xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 615c36f5)
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 <asm/system.h>
18 #include <linux/bitops.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/jiffies.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
27 #include <linux/in.h>
28 #include <linux/inet.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/proc_fs.h>
33 #include <linux/skbuff.h>
34 #include <linux/init.h>
35 #include <linux/slab.h>
36 
37 #include <net/arp.h>
38 #include <net/ip.h>
39 #include <net/protocol.h>
40 #include <net/route.h>
41 #include <net/tcp.h>
42 #include <net/sock.h>
43 #include <net/ip_fib.h>
44 #include <net/netlink.h>
45 #include <net/nexthop.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 /* Release a nexthop info record */
145 static void free_fib_info_rcu(struct rcu_head *head)
146 {
147 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
148 
149 	if (fi->fib_metrics != (u32 *) dst_default_metrics)
150 		kfree(fi->fib_metrics);
151 	kfree(fi);
152 }
153 
154 void free_fib_info(struct fib_info *fi)
155 {
156 	if (fi->fib_dead == 0) {
157 		pr_warning("Freeing alive fib_info %p\n", fi);
158 		return;
159 	}
160 	change_nexthops(fi) {
161 		if (nexthop_nh->nh_dev)
162 			dev_put(nexthop_nh->nh_dev);
163 		nexthop_nh->nh_dev = NULL;
164 	} endfor_nexthops(fi);
165 	fib_info_cnt--;
166 	release_net(fi->fib_net);
167 	call_rcu(&fi->rcu, free_fib_info_rcu);
168 }
169 
170 void fib_release_info(struct fib_info *fi)
171 {
172 	spin_lock_bh(&fib_info_lock);
173 	if (fi && --fi->fib_treeref == 0) {
174 		hlist_del(&fi->fib_hash);
175 		if (fi->fib_prefsrc)
176 			hlist_del(&fi->fib_lhash);
177 		change_nexthops(fi) {
178 			if (!nexthop_nh->nh_dev)
179 				continue;
180 			hlist_del(&nexthop_nh->nh_hash);
181 		} endfor_nexthops(fi)
182 		fi->fib_dead = 1;
183 		fib_info_put(fi);
184 	}
185 	spin_unlock_bh(&fib_info_lock);
186 }
187 
188 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
189 {
190 	const struct fib_nh *onh = ofi->fib_nh;
191 
192 	for_nexthops(fi) {
193 		if (nh->nh_oif != onh->nh_oif ||
194 		    nh->nh_gw  != onh->nh_gw ||
195 		    nh->nh_scope != onh->nh_scope ||
196 #ifdef CONFIG_IP_ROUTE_MULTIPATH
197 		    nh->nh_weight != onh->nh_weight ||
198 #endif
199 #ifdef CONFIG_IP_ROUTE_CLASSID
200 		    nh->nh_tclassid != onh->nh_tclassid ||
201 #endif
202 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
203 			return -1;
204 		onh++;
205 	} endfor_nexthops(fi);
206 	return 0;
207 }
208 
209 static inline unsigned int fib_devindex_hashfn(unsigned int val)
210 {
211 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
212 
213 	return (val ^
214 		(val >> DEVINDEX_HASHBITS) ^
215 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
216 }
217 
218 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
219 {
220 	unsigned int mask = (fib_info_hash_size - 1);
221 	unsigned int val = fi->fib_nhs;
222 
223 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
224 	val ^= (__force u32)fi->fib_prefsrc;
225 	val ^= fi->fib_priority;
226 	for_nexthops(fi) {
227 		val ^= fib_devindex_hashfn(nh->nh_oif);
228 	} endfor_nexthops(fi)
229 
230 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
231 }
232 
233 static struct fib_info *fib_find_info(const struct fib_info *nfi)
234 {
235 	struct hlist_head *head;
236 	struct hlist_node *node;
237 	struct fib_info *fi;
238 	unsigned int hash;
239 
240 	hash = fib_info_hashfn(nfi);
241 	head = &fib_info_hash[hash];
242 
243 	hlist_for_each_entry(fi, node, head, fib_hash) {
244 		if (!net_eq(fi->fib_net, nfi->fib_net))
245 			continue;
246 		if (fi->fib_nhs != nfi->fib_nhs)
247 			continue;
248 		if (nfi->fib_protocol == fi->fib_protocol &&
249 		    nfi->fib_scope == fi->fib_scope &&
250 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
251 		    nfi->fib_priority == fi->fib_priority &&
252 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
253 			   sizeof(u32) * RTAX_MAX) == 0 &&
254 		    ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
255 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
256 			return fi;
257 	}
258 
259 	return NULL;
260 }
261 
262 /* Check, that the gateway is already configured.
