xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 81d67439)
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 
146 void free_fib_info(struct fib_info *fi)
147 {
148 	if (fi->fib_dead == 0) {
149 		pr_warning("Freeing alive fib_info %p\n", fi);
150 		return;
151 	}
152 	change_nexthops(fi) {
153 		if (nexthop_nh->nh_dev)
154 			dev_put(nexthop_nh->nh_dev);
155 		nexthop_nh->nh_dev = NULL;
156 	} endfor_nexthops(fi);
157 	fib_info_cnt--;
158 	release_net(fi->fib_net);
159 	kfree_rcu(fi, rcu);
160 }
161 
162 void fib_release_info(struct fib_info *fi)
163 {
164 	spin_lock_bh(&fib_info_lock);
165 	if (fi && --fi->fib_treeref == 0) {
166 		hlist_del(&fi->fib_hash);
167 		if (fi->fib_prefsrc)
168 			hlist_del(&fi->fib_lhash);
169 		change_nexthops(fi) {
170 			if (!nexthop_nh->nh_dev)
171 				continue;
172 			hlist_del(&nexthop_nh->nh_hash);
173 		} endfor_nexthops(fi)
174 		fi->fib_dead = 1;
175 		fib_info_put(fi);
176 	}
177 	spin_unlock_bh(&fib_info_lock);
178 }
179 
180 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
181 {
182 	const struct fib_nh *onh = ofi->fib_nh;
183 
184 	for_nexthops(fi) {
185 		if (nh->nh_oif != onh->nh_oif ||
186 		    nh->nh_gw  != onh->nh_gw ||
187 		    nh->nh_scope != onh->nh_scope ||
188 #ifdef CONFIG_IP_ROUTE_MULTIPATH
189 		    nh->nh_weight != onh->nh_weight ||
190 #endif
191 #ifdef CONFIG_IP_ROUTE_CLASSID
192 		    nh->nh_tclassid != onh->nh_tclassid ||
193 #endif
194 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
195 			return -1;
196 		onh++;
197 	} endfor_nexthops(fi);
198 	return 0;
199 }
200 
201 static inline unsigned int fib_devindex_hashfn(unsigned int val)
202 {
203 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
204 
205 	return (val ^
206 		(val >> DEVINDEX_HASHBITS) ^
207 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
208 }
209 
210 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
211 {
212 	unsigned int mask = (fib_info_hash_size - 1);
213 	unsigned int val = fi->fib_nhs;
214 
215 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
216 	val ^= (__force u32)fi->fib_prefsrc;
217 	val ^= fi->fib_priority;
218 	for_nexthops(fi) {
219 		val ^= fib_devindex_hashfn(nh->nh_oif);
220 	} endfor_nexthops(fi)
221 
222 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
223 }
224 
225 static struct fib_info *fib_find_info(const struct fib_info *nfi)
226 {
227 	struct hlist_head *head;
228 	struct hlist_node *node;
229 	struct fib_info *fi;
230 	unsigned int hash;
231 
232 	hash = fib_info_hashfn(nfi);
233 	head = &fib_info_hash[hash];
234 
235 	hlist_for_each_entry(fi, node, head, fib_hash) {
236 		if (!net_eq(fi->fib_net, nfi->fib_net))
237 			continue;
238 		if (fi->fib_nhs != nfi->fib_nhs)
239 			continue;
240 		if (nfi->fib_protocol == fi->fib_protocol &&
241 		    nfi->fib_scope == fi->fib_scope &&
242 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
243 		    nfi->fib_priority == fi->fib_priority &&
244 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
245 			   sizeof(u32) * RTAX_MAX) == 0 &&
246 		    ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
247 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
248 			return fi;
249 	}
250 
251 	return NULL;
252 }
253 
254 /* Check, that the gateway is already configured.
255  * Used only by redirect accept routine.
