xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 565d76cb)
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 
145 /* Release a nexthop info record */
146 
147 static void free_fib_info_rcu(struct rcu_head *head)
148 {
149 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
150 
151 	if (fi->fib_metrics != (u32 *) dst_default_metrics)
152 		kfree(fi->fib_metrics);
153 	kfree(fi);
154 }
155 
156 void free_fib_info(struct fib_info *fi)
157 {
158 	if (fi->fib_dead == 0) {
159 		pr_warning("Freeing alive fib_info %p\n", fi);
160 		return;
161 	}
162 	change_nexthops(fi) {
163 		if (nexthop_nh->nh_dev)
164 			dev_put(nexthop_nh->nh_dev);
165 		nexthop_nh->nh_dev = NULL;
166 	} endfor_nexthops(fi);
167 	fib_info_cnt--;
168 	release_net(fi->fib_net);
169 	call_rcu(&fi->rcu, free_fib_info_rcu);
170 }
171 
172 void fib_release_info(struct fib_info *fi)
173 {
174 	spin_lock_bh(&fib_info_lock);
175 	if (fi && --fi->fib_treeref == 0) {
176 		hlist_del(&fi->fib_hash);
177 		if (fi->fib_prefsrc)
178 			hlist_del(&fi->fib_lhash);
179 		change_nexthops(fi) {
180 			if (!nexthop_nh->nh_dev)
181 				continue;
182 			hlist_del(&nexthop_nh->nh_hash);
183 		} endfor_nexthops(fi)
184 		fi->fib_dead = 1;
185 		fib_info_put(fi);
186 	}
187 	spin_unlock_bh(&fib_info_lock);
188 }
189 
190 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
191 {
192 	const struct fib_nh *onh = ofi->fib_nh;
193 
194 	for_nexthops(fi) {
195 		if (nh->nh_oif != onh->nh_oif ||
196 		    nh->nh_gw  != onh->nh_gw ||
197 		    nh->nh_scope != onh->nh_scope ||
198 #ifdef CONFIG_IP_ROUTE_MULTIPATH
199 		    nh->nh_weight != onh->nh_weight ||
200 #endif
201 #ifdef CONFIG_IP_ROUTE_CLASSID
202 		    nh->nh_tclassid != onh->nh_tclassid ||
203 #endif
204 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
205 			return -1;
206 		onh++;
207 	} endfor_nexthops(fi);
208 	return 0;
209 }
210 
211 static inline unsigned int fib_devindex_hashfn(unsigned int val)
212 {
213 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
214 
215 	return (val ^
216 		(val >> DEVINDEX_HASHBITS) ^
217 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
218 }
219 
220 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
221 {
222 	unsigned int mask = (fib_info_hash_size - 1);
223 	unsigned int val = fi->fib_nhs;
224 
225 	val ^= fi->fib_protocol;
226 	val ^= (__force u32)fi->fib_prefsrc;
227 	val ^= fi->fib_priority;
228 	for_nexthops(fi) {
229 		val ^= fib_devindex_hashfn(nh->nh_oif);
230 	} endfor_nexthops(fi)
231 
232 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
233 }
234 
235 static struct fib_info *fib_find_info(const struct fib_info *nfi)
236 {
237 	struct hlist_head *head;
238 	struct hlist_node *node;
239 	struct fib_info *fi;
240 	unsigned int hash;
241 
242 	hash = fib_info_hashfn(nfi);
243 	head = &fib_info_hash[hash];
244 
245 	hlist_for_each_entry(fi, node, head, fib_hash) {
246 		if (!net_eq(fi->fib_net, nfi->fib_net))
247 			continue;
248 		if (fi->fib_nhs != nfi->fib_nhs)
249 			continue;
250 		if (nfi->fib_protocol == fi->fib_protocol &&
251 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
252 		    nfi->fib_priority == fi->fib_priority &&
253 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
254 			   sizeof(fi->fib_metrics)) == 0 &&
255 		    ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
256 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
257 			return fi;
258 	}
259 
260 	return NULL;
261 }
262 
263 /* Check, that the gateway is already configured.
264  * Used only by redirect accept routine.