263  * Used only by redirect accept routine.
264  */
265 int ip_fib_check_default(__be32 gw, struct net_device *dev)
266 {
267 	struct hlist_head *head;
268 	struct hlist_node *node;
269 	struct fib_nh *nh;
270 	unsigned int hash;
271 
272 	spin_lock(&fib_info_lock);
273 
274 	hash = fib_devindex_hashfn(dev->ifindex);
275 	head = &fib_info_devhash[hash];
276 	hlist_for_each_entry(nh, node, head, nh_hash) {
277 		if (nh->nh_dev == dev &&
278 		    nh->nh_gw == gw &&
279 		    !(nh->nh_flags & RTNH_F_DEAD)) {
280 			spin_unlock(&fib_info_lock);
281 			return 0;
282 		}
283 	}
284 
285 	spin_unlock(&fib_info_lock);
286 
287 	return -1;
288 }
289 
290 static inline size_t fib_nlmsg_size(struct fib_info *fi)
291 {
292 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
293 			 + nla_total_size(4) /* RTA_TABLE */
294 			 + nla_total_size(4) /* RTA_DST */
295 			 + nla_total_size(4) /* RTA_PRIORITY */
296 			 + nla_total_size(4); /* RTA_PREFSRC */
297 
298 	/* space for nested metrics */
299 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
300 
301 	if (fi->fib_nhs) {
302 		/* Also handles the special case fib_nhs == 1 */
303 
304 		/* each nexthop is packed in an attribute */
305 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
306 
307 		/* may contain flow and gateway attribute */
308 		nhsize += 2 * nla_total_size(4);
309 
310 		/* all nexthops are packed in a nested attribute */
311 		payload += nla_total_size(fi->fib_nhs * nhsize);
312 	}
313 
314 	return payload;
315 }
316 
317 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
318 	       int dst_len, u32 tb_id, struct nl_info *info,
319 	       unsigned int nlm_flags)
320 {
321 	struct sk_buff *skb;
322 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
323 	int err = -ENOBUFS;
324 
325 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
326 	if (skb == NULL)
327 		goto errout;
328 
329 	err = fib_dump_info(skb, info->pid, seq, event, tb_id,
330 			    fa->fa_type, key, dst_len,
331 			    fa->fa_tos, fa->fa_info, nlm_flags);
332 	if (err < 0) {
333 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
334 		WARN_ON(err == -EMSGSIZE);
335 		kfree_skb(skb);
336 		goto errout;
337 	}
338 	rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
339 		    info->nlh, GFP_KERNEL);
340 	return;
341 errout:
342 	if (err < 0)
343 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
344 }
345 
346 /* Return the first fib alias matching TOS with
347  * priority less than or equal to PRIO.
348  */
349 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
350 {
351 	if (fah) {
352 		struct fib_alias *fa;
353 		list_for_each_entry(fa, fah, fa_list) {
354 			if (fa->fa_tos > tos)
355 				continue;
356 			if (fa->fa_info->fib_priority >= prio ||
357 			    fa->fa_tos < tos)
358 				return fa;
359 		}
360 	}
361 	return NULL;
362 }
363 
364 int fib_detect_death(struct fib_info *fi, int order,
365 		     struct fib_info **last_resort, int *last_idx, int dflt)
366 {
367 	struct neighbour *n;
368 	int state = NUD_NONE;
369 
370 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
371 	if (n) {
372 		state = n->nud_state;
373 		neigh_release(n);
374 	}
375 	if (state == NUD_REACHABLE)
376 		return 0;
377 	if ((state & NUD_VALID) && order != dflt)
378 		return 0;
379 	if ((state & NUD_VALID) ||
380 	    (*last_idx < 0 && order > dflt)) {
381 		*last_resort = fi;
382 		*last_idx = order;
383 	}
384 	return 1;
385 }
386 
387 #ifdef CONFIG_IP_ROUTE_MULTIPATH
388 
389 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
390 {
391 	int nhs = 0;
392 
393 	while (rtnh_ok(rtnh, remaining)) {
394 		nhs++;
395 		rtnh = rtnh_next(rtnh, &remaining);
396 	}
397 
398 	/* leftover implies invalid nexthop configuration, discard it */
399 	return remaining > 0 ? 0 : nhs;
400 }
401 
402 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