256  */
257 int ip_fib_check_default(__be32 gw, struct net_device *dev)
258 {
259 	struct hlist_head *head;
260 	struct hlist_node *node;
261 	struct fib_nh *nh;
262 	unsigned int hash;
263 
264 	spin_lock(&fib_info_lock);
265 
266 	hash = fib_devindex_hashfn(dev->ifindex);
267 	head = &fib_info_devhash[hash];
268 	hlist_for_each_entry(nh, node, head, nh_hash) {
269 		if (nh->nh_dev == dev &&
270 		    nh->nh_gw == gw &&
271 		    !(nh->nh_flags & RTNH_F_DEAD)) {
272 			spin_unlock(&fib_info_lock);
273 			return 0;
274 		}
275 	}
276 
277 	spin_unlock(&fib_info_lock);
278 
279 	return -1;
280 }
281 
282 static inline size_t fib_nlmsg_size(struct fib_info *fi)
283 {
284 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
285 			 + nla_total_size(4) /* RTA_TABLE */
286 			 + nla_total_size(4) /* RTA_DST */
287 			 + nla_total_size(4) /* RTA_PRIORITY */
288 			 + nla_total_size(4); /* RTA_PREFSRC */
289 
290 	/* space for nested metrics */
291 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
292 
293 	if (fi->fib_nhs) {
294 		/* Also handles the special case fib_nhs == 1 */
295 
296 		/* each nexthop is packed in an attribute */
297 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
298 
299 		/* may contain flow and gateway attribute */
300 		nhsize += 2 * nla_total_size(4);
301 
302 		/* all nexthops are packed in a nested attribute */
303 		payload += nla_total_size(fi->fib_nhs * nhsize);
304 	}
305 
306 	return payload;
307 }
308 
309 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
310 	       int dst_len, u32 tb_id, struct nl_info *info,
311 	       unsigned int nlm_flags)
312 {
313 	struct sk_buff *skb;
314 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
315 	int err = -ENOBUFS;
316 
317 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
318 	if (skb == NULL)
319 		goto errout;
320 
321 	err = fib_dump_info(skb, info->pid, seq, event, tb_id,
322 			    fa->fa_type, key, dst_len,
323 			    fa->fa_tos, fa->fa_info, nlm_flags);
324 	if (err < 0) {
325 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
326 		WARN_ON(err == -EMSGSIZE);
327 		kfree_skb(skb);
328 		goto errout;
329 	}
330 	rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
331 		    info->nlh, GFP_KERNEL);
332 	return;
333 errout:
334 	if (err < 0)
335 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
336 }
337 
338 /* Return the first fib alias matching TOS with
339  * priority less than or equal to PRIO.
340  */
341 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
342 {
343 	if (fah) {
344 		struct fib_alias *fa;
345 		list_for_each_entry(fa, fah, fa_list) {
346 			if (fa->fa_tos > tos)
347 				continue;
348 			if (fa->fa_info->fib_priority >= prio ||
349 			    fa->fa_tos < tos)
350 				return fa;
351 		}
352 	}
353 	return NULL;
354 }
355 
356 int fib_detect_death(struct fib_info *fi, int order,
357 		     struct fib_info **last_resort, int *last_idx, int dflt)
358 {
359 	struct neighbour *n;
360 	int state = NUD_NONE;
361 
362 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
363 	if (n) {
364 		state = n->nud_state;
365 		neigh_release(n);
366 	}
367 	if (state == NUD_REACHABLE)
368 		return 0;
369 	if ((state & NUD_VALID) && order != dflt)
370 		return 0;
371 	if ((state & NUD_VALID) ||
372 	    (*last_idx < 0 && order > dflt)) {
373 		*last_resort = fi;
374 		*last_idx = order;
375 	}
376 	return 1;
377 }
378 
379 #ifdef CONFIG_IP_ROUTE_MULTIPATH
380 
381 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
382 {
383 	int nhs = 0;
384 
385 	while (rtnh_ok(rtnh, remaining)) {
386 		nhs++;
387 		rtnh = rtnh_next(rtnh, &remaining);
388 	}
389 
390 	/* leftover implies invalid nexthop configuration, discard it */
391 	return remaining > 0 ? 0 : nhs;
392 }
393 
394 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
395 		       int remaining, struct fib_config *cfg)
396 {
397 	change_nexthops(fi) {
398 		int attrlen;
399 
400 		if (!rtnh_ok(rtnh, remaining))
401 			return -EINVAL;
402 
403 		nexthop_nh->nh_flags =
404 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
405 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