265  */
266 int ip_fib_check_default(__be32 gw, struct net_device *dev)
267 {
268 	struct hlist_head *head;
269 	struct hlist_node *node;
270 	struct fib_nh *nh;
271 	unsigned int hash;
272 
273 	spin_lock(&fib_info_lock);
274 
275 	hash = fib_devindex_hashfn(dev->ifindex);
276 	head = &fib_info_devhash[hash];
277 	hlist_for_each_entry(nh, node, head, nh_hash) {
278 		if (nh->nh_dev == dev &&
279 		    nh->nh_gw == gw &&
280 		    !(nh->nh_flags & RTNH_F_DEAD)) {
281 			spin_unlock(&fib_info_lock);
282 			return 0;
283 		}
284 	}
285 
286 	spin_unlock(&fib_info_lock);
287 
288 	return -1;
289 }
290 
291 static inline size_t fib_nlmsg_size(struct fib_info *fi)
292 {
293 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
294 			 + nla_total_size(4) /* RTA_TABLE */
295 			 + nla_total_size(4) /* RTA_DST */
296 			 + nla_total_size(4) /* RTA_PRIORITY */
297 			 + nla_total_size(4); /* RTA_PREFSRC */
298 
299 	/* space for nested metrics */
300 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
301 
302 	if (fi->fib_nhs) {
303 		/* Also handles the special case fib_nhs == 1 */
304 
305 		/* each nexthop is packed in an attribute */
306 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
307 
308 		/* may contain flow and gateway attribute */
309 		nhsize += 2 * nla_total_size(4);
310 
311 		/* all nexthops are packed in a nested attribute */
312 		payload += nla_total_size(fi->fib_nhs * nhsize);
313 	}
314 
315 	return payload;
316 }
317 
318 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
319 	       int dst_len, u32 tb_id, struct nl_info *info,
320 	       unsigned int nlm_flags)
321 {
322 	struct sk_buff *skb;
323 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
324 	int err = -ENOBUFS;
325 
326 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
327 	if (skb == NULL)
328 		goto errout;
329 
330 	err = fib_dump_info(skb, info->pid, seq, event, tb_id,
331 			    fa->fa_type, fa->fa_scope, key, dst_len,
332 			    fa->fa_tos, fa->fa_info, nlm_flags);
333 	if (err < 0) {
334 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
335 		WARN_ON(err == -EMSGSIZE);
336 		kfree_skb(skb);
337 		goto errout;
338 	}
339 	rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
340 		    info->nlh, GFP_KERNEL);
341 	return;
342 errout:
343 	if (err < 0)
344 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
345 }
346 
347 /* Return the first fib alias matching TOS with
348  * priority less than or equal to PRIO.
349  */
350 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
351 {
352 	if (fah) {
353 		struct fib_alias *fa;
354 		list_for_each_entry(fa, fah, fa_list) {
355 			if (fa->fa_tos > tos)
356 				continue;
357 			if (fa->fa_info->fib_priority >= prio ||
358 			    fa->fa_tos < tos)
359 				return fa;
360 		}
361 	}
362 	return NULL;
363 }
364 
365 int fib_detect_death(struct fib_info *fi, int order,
366 		     struct fib_info **last_resort, int *last_idx, int dflt)
367 {
368 	struct neighbour *n;
369 	int state = NUD_NONE;
370 
371 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
372 	if (n) {
373 		state = n->nud_state;
374 		neigh_release(n);
375 	}
376 	if (state == NUD_REACHABLE)
377 		return 0;
378 	if ((state & NUD_VALID) && order != dflt)
379 		return 0;
380 	if ((state & NUD_VALID) ||
381 	    (*last_idx < 0 && order > dflt)) {
382 		*last_resort = fi;
383 		*last_idx = order;
384 	}
385 	return 1;
386 }
387 
388 #ifdef CONFIG_IP_ROUTE_MULTIPATH
389 
390 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
391 {
392 	int nhs = 0;
393 
394 	while (rtnh_ok(rtnh, remaining)) {
395 		nhs++;
396 		rtnh = rtnh_next(rtnh, &remaining);
397 	}
398 
399 	/* leftover implies invalid nexthop configuration, discard it */
400 	return remaining > 0 ? 0 : nhs;
401 }
402 
403 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
404 		       int remaining, struct fib_config *cfg)
405 {
406 	change_nexthops(fi) {
407 		int attrlen;
408 
409 		if (!rtnh_ok(rtnh, remaining))
410 			return -EINVAL;
411 
412 		nexthop_nh->nh_flags =
413 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
414 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
