)
1055 			goto nla_put_failure;
1056 
1057 		for_nexthops(fi) {
1058 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1059 			if (rtnh == NULL)
1060 				goto nla_put_failure;
1061 
1062 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1063 			rtnh->rtnh_hops = nh->nh_weight - 1;
1064 			rtnh->rtnh_ifindex = nh->nh_oif;
1065 
1066 			if (nh->nh_gw &&
1067 			    nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw))
1068 				goto nla_put_failure;
1069 #ifdef CONFIG_IP_ROUTE_CLASSID
1070 			if (nh->nh_tclassid &&
1071 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1072 				goto nla_put_failure;
1073 #endif
1074 			/* length of rtnetlink header + attributes */
1075 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1076 		} endfor_nexthops(fi);
1077 
1078 		nla_nest_end(skb, mp);
1079 	}
1080 #endif
1081 	return nlmsg_end(skb, nlh);
1082 
1083 nla_put_failure:
1084 	nlmsg_cancel(skb, nlh);
1085 	return -EMSGSIZE;
1086 }
1087 
1088 /*
1089  * Update FIB if:
1090  * - local address disappeared -> we must delete all the entries
1091  *   referring to it.
1092  * - device went down -> we must shutdown all nexthops going via it.
1093  */
1094 int fib_sync_down_addr(struct net *net, __be32 local)
1095 {
1096 	int ret = 0;
1097 	unsigned int hash = fib_laddr_hashfn(local);
1098 	struct hlist_head *head = &fib_info_laddrhash[hash];
1099 	struct hlist_node *node;
1100 	struct fib_info *fi;
1101 
1102 	if (fib_info_laddrhash == NULL || local == 0)
1103 		return 0;
1104 
1105 	hlist_for_each_entry(fi, node, head, fib_lhash) {
1106 		if (!net_eq(fi->fib_net, net))
1107 			continue;
1108 		if (fi->fib_prefsrc == local) {
1109 			fi->fib_flags |= RTNH_F_DEAD;
1110 			ret++;
1111 		}
1112 	}
1113 	return ret;
1114 }
1115 
1116 int fib_sync_down_dev(struct net_device *dev, int force)
1117 {
1118 	int ret = 0;
1119 	int scope = RT_SCOPE_NOWHERE;
1120 	struct fib_info *prev_fi = NULL;
1121 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1122 	struct hlist_head *head = &fib_info_devhash[hash];
1123 	struct hlist_node *node;
1124 	struct fib_nh *nh;
1125 
1126 	if (force)
1127 		scope = -1;
1128 
1129 	hlist_for_each_entry(nh, node, head, nh_hash) {
1130 		struct fib_info *fi = nh->nh_parent;
1131 		int dead;
1132 
1133 		BUG_ON(!fi->fib_nhs);
1134 		if (nh->nh_dev != dev || fi == prev_fi)
1135 			continue;
1136 		prev_fi = fi;
1137 		dead = 0;
1138 		change_nexthops(fi) {
1139 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1140 				dead++;
1141 			else if (nexthop_nh->nh_dev == dev &&
1142 				 nexthop_nh->nh_scope != scope) {
1143 				nexthop_nh->nh_flags |= RTNH_F_DEAD;
1144 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1145 				spin_lock_bh(&fib_multipath_lock);
1146 				fi->fib_power -= nexthop_nh->nh_power;
1147 				nexthop_nh->nh_power = 0;
1148 				spin_unlock_bh(&fib_multipath_lock);
1149 #endif
1150 				dead++;
1151 			}
1152 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1153 			if (force > 1 && nexthop_nh->nh_dev == dev) {
1154 				dead = fi->fib_nhs;
1155 				break;
1156 			}
1157 #endif
1158 		} endfor_nexthops(fi)
1159 		if (dead == fi->fib_nhs) {
1160 			fi->fib_flags |= RTNH_F_DEAD;
1161 			ret++;
1162 		}
1163 	}
1164 
1165 	return ret;
1166 }
1167 
1168 /* Must be invoked inside of an RCU protected region.  */
1169 void fib_select_default(struct fib_result *res)
1170 {
1171 	struct fib_info *fi = NULL, *last_resort = NULL;
1172 	struct list_head *fa_head = res->fa_head;
1173 	struct fib_table *tb = res->table;
1174 	int order = -1, last_idx = -1;
1175 	struct fib_alias *fa;
1176 
1177 	list_for_each_entry_rcu(fa, fa_head, fa_list) {
1178 		struct fib_info *next_fi = fa->fa_info;
1179 
1180 		if (next_fi->fib_scope != res->scope ||
1181 		    fa->fa_type != RTN_UNICAST)
1182 			continue;
1183 
1184 		if (next_fi->fib_priority > res->fi->fib_priority)
1185 			break;
1186 		if (!next_fi->fib_nh[0].nh_gw ||
1187 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1188 			continue;
1189 
1190 		fib_alias_accessed(fa);
1191 
1192 		if (fi == NULL) {
1193 			if (next_fi != res->fi)
1194 				break;
1195 		} else if (!fib_detect_death(fi, order, &last_resort,
1196 					     &last_idx, tb->tb_default)) {
1197 			fib_result_assign(res, fi);
1198 			tb->tb_default = order;
1199 			goto out;
1200 		}
1201 		fi = next_fi;
1202 		order++;
1203 	}
1204 
1205 	if (order <= 0 || fi == NULL) {
1206 		tb->tb_default = -1;
1207 		goto out;
1208 	}
1209 
1210 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1211 				tb->tb_default)) {
1212 		fib_result_assign(res, fi);
1213 		tb->tb_default = order;
1214 		goto out;
1215 	}
1216 
1217 	if (last_idx >= 0)
1218 		fib_result_assign(res, last_resort);
1219 	tb->tb_default = last_idx;
1220 out:
1221 	return;
1222 }
1223 
1224 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1225 
1226 /*
1227  * Dead device goes up. We wake up dead nexthops.
1228  * It takes sense only on multipath routes.
1229  */
1230 int fib_sync_up(struct net_device *dev)
1231 {
1232 	struct fib_info *prev_fi;
1233 	unsigned int hash;
1234 	struct hlist_head *head;
1235 	struct hlist_node *node;
1236 	struct fib_nh *nh;
1237 	int ret;
1238 
1239 	if (!(dev->flags & IFF_UP))
1240 		return 0;
1241 
1242 	prev_fi = NULL;
1243 	hash = fib_devindex_hashfn(dev->ifindex);
1244 	head = &fib_info_devhash[hash];
1245 	ret = 0;
1246 
1247 	hlist_for_each_entry(nh, node, head, nh_hash) {
1248 		struct fib_info *fi = nh->nh_parent;
1249 		int alive;
1250 
1251 		BUG_ON(!fi->fib_nhs);
1252 		if (nh->nh_dev != dev || fi == prev_fi)
1253 			continue;
1254 
1255 		prev_fi = fi;
1256 		alive = 0;
1257 		change_nexthops(fi) {
1258 			if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1259 				alive++;
1260 				continue;
1261 			}
1262 			if (nexthop_nh->nh_dev == NULL ||
1263 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1264 				continue;
1265 			if (nexthop_nh->nh_dev != dev ||
1266 			    !__in_dev_get_rtnl(dev))
1267 				continue;
1268 			alive++;
1269 			spin_lock_bh(&fib_multipath_lock);
1270 			nexthop_nh->nh_power = 0;
1271 			nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1272 			spin_unlock_bh(&fib_multipath_lock);
1273 		} endfor_nexthops(fi)
1274 
1275 		if (alive > 0) {
1276 			fi->fib_flags &= ~RTNH_F_DEAD;
1277 			ret++;
1278 		}
1279 	}
1280 
1281 	return ret;
1282 }
1283 
1284 /*
1285  * The algorithm is suboptimal, but it provides really
1286  * fair weighted route distribution.
1287  */
1288 void fib_select_multipath(struct fib_result *res)
1289 {
1290 	struct fib_info *fi = res->fi;
1291 	int w;
1292 
1293 	spin_lock_bh(&fib_multipath_lock);
1294 	if (fi->fib_power <= 0) {
1295 		int power = 0;
1296 		change_nexthops(fi) {
1297 			if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1298 				power += nexthop_nh->nh_weight;
1299 				nexthop_nh->nh_power = nexthop_nh->nh_weight;
1300 			}
1301 		} endfor_nexthops(fi);
1302 		fi->fib_power = power;
1303 		if (power <= 0) {
1304 			spin_unlock_bh(&fib_multipath_lock);
1305 			/* Race condition: route has just become dead. */
1306 			res->nh_sel = 0;
1307 			return;
1308 		}
1309 	}
1310 
1311 
1312 	/* w should be random number [0..fi->fib_power-1],
1313 	 * it is pretty bad approximation.
1314 	 */
1315 
1316 	w = jiffies % fi->fib_power;
1317 
1318 	change_nexthops(fi) {
1319 		if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1320 		    nexthop_nh->nh_power) {
1321 			w -= nexthop_nh->nh_power;
1322 			if (w <= 0) {
1323 				nexthop_nh->nh_power--;
1324 				fi->fib_power--;
1325 				res->nh_sel = nhsel;
1326 				spin_unlock_bh(&fib_multipath_lock);
1327 				return;
1328 			}
1329 		}
1330 	} endfor_nexthops(fi);
1331 
1332 	/* Race condition: route has just become dead. */
1333 	res->nh_sel = 0;
1334 	spin_unlock_bh(&fib_multipath_lock);
1335 }
1336 #endif
1337