xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 6d99a79c)
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 <linux/uaccess.h>
17 #include <linux/bitops.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/jiffies.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/errno.h>
26 #include <linux/in.h>
27 #include <linux/inet.h>
28 #include <linux/inetdevice.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_arp.h>
31 #include <linux/proc_fs.h>
32 #include <linux/skbuff.h>
33 #include <linux/init.h>
34 #include <linux/slab.h>
35 #include <linux/netlink.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 #include <net/lwtunnel.h>
47 #include <net/fib_notifier.h>
48 
49 #include "fib_lookup.h"
50 
51 static DEFINE_SPINLOCK(fib_info_lock);
52 static struct hlist_head *fib_info_hash;
53 static struct hlist_head *fib_info_laddrhash;
54 static unsigned int fib_info_hash_size;
55 static unsigned int fib_info_cnt;
56 
57 #define DEVINDEX_HASHBITS 8
58 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
59 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
60 
61 #ifdef CONFIG_IP_ROUTE_MULTIPATH
62 
63 #define for_nexthops(fi) {						\
64 	int nhsel; const struct fib_nh *nh;				\
65 	for (nhsel = 0, nh = (fi)->fib_nh;				\
66 	     nhsel < (fi)->fib_nhs;					\
67 	     nh++, nhsel++)
68 
69 #define change_nexthops(fi) {						\
70 	int nhsel; struct fib_nh *nexthop_nh;				\
71 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
72 	     nhsel < (fi)->fib_nhs;					\
73 	     nexthop_nh++, nhsel++)
74 
75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
76 
77 /* Hope, that gcc will optimize it to get rid of dummy loop */
78 
79 #define for_nexthops(fi) {						\
80 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
81 	for (nhsel = 0; nhsel < 1; nhsel++)
82 
83 #define change_nexthops(fi) {						\
84 	int nhsel;							\
85 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
86 	for (nhsel = 0; nhsel < 1; nhsel++)
87 
88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
89 
90 #define endfor_nexthops(fi) }
91 
92 
93 const struct fib_prop fib_props[RTN_MAX + 1] = {
94 	[RTN_UNSPEC] = {
95 		.error	= 0,
96 		.scope	= RT_SCOPE_NOWHERE,
97 	},
98 	[RTN_UNICAST] = {
99 		.error	= 0,
100 		.scope	= RT_SCOPE_UNIVERSE,
101 	},
102 	[RTN_LOCAL] = {
103 		.error	= 0,
104 		.scope	= RT_SCOPE_HOST,
105 	},
106 	[RTN_BROADCAST] = {
107 		.error	= 0,
108 		.scope	= RT_SCOPE_LINK,
109 	},
110 	[RTN_ANYCAST] = {
111 		.error	= 0,
112 		.scope	= RT_SCOPE_LINK,
113 	},
114 	[RTN_MULTICAST] = {
115 		.error	= 0,
116 		.scope	= RT_SCOPE_UNIVERSE,
117 	},
118 	[RTN_BLACKHOLE] = {
119 		.error	= -EINVAL,
120 		.scope	= RT_SCOPE_UNIVERSE,
121 	},
122 	[RTN_UNREACHABLE] = {
123 		.error	= -EHOSTUNREACH,
124 		.scope	= RT_SCOPE_UNIVERSE,
125 	},
126 	[RTN_PROHIBIT] = {
127 		.error	= -EACCES,
128 		.scope	= RT_SCOPE_UNIVERSE,
129 	},
130 	[RTN_THROW] = {
131 		.error	= -EAGAIN,
132 		.scope	= RT_SCOPE_UNIVERSE,
133 	},
134 	[RTN_NAT] = {
135 		.error	= -EINVAL,
136 		.scope	= RT_SCOPE_NOWHERE,
137 	},
138 	[RTN_XRESOLVE] = {
139 		.error	= -EINVAL,
140 		.scope	= RT_SCOPE_NOWHERE,
141 	},
142 };
143 
144 static void rt_fibinfo_free(struct rtable __rcu **rtp)
145 {
146 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
147 
148 	if (!rt)
149 		return;
150 
151 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
152 	 * because we waited an RCU grace period before calling
153 	 * free_fib_info_rcu()
154 	 */
155 
156 	dst_dev_put(&rt->dst);
157 	dst_release_immediate(&rt->dst);
158 }
159 
160 static void free_nh_exceptions(struct fib_nh *nh)
161 {
162 	struct fnhe_hash_bucket *hash;
163 	int i;
164 
165 	hash = rcu_dereference_protected(nh->nh_exceptions, 1);
166 	if (!hash)
167 		return;
168 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
169 		struct fib_nh_exception *fnhe;
170 
171 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
172 		while (fnhe) {
173 			struct fib_nh_exception *next;
174 
175 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
176 
177 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
178 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
179 
180 			kfree(fnhe);
181 
182 			fnhe = next;
183 		}
184 	}
185 	kfree(hash);
186 }
187 