415 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
416 
417 		attrlen = rtnh_attrlen(rtnh);
418 		if (attrlen > 0) {
419 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
420 
421 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
422 			nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
423 #ifdef CONFIG_IP_ROUTE_CLASSID
424 			nla = nla_find(attrs, attrlen, RTA_FLOW);
425 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
426 #endif
427 		}
428 
429 		rtnh = rtnh_next(rtnh, &remaining);
430 	} endfor_nexthops(fi);
431 
432 	return 0;
433 }
434 
435 #endif
436 
437 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
438 {
439 #ifdef CONFIG_IP_ROUTE_MULTIPATH
440 	struct rtnexthop *rtnh;
441 	int remaining;
442 #endif
443 
444 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
445 		return 1;
446 
447 	if (cfg->fc_oif || cfg->fc_gw) {
448 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
449 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
450 			return 0;
451 		return 1;
452 	}
453 
454 #ifdef CONFIG_IP_ROUTE_MULTIPATH
455 	if (cfg->fc_mp == NULL)
456 		return 0;
457 
458 	rtnh = cfg->fc_mp;
459 	remaining = cfg->fc_mp_len;
460 
461 	for_nexthops(fi) {
462 		int attrlen;
463 
464 		if (!rtnh_ok(rtnh, remaining))
465 			return -EINVAL;
466 
467 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
468 			return 1;
469 
470 		attrlen = rtnh_attrlen(rtnh);
471 		if (attrlen < 0) {
472 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
473 
474 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
475 			if (nla && nla_get_be32(nla) != nh->nh_gw)
476 				return 1;
477 #ifdef CONFIG_IP_ROUTE_CLASSID
478 			nla = nla_find(attrs, attrlen, RTA_FLOW);
479 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
480 				return 1;
481 #endif
482 		}
483 
484 		rtnh = rtnh_next(rtnh, &remaining);
485 	} endfor_nexthops(fi);
486 #endif
487 	return 0;
488 }
489 
490 
491 /*
492  * Picture
493  * -------
494  *
495  * Semantics of nexthop is very messy by historical reasons.
496  * We have to take into account, that:
497  * a) gateway can be actually local interface address,
498  *    so that gatewayed route is direct.
499  * b) gateway must be on-link address, possibly
500  *    described not by an ifaddr, but also by a direct route.
501  * c) If both gateway and interface are specified, they should not
502  *    contradict.
503  * d) If we use tunnel routes, gateway could be not on-link.
504  *
505  * Attempt to reconcile all of these (alas, self-contradictory) conditions
506  * results in pretty ugly and hairy code with obscure logic.
507  *
508  * I chose to generalized it instead, so that the size
509  * of code does not increase practically, but it becomes
510  * much more general.
511  * Every prefix is assigned a "scope" value: "host" is local address,
512  * "link" is direct route,
513  * [ ... "site" ... "interior" ... ]
514  * and "universe" is true gateway route with global meaning.
515  *
516  * Every prefix refers to a set of "nexthop"s (gw, oif),
517  * where gw must have narrower scope. This recursion stops
518  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
519  * which means that gw is forced to be on link.
520  *
521  * Code is still hairy, but now it is apparently logically
522  * consistent and very flexible. F.e. as by-product it allows
523  * to co-exists in peace independent exterior and interior
524  * routing processes.
525  *
526  * Normally it looks as following.
527  *
528  * {universe prefix}  -> (gw, oif) [scope link]
529  *		  |
530  *		  |-> {link prefix} -> (gw, oif) [scope local]
531  *					|
532  *					|-> {local prefix} (terminal node)
533  */
534 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
535 			struct fib_nh *nh)
536 {
537 	int err;
538 	struct net *net;
539 	struct net_device *dev;
540 
541 	net = cfg->fc_nlinfo.nl_net;
542 	if (nh->nh_gw) {
543 		struct fib_result res;
544 
545 		if (nh->nh_flags & RTNH_F_ONLINK) {
546 
547 			if (cfg->fc_scope >= RT_SCOPE_LINK)
548 				return -EINVAL;
549 			if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