406 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
407 
408 		attrlen = rtnh_attrlen(rtnh);
409 		if (attrlen > 0) {
410 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
411 
412 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
413 			nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
414 #ifdef CONFIG_IP_ROUTE_CLASSID
415 			nla = nla_find(attrs, attrlen, RTA_FLOW);
416 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
417 #endif
418 		}
419 
420 		rtnh = rtnh_next(rtnh, &remaining);
421 	} endfor_nexthops(fi);
422 
423 	return 0;
424 }
425 
426 #endif
427 
428 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
429 {
430 #ifdef CONFIG_IP_ROUTE_MULTIPATH
431 	struct rtnexthop *rtnh;
432 	int remaining;
433 #endif
434 
435 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
436 		return 1;
437 
438 	if (cfg->fc_oif || cfg->fc_gw) {
439 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
440 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
441 			return 0;
442 		return 1;
443 	}
444 
445 #ifdef CONFIG_IP_ROUTE_MULTIPATH
446 	if (cfg->fc_mp == NULL)
447 		return 0;
448 
449 	rtnh = cfg->fc_mp;
450 	remaining = cfg->fc_mp_len;
451 
452 	for_nexthops(fi) {
453 		int attrlen;
454 
455 		if (!rtnh_ok(rtnh, remaining))
456 			return -EINVAL;
457 
458 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
459 			return 1;
460 
461 		attrlen = rtnh_attrlen(rtnh);
462 		if (attrlen < 0) {
463 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
464 
465 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
466 			if (nla && nla_get_be32(nla) != nh->nh_gw)
467 				return 1;
468 #ifdef CONFIG_IP_ROUTE_CLASSID
469 			nla = nla_find(attrs, attrlen, RTA_FLOW);
470 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
471 				return 1;
472 #endif
473 		}
474 
475 		rtnh = rtnh_next(rtnh, &remaining);
476 	} endfor_nexthops(fi);
477 #endif
478 	return 0;
479 }
480 
481 
482 /*
483  * Picture
484  * -------
485  *
486  * Semantics of nexthop is very messy by historical reasons.
487  * We have to take into account, that:
488  * a) gateway can be actually local interface address,
489  *    so that gatewayed route is direct.
490  * b) gateway must be on-link address, possibly
491  *    described not by an ifaddr, but also by a direct route.
492  * c) If both gateway and interface are specified, they should not
493  *    contradict.
494  * d) If we use tunnel routes, gateway could be not on-link.
495  *
496  * Attempt to reconcile all of these (alas, self-contradictory) conditions
497  * results in pretty ugly and hairy code with obscure logic.
498  *
499  * I chose to generalized it instead, so that the size
500  * of code does not increase practically, but it becomes
501  * much more general.
502  * Every prefix is assigned a "scope" value: "host" is local address,
503  * "link" is direct route,
504  * [ ... "site" ... "interior" ... ]
505  * and "universe" is true gateway route with global meaning.
506  *
507  * Every prefix refers to a set of "nexthop"s (gw, oif),
508  * where gw must have narrower scope. This recursion stops
509  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
510  * which means that gw is forced to be on link.
511  *
512  * Code is still hairy, but now it is apparently logically
513  * consistent and very flexible. F.e. as by-product it allows
514  * to co-exists in peace independent exterior and interior
515  * routing processes.
516  *
517  * Normally it looks as following.
518  *
519  * {universe prefix}  -> (gw, oif) [scope link]
520  *		  |
521  *		  |-> {link prefix} -> (gw, oif) [scope local]
522  *					|
523  *					|-> {local prefix} (terminal node)
524  */
525 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
526 			struct fib_nh *nh)
527 {
528 	int err;
529 	struct net *net;
530 	struct net_device *dev;
531 
532 	net = cfg->fc_nlinfo.nl_net;
533 	if (nh->nh_gw) {
534 		struct fib_result res;
535 
536 		if (nh->nh_flags & RTNH_F_ONLINK) {
537 
538 			if (cfg->fc_scope >= RT_SCOPE_LINK)
539 				return -EINVAL;
540 			if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
541 				return -EINVAL;
542 			dev = __dev_get_by_index(net, nh->nh_oif);