188 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
189 {
190 	int cpu;
191 
192 	if (!rtp)
193 		return;
194 
195 	for_each_possible_cpu(cpu) {
196 		struct rtable *rt;
197 
198 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
199 		if (rt) {
200 			dst_dev_put(&rt->dst);
201 			dst_release_immediate(&rt->dst);
202 		}
203 	}
204 	free_percpu(rtp);
205 }
206 
207 /* Release a nexthop info record */
208 static void free_fib_info_rcu(struct rcu_head *head)
209 {
210 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
211 
212 	change_nexthops(fi) {
213 		if (nexthop_nh->nh_dev)
214 			dev_put(nexthop_nh->nh_dev);
215 		lwtstate_put(nexthop_nh->nh_lwtstate);
216 		free_nh_exceptions(nexthop_nh);
217 		rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
218 		rt_fibinfo_free(&nexthop_nh->nh_rth_input);
219 	} endfor_nexthops(fi);
220 
221 	ip_fib_metrics_put(fi->fib_metrics);
222 
223 	kfree(fi);
224 }
225 
226 void free_fib_info(struct fib_info *fi)
227 {
228 	if (fi->fib_dead == 0) {
229 		pr_warn("Freeing alive fib_info %p\n", fi);
230 		return;
231 	}
232 	fib_info_cnt--;
233 #ifdef CONFIG_IP_ROUTE_CLASSID
234 	change_nexthops(fi) {
235 		if (nexthop_nh->nh_tclassid)
236 			fi->fib_net->ipv4.fib_num_tclassid_users--;
237 	} endfor_nexthops(fi);
238 #endif
239 	call_rcu(&fi->rcu, free_fib_info_rcu);
240 }
241 EXPORT_SYMBOL_GPL(free_fib_info);
242 
243 void fib_release_info(struct fib_info *fi)
244 {
245 	spin_lock_bh(&fib_info_lock);
246 	if (fi && --fi->fib_treeref == 0) {
247 		hlist_del(&fi->fib_hash);
248 		if (fi->fib_prefsrc)
249 			hlist_del(&fi->fib_lhash);
250 		change_nexthops(fi) {
251 			if (!nexthop_nh->nh_dev)
252 				continue;
253 			hlist_del(&nexthop_nh->nh_hash);
254 		} endfor_nexthops(fi)
255 		fi->fib_dead = 1;
256 		fib_info_put(fi);
257 	}
258 	spin_unlock_bh(&fib_info_lock);
259 }
260 
261 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
262 {
263 	const struct fib_nh *onh = ofi->fib_nh;
264 
265 	for_nexthops(fi) {
266 		if (nh->nh_oif != onh->nh_oif ||
267 		    nh->nh_gw  != onh->nh_gw ||
268 		    nh->nh_scope != onh->nh_scope ||
269 #ifdef CONFIG_IP_ROUTE_MULTIPATH
270 		    nh->nh_weight != onh->nh_weight ||
271 #endif
272 #ifdef CONFIG_IP_ROUTE_CLASSID
273 		    nh->nh_tclassid != onh->nh_tclassid ||
274 #endif
275 		    lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
276 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
277 			return -1;
278 		onh++;
279 	} endfor_nexthops(fi);
280 	return 0;
281 }
282 
283 static inline unsigned int fib_devindex_hashfn(unsigned int val)
284 {
285 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
286 
287 	return (val ^
288 		(val >> DEVINDEX_HASHBITS) ^
289 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
290 }
291 
292 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
293 {
294 	unsigned int mask = (fib_info_hash_size - 1);
295 	unsigned int val = fi->fib_nhs;
296 
297 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
298 	val ^= (__force u32)fi->fib_prefsrc;
299 	val ^= fi->fib_priority;
300 	for_nexthops(fi) {
301 		val ^= fib_devindex_hashfn(nh->nh_oif);
302 	} endfor_nexthops(fi)
303 
304 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
305 }
306 
307 static struct fib_info *fib_find_info(const struct fib_info *nfi)
308 {
309 	struct hlist_head *head;
310 	struct fib_info *fi;
311 	unsigned int hash;
312 
313 	hash = fib_info_hashfn(nfi);
314 	head = &fib_info_hash[hash];
315 
316 	hlist_for_each_entry(fi, head, fib_hash) {
317 		if (!net_eq(fi->fib_net, nfi->fib_net))
318 			continue;
319 		if (fi->fib_nhs != nfi->fib_nhs)
320 			continue;
321 		if (nfi->fib_protocol == fi->fib_protocol &&
322 		    nfi->fib_scope == fi->fib_scope &&
323 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
324 		    nfi->fib_priority == fi->fib_priority &&
325 		    nfi->fib_type == fi->fib_type &&
326 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
327 			   sizeof(u32) * RTAX_MAX) == 0 &&
328 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
329 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
330 			return fi;
331 	}
332 
333 	return NULL;
334 }
335 
336 /* Check, that the gateway is already configured.
337  * Used only by redirect accept routine.