550 				return -EINVAL;
551 			dev = __dev_get_by_index(net, nh->nh_oif);
552 			if (!dev)
553 				return -ENODEV;
554 			if (!(dev->flags & IFF_UP))
555 				return -ENETDOWN;
556 			nh->nh_dev = dev;
557 			dev_hold(dev);
558 			nh->nh_scope = RT_SCOPE_LINK;
559 			return 0;
560 		}
561 		rcu_read_lock();
562 		{
563 			struct flowi4 fl4 = {
564 				.daddr = nh->nh_gw,
565 				.flowi4_scope = cfg->fc_scope + 1,
566 				.flowi4_oif = nh->nh_oif,
567 			};
568 
569 			/* It is not necessary, but requires a bit of thinking */
570 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
571 				fl4.flowi4_scope = RT_SCOPE_LINK;
572 			err = fib_lookup(net, &fl4, &res);
573 			if (err) {
574 				rcu_read_unlock();
575 				return err;
576 			}
577 		}
578 		err = -EINVAL;
579 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
580 			goto out;
581 		nh->nh_scope = res.scope;
582 		nh->nh_oif = FIB_RES_OIF(res);
583 		nh->nh_dev = dev = FIB_RES_DEV(res);
584 		if (!dev)
585 			goto out;
586 		dev_hold(dev);
587 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
588 	} else {
589 		struct in_device *in_dev;
590 
591 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
592 			return -EINVAL;
593 
594 		rcu_read_lock();
595 		err = -ENODEV;
596 		in_dev = inetdev_by_index(net, nh->nh_oif);
597 		if (in_dev == NULL)
598 			goto out;
599 		err = -ENETDOWN;
600 		if (!(in_dev->dev->flags & IFF_UP))
601 			goto out;
602 		nh->nh_dev = in_dev->dev;
603 		dev_hold(nh->nh_dev);
604 		nh->nh_scope = RT_SCOPE_HOST;
605 		err = 0;
606 	}
607 out:
608 	rcu_read_unlock();
609 	return err;
610 }
611 
612 static inline unsigned int fib_laddr_hashfn(__be32 val)
613 {
614 	unsigned int mask = (fib_info_hash_size - 1);
615 
616 	return ((__force u32)val ^
617 		((__force u32)val >> 7) ^
618 		((__force u32)val >> 14)) & mask;
619 }
620 
621 static struct hlist_head *fib_info_hash_alloc(int bytes)
622 {
623 	if (bytes <= PAGE_SIZE)
624 		return kzalloc(bytes, GFP_KERNEL);
625 	else
626 		return (struct hlist_head *)
627 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
628 					 get_order(bytes));
629 }
630 
631 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
632 {
633 	if (!hash)
634 		return;
635 
636 	if (bytes <= PAGE_SIZE)
637 		kfree(hash);
638 	else
639 		free_pages((unsigned long) hash, get_order(bytes));
640 }
641 
642 static void fib_info_hash_move(struct hlist_head *new_info_hash,
643 			       struct hlist_head *new_laddrhash,
644 			       unsigned int new_size)
645 {
646 	struct hlist_head *old_info_hash, *old_laddrhash;
647 	unsigned int old_size = fib_info_hash_size;
648 	unsigned int i, bytes;
649 
650 	spin_lock_bh(&fib_info_lock);
651 	old_info_hash = fib_info_hash;
652 	old_laddrhash = fib_info_laddrhash;
653 	fib_info_hash_size = new_size;
654 
655 	for (i = 0; i < old_size; i++) {
656 		struct hlist_head *head = &fib_info_hash[i];
657 		struct hlist_node *node, *n;
658 		struct fib_info *fi;
659 
660 		hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
661 			struct hlist_head *dest;
662 			unsigned int new_hash;
663 
664 			hlist_del(&fi->fib_hash);
665 
666 			new_hash = fib_info_hashfn(fi);
667 			dest = &new_info_hash[new_hash];
668 			hlist_add_head(&fi->fib_hash, dest);
669 		}
670 	}
671 	fib_info_hash = new_info_hash;
672 
673 	for (i = 0; i < old_size; i++) {
674 		struct hlist_head *lhead = &fib_info_laddrhash[i];
675 		struct hlist_node *node, *n;
676 		struct fib_info *fi;
677 
678 		hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
679 			struct hlist_head *ldest;
680 			unsigned int new_hash;
681 
682 			hlist_del(&fi->fib_lhash);
683 
684 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
685 			ldest = &new_laddrhash[new_hash];
686 			hlist_add_head(&fi->fib_lhash, ldest);
687 		}
688 	}
689 	fib_info_laddrhash = new_laddrhash;
690 
691 	spin_unlock_bh(&fib_info_lock);
692 
693 	bytes = old_size * sizeof(struct hlist_head *);
694 	fib_info_hash_free(old_info_hash, bytes);
695 	fib_info_hash_free(old_laddrhash, bytes);
696 }
697 
698 struct fib_info *fib_create_info(struct fib_config *cfg)