543 			if (!dev)
544 				return -ENODEV;
545 			if (!(dev->flags & IFF_UP))
546 				return -ENETDOWN;
547 			nh->nh_dev = dev;
548 			dev_hold(dev);
549 			nh->nh_scope = RT_SCOPE_LINK;
550 			return 0;
551 		}
552 		rcu_read_lock();
553 		{
554 			struct flowi4 fl4 = {
555 				.daddr = nh->nh_gw,
556 				.flowi4_scope = cfg->fc_scope + 1,
557 				.flowi4_oif = nh->nh_oif,
558 			};
559 
560 			/* It is not necessary, but requires a bit of thinking */
561 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
562 				fl4.flowi4_scope = RT_SCOPE_LINK;
563 			err = fib_lookup(net, &fl4, &res);
564 			if (err) {
565 				rcu_read_unlock();
566 				return err;
567 			}
568 		}
569 		err = -EINVAL;
570 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
571 			goto out;
572 		nh->nh_scope = res.scope;
573 		nh->nh_oif = FIB_RES_OIF(res);
574 		nh->nh_dev = dev = FIB_RES_DEV(res);
575 		if (!dev)
576 			goto out;
577 		dev_hold(dev);
578 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
579 	} else {
580 		struct in_device *in_dev;
581 
582 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
583 			return -EINVAL;
584 
585 		rcu_read_lock();
586 		err = -ENODEV;
587 		in_dev = inetdev_by_index(net, nh->nh_oif);
588 		if (in_dev == NULL)
589 			goto out;
590 		err = -ENETDOWN;
591 		if (!(in_dev->dev->flags & IFF_UP))
592 			goto out;
593 		nh->nh_dev = in_dev->dev;
594 		dev_hold(nh->nh_dev);
595 		nh->nh_scope = RT_SCOPE_HOST;
596 		err = 0;
597 	}
598 out:
599 	rcu_read_unlock();
600 	return err;
601 }
602 
603 static inline unsigned int fib_laddr_hashfn(__be32 val)
604 {
605 	unsigned int mask = (fib_info_hash_size - 1);
606 
607 	return ((__force u32)val ^
608 		((__force u32)val >> 7) ^
609 		((__force u32)val >> 14)) & mask;
610 }
611 
612 static struct hlist_head *fib_info_hash_alloc(int bytes)
613 {
614 	if (bytes <= PAGE_SIZE)
615 		return kzalloc(bytes, GFP_KERNEL);
616 	else
617 		return (struct hlist_head *)
618 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
619 					 get_order(bytes));
620 }
621 
622 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
623 {
624 	if (!hash)
625 		return;
626 
627 	if (bytes <= PAGE_SIZE)
628 		kfree(hash);
629 	else
630 		free_pages((unsigned long) hash, get_order(bytes));
631 }
632 
633 static void fib_info_hash_move(struct hlist_head *new_info_hash,
634 			       struct hlist_head *new_laddrhash,
635 			       unsigned int new_size)
636 {
637 	struct hlist_head *old_info_hash, *old_laddrhash;
638 	unsigned int old_size = fib_info_hash_size;
639 	unsigned int i, bytes;
640 
641 	spin_lock_bh(&fib_info_lock);
642 	old_info_hash = fib_info_hash;
643 	old_laddrhash = fib_info_laddrhash;
644 	fib_info_hash_size = new_size;
645 
646 	for (i = 0; i < old_size; i++) {
647 		struct hlist_head *head = &fib_info_hash[i];
648 		struct hlist_node *node, *n;
649 		struct fib_info *fi;
650 
651 		hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
652 			struct hlist_head *dest;
653 			unsigned int new_hash;
654 
655 			hlist_del(&fi->fib_hash);
656 
657 			new_hash = fib_info_hashfn(fi);
658 			dest = &new_info_hash[new_hash];
659 			hlist_add_head(&fi->fib_hash, dest);
660 		}
661 	}
662 	fib_info_hash = new_info_hash;
663 
664 	for (i = 0; i < old_size; i++) {
665 		struct hlist_head *lhead = &fib_info_laddrhash[i];
666 		struct hlist_node *node, *n;
667 		struct fib_info *fi;
668 
669 		hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
670 			struct hlist_head *ldest;
671 			unsigned int new_hash;
672 
673 			hlist_del(&fi->fib_lhash);
674 
675 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
676 			ldest = &new_laddrhash[new_hash];
677 			hlist_add_head(&fi->fib_lhash, ldest);
678 		}
679 	}
680 	fib_info_laddrhash = new_laddrhash;
681 
682 	spin_unlock_bh(&fib_info_lock);
683 
684 	bytes = old_size * sizeof(struct hlist_head *);
685 	fib_info_hash_free(old_info_hash, bytes);
686 	fib_info_hash_free(old_laddrhash, bytes);
687 }
688 
689 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
690 {
691 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
692 					nh->nh_gw,