338  */
339 int ip_fib_check_default(__be32 gw, struct net_device *dev)
340 {
341 	struct hlist_head *head;
342 	struct fib_nh *nh;
343 	unsigned int hash;
344 
345 	spin_lock(&fib_info_lock);
346 
347 	hash = fib_devindex_hashfn(dev->ifindex);
348 	head = &fib_info_devhash[hash];
349 	hlist_for_each_entry(nh, head, nh_hash) {
350 		if (nh->nh_dev == dev &&
351 		    nh->nh_gw == gw &&
352 		    !(nh->nh_flags & RTNH_F_DEAD)) {
353 			spin_unlock(&fib_info_lock);
354 			return 0;
355 		}
356 	}
357 
358 	spin_unlock(&fib_info_lock);
359 
360 	return -1;
361 }
362 
363 static inline size_t fib_nlmsg_size(struct fib_info *fi)
364 {
365 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
366 			 + nla_total_size(4) /* RTA_TABLE */
367 			 + nla_total_size(4) /* RTA_DST */
368 			 + nla_total_size(4) /* RTA_PRIORITY */
369 			 + nla_total_size(4) /* RTA_PREFSRC */
370 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
371 
372 	/* space for nested metrics */
373 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
374 
375 	if (fi->fib_nhs) {
376 		size_t nh_encapsize = 0;
377 		/* Also handles the special case fib_nhs == 1 */
378 
379 		/* each nexthop is packed in an attribute */
380 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
381 
382 		/* may contain flow and gateway attribute */
383 		nhsize += 2 * nla_total_size(4);
384 
385 		/* grab encap info */
386 		for_nexthops(fi) {
387 			if (nh->nh_lwtstate) {
388 				/* RTA_ENCAP_TYPE */
389 				nh_encapsize += lwtunnel_get_encap_size(
390 						nh->nh_lwtstate);
391 				/* RTA_ENCAP */
392 				nh_encapsize +=  nla_total_size(2);
393 			}
394 		} endfor_nexthops(fi);
395 
396 		/* all nexthops are packed in a nested attribute */
397 		payload += nla_total_size((fi->fib_nhs * nhsize) +
398 					  nh_encapsize);
399 
400 	}
401 
402 	return payload;
403 }
404 
405 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
406 	       int dst_len, u32 tb_id, const struct nl_info *info,
407 	       unsigned int nlm_flags)
408 {
409 	struct sk_buff *skb;
410 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
411 	int err = -ENOBUFS;
412 
413 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
414 	if (!skb)
415 		goto errout;
416 
417 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
418 			    fa->fa_type, key, dst_len,
419 			    fa->fa_tos, fa->fa_info, nlm_flags);
420 	if (err < 0) {
421 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
422 		WARN_ON(err == -EMSGSIZE);
423 		kfree_skb(skb);
424 		goto errout;
425 	}
426 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
427 		    info->nlh, GFP_KERNEL);
428 	return;
429 errout:
430 	if (err < 0)
431 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
432 }
433 
434 static int fib_detect_death(struct fib_info *fi, int order,
435 			    struct fib_info **last_resort, int *last_idx,
436 			    int dflt)
437 {
438 	struct neighbour *n;
439 	int state = NUD_NONE;
440 
441 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
442 	if (n) {
443 		state = n->nud_state;
444 		neigh_release(n);
445 	} else {
446 		return 0;
447 	}
448 	if (state == NUD_REACHABLE)
449 		return 0;
450 	if ((state & NUD_VALID) && order != dflt)
451 		return 0;
452 	if ((state & NUD_VALID) ||
453 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
454 		*last_resort = fi;
455 		*last_idx = order;
456 	}
457 	return 1;
458 }
459 
460 #ifdef CONFIG_IP_ROUTE_MULTIPATH
461 
462 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
463 			      struct netlink_ext_ack *extack)
464 {
465 	int nhs = 0;
466 
467 	while (rtnh_ok(rtnh, remaining)) {
468 		nhs++;
469 		rtnh = rtnh_next(rtnh, &remaining);
470 	}
471 
472 	/* leftover implies invalid nexthop configuration, discard it */
473 	if (remaining > 0) {
474 		NL_SET_ERR_MSG(extack,
475 			       "Invalid nexthop configuration - extra data after nexthops");
476 		nhs = 0;
477 	}
478 
479 	return nhs;
480 }
481 
482 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
483 		       int remaining, struct fib_config *cfg,
484 		       struct netlink_ext_ack *extack)
485 {
486 	int ret;
487 
488 	change_nexthops(fi) {
489 		int attrlen;
490 
491 		if (!rtnh_ok(rtnh, remaining)) {
492 			NL_SET_ERR_MSG(extack,
493 				       "Invalid nexthop configuration - extra data after nexthop");
494 			return -EINVAL;
495 		}
496 
497 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
498 			NL_SET_ERR_MSG(extack,
499 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
500 			return -EINVAL;
501 		}
502 
503 		nexthop_nh->nh_flags =
504 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
505 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
506 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
507 
508 		attrlen = rtnh_attrlen(rtnh);
509 		if (attrlen > 0) {
510 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
511 
512 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
513 			nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
514 #ifdef CONFIG_IP_ROUTE_CLASSID
515 			nla = nla_find(attrs, attrlen, RTA_FLOW);
516 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
517 			if (nexthop_nh->nh_tclassid)
518 				fi->fib_net->ipv4.fib_num_tclassid_users++;
519 #endif
520 			nla = nla_find(attrs, attrlen, RTA_ENCAP);
521 			if (nla) {
522 				struct lwtunnel_state *lwtstate;
523 				struct nlattr *nla_entype;
524 
525 				nla_entype = nla_find(attrs, attrlen,
526 						      RTA_ENCAP_TYPE);
527 				if (!nla_entype) {
528 					NL_SET_BAD_ATTR(extack, nla);
529 					NL_SET_ERR_MSG(extack,
530 						       "Encap type is missing");
531 					goto err_inval;
532 				}
533 
534 				ret = lwtunnel_build_state(nla_get_u16(
535 							   nla_entype),
536 							   nla,  AF_INET, cfg,
537 							   &lwtstate, extack);
538 				if (ret)
539 					goto errout;
540 				nexthop_nh->nh_lwtstate =
541 					lwtstate_get(lwtstate);
542 			}
543 		}
544 
545 		rtnh = rtnh_next(rtnh, &remaining);
546 	} endfor_nexthops(fi);
547 
548 	return 0;
549 
550 err_inval:
551 	ret = -EINVAL;
552 
553 errout:
554 	return ret;