699 {
700 	int err;
701 	struct fib_info *fi = NULL;
702 	struct fib_info *ofi;
703 	int nhs = 1;
704 	struct net *net = cfg->fc_nlinfo.nl_net;
705 
706 	if (cfg->fc_type > RTN_MAX)
707 		goto err_inval;
708 
709 	/* Fast check to catch the most weird cases */
710 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
711 		goto err_inval;
712 
713 #ifdef CONFIG_IP_ROUTE_MULTIPATH
714 	if (cfg->fc_mp) {
715 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
716 		if (nhs == 0)
717 			goto err_inval;
718 	}
719 #endif
720 
721 	err = -ENOBUFS;
722 	if (fib_info_cnt >= fib_info_hash_size) {
723 		unsigned int new_size = fib_info_hash_size << 1;
724 		struct hlist_head *new_info_hash;
725 		struct hlist_head *new_laddrhash;
726 		unsigned int bytes;
727 
728 		if (!new_size)
729 			new_size = 1;
730 		bytes = new_size * sizeof(struct hlist_head *);
731 		new_info_hash = fib_info_hash_alloc(bytes);
732 		new_laddrhash = fib_info_hash_alloc(bytes);
733 		if (!new_info_hash || !new_laddrhash) {
734 			fib_info_hash_free(new_info_hash, bytes);
735 			fib_info_hash_free(new_laddrhash, bytes);
736 		} else
737 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
738 
739 		if (!fib_info_hash_size)
740 			goto failure;
741 	}
742 
743 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
744 	if (fi == NULL)
745 		goto failure;
746 	if (cfg->fc_mx) {
747 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
748 		if (!fi->fib_metrics)
749 			goto failure;
750 	} else
751 		fi->fib_metrics = (u32 *) dst_default_metrics;
752 	fib_info_cnt++;
753 
754 	fi->fib_net = hold_net(net);
755 	fi->fib_protocol = cfg->fc_protocol;
756 	fi->fib_flags = cfg->fc_flags;
757 	fi->fib_priority = cfg->fc_priority;
758 	fi->fib_prefsrc = cfg->fc_prefsrc;
759 
760 	fi->fib_nhs = nhs;
761 	change_nexthops(fi) {
762 		nexthop_nh->nh_parent = fi;
763 	} endfor_nexthops(fi)
764 
765 	if (cfg->fc_mx) {
766 		struct nlattr *nla;
767 		int remaining;
768 
769 		nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
770 			int type = nla_type(nla);
771 
772 			if (type) {
773 				if (type > RTAX_MAX)
774 					goto err_inval;
775 				fi->fib_metrics[type - 1] = nla_get_u32(nla);
776 			}
777 		}
778 	}
779 
780 	if (cfg->fc_mp) {
781 #ifdef CONFIG_IP_ROUTE_MULTIPATH
782 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
783 		if (err != 0)
784 			goto failure;
785 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
786 			goto err_inval;
787 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
788 			goto err_inval;
789 #ifdef CONFIG_IP_ROUTE_CLASSID
790 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
791 			goto err_inval;
792 #endif
793 #else
794 		goto err_inval;
795 #endif
796 	} else {
797 		struct fib_nh *nh = fi->fib_nh;
798 
799 		nh->nh_oif = cfg->fc_oif;
800 		nh->nh_gw = cfg->fc_gw;
801 		nh->nh_flags = cfg->fc_flags;
802 #ifdef CONFIG_IP_ROUTE_CLASSID
803 		nh->nh_tclassid = cfg->fc_flow;
804 #endif
805 #ifdef CONFIG_IP_ROUTE_MULTIPATH
806 		nh->nh_weight = 1;
807 #endif
808 	}
809 
810 	if (fib_props[cfg->fc_type].error) {
811 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
812 			goto err_inval;
813 		goto link_it;
814 	} else {
815 		switch (cfg->fc_type) {
816 		case RTN_UNICAST:
817 		case RTN_LOCAL:
818 		case RTN_BROADCAST:
819 		case RTN_ANYCAST:
820 		case RTN_MULTICAST:
821 			break;
822 		default:
823 			goto err_inval;
824 		}
825 	}
826 
827 	if (cfg->fc_scope > RT_SCOPE_HOST)
828 		goto err_inval;
829 
830 	if (cfg->fc_scope == RT_SCOPE_HOST) {
831 		struct fib_nh *nh = fi->fib_nh;
832 
833 		/* Local address is added. */
834 		if (nhs != 1 || nh->nh_gw)
835 			goto err_inval;
836 		nh->nh_scope = RT_SCOPE_NOWHERE;
837 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
838 		err = -ENODEV;
839 		if (nh->nh_dev == NULL)
840 			goto failure;
841 	} else {
842 		change_nexthops(fi) {
843 			err = fib_check_nh(cfg, fi, nexthop_nh);
844 			if (err != 0)
845 				goto failure;
846 		} endfor_nexthops(fi)
847 	}
848 
849 	if (fi->fib_prefsrc) {
850 		if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
851 		    fi->fib_prefsrc != cfg->fc_dst)
852 			if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
853 				goto err_inval;
854 	}
855 
856 	change_nexthops(fi) {
857 		nexthop_nh->nh_cfg_scope = cfg->fc_scope;
858 		nexthop_nh->nh_saddr = inet_select_addr(nexthop_nh->nh_dev,
859 							nexthop_nh->nh_gw,
860 							nexthop_nh->nh_cfg_scope);
861 	} endfor_nexthops(fi)
862 
863 link_it:
864 	ofi = fib_find_info(fi);
865 	if (ofi) {
866 		fi->fib_dead = 1;
867 		free_fib_info(fi);
868 		ofi->fib_treeref++;
869 		return ofi;
870 	}
871 
872 	fi->fib_treeref++;
873 	atomic_inc(&fi->fib_clntref);
874 	spin_lock_bh(&fib_info_lock);
875 	hlist_add_head(&fi->fib_hash,
876 		       &fib_info_hash[fib_info_hashfn(fi)]);
877 	if (fi->fib_prefsrc) {
878 		struct hlist_head *head;
879 
880 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
881 		hlist_add_head(&fi->fib_lhash, head);
882 	}
883 	change_nexthops(fi) {
884 		struct hlist_head *head;
885 		unsigned int hash;
886 
887 		if (!nexthop_nh->nh_dev)
888 			continue;
889 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
890 		head = &fib_info_devhash[hash];
891 		hlist_add_head(&nexthop_nh->nh_hash, head);
892 	} endfor_nexthops(fi)
893 	spin_unlock_bh(&fib_info_lock);
894 	return fi;
895 
896 err_inval:
897 	err = -EINVAL;
898 
899 failure:
900 	if (fi) {
901 		fi->fib_dead = 1;
902 		free_fib_info(fi);
903 	}
904 
905 	return ERR_PTR(err);
906 }
907 
908 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
909 		  u32 tb_id, u8 type, u8 scope, __be32 dst, int dst_len, u8 tos,
910 		  struct fib_info *fi, unsigned int flags)
911 {
912 	struct nlmsghdr *nlh;
913 	struct rtmsg *rtm;
914 
915 	nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
916 	if (nlh == NULL)
917 		return -EMSGSIZE;
918 
919 	rtm = nlmsg_data(nlh);
920 	rtm->rtm_family = AF_INET;
921 	rtm->rtm_dst_len = dst_len;
922 	rtm->rtm_src_len = 0;
923 	rtm->rtm_tos = tos;
924 	if (tb_id < 256)
925 		rtm->rtm_table = tb_id;
926 	else
927 		rtm->rtm_table = RT_TABLE_COMPAT;
928 	NLA_PUT_U32(skb, RTA_TABLE, tb_id);
929 	rtm->rtm_type = type;
930 	rtm->rtm_flags = fi->fib_flags;
931 	rtm->rtm_scope = scope;
932 	rtm->rtm_protocol = fi->fib_protocol;
933 
934 	if (rtm->rtm_dst_len)
935 		NLA_PUT_BE32(skb, RTA_DST, dst);
936 
937 	if (fi->fib_priority)
938 		NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
939 
940 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
941 		goto nla_put_failure;
942 
943 	if (fi->fib_prefsrc)
944 		NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
945 
946 	if (fi->fib_nhs == 1) {
947 		if (fi->fib_nh->nh_gw)
948 			NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
949 
950 		if (fi->fib_nh->nh_oif)
951 			NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
952 #ifdef CONFIG_IP_ROUTE_CLASSID
953 		if (fi->fib_nh[0].nh_tclassid)
954 			NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
955 #endif
956 	}
957 #ifdef CONFIG_IP_ROUTE_MULTIPATH
958 	if (fi->fib_nhs > 1) {
959 		struct rtnexthop *rtnh;
960 		struct nlattr *mp;
961 
962 		mp = nla_nest_start(skb, RTA_MULTIPATH);
963 		if (mp == NULL)
964 			goto nla_put_failure;
965 
966 		for_nexthops(fi) {
967 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
968 			if (rtnh == NULL)
969 				goto nla_put_failure;
970 
971 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
972 			rtnh->rtnh_hops = nh->nh_weight - 1;
973 			rtnh->rtnh_ifindex = nh->nh_oif;
974 
975 			if (nh->nh_gw)
976 				NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
977 #ifdef CONFIG_IP_ROUTE_CLASSID
978 			if (nh->nh_tclassid)
979 				NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
980 #endif
981 			/* length of rtnetlink header + attributes */
982 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
983 		} endfor_nexthops(fi);
984 
985 		nla_nest_end(skb, mp);
986 	}
987 #endif
988 	return nlmsg_end(skb, nlh);
989 
990 nla_put_failure:
991 	nlmsg_cancel(skb, nlh);
992 	return -EMSGSIZE;
993 }
994 
995 /*
996  * Update FIB if:
997  * - local address disappeared -> we must delete all the entries
998  *   referring to it.
999  * - device went down -> we must shutdown all nexthops going via it.
1000  */
1001 int fib_sync_down_addr(struct net *net, __be32 local)
1002 {
1003 	int ret = 0;
1004 	unsigned int hash = fib_laddr_hashfn(local);
1005 	struct hlist_head *head = &fib_info_laddrhash[hash];
1006 	struct hlist_node *node;
1007 	struct fib_info *fi;
1008 
1009 	if (fib_info_laddrhash == NULL || local == 0)
1010 		return 0;
1011 
1012 	hlist_for_each_entry(fi, node, head, fib_lhash) {
1013 		if (!net_eq(fi->fib_net, net))
1014 			continue;
1015 		if (fi->fib_prefsrc == local) {
1016 			fi->fib_flags |= RTNH_F_DEAD;
1017 			ret++;
1018 		}
1019 	}
1020 	return ret;
1021 }
1022 
1023 int fib_sync_down_dev(struct net_device *dev, int force)
1024 {
1025 	int ret = 0;
1026 	int scope = RT_SCOPE_NOWHERE;
1027 	struct fib_info *prev_fi = NULL;
1028 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1029 	struct hlist_head *head = &fib_info_devhash[hash];
1030 	struct hlist_node *node;
1031 	struct fib_nh *nh;
1032 
1033 	if (force)
1034 		scope = -1;
1035 
1036 	hlist_for_each_entry(nh, node, head, nh_hash) {
1037 		struct fib_info *fi = nh->nh_parent;
1038 		int dead;
1039 
1040 		BUG_ON(!fi->fib_nhs);
1041 		if (nh->nh_dev != dev || fi == prev_fi)
1042 			continue;
1043 		prev_fi = fi;
1044 		dead = 0;
1045 		change_nexthops(fi) {
1046 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1047 				dead++;
1048 			else if (nexthop_nh->nh_dev == dev &&
1049 				 nexthop_nh->nh_scope != scope) {
1050 				nexthop_nh->nh_flags |= RTNH_F_DEAD;
1051 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1052 				spin_lock_bh(&fib_multipath_lock);
1053 				fi->fib_power -= nexthop_nh->nh_power;
1054 				nexthop_nh->nh_power = 0;
1055 				spin_unlock_bh(&fib_multipath_lock);
1056 #endif
1057 				dead++;
1058 			}
1059 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1060 			if (force > 1 && nexthop_nh->nh_dev == dev) {
1061 				dead = fi->fib_nhs;
1062 				break;
1063 			}
1064 #endif
1065 		} endfor_nexthops(fi)
1066 		if (dead == fi->fib_nhs) {
1067 			fi->fib_flags |= RTNH_F_DEAD;
1068 			ret++;
1069 		}
1070 	}
1071 
1072 	return ret;
1073 }
1074 
1075 /* Must be invoked inside of an RCU protected region.  */
1076 void fib_select_default(struct fib_result *res)
1077 {
1078 	struct fib_info *fi = NULL, *last_resort = NULL;
1079 	struct list_head *fa_head = res->fa_head;
1080 	struct fib_table *tb = res->table;
1081 	int order = -1, last_idx = -1;
1082 	struct fib_alias *fa;
1083 
1084 	list_for_each_entry_rcu(fa, fa_head, fa_list) {
1085 		struct fib_info *next_fi = fa->fa_info;
1086 
1087 		if (fa->fa_scope != res->scope ||
1088 		    fa->fa_type != RTN_UNICAST)
1089 			continue;
1090 
1091 		if (next_fi->fib_priority > res->fi->fib_priority)
1092 			break;
1093 		if (!next_fi->fib_nh[0].nh_gw ||
1094 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1095 			continue;
1096 
1097 		fib_alias_accessed(fa);
1098 
1099 		if (fi == NULL) {
1100 			if (next_fi != res->fi)
1101 				break;
1102 		} else if (!fib_detect_death(fi, order, &last_resort,
1103 					     &last_idx, tb->tb_default)) {
1104 			fib_result_assign(res, fi);
1105 			tb->tb_default = order;
1106 			goto out;
1107 		}
1108 		fi = next_fi;
1109 		order++;
1110 	}
1111 
1112 	if (order <= 0 || fi == NULL) {
1113 		tb->tb_default = -1;
1114 		goto out;
1115 	}
1116 
1117 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1118 				tb->tb_default)) {
1119 		fib_result_assign(res, fi);
1120 		tb->tb_default = order;
1121 		goto out;
1122 	}
1123 
1124 	if (last_idx >= 0)
1125 		fib_result_assign(res, last_resort);
1126 	tb->tb_default = last_idx;
1127 out:
1128 	return;
1129 }
1130 
1131 void fib_update_nh_saddrs(struct net_device *dev)
1132 {
1133 	struct hlist_head *head;
1134 	struct hlist_node *node;
1135 	struct fib_nh *nh;
1136 	unsigned int hash;
1137 
1138 	hash = fib_devindex_hashfn(dev->ifindex);
1139 	head = &fib_info_devhash[hash];
1140 	hlist_for_each_entry(nh, node, head, nh_hash) {
1141 		if (nh->nh_dev != dev)
1142 			continue;
1143 		nh->nh_saddr = inet_select_addr(nh->nh_dev,
1144 						nh->nh_gw,
1145 						nh->nh_cfg_scope);
1146 	}
1147 }
1148 
1149 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1150 
1151 /*
1152  * Dead device goes up. We wake up dead nexthops.