693 					nh->nh_parent->fib_scope);
694 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
695 
696 	return nh->nh_saddr;
697 }
698 
699 struct fib_info *fib_create_info(struct fib_config *cfg)
700 {
701 	int err;
702 	struct fib_info *fi = NULL;
703 	struct fib_info *ofi;
704 	int nhs = 1;
705 	struct net *net = cfg->fc_nlinfo.nl_net;
706 
707 	if (cfg->fc_type > RTN_MAX)
708 		goto err_inval;
709 
710 	/* Fast check to catch the most weird cases */
711 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
712 		goto err_inval;
713 
714 #ifdef CONFIG_IP_ROUTE_MULTIPATH
715 	if (cfg->fc_mp) {
716 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
717 		if (nhs == 0)
718 			goto err_inval;
719 	}
720 #endif
721 
722 	err = -ENOBUFS;
723 	if (fib_info_cnt >= fib_info_hash_size) {
724 		unsigned int new_size = fib_info_hash_size << 1;
725 		struct hlist_head *new_info_hash;
726 		struct hlist_head *new_laddrhash;
727 		unsigned int bytes;
728 
729 		if (!new_size)
730 			new_size = 1;
731 		bytes = new_size * sizeof(struct hlist_head *);
732 		new_info_hash = fib_info_hash_alloc(bytes);
733 		new_laddrhash = fib_info_hash_alloc(bytes);
734 		if (!new_info_hash || !new_laddrhash) {
735 			fib_info_hash_free(new_info_hash, bytes);
736 			fib_info_hash_free(new_laddrhash, bytes);
737 		} else
738 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
739 
740 		if (!fib_info_hash_size)
741 			goto failure;
742 	}
743 
744 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
745 	if (fi == NULL)
746 		goto failure;
747 	if (cfg->fc_mx) {
748 		fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
749 		if (!fi->fib_metrics)
750 			goto failure;
751 	} else
752 		fi->fib_metrics = (u32 *) dst_default_metrics;
753 	fib_info_cnt++;
754 
755 	fi->fib_net = hold_net(net);
756 	fi->fib_protocol = cfg->fc_protocol;
757 	fi->fib_scope = cfg->fc_scope;
758 	fi->fib_flags = cfg->fc_flags;
759 	fi->fib_priority = cfg->fc_priority;
760 	fi->fib_prefsrc = cfg->fc_prefsrc;
761 
762 	fi->fib_nhs = nhs;
763 	change_nexthops(fi) {
764 		nexthop_nh->nh_parent = fi;
765 	} endfor_nexthops(fi)
766 
767 	if (cfg->fc_mx) {
768 		struct nlattr *nla;
769 		int remaining;
770 
771 		nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
772 			int type = nla_type(nla);
773 
774 			if (type) {
775 				if (type > RTAX_MAX)
776 					goto err_inval;
777 				fi->fib_metrics[type - 1] = nla_get_u32(nla);
778 			}
779 		}
780 	}
781 
782 	if (cfg->fc_mp) {
783 #ifdef CONFIG_IP_ROUTE_MULTIPATH
784 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
785 		if (err != 0)
786 			goto failure;
787 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
788 			goto err_inval;
789 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
790 			goto err_inval;
791 #ifdef CONFIG_IP_ROUTE_CLASSID
792 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
793 			goto err_inval;
794 #endif
795 #else
796 		goto err_inval;
797 #endif
798 	} else {
799 		struct fib_nh *nh = fi->fib_nh;
800 
801 		nh->nh_oif = cfg->fc_oif;
802 		nh->nh_gw = cfg->fc_gw;
803 		nh->nh_flags = cfg->fc_flags;
804 #ifdef CONFIG_IP_ROUTE_CLASSID
805 		nh->nh_tclassid = cfg->fc_flow;
806 #endif
807 #ifdef CONFIG_IP_ROUTE_MULTIPATH
808 		nh->nh_weight = 1;
809 #endif
810 	}
811 
812 	if (fib_props[cfg->fc_type].error) {
813 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
814 			goto err_inval;
815 		goto link_it;
816 	} else {
817 		switch (cfg->fc_type) {
818 		case RTN_UNICAST:
819 		case RTN_LOCAL:
820 		case RTN_BROADCAST:
821 		case RTN_ANYCAST:
822 		case RTN_MULTICAST:
823 			break;
824 		default:
825 			goto err_inval;
826 		}
827 	}
828 
829 	if (cfg->fc_scope > RT_SCOPE_HOST)
830 		goto err_inval;
831 
832 	if (cfg->fc_scope == RT_SCOPE_HOST) {
833 		struct fib_nh *nh = fi->fib_nh;
834 
835 		/* Local address is added. */
836 		if (nhs != 1 || nh->nh_gw)
837 			goto err_inval;
838 		nh->nh_scope = RT_SCOPE_NOWHERE;
839 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
840 		err = -ENODEV;
841 		if (nh->nh_dev == NULL)
842 			goto failure;
843 	} else {
844 		change_nexthops(fi) {
845 			err = fib_check_nh(cfg, fi, nexthop_nh);