555 }
556 
557 static void fib_rebalance(struct fib_info *fi)
558 {
559 	int total;
560 	int w;
561 	struct in_device *in_dev;
562 
563 	if (fi->fib_nhs < 2)
564 		return;
565 
566 	total = 0;
567 	for_nexthops(fi) {
568 		if (nh->nh_flags & RTNH_F_DEAD)
569 			continue;
570 
571 		in_dev = __in_dev_get_rtnl(nh->nh_dev);
572 
573 		if (in_dev &&
574 		    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
575 		    nh->nh_flags & RTNH_F_LINKDOWN)
576 			continue;
577 
578 		total += nh->nh_weight;
579 	} endfor_nexthops(fi);
580 
581 	w = 0;
582 	change_nexthops(fi) {
583 		int upper_bound;
584 
585 		in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
586 
587 		if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
588 			upper_bound = -1;
589 		} else if (in_dev &&
590 			   IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
591 			   nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
592 			upper_bound = -1;
593 		} else {
594 			w += nexthop_nh->nh_weight;
595 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
596 							    total) - 1;
597 		}
598 
599 		atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
600 	} endfor_nexthops(fi);
601 }
602 #else /* CONFIG_IP_ROUTE_MULTIPATH */
603 
604 #define fib_rebalance(fi) do { } while (0)
605 
606 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
607 
608 static int fib_encap_match(u16 encap_type,
609 			   struct nlattr *encap,
610 			   const struct fib_nh *nh,
611 			   const struct fib_config *cfg,
612 			   struct netlink_ext_ack *extack)
613 {
614 	struct lwtunnel_state *lwtstate;
615 	int ret, result = 0;
616 
617 	if (encap_type == LWTUNNEL_ENCAP_NONE)
618 		return 0;
619 
620 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
621 				   cfg, &lwtstate, extack);
622 	if (!ret) {
623 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
624 		lwtstate_free(lwtstate);
625 	}
626 
627 	return result;
628 }
629 
630 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
631 		 struct netlink_ext_ack *extack)
632 {
633 #ifdef CONFIG_IP_ROUTE_MULTIPATH
634 	struct rtnexthop *rtnh;
635 	int remaining;
636 #endif
637 
638 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
639 		return 1;
640 
641 	if (cfg->fc_oif || cfg->fc_gw) {
642 		if (cfg->fc_encap) {
643 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
644 					    fi->fib_nh, cfg, extack))
645 				return 1;
646 		}
647 #ifdef CONFIG_IP_ROUTE_CLASSID
648 		if (cfg->fc_flow &&
649 		    cfg->fc_flow != fi->fib_nh->nh_tclassid)
650 			return 1;
651 #endif
652 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
653 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
654 			return 0;
655 		return 1;
656 	}
657 
658 #ifdef CONFIG_IP_ROUTE_MULTIPATH
659 	if (!cfg->fc_mp)
660 		return 0;
661 
662 	rtnh = cfg->fc_mp;
663 	remaining = cfg->fc_mp_len;
664 
665 	for_nexthops(fi) {
666 		int attrlen;
667 
668 		if (!rtnh_ok(rtnh, remaining))
669 			return -EINVAL;
670 
671 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
672 			return 1;
673 
674 		attrlen = rtnh_attrlen(rtnh);
675 		if (attrlen > 0) {
676 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
677 
678 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
679 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
680 				return 1;
681 #ifdef CONFIG_IP_ROUTE_CLASSID
682 			nla = nla_find(attrs, attrlen, RTA_FLOW);
683 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
684 				return 1;
685 #endif
686 		}
687 
688 		rtnh = rtnh_next(rtnh, &remaining);
689 	} endfor_nexthops(fi);
690 #endif
691 	return 0;
692 }
693 
694 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
695 {
696 	struct nlattr *nla;
697 	int remaining;
698 
699 	if (!cfg->fc_mx)
700 		return true;
701 
702 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
703 		int type = nla_type(nla);
704 		u32 fi_val, val;
705 
706 		if (!type)
707 			continue;
708 		if (type > RTAX_MAX)
709 			return false;
710 
711 		if (type == RTAX_CC_ALGO) {
712 			char tmp[TCP_CA_NAME_MAX];
713 			bool ecn_ca = false;
714 
715 			nla_strlcpy(tmp, nla, sizeof(tmp));
716 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
717 		} else {
718 			if (nla_len(nla) != sizeof(u32))
719 				return false;
720 			val = nla_get_u32(nla);
721 		}
722 
723 		fi_val = fi->fib_metrics->metrics[type - 1];
724 		if (type == RTAX_FEATURES)
725 			fi_val &= ~DST_FEATURE_ECN_CA;
726 
727 		if (fi_val != val)
728 			return false;
729 	}
730 
731 	return true;
732 }
733 
734 
735 /*
736  * Picture
737  * -------
738  *
739  * Semantics of nexthop is very messy by historical reasons.
740  * We have to take into account, that:
741  * a) gateway can be actually local interface address,
742  *    so that gatewayed route is direct.
743  * b) gateway must be on-link address, possibly
744  *    described not by an ifaddr, but also by a direct route.
745  * c) If both gateway and interface are specified, they should not
746  *    contradict.
747  * d) If we use tunnel routes, gateway could be not on-link.
748  *
749  * Attempt to reconcile all of these (alas, self-contradictory) conditions
750  * results in pretty ugly and hairy code with obscure logic.
751  *
752  * I chose to generalized it instead, so that the size
753  * of code does not increase practically, but it becomes
754  * much more general.
755  * Every prefix is assigned a "scope" value: "host" is local address,
756  * "link" is direct route,
757  * [ ... "site" ... "interior" ... ]
758  * and "universe" is true gateway route with global meaning.
759  *
760  * Every prefix refers to a set of "nexthop"s (gw, oif),
761  * where gw must have narrower scope. This recursion stops
762  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
763  * which means that gw is forced to be on link.
764  *
765  * Code is still hairy, but now it is apparently logically
766  * consistent and very flexible. F.e. as by-product it allows
767  * to co-exists in peace independent exterior and interior
768  * routing processes.
769  *
770  * Normally it looks as following.