403 		       int remaining, struct fib_config *cfg)
404 {
405 	change_nexthops(fi) {
406 		int attrlen;
407 
408 		if (!rtnh_ok(rtnh, remaining))
409 			return -EINVAL;
410 
411 		nexthop_nh->nh_flags =
412 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
413 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
414 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
415 
416 		attrlen = rtnh_attrlen(rtnh);
417 		if (attrlen > 0) {
418 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
419 
420 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
421 			nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
422 #ifdef CONFIG_IP_ROUTE_CLASSID
423 			nla = nla_find(attrs, attrlen, RTA_FLOW);
424 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
425 #endif
426 		}
427 
428 		rtnh = rtnh_next(rtnh, &remaining);
429 	} endfor_nexthops(fi);
430 
431 	return 0;
432 }
433 
434 #endif
435 
436 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
437 {
438 #ifdef CONFIG_IP_ROUTE_MULTIPATH
439 	struct rtnexthop *rtnh;
440 	int remaining;
441 #endif
442 
443 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
444 		return 1;
445 
446 	if (cfg->fc_oif || cfg->fc_gw) {
447 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
448 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
449 			return 0;
450 		return 1;
451 	}
452 
453 #ifdef CONFIG_IP_ROUTE_MULTIPATH
454 	if (cfg->fc_mp == NULL)
455 		return 0;
456 
457 	rtnh = cfg->fc_mp;
458 	remaining = cfg->fc_mp_len;
459 
460 	for_nexthops(fi) {
461 		int attrlen;
462 
463 		if (!rtnh_ok(rtnh, remaining))
464 			return -EINVAL;
465 
466 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
467 			return 1;
468 
469 		attrlen = rtnh_attrlen(rtnh);
470 		if (attrlen < 0) {
471 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
472 
473 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
474 			if (nla && nla_get_be32(nla) != nh->nh_gw)
475 				return 1;
476 #ifdef CONFIG_IP_ROUTE_CLASSID
477 			nla = nla_find(attrs, attrlen, RTA_FLOW);
478 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
479 				return 1;
480 #endif
481 		}
482 
483 		rtnh = rtnh_next(rtnh, &remaining);
484 	} endfor_nexthops(fi);
485 #endif
486 	return 0;
487 }
488 
489 
490 /*
491  * Picture
492  * -------
493  *
494  * Semantics of nexthop is very messy by historical reasons.
495  * We have to take into account, that:
496  * a) gateway can be actually local interface address,
497  *    so that gatewayed route is direct.
498  * b) gateway must be on-link address, possibly
499  *    described not by an ifaddr, but also by a direct route.
500  * c) If both gateway and interface are specified, they should not
501  *    contradict.
502  * d) If we use tunnel routes, gateway could be not on-link.
503  *
504  * Attempt to reconcile all of these (alas, self-contradictory) conditions
505  * results in pretty ugly and hairy code with obscure logic.
506  *
507  * I chose to generalized it instead, so that the size
508  * of code does not increase practically, but it becomes
509  * much more general.
510  * Every prefix is assigned a "scope" value: "host" is local address,
511  * "link" is direct route,
512  * [ ... "site" ... "interior" ... ]
513  * and "universe" is true gateway route with global meaning.
514  *
515  * Every prefix refers to a set of "nexthop"s (gw, oif),
516  * where gw must have narrower scope. This recursion stops
517  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
518  * which means that gw is forced to be on link.
519  *
520  * Code is still hairy, but now it is apparently logically
521  * consistent and very flexible. F.e. as by-product it allows
522  * to co-exists in peace independent exterior and interior
523  * routing processes.
524  *
525  * Normally it looks as following.
526  *
527  * {universe prefix}  -> (gw, oif) [scope link]
528  *		  |
529  *		  |-> {link prefix} -> (gw, oif) [scope local]
530  *					|
531  *					|-> {local prefix} (terminal node)
532  */
533 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
534 			struct fib_nh *nh)
535 {
536 	int err;
537 	struct net *net;
538 	struct net_device *dev;
539 
540 	net = cfg->fc_nlinfo.nl_net;
541 	if (nh->nh_gw) {
542 		struct fib_result res;
543 
544 		if (nh->nh_flags & RTNH_F_ONLINK) {
545 
546 			if (cfg->fc_scope >= RT_SCOPE_LINK)
547 				return -EINVAL;
548 			if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
549 				return -EINVAL;
550 			dev = __dev_get_by_index(net, nh->nh_oif);
551 			if (!dev)
552 				return -ENODEV;
553 			if (!(dev->flags & IFF_UP))
554 				return -ENETDOWN;
555 			nh->nh_dev = dev;
556 			dev_hold(dev);
557 			nh->nh_scope = RT_SCOPE_LINK;
558 			return 0;
559 		}
560 		rcu_read_lock();
561 		{
562 			struct flowi4 fl4 = {
563 				.daddr = nh->nh_gw,
564 				.flowi4_scope = cfg->fc_scope + 1,
565 				.flowi4_oif = nh->nh_oif,
566 			};
567 
568 			/* It is not necessary, but requires a bit of thinking */
569 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
570 				fl4.flowi4_scope = RT_SCOPE_LINK;
571 			err = fib_lookup(net, &fl4, &res);
572 			if (err) {
573 				rcu_read_unlock();
574 				return err;
575 			}
576 		}
577 		err = -EINVAL;
578 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
579 			goto out;
580 		nh->nh_scope = res.scope;
581 		nh->nh_oif = FIB_RES_OIF(res);
582 		nh->nh_dev = dev = FIB_RES_DEV(res);
583 		if (!dev)
584 			goto out;
585 		dev_hold(dev);
586 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
587 	} else {
588 		struct in_device *in_dev;
589 
590 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
591 			return -EINVAL;
592 
593 		rcu_read_lock();
594 		err = -ENODEV;
595 		in_dev = inetdev_by_index(net, nh->nh_oif);
596 		if (in_dev == NULL)
597 			goto out;
598 		err = -ENETDOWN;
599 		if (!(in_dev->dev->flags & IFF_UP))
600 			goto out;
601 		nh->nh_dev = in_dev->dev;
602 		dev_hold(nh->nh_dev);
603 		nh->nh_scope = RT_SCOPE_HOST;
604 		err = 0;
605 	}
606 out:
607 	rcu_read_unlock();
608 	return err;
609 }
610 
611 static inline unsigned int fib_laddr_hashfn(__be32 val)
612 {
613 	unsigned int mask = (fib_info_hash_size - 1);
614 
615 	return ((__force u32)val ^
616 		((__force u32)val >> 7) ^
617 		((__force u32)val >> 14)) & mask;
618 }
619 
620 static struct hlist_head *fib_info_hash_alloc(int bytes)
621 {
622 	if (bytes <= PAGE_SIZE)
623 		return kzalloc(bytes, GFP_KERNEL);
624 	else
625 		return (struct hlist_head *)
626 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
627 					 get_order(bytes));
628 }
629 
630 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
631 {
632 	if (!hash)
633 		return;
634 
635 	if (bytes <= PAGE_SIZE)
636 		kfree(hash);
637 	else
638 		free_pages((unsigned long) hash, get_order(bytes));
639 }
640 
641 static void fib_info_hash_move(struct hlist_head *new_info_hash,
642 			       struct hlist_head *new_laddrhash,
643 			       unsigned int new_size)
644 {
645 	struct hlist_head *old_info_hash, *old_laddrhash;
646 	unsigned int old_size = fib_info_hash_size;
647 	unsigned int i, bytes;
648 
649 	spin_lock_bh(&fib_info_lock);
650 	old_info_hash = fib_info_hash;
651 	old_laddrhash = fib_info_laddrhash;
652 	fib_info_hash_size = new_size;
653 
654 	for (i = 0; i < old_size; i++) {
655 		struct hlist_head *head = &fib_info_hash[i];
656 		struct hlist_node *node, *n;
657 		struct fib_info *fi;
658 
659 		hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
660 			struct hlist_head *dest;
661 			unsigned int new_hash;
662 
663 			hlist_del(&fi->fib_hash);
664 
665 			new_hash = fib_info_hashfn(fi);
666 			dest = &new_info_hash[new_hash];
667 			hlist_add_head(&fi->fib_hash, dest);
668 		}
669 	}
670 	fib_info_hash = new_info_hash;
671 
672 	for (i = 0; i < old_size; i++) {
673 		struct hlist_head *lhead = &fib_info_laddrhash[i];
674 		struct hlist_node *node, *n;
675 		struct fib_info *fi;
676 
677 		hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
678 			struct hlist_head *ldest;
679 			unsigned int new_hash;
680 
681 			hlist_del(&fi->fib_lhash);
682 
683 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
684 			ldest = &new_laddrhash[new_hash];
685 			hlist_add_head(&fi->fib_lhash, ldest);
686 		}
687 	}
688 	fib_info_laddrhash = new_laddrhash;
689 
690 	spin_unlock_bh(&fib_info_lock);
691 
692 	bytes = old_size * sizeof(struct hlist_head *);
693 	fib_info_hash_free(old_info_hash, bytes);
694 	fib_info_hash_free(old_laddrhash, bytes);
695 }
696 
697 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
698 {
699 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
700 					nh->nh_gw,
701 					nh->nh_parent->fib_scope);
702 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
703 
704 	return nh->nh_saddr;
705 }
706 
707 struct fib_info *fib_create_info(struct fib_config *cfg)
708 {
709 	int err;
710 	struct fib_info *fi = NULL;
711 	struct fib_info *ofi;
712 	int nhs = 1;
713 	struct net *net = cfg->fc_nlinfo.nl_net;
714 
715 	if (cfg->fc_type > RTN_MAX)
716 		goto err_inval;
717 
718 	/* Fast check to catch the most weird cases */
719 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
720 		goto err_inval;
721 
722 #ifdef CONFIG_IP_ROUTE_MULTIPATH
723 	if (cfg->fc_mp) {
724 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
725 		if (nhs == 0)
726 			goto err_inval;
727 	}
728 #endif
729 
730 	err = -ENOBUFS;
731 	if (fib_info_cnt >= fib_info_hash_size) {
732 		unsigned int new_size = fib_info_hash_size << 1;
733 		struct hlist_head *new_info_hash;
734 		struct hlist_head *new_laddrhash;
735 		unsigned int bytes;
736 
737 		if (!new_size)
738 			new_size = 1;
739 		bytes = new_size * sizeof(struct hlist_head *);
740 		new_info_hash = fib_info_hash_alloc(bytes);
741 		new_laddrhash = fib_info_hash_alloc(bytes);
742 		if (!new_info_hash || !new_laddrhash) {
743 			fib_info_hash_free(new_info_hash, bytes);
744 			fib_info_hash_free(new_laddrhash, bytes);
745 		} else
746 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
747 
748 		if (!fib_info_hash_size)
749 			goto failure;
750 	}
751 
752 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
753 	if (fi == NULL)
754 		goto failure;
755 	if (cfg->fc_mx) {
756 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
757 		if (!fi->fib_metrics)
758 			goto failure;
759 	} else
760 		fi->fib_metrics = (u32 *) dst_default_metrics;
761 	fib_info_cnt++;
762 
763 	fi->fib_net = hold_net(net);
764 	fi->fib_protocol = cfg->fc_protocol;
765 	fi->fib_scope = cfg->fc_scope;
766 	fi->fib_flags = cfg->fc_flags;
767 	fi->fib_priority = cfg->fc_priority;
768 	fi->fib_prefsrc = cfg->fc_prefsrc;
769 
770 	fi->fib_nhs = nhs;
771 	change_nexthops(fi) {
772 		nexthop_nh->nh_parent = fi;
773 	} endfor_nexthops(fi)
774 
775 	if (cfg->fc_mx) {
776 		struct nlattr *nla;
777 		int remaining;
778 
779 		nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
780 			int type = nla_type(nla);
781 
782 			if (type) {
783 				if (type > RTAX_MAX)
784 					goto err_inval;
785 				fi->fib_metrics[type - 1] = nla_get_u32(nla);
786 			}
787 		}
788 	}
789 
790 	if (cfg->fc_mp) {
791 #ifdef CONFIG_IP_ROUTE_MULTIPATH
792 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
793 		if (err != 0)
794 			goto failure;
795 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
796 			goto err_inval;
797 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
798 			goto err_inval;
799 #ifdef CONFIG_IP_ROUTE_CLASSID
800 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
801 			goto err_inval;
802 #endif
803 #else
804 		goto err_inval;
805 #endif
806 	} else {
807 		struct fib_nh *nh = fi->fib_nh;
808 
809 		nh->nh_oif = cfg->fc_oif;
810 		nh->nh_gw = cfg->fc_gw;
811 		nh->nh_flags = cfg->fc_flags;
812 #ifdef CONFIG_IP_ROUTE_CLASSID
813 		nh->nh_tclassid = cfg->fc_flow;
814 #endif
815 #ifdef CONFIG_IP_ROUTE_MULTIPATH
816 		nh->nh_weight = 1;
817 #endif
818 	}
819 
820 	if (fib_props[cfg->fc_type].error) {
821 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
822 			goto err_inval;
823 		goto link_it;
824 	} else {
825 		switch (cfg->fc_type) {
826 		case RTN_UNICAST:
827 		case RTN_LOCAL:
828 		case RTN_BROADCAST:
829 		case RTN_ANYCAST:
830 		case RTN_MULTICAST:
831 			break;
832 		default:
833 			goto err_inval;
834 		}
835 	}
836 
837 	if (cfg->fc_scope > RT_SCOPE_HOST)
838 		goto err_inval;
839 
840 	if (cfg->fc_scope == RT_SCOPE_HOST) {
841 		struct fib_nh *nh = fi->fib_nh;
842 
843 		/* Local address is added. */
844 		if (nhs != 1 || nh->nh_gw)
845 			goto err_inval;
846 		nh->nh_scope = RT_SCOPE_NOWHERE;
847 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
848 		err = -ENODEV;
849 		if (nh->nh_dev == NULL)
850 			goto failure;
851 	} else {
852 		change_nexthops(fi) {
853 			err = fib_check_nh(cfg, fi, nexthop_nh);
854 			if (err != 0)
855 				goto failure;
856 		} endfor_nexthops(fi)
857 	}
858 
859 	if (fi->fib_prefsrc) {
860 		if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
861 		    fi->fib_prefsrc != cfg->fc_dst)
862 			if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
863 				goto err_inval;
864 	}
865 
866 	change_nexthops(fi) {
867 		fib_info_update_nh_saddr(net, nexthop_nh);
868 	} endfor_nexthops(fi)
869 
870 link_it:
871 	ofi = fib_find_info(fi);
872 	if (ofi) {
873 		fi->fib_dead = 1;
874 		free_fib_info(fi);
875 		ofi->fib_treeref++;
876 		return ofi;
877 	}
878 
879 	fi->fib_treeref++;
880 	atomic_inc(&fi->fib_clntref);
881 	spin_lock_bh(&fib_info_lock);
882 	hlist_add_head(&fi->fib_hash,
883 		       &fib_info_hash[fib_info_hashfn(fi)]);
884 	if (fi->fib_prefsrc) {
885 		struct hlist_head *head;
886 
887 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
888 		hlist_add_head(&fi->fib_lhash, head);
889 	}
890 	change_nexthops(fi) {
891 		struct hlist_head *head;
892 		unsigned int hash;
893 
894 		if (!nexthop_nh->nh_dev)
895 			continue;
896 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
897 		head = &fib_info_devhash[hash];
898 		hlist_add_head(&nexthop_nh->nh_hash, head);
899 	} endfor_nexthops(fi)
900 	spin_unlock_bh(&fib_info_lock);
901 	return fi;
902 
903 err_inval:
904 	err = -EINVAL;
905 
906 failure:
907 	if (fi) {
908 		fi->fib_dead = 1;
909 		free_fib_info(fi);
910 	}
911 
912 	return ERR_PTR(err);
913 }
914 
915 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
916 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
917 		  struct fib_info *fi, unsigned int flags)
918 {
919 	struct nlmsghdr *nlh;
920 	struct rtmsg *rtm;
921 
922 	nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
923 	if (nlh == NULL)
924 		return -EMSGSIZE;
925 
926 	rtm = nlmsg_data(nlh);
927 	rtm->rtm_family = AF_INET;
928 	rtm->rtm_dst_len = dst_len;
929 	rtm->rtm_src_len = 0;
930 	rtm->rtm_tos = tos;
931 	if (tb_id < 256)
932 		rtm->rtm_table = tb_id;
933 	else
934 		rtm->rtm_table = RT_TABLE_COMPAT;
935 	NLA_PUT_U32(skb, RTA_TABLE, tb_id);
936 	rtm->rtm_type = type;
937 	rtm->rtm_flags = fi->fib_flags;
938 	rtm->rtm_scope = fi->fib_scope;
939 	rtm->rtm_protocol = fi->fib_protocol;
940 
941 	if (rtm->rtm_dst_len)
942 		NLA_PUT_BE32(skb, RTA_DST, dst);
943 
944 	if (fi->fib_priority)
945 		NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
946 
947 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
948 		goto nla_put_failure;
949 
950 	if (fi->fib_prefsrc)
951 		NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
952 
953 	if (fi->fib_nhs == 1) {
954 		if (fi->fib_nh->nh_gw)
955 			NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
956 
957 		if (fi->fib_nh->nh_oif)
958 			NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
959 #ifdef CONFIG_IP_ROUTE_CLASSID
960 		if (fi->fib_nh[0].nh_tclassid)
961 			NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
962 #endif
963 	}
964 #ifdef CONFIG_IP_ROUTE_MULTIPATH
965 	if (fi->fib_nhs > 1) {
966 		struct rtnexthop *rtnh;
967 		struct nlattr *mp;
968 
969 		mp = nla_nest_start(skb, RTA_MULTIPATH);
970 		if (mp == NULL)
971 			goto nla_put_failure;
972 
973 		for_nexthops(fi) {
974 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
975 			if (rtnh == NULL)
976 				goto nla_put_failure;
977 
978 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
979 			rtnh->rtnh_hops = nh->nh_weight - 1;
980 			rtnh->rtnh_ifindex = nh->nh_oif;
981 
982 			if (nh->nh_gw)
983 				NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
984 #ifdef CONFIG_IP_ROUTE_CLASSID
985 			if (nh->nh_tclassid)
986 				NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
987 #endif
988 			/* length of rtnetlink header + attributes */
989 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
990 		} endfor_nexthops(fi);
991 
992 		nla_nest_end(skb, mp);
993 	}
994 #endif
995 	return nlmsg_end(skb, nlh);
996 
997 nla_put_failure:
998 	nlmsg_cancel(skb, nlh);
999 	return -EMSGSIZE;
1000 }
1001 
1002 /*
1003  * Update FIB if:
1004  * - local address disappeared -> we must delete all the entries
1005  *   referring to it.
1006  * - device went down -> we must shutdown all nexthops going via it.
1007  */
1008 int fib_sync_down_addr(struct net *net, __be32 local)
1009 {
1010 	int ret = 0;
1011 	unsigned int hash = fib_laddr_hashfn(local);
1012 	struct hlist_head *head = &fib_info_laddrhash[hash];
1013 	struct hlist_node *node;
1014 	struct fib_info *fi;
1015 
1016 	if (fib_info_laddrhash == NULL || local == 0)
1017 		return 0;
1018 
1019 	hlist_for_each_entry(fi, node, head, fib_lhash) {
1020 		if (!net_eq(fi->fib_net, net))
1021 			continue;
1022 		if (fi->fib_prefsrc == local) {
1023 			fi->fib_flags |= RTNH_F_DEAD;
1024 			ret++;
1025 		}
1026 	}
1027 	return ret;
1028 }
1029 
1030 int fib_sync_down_dev(struct net_device *dev, int force)
1031 {
1032 	int ret = 0;
1033 	int scope = RT_SCOPE_NOWHERE;
1034 	struct fib_info *prev_fi = NULL;
1035 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1036 	struct hlist_head *head = &fib_info_devhash[hash];
1037 	struct hlist_node *node;
1038 	struct fib_nh *nh;
1039 
1040 	if (force)
1041 		scope = -1;
1042 
1043 	hlist_for_each_entry(nh, node, head, nh_hash) {
1044 		struct fib_info *fi = nh->nh_parent;
1045 		int dead;
1046 
1047 		BUG_ON(!fi->fib_nhs);
1048 		if (nh->nh_dev != dev || fi == prev_fi)
1049 			continue;
1050 		prev_fi = fi;
1051 		dead = 0;
1052 		change_nexthops(fi) {
1053 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1054 				dead++;
1055 			else if (nexthop_nh->nh_dev == dev &&
1056 				 nexthop_nh->nh_scope != scope) {
1057 				nexthop_nh->nh_flags |= RTNH_F_DEAD;
1058 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1059 				spin_lock_bh(&fib_multipath_lock);
1060 				fi->fib_power -= nexthop_nh->nh_power;
1061 				nexthop_nh->nh_power = 0;
1062 				spin_unlock_bh(&fib_multipath_lock);
1063 #endif
1064 				dead++;
1065 			}
1066 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1067 			if (force > 1 && nexthop_nh->nh_dev == dev) {
1068 				dead = fi->fib_nhs;
1069 				break;
1070 			}
1071 #endif
1072 		} endfor_nexthops(fi)
1073 		if (dead == fi->fib_nhs) {
1074 			fi->fib_flags |= RTNH_F_DEAD;
1075 			ret++;
1076 		}
1077 	}
1078 
1079 	return ret;
1080 }
1081 
1082 /* Must be invoked inside of an RCU protected region.  */
1083 void fib_select_default(struct fib_result *res)
1084 {
1085 	struct fib_info *fi = NULL, *last_resort = NULL;
1086 	struct list_head *fa_head = res->fa_head;
1087 	struct fib_table *tb = res->table;
1088 	int order = -1, last_idx = -1;
1089 	struct fib_alias *fa;
1090 
1091 	list_for_each_entry_rcu(fa, fa_head, fa_list) {
1092 		struct fib_info *next_fi = fa->fa_info;
1093 
1094 		if (next_fi->fib_scope != res->scope ||
1095 		    fa->fa_type != RTN_UNICAST)
1096 			continue;
1097 
1098 		if (next_fi->fib_priority > res->fi->fib_priority)
1099 			break;
1100 		if (!next_fi->fib_nh[0].nh_gw ||
1101 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1102 			continue;
1103 
1104 		fib_alias_accessed(fa);
1105 
1106 		if (fi == NULL) {
1107 			if (next_fi != res->fi)
1108 				break;
1109 		} else if (!fib_detect_death(fi, order, &last_resort,
1110 					     &last_idx, tb->tb_default)) {
1111 			fib_result_assign(res, fi);
1112 			tb->tb_default = order;
1113 			goto out;
1114 		}
1115 		fi = next_fi;
1116 		order++;
1117 	}
1118 
1119 	if (order <= 0 || fi == NULL) {
1120 		tb->tb_default = -1;
1121 		goto out;
1122 	}
1123 
1124 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1125 				tb->tb_default)) {
1126 		fib_result_assign(res, fi);
1127 		tb->tb_default = order;
1128 		goto out;
1129 	}
1130 
1131 	if (last_idx >= 0)
1132 		fib_result_assign(res, last_resort);
1133 	tb->tb_default = last_idx;
1134 out:
1135 	return;
1136 }
1137 
1138 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1139 
1140 /*
1141  * Dead device goes up. We wake up dead nexthops.
1142  * It takes sense only on multipath routes.
1143  */
1144 int fib_sync_up(struct net_device *dev)
1145 {
1146 	struct fib_info *prev_fi;
1147 	unsigned int hash;
1148 	struct hlist_head *head;
1149 	struct hlist_node *node;
1150 	struct fib_nh *nh;
1151 	int ret;
1152 
1153 	if (!(dev->flags & IFF_UP))
1154 		return 0;
1155 
1156 	prev_fi = NULL;
1157 	hash = fib_devindex_hashfn(dev->ifindex);
1158 	head = &fib_info_devhash[hash];
1159 	ret = 0;
1160 
1161 	hlist_for_each_entry(nh, node, head, nh_hash) {
1162 		struct fib_info *fi = nh->nh_parent;
1163 		int alive;
1164 
1165 		BUG_ON(!fi->fib_nhs);
1166 		if (nh->nh_dev != dev || fi == prev_fi)
1167 			continue;
1168 
1169 		prev_fi = fi;
1170 		alive = 0;
1171 		change_nexthops(fi) {
1172 			if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1173 				alive++;
1174 				continue;
1175 			}
1176 			if (nexthop_nh->nh_dev == NULL ||
1177 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1178 				continue;
1179 			if (nexthop_nh->nh_dev != dev ||
1180 			    !__in_dev_get_rtnl(dev))
1181 				continue;
1182 			alive++;
1183 			spin_lock_bh(&fib_multipath_lock);
1184 			nexthop_nh->nh_power = 0;
1185 			nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1186 			spin_unlock_bh(&fib_multipath_lock);
1187 		} endfor_nexthops(fi)
1188 
1189 		if (alive > 0) {
1190 			fi->fib_flags &= ~RTNH_F_DEAD;
1191 			ret++;
1192 		}
1193 	}
1194 
1195 	return ret;
1196 }
1197 
1198 /*
1199  * The algorithm is suboptimal, but it provides really
1200  * fair weighted route distribution.
1201  */
1202 void fib_select_multipath(struct fib_result *res)
1203 {
1204 	struct fib_info *fi = res->fi;
1205 	int w;
1206 
1207 	spin_lock_bh(&fib_multipath_lock);
1208 	if (fi->fib_power <= 0) {
1209 		int power = 0;
1210 		change_nexthops(fi) {
1211 			if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1212 				power += nexthop_nh->nh_weight;
1213 				nexthop_nh->nh_power = nexthop_nh->nh_weight;
1214 			}
1215 		} endfor_nexthops(fi);
1216 		fi->fib_power = power;
1217 		if (power <= 0) {
1218 			spin_unlock_bh(&fib_multipath_lock);
1219 			/* Race condition: route has just become dead. */
1220 			res->nh_sel = 0;
1221 			return;
1222 		}
1223 	}
1224 
1225 
1226 	/* w should be random number [0..fi->fib_power-1],
1227 	 * it is pretty bad approximation.
1228 	 */
1229 
1230 	w = jiffies % fi->fib_power;
1231 
1232 	change_nexthops(fi) {
1233 		if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1234 		    nexthop_nh->nh_power) {
1235 			w -= nexthop_nh->nh_power;
1236 			if (w <= 0) {
1237 				nexthop_nh->nh_power--;
1238 				fi->fib_power--;
1239 				res->nh_sel = nhsel;
1240 				spin_unlock_bh(&fib_multipath_lock);
1241 				return;
1242 			}
1243 		}
1244 	} endfor_nexthops(fi);
1245 
1246 	/* Race condition: route has just become dead. */
1247 	res->nh_sel = 0;
1248 	spin_unlock_bh(&fib_multipath_lock);
1249 }
1250 #endif
1251