846 			if (err != 0)
847 				goto failure;
848 		} endfor_nexthops(fi)
849 	}
850 
851 	if (fi->fib_prefsrc) {
852 		if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
853 		    fi->fib_prefsrc != cfg->fc_dst)
854 			if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
855 				goto err_inval;
856 	}
857 
858 	change_nexthops(fi) {
859 		fib_info_update_nh_saddr(net, nexthop_nh);
860 	} endfor_nexthops(fi)
861 
862 link_it:
863 	ofi = fib_find_info(fi);
864 	if (ofi) {
865 		fi->fib_dead = 1;
866 		free_fib_info(fi);
867 		ofi->fib_treeref++;
868 		return ofi;
869 	}
870 
871 	fi->fib_treeref++;
872 	atomic_inc(&fi->fib_clntref);
873 	spin_lock_bh(&fib_info_lock);
874 	hlist_add_head(&fi->fib_hash,
875 		       &fib_info_hash[fib_info_hashfn(fi)]);
876 	if (fi->fib_prefsrc) {
877 		struct hlist_head *head;
878 
879 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
880 		hlist_add_head(&fi->fib_lhash, head);
881 	}
882 	change_nexthops(fi) {
883 		struct hlist_head *head;
884 		unsigned int hash;
885 
886 		if (!nexthop_nh->nh_dev)
887 			continue;
888 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
889 		head = &fib_info_devhash[hash];
890 		hlist_add_head(&nexthop_nh->nh_hash, head);
891 	} endfor_nexthops(fi)
892 	spin_unlock_bh(&fib_info_lock);
893 	return fi;
894 
895 err_inval:
896 	err = -EINVAL;
897 
898 failure:
899 	if (fi) {
900 		fi->fib_dead = 1;
901 		free_fib_info(fi);
902 	}
903 
904 	return ERR_PTR(err);
905 }
906 
907 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
908 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
909 		  struct fib_info *fi, unsigned int flags)
910 {
911 	struct nlmsghdr *nlh;
912 	struct rtmsg *rtm;
913 
914 	nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
915 	if (nlh == NULL)
916 		return -EMSGSIZE;
917 
918 	rtm = nlmsg_data(nlh);
919 	rtm->rtm_family = AF_INET;
920 	rtm->rtm_dst_len = dst_len;
921 	rtm->rtm_src_len = 0;
922 	rtm->rtm_tos = tos;
923 	if (tb_id < 256)
924 		rtm->rtm_table = tb_id;
925 	else
926 		rtm->rtm_table = RT_TABLE_COMPAT;
927 	NLA_PUT_U32(skb, RTA_TABLE, tb_id);
928 	rtm->rtm_type = type;
929 	rtm->rtm_flags = fi->fib_flags;
930 	rtm->rtm_scope = fi->fib_scope;
931 	rtm->rtm_protocol = fi->fib_protocol;
932 
933 	if (rtm->rtm_dst_len)
934 		NLA_PUT_BE32(skb, RTA_DST, dst);
935 
936 	if (fi->fib_priority)
937 		NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
938 
939 	if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
940 		goto nla_put_failure;
941 
942 	if (fi->fib_prefsrc)
943 		NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
944 
945 	if (fi->fib_nhs == 1) {
946 		if (fi->fib_nh->nh_gw)
947 			NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
948 
949 		if (fi->fib_nh->nh_oif)
950 			NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
951 #ifdef CONFIG_IP_ROUTE_CLASSID
952 		if (fi->fib_nh[0].nh_tclassid)
953 			NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
954 #endif
955 	}
956 #ifdef CONFIG_IP_ROUTE_MULTIPATH
957 	if (fi->fib_nhs > 1) {
958 		struct rtnexthop *rtnh;
959 		struct nlattr *mp;
960 
961 		mp = nla_nest_start(skb, RTA_MULTIPATH);
962 		if (mp == NULL)
963 			goto nla_put_failure;
964 
965 		for_nexthops(fi) {
966 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
967 			if (rtnh == NULL)
968 				goto nla_put_failure;
969 
970 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
971 			rtnh->rtnh_hops = nh->nh_weight - 1;
972 			rtnh->rtnh_ifindex = nh->nh_oif;
973 
974 			if (nh->nh_gw)
975 				NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
976 #ifdef CONFIG_IP_ROUTE_CLASSID
977 			if (nh->nh_tclassid)
978 				NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
979 #endif
980 			/* length of rtnetlink header + attributes */
981 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
982 		} endfor_nexthops(fi);
983 
984 		nla_nest_end(skb, mp);
985 	}
986 #endif
987 	return nlmsg_end(skb, nlh);
988 
989 nla_put_failure:
990 	nlmsg_cancel(skb, nlh);
991 	return -EMSGSIZE;
992 }
993 
994 /*
995  * Update FIB if:
996  * - local address disappeared -> we must delete all the entries
997  *   referring to it.
998  * - device went down -> we must shutdown all nexthops going via it.
999  */
1000 int fib_sync_down_addr(struct net *net, __be32 local)
1001 {
1002 	int ret = 0;
1003 	unsigned int hash = fib_laddr_hashfn(local);
1004 	struct hlist_head *head = &fib_info_laddrhash[hash];
1005 	struct hlist_node *node;
1006 	struct fib_info *fi;
1007 
1008 	if (fib_info_laddrhash == NULL || local == 0)
1009 		return 0;
1010 
1011 	hlist_for_each_entry(fi, node, head, fib_lhash) {
1012 		if (!net_eq(fi->fib_net, net))
1013 			continue;
1014 		if (fi->fib_prefsrc == local) {
1015 			fi->fib_flags |= RTNH_F_DEAD;
1016 			ret++;
1017 		}
1018 	}
1019 	return ret;
1020 }
1021 
1022 int fib_sync_down_dev(struct net_device *dev, int force)
1023 {
1024 	int ret = 0;
1025 	int scope = RT_SCOPE_NOWHERE;
1026 	struct fib_info *prev_fi = NULL;
1027 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1028 	struct hlist_head *head = &fib_info_devhash[hash];
1029 	struct hlist_node *node;
1030 	struct fib_nh *nh;
1031 
1032 	if (force)
1033 		scope = -1;
1034 
1035 	hlist_for_each_entry(nh, node, head, nh_hash) {
1036 		struct fib_info *fi = nh->nh_parent;
1037 		int dead;
1038 
1039 		BUG_ON(!fi->fib_nhs);
1040 		if (nh->nh_dev != dev || fi == prev_fi)
1041 			continue;
1042 		prev_fi = fi;
1043 		dead = 0;
1044 		change_nexthops(fi) {
1045 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1046 				dead++;
1047 			else if (nexthop_nh->nh_dev == dev &&
1048 				 nexthop_nh->nh_scope != scope) {
1049 				nexthop_nh->nh_flags |= RTNH_F_DEAD;
1050 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1051 				spin_lock_bh(&fib_multipath_lock);
1052 				fi->fib_power -= nexthop_nh->nh_power;
1053 				nexthop_nh->nh_power = 0;
1054 				spin_unlock_bh(&fib_multipath_lock);
1055 #endif
1056 				dead++;
1057 			}
1058 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1059 			if (force > 1 && nexthop_nh->nh_dev == dev) {
1060 				dead = fi->fib_nhs;
1061 				break;
1062 			}
1063 #endif
1064 		} endfor_nexthops(fi)
1065 		if (dead == fi->fib_nhs) {
1066 			fi->fib_flags |= RTNH_F_DEAD;
1067 			ret++;
1068 		}
1069 	}
1070 
1071 	return ret;
1072 }
1073 
1074 /* Must be invoked inside of an RCU protected region.  */
1075 void fib_select_default(struct fib_result *res)
1076 {
1077 	struct fib_info *fi = NULL, *last_resort = NULL;
1078 	struct list_head *fa_head = res->fa_head;
1079 	struct fib_table *tb = res->table;
1080 	int order = -1, last_idx = -1;
1081 	struct fib_alias *fa;
1082 
1083 	list_for_each_entry_rcu(fa, fa_head, fa_list) {
1084 		struct fib_info *next_fi = fa->fa_info;
1085 
1086 		if (next_fi->fib_scope != res->scope ||
1087 		    fa->fa_type != RTN_UNICAST)
1088 			continue;
1089 
1090 		if (next_fi->fib_priority > res->fi->fib_priority)
1091 			break;
1092 		if (!next_fi->fib_nh[0].nh_gw ||
1093 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1094 			continue;
1095 
1096 		fib_alias_accessed(fa);
1097 
1098 		if (fi == NULL) {
1099 			if (next_fi != res->fi)
1100 				break;
1101 		} else if (!fib_detect_death(fi, order, &last_resort,
1102 					     &last_idx, tb->tb_default)) {
1103 			fib_result_assign(res, fi);
1104 			tb->tb_default = order;
1105 			goto out;
1106 		}
1107 		fi = next_fi;
1108 		order++;
1109 	}
1110 
1111 	if (order <= 0 || fi == NULL) {
1112 		tb->tb_default = -1;
1113 		goto out;
1114 	}
1115 
1116 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1117 				tb->tb_default)) {
1118 		fib_result_assign(res, fi);
1119 		tb->tb_default = order;
1120 		goto out;
1121 	}
1122 
1123 	if (last_idx >= 0)
1124 		fib_result_assign(res, last_resort);
1125 	tb->tb_default = last_idx;
1126 out:
1127 	return;
1128 }
1129 
1130 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1131 
1132 /*
1133  * Dead device goes up. We wake up dead nexthops.
1134  * It takes sense only on multipath routes.
1135  */
1136 int fib_sync_up(struct net_device *dev)
1137 {
1138 	struct fib_info *prev_fi;
1139 	unsigned int hash;
1140 	struct hlist_head *head;
1141 	struct hlist_node *node;
1142 	struct fib_nh *nh;
1143 	int ret;
1144 
1145 	if (!(dev->flags & IFF_UP))
1146 		return 0;
1147 
1148 	prev_fi = NULL;
1149 	hash = fib_devindex_hashfn(dev->ifindex);
1150 	head = &fib_info_devhash[hash];
1151 	ret = 0;
1152 
1153 	hlist_for_each_entry(nh, node, head, nh_hash) {
1154 		struct fib_info *fi = nh->nh_parent;
1155 		int alive;
1156 
1157 		BUG_ON(!fi->fib_nhs);
1158 		if (nh->nh_dev != dev || fi == prev_fi)
1159 			continue;
1160 
1161 		prev_fi = fi;
1162 		alive = 0;
1163 		change_nexthops(fi) {
1164 			if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1165 				alive++;
1166 				continue;
1167 			}
1168 			if (nexthop_nh->nh_dev == NULL ||
1169 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1170 				continue;
1171 			if (nexthop_nh->nh_dev != dev ||
1172 			    !__in_dev_get_rtnl(dev))
1173 				continue;
1174 			alive++;
1175 			spin_lock_bh(&fib_multipath_lock);
1176 			nexthop_nh->nh_power = 0;
1177 			nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1178 			spin_unlock_bh(&fib_multipath_lock);
1179 		} endfor_nexthops(fi)
1180 
1181 		if (alive > 0) {
1182 			fi->fib_flags &= ~RTNH_F_DEAD;
1183 			ret++;
1184 		}
1185 	}
1186 
1187 	return ret;
1188 }
1189 
1190 /*
1191  * The algorithm is suboptimal, but it provides really
1192  * fair weighted route distribution.
1193  */
1194 void fib_select_multipath(struct fib_result *res)
1195 {
1196 	struct fib_info *fi = res->fi;
1197 	int w;
1198 
1199 	spin_lock_bh(&fib_multipath_lock);
1200 	if (fi->fib_power <= 0) {
1201 		int power = 0;
1202 		change_nexthops(fi) {
1203 			if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1204 				power += nexthop_nh->nh_weight;
1205 				nexthop_nh->nh_power = nexthop_nh->nh_weight;
1206 			}
1207 		} endfor_nexthops(fi);
1208 		fi->fib_power = power;
1209 		if (power <= 0) {
1210 			spin_unlock_bh(&fib_multipath_lock);
1211 			/* Race condition: route has just become dead. */
1212 			res->nh_sel = 0;
1213 			return;
1214 		}
1215 	}
1216 
1217 
1218 	/* w should be random number [0..fi->fib_power-1],
1219 	 * it is pretty bad approximation.
1220 	 */
1221 
1222 	w = jiffies % fi->fib_power;
1223 
1224 	change_nexthops(fi) {
1225 		if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1226 		    nexthop_nh->nh_power) {
1227 			w -= nexthop_nh->nh_power;
1228 			if (w <= 0) {
1229 				nexthop_nh->nh_power--;
1230 				fi->fib_power--;
1231 				res->nh_sel = nhsel;
1232 				spin_unlock_bh(&fib_multipath_lock);
1233 				return;
1234 			}
1235 		}
1236 	} endfor_nexthops(fi);
1237 
1238 	/* Race condition: route has just become dead. */
1239 	res->nh_sel = 0;
1240 	spin_unlock_bh(&fib_multipath_lock);
1241 }
1242 #endif
1243