771  *
772  * {universe prefix}  -> (gw, oif) [scope link]
773  *		  |
774  *		  |-> {link prefix} -> (gw, oif) [scope local]
775  *					|
776  *					|-> {local prefix} (terminal node)
777  */
778 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
779 			struct netlink_ext_ack *extack)
780 {
781 	int err = 0;
782 	struct net *net;
783 	struct net_device *dev;
784 
785 	net = cfg->fc_nlinfo.nl_net;
786 	if (nh->nh_gw) {
787 		struct fib_result res;
788 
789 		if (nh->nh_flags & RTNH_F_ONLINK) {
790 			unsigned int addr_type;
791 
792 			if (cfg->fc_scope >= RT_SCOPE_LINK) {
793 				NL_SET_ERR_MSG(extack,
794 					       "Nexthop has invalid scope");
795 				return -EINVAL;
796 			}
797 			dev = __dev_get_by_index(net, nh->nh_oif);
798 			if (!dev) {
799 				NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
800 				return -ENODEV;
801 			}
802 			if (!(dev->flags & IFF_UP)) {
803 				NL_SET_ERR_MSG(extack,
804 					       "Nexthop device is not up");
805 				return -ENETDOWN;
806 			}
807 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
808 			if (addr_type != RTN_UNICAST) {
809 				NL_SET_ERR_MSG(extack,
810 					       "Nexthop has invalid gateway");
811 				return -EINVAL;
812 			}
813 			if (!netif_carrier_ok(dev))
814 				nh->nh_flags |= RTNH_F_LINKDOWN;
815 			nh->nh_dev = dev;
816 			dev_hold(dev);
817 			nh->nh_scope = RT_SCOPE_LINK;
818 			return 0;
819 		}
820 		rcu_read_lock();
821 		{
822 			struct fib_table *tbl = NULL;
823 			struct flowi4 fl4 = {
824 				.daddr = nh->nh_gw,
825 				.flowi4_scope = cfg->fc_scope + 1,
826 				.flowi4_oif = nh->nh_oif,
827 				.flowi4_iif = LOOPBACK_IFINDEX,
828 			};
829 
830 			/* It is not necessary, but requires a bit of thinking */
831 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
832 				fl4.flowi4_scope = RT_SCOPE_LINK;
833 
834 			if (cfg->fc_table)
835 				tbl = fib_get_table(net, cfg->fc_table);
836 
837 			if (tbl)
838 				err = fib_table_lookup(tbl, &fl4, &res,
839 						       FIB_LOOKUP_IGNORE_LINKSTATE |
840 						       FIB_LOOKUP_NOREF);
841 
842 			/* on error or if no table given do full lookup. This
843 			 * is needed for example when nexthops are in the local
844 			 * table rather than the given table
845 			 */
846 			if (!tbl || err) {
847 				err = fib_lookup(net, &fl4, &res,
848 						 FIB_LOOKUP_IGNORE_LINKSTATE);
849 			}
850 
851 			if (err) {
852 				NL_SET_ERR_MSG(extack,
853 					       "Nexthop has invalid gateway");
854 				rcu_read_unlock();
855 				return err;
856 			}
857 		}
858 		err = -EINVAL;
859 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
860 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
861 			goto out;
862 		}
863 		nh->nh_scope = res.scope;
864 		nh->nh_oif = FIB_RES_OIF(res);
865 		nh->nh_dev = dev = FIB_RES_DEV(res);
866 		if (!dev) {
867 			NL_SET_ERR_MSG(extack,
868 				       "No egress device for nexthop gateway");
869 			goto out;
870 		}
871 		dev_hold(dev);
872 		if (!netif_carrier_ok(dev))
873 			nh->nh_flags |= RTNH_F_LINKDOWN;
874 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
875 	} else {
876 		struct in_device *in_dev;
877 
878 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
879 			NL_SET_ERR_MSG(extack,
880 				       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
881 			return -EINVAL;
882 		}
883 		rcu_read_lock();
884 		err = -ENODEV;
885 		in_dev = inetdev_by_index(net, nh->nh_oif);
886 		if (!in_dev)
887 			goto out;
888 		err = -ENETDOWN;
889 		if (!(in_dev->dev->flags & IFF_UP)) {
890 			NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
891 			goto out;
892 		}
893 		nh->nh_dev = in_dev->dev;
894 		dev_hold(nh->nh_dev);
895 		nh->nh_scope = RT_SCOPE_HOST;
896 		if (!netif_carrier_ok(nh->nh_dev))
897 			nh->nh_flags |= RTNH_F_LINKDOWN;
898 		err = 0;
899 	}
900 out:
901 	rcu_read_unlock();
902 	return err;
903 }
904 
905 static inline unsigned int fib_laddr_hashfn(__be32 val)
906 {
907 	unsigned int mask = (fib_info_hash_size - 1);
908 
909 	return ((__force u32)val ^
910 		((__force u32)val >> 7) ^
911 		((__force u32)val >> 14)) & mask;
912 }
913 
914 static struct hlist_head *fib_info_hash_alloc(int bytes)
915 {
916 	if (bytes <= PAGE_SIZE)
917 		return kzalloc(bytes, GFP_KERNEL);
918 	else
919 		return (struct hlist_head *)
920 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
921 					 get_order(bytes));
922 }
923 
924 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
925 {
926 	if (!hash)
927 		return;
928 
929 	if (bytes <= PAGE_SIZE)
930 		kfree(hash);
931 	else
932 		free_pages((unsigned long) hash, get_order(bytes));
933 }
934 
935 static void fib_info_hash_move(struct hlist_head *new_info_hash,
936 			       struct hlist_head *new_laddrhash,
937 			       unsigned int new_size)
938 {
939 	struct hlist_head *old_info_hash, *old_laddrhash;
940 	unsigned int old_size = fib_info_hash_size;
941 	unsigned int i, bytes;
942 
943 	spin_lock_bh(&fib_info_lock);
944 	old_info_hash = fib_info_hash;
945 	old_laddrhash = fib_info_laddrhash;
946 	fib_info_hash_size = new_size;
947 
948 	for (i = 0; i < old_size; i++) {
949 		struct hlist_head *head = &fib_info_hash[i];