1153  * It takes sense only on multipath routes.
1154  */
1155 int fib_sync_up(struct net_device *dev)
1156 {
1157 	struct fib_info *prev_fi;
1158 	unsigned int hash;
1159 	struct hlist_head *head;
1160 	struct hlist_node *node;
1161 	struct fib_nh *nh;
1162 	int ret;
1163 
1164 	if (!(dev->flags & IFF_UP))
1165 		return 0;
1166 
1167 	prev_fi = NULL;
1168 	hash = fib_devindex_hashfn(dev->ifindex);
1169 	head = &fib_info_devhash[hash];
1170 	ret = 0;
1171 
1172 	hlist_for_each_entry(nh, node, head, nh_hash) {
1173 		struct fib_info *fi = nh->nh_parent;
1174 		int alive;
1175 
1176 		BUG_ON(!fi->fib_nhs);
1177 		if (nh->nh_dev != dev || fi == prev_fi)
1178 			continue;
1179 
1180 		prev_fi = fi;
1181 		alive = 0;
1182 		change_nexthops(fi) {
1183 			if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1184 				alive++;
1185 				continue;
1186 			}
1187 			if (nexthop_nh->nh_dev == NULL ||
1188 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1189 				continue;
1190 			if (nexthop_nh->nh_dev != dev ||
1191 			    !__in_dev_get_rtnl(dev))
1192 				continue;
1193 			alive++;
1194 			spin_lock_bh(&fib_multipath_lock);
1195 			nexthop_nh->nh_power = 0;
1196 			nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1197 			spin_unlock_bh(&fib_multipath_lock);
1198 		} endfor_nexthops(fi)
1199 
1200 		if (alive > 0) {
1201 			fi->fib_flags &= ~RTNH_F_DEAD;
1202 			ret++;
1203 		}
1204 	}
1205 
1206 	return ret;
1207 }
1208 
1209 /*
1210  * The algorithm is suboptimal, but it provides really
1211  * fair weighted route distribution.
1212  */
1213 void fib_select_multipath(struct fib_result *res)
1214 {
1215 	struct fib_info *fi = res->fi;
1216 	int w;
1217 
1218 	spin_lock_bh(&fib_multipath_lock);
1219 	if (fi->fib_power <= 0) {
1220 		int power = 0;
1221 		change_nexthops(fi) {
1222 			if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1223 				power += nexthop_nh->nh_weight;
1224 				nexthop_nh->nh_power = nexthop_nh->nh_weight;
1225 			}
1226 		} endfor_nexthops(fi);
1227 		fi->fib_power = power;
1228 		if (power <= 0) {
1229 			spin_unlock_bh(&fib_multipath_lock);
1230 			/* Race condition: route has just become dead. */
1231 			res->nh_sel = 0;
1232 			return;
1233 		}
1234 	}
1235 
1236 
1237 	/* w should be random number [0..fi->fib_power-1],
1238 	 * it is pretty bad approximation.
1239 	 */
1240 
1241 	w = jiffies % fi->fib_power;
1242 
1243 	change_nexthops(fi) {
1244 		if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1245 		    nexthop_nh->nh_power) {
1246 			w -= nexthop_nh->nh_power;
1247 			if (w <= 0) {
1248 				nexthop_nh->nh_power--;
1249 				fi->fib_power--;
1250 				res->nh_sel = nhsel;
1251 				spin_unlock_bh(&fib_multipath_lock);
1252 				return;
1253 			}
1254 		}
1255 	} endfor_nexthops(fi);
1256 
1257 	/* Race condition: route has just become dead. */
1258 	res->nh_sel = 0;
1259 	spin_unlock_bh(&fib_multipath_lock);
1260 }
1261 #endif
1262