950 		struct hlist_node *n;
951 		struct fib_info *fi;
952 
953 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
954 			struct hlist_head *dest;
955 			unsigned int new_hash;
956 
957 			new_hash = fib_info_hashfn(fi);
958 			dest = &new_info_hash[new_hash];
959 			hlist_add_head(&fi->fib_hash, dest);
960 		}
961 	}
962 	fib_info_hash = new_info_hash;
963 
964 	for (i = 0; i < old_size; i++) {
965 		struct hlist_head *lhead = &fib_info_laddrhash[i];
966 		struct hlist_node *n;
967 		struct fib_info *fi;
968 
969 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
970 			struct hlist_head *ldest;
971 			unsigned int new_hash;
972 
973 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
974 			ldest = &new_laddrhash[new_hash];
975 			hlist_add_head(&fi->fib_lhash, ldest);
976 		}
977 	}
978 	fib_info_laddrhash = new_laddrhash;
979 
980 	spin_unlock_bh(&fib_info_lock);
981 
982 	bytes = old_size * sizeof(struct hlist_head *);
983 	fib_info_hash_free(old_info_hash, bytes);
984 	fib_info_hash_free(old_laddrhash, bytes);
985 }
986 
987 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
988 {
989 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
990 					nh->nh_gw,
991 					nh->nh_parent->fib_scope);
992 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
993 
994 	return nh->nh_saddr;
995 }
996 
997 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
998 {
999 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1000 	    fib_prefsrc != cfg->fc_dst) {
1001 		u32 tb_id = cfg->fc_table;
1002 		int rc;
1003 
1004 		if (tb_id == RT_TABLE_MAIN)
1005 			tb_id = RT_TABLE_LOCAL;
1006 
1007 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1008 					  fib_prefsrc, tb_id);
1009 
1010 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1011 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1012 						  fib_prefsrc, RT_TABLE_LOCAL);
1013 		}
1014 
1015 		if (rc != RTN_LOCAL)
1016 			return false;
1017 	}
1018 	return true;
1019 }
1020 
1021 struct fib_info *fib_create_info(struct fib_config *cfg,
1022 				 struct netlink_ext_ack *extack)
1023 {
1024 	int err;
1025 	struct fib_info *fi = NULL;
1026 	struct fib_info *ofi;
1027 	int nhs = 1;
1028 	struct net *net = cfg->fc_nlinfo.nl_net;
1029 
1030 	if (cfg->fc_type > RTN_MAX)
1031 		goto err_inval;
1032 
1033 	/* Fast check to catch the most weird cases */
1034 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1035 		NL_SET_ERR_MSG(extack, "Invalid scope");
1036 		goto err_inval;
1037 	}
1038 
1039 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1040 		NL_SET_ERR_MSG(extack,
1041 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1042 		goto err_inval;
1043 	}
1044 
1045 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1046 	if (cfg->fc_mp) {
1047 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1048 		if (nhs == 0)
1049 			goto err_inval;
1050 	}
1051 #endif
1052 
1053 	err = -ENOBUFS;
1054 	if (fib_info_cnt >= fib_info_hash_size) {
1055 		unsigned int new_size = fib_info_hash_size << 1;
1056 		struct hlist_head *new_info_hash;
1057 		struct hlist_head *new_laddrhash;
1058 		unsigned int bytes;
1059 
1060 		if (!new_size)
1061 			new_size = 16;
1062 		bytes = new_size * sizeof(struct hlist_head *);
1063 		new_info_hash = fib_info_hash_alloc(bytes);
1064 		new_laddrhash = fib_info_hash_alloc(bytes);
1065 		if (!new_info_hash || !new_laddrhash) {
1066 			fib_info_hash_free(new_info_hash, bytes);
1067 			fib_info_hash_free(new_laddrhash, bytes);
1068 		} else
1069 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1070 
1071 		if (!fib_info_hash_size)
1072 			goto failure;
1073 	}
1074 
1075 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1076 	if (!fi)
1077 		goto failure;
1078 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1079 					      cfg->fc_mx_len);
1080 	if (unlikely(IS_ERR(fi->fib_metrics))) {
1081 		err = PTR_ERR(fi->fib_metrics);
1082 		kfree(fi);
1083 		return ERR_PTR(err);
1084 	}
1085 
1086 	fib_info_cnt++;
1087 	fi->fib_net = net;
1088 	fi->fib_protocol = cfg->fc_protocol;
1089 	fi->fib_scope = cfg->fc_scope;
1090 	fi->fib_flags = cfg->fc_flags;
1091 	fi->fib_priority = cfg->fc_priority;
1092 	fi->fib_prefsrc = cfg->fc_prefsrc;
1093 	fi->fib_type = cfg->fc_type;
1094 	fi->fib_tb_id = cfg->fc_table;
1095 
1096 	fi->fib_nhs = nhs;
1097 	change_nexthops(fi) {
1098 		nexthop_nh->nh_parent = fi;
1099 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1100 		if (!nexthop_nh->nh_pcpu_rth_output)
1101 			goto failure;
1102 	} endfor_nexthops(fi)
1103 
1104 	if (cfg->fc_mp) {
1105 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1106 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1107 		if (err != 0)
1108 			goto failure;
1109 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1110 			NL_SET_ERR_MSG(extack,
1111 				       "Nexthop device index does not match RTA_OIF");
1112 			goto err_inval;
1113 		}
1114 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1115 			NL_SET_ERR_MSG(extack,
1116 				       "Nexthop gateway does not match RTA_GATEWAY");
1117 			goto err_inval;
1118 		}
1119 #ifdef CONFIG_IP_ROUTE_CLASSID
1120 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1121 			NL_SET_ERR_MSG(extack,
1122 				       "Nexthop class id does not match RTA_FLOW");
1123 			goto err_inval;
1124 		}
1125 #endif
1126 #else
1127 		NL_SET_ERR_MSG(extack,
1128 			       "Multipath support not enabled in kernel");
1129 		goto err_inval;
1130 #endif
1131 	} else {
1132 		struct fib_nh *nh = fi->fib_nh;
1133 
1134 		if (cfg->fc_encap) {
1135 			struct lwtunnel_state *lwtstate;
1136 
1137 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1138 				NL_SET_ERR_MSG(extack,
1139 					       "LWT encap type not specified");
1140 				goto err_inval;
1141 			}
1142 			err = lwtunnel_build_state(cfg->fc_encap_type,
1143 						   cfg->fc_encap, AF_INET, cfg,
1144 						   &lwtstate, extack);
1145 			if (err)
1146 				goto failure;
1147 
1148 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1149 		}
1150 		nh->nh_oif = cfg->fc_oif;
1151 		nh->nh_gw = cfg->fc_gw;
1152 		nh->nh_flags = cfg->fc_flags;
1153 #ifdef CONFIG_IP_ROUTE_CLASSID
1154 		nh->nh_tclassid = cfg->fc_flow;
1155 		if (nh->nh_tclassid)
1156 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1157 #endif
1158 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1159 		nh->nh_weight = 1;
1160 #endif
1161 	}
1162 
1163 	if (fib_props[cfg->fc_type].error) {
1164 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1165 			NL_SET_ERR_MSG(extack,
1166 				       "Gateway, device and multipath can not be specified for this route type");
1167 			goto err_inval;
1168 		}
1169 		goto link_it;
1170 	} else {
1171 		switch (cfg->fc_type) {
1172 		case RTN_UNICAST:
1173 		case RTN_LOCAL:
1174 		case RTN_BROADCAST:
1175 		case RTN_ANYCAST:
1176 		case RTN_MULTICAST:
1177 			break;
1178 		default:
1179 			NL_SET_ERR_MSG(extack, "Invalid route type");
1180 			goto err_inval;
1181 		}
1182 	}
1183 
1184 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1185 		NL_SET_ERR_MSG(extack, "Invalid scope");
1186 		goto err_inval;
1187 	}
1188 
1189 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1190 		struct fib_nh *nh = fi->fib_nh;
1191 
1192 		/* Local address is added. */
1193 		if (nhs != 1) {
1194 			NL_SET_ERR_MSG(extack,
1195 				       "Route with host scope can not have multiple nexthops");
1196 			goto err_inval;
1197 		}
1198 		if (nh->nh_gw) {
1199 			NL_SET_ERR_MSG(extack,
1200 				       "Route with host scope can not have a gateway");
1201 			goto err_inval;
1202 		}
1203 		nh->nh_scope = RT_SCOPE_NOWHERE;
1204 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1205 		err = -ENODEV;
1206 		if (!nh->nh_dev)
1207 			goto failure;
1208 	} else {
1209 		int linkdown = 0;
1210 
1211 		change_nexthops(fi) {
1212 			err = fib_check_nh(cfg, nexthop_nh, extack);
1213 			if (err != 0)
1214 				goto failure;
1215 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1216 				linkdown++;
1217 		} endfor_nexthops(fi)
1218 		if (linkdown == fi->fib_nhs)
1219 			fi->fib_flags |= RTNH_F_LINKDOWN;
1220 	}
1221 
1222 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1223 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1224 		goto err_inval;
1225 	}
1226 
1227 	change_nexthops(fi) {
1228 		fib_info_update_nh_saddr(net, nexthop_nh);
1229 	} endfor_nexthops(fi)
1230 
1231 	fib_rebalance(fi);
1232 
1233 link_it:
1234 	ofi = fib_find_info(fi);
1235 	if (ofi) {
1236 		fi->fib_dead = 1;
1237 		free_fib_info(fi);
1238 		ofi->fib_treeref++;
1239 		return ofi;
1240 	}
1241 
1242 	fi->fib_treeref++;
1243 	refcount_set(&fi->fib_clntref, 1);
1244 	spin_lock_bh(&fib_info_lock);
1245 	hlist_add_head(&fi->fib_hash,
1246 		       &fib_info_hash[fib_info_hashfn(fi)]);
1247 	if (fi->fib_prefsrc) {
1248 		struct hlist_head *head;
1249 
1250 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1251 		hlist_add_head(&fi->fib_lhash, head);
1252 	}
1253 	change_nexthops(fi) {
1254 		struct hlist_head *head;
1255 		unsigned int hash;
1256 
1257 		if (!nexthop_nh->nh_dev)
1258 			continue;
1259 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1260 		head = &fib_info_devhash[hash];
1261 		hlist_add_head(&nexthop_nh->nh_hash, head);
1262 	} endfor_nexthops(fi)
1263 	spin_unlock_bh(&fib_info_lock);
1264 	return fi;
1265 
1266 err_inval:
1267 	err = -EINVAL;
1268 
1269 failure:
1270 	if (fi) {
1271 		fi->fib_dead = 1;
1272 		free_fib_info(fi);
1273 	}
1274 
1275 	return ERR_PTR(err);
1276 }
1277 
1278 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1279 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1280 		  struct fib_info *fi, unsigned int flags)
1281 {
1282 	struct nlmsghdr *nlh;
1283 	struct rtmsg *rtm;
1284 
1285 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1286 	if (!nlh)
1287 		return -EMSGSIZE;
1288 
1289 	rtm = nlmsg_data(nlh);
1290 	rtm->rtm_family = AF_INET;
1291 	rtm->rtm_dst_len = dst_len;
1292 	rtm->rtm_src_len = 0;
1293 	rtm->rtm_tos = tos;
1294 	if (tb_id < 256)
1295 		rtm->rtm_table = tb_id;
1296 	else
1297 		rtm->rtm_table = RT_TABLE_COMPAT;
1298 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1299 		goto nla_put_failure;
1300 	rtm->rtm_type = type;
1301 	rtm->rtm_flags = fi->fib_flags;
1302 	rtm->rtm_scope = fi->fib_scope;
1303 	rtm->rtm_protocol = fi->fib_protocol;
1304 
1305 	if (rtm->rtm_dst_len &&
1306 	    nla_put_in_addr(skb, RTA_DST, dst))
1307 		goto nla_put_failure;
1308 	if (fi->fib_priority &&
1309 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1310 		goto nla_put_failure;
1311 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1312 		goto nla_put_failure;
1313 
1314 	if (fi->fib_prefsrc &&
1315 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1316 		goto nla_put_failure;
1317 	if (fi->fib_nhs == 1) {
1318 		if (fi->fib_nh->nh_gw &&
1319 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1320 			goto nla_put_failure;
1321 		if (fi->fib_nh->nh_oif &&
1322 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1323 			goto nla_put_failure;
1324 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1325 			struct in_device *in_dev;
1326 
1327 			rcu_read_lock();
1328 			in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev);
1329 			if (in_dev &&
1330 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1331 				rtm->rtm_flags |= RTNH_F_DEAD;
1332 			rcu_read_unlock();
1333 		}
1334 		if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1335 			rtm->rtm_flags |= RTNH_F_OFFLOAD;
1336 #ifdef CONFIG_IP_ROUTE_CLASSID
1337 		if (fi->fib_nh[0].nh_tclassid &&
1338 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1339 			goto nla_put_failure;
1340 #endif
1341 		if (fi->fib_nh->nh_lwtstate &&
1342 		    lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1343 			goto nla_put_failure;
1344 	}
1345 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1346 	if (fi->fib_nhs > 1) {
1347 		struct rtnexthop *rtnh;
1348 		struct nlattr *mp;
1349 
1350 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1351 		if (!mp)
1352 			goto nla_put_failure;
1353 
1354 		for_nexthops(fi) {
1355 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1356 			if (!rtnh)
1357 				goto nla_put_failure;
1358 
1359 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1360 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1361 				struct in_device *in_dev;
1362 
1363 				rcu_read_lock();
1364 				in_dev = __in_dev_get_rcu(nh->nh_dev);
1365 				if (in_dev &&
1366 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1367 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1368 				rcu_read_unlock();
1369 			}
1370 			rtnh->rtnh_hops = nh->nh_weight - 1;
1371 			rtnh->rtnh_ifindex = nh->nh_oif;
1372 
1373 			if (nh->nh_gw &&
1374 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1375 				goto nla_put_failure;
1376 #ifdef CONFIG_IP_ROUTE_CLASSID
1377 			if (nh->nh_tclassid &&
1378 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1379 				goto nla_put_failure;
1380 #endif
1381 			if (nh->nh_lwtstate &&
1382 			    lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1383 				goto nla_put_failure;
1384 
1385 			/* length of rtnetlink header + attributes */
1386 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1387 		} endfor_nexthops(fi);
1388 
1389 		nla_nest_end(skb, mp);
1390 	}
1391 #endif
1392 	nlmsg_end(skb, nlh);
1393 	return 0;
1394 
1395 nla_put_failure:
1396 	nlmsg_cancel(skb, nlh);
1397 	return -EMSGSIZE;
1398 }
1399 
1400 /*
1401  * Update FIB if:
1402  * - local address disappeared -> we must delete all the entries
1403  *   referring to it.
1404  * - device went down -> we must shutdown all nexthops going via it.
1405  */
1406 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1407 {
1408 	int ret = 0;
1409 	unsigned int hash = fib_laddr_hashfn(local);
1410 	struct hlist_head *head = &fib_info_laddrhash[hash];
1411 	struct net *net = dev_net(dev);
1412 	int tb_id = l3mdev_fib_table(dev);
1413 	struct fib_info *fi;
1414 
1415 	if (!fib_info_laddrhash || local == 0)
1416 		return 0;
1417 
1418 	hlist_for_each_entry(fi, head, fib_lhash) {
1419 		if (!net_eq(fi->fib_net, net) ||
1420 		    fi->fib_tb_id != tb_id)
1421 			continue;
1422 		if (fi->fib_prefsrc == local) {
1423 			fi->fib_flags |= RTNH_F_DEAD;
1424 			ret++;
1425 		}
1426 	}
1427 	return ret;
1428 }
1429 
1430 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1431 				 enum fib_event_type event_type)
1432 {
1433 	struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1434 	struct fib_nh_notifier_info info = {
1435 		.fib_nh = fib_nh,
1436 	};
1437 
1438 	switch (event_type) {
1439 	case FIB_EVENT_NH_ADD:
1440 		if (fib_nh->nh_flags & RTNH_F_DEAD)
1441 			break;
1442 		if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1443 		    fib_nh->nh_flags & RTNH_F_LINKDOWN)
1444 			break;
1445 		return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1446 					   &info.info);
1447 	case FIB_EVENT_NH_DEL:
1448 		if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1449 		     fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1450 		    (fib_nh->nh_flags & RTNH_F_DEAD))
1451 			return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1452 						   event_type, &info.info);
1453 	default:
1454 		break;
1455 	}
1456 
1457 	return NOTIFY_DONE;
1458 }
1459 
1460 /* Update the PMTU of exceptions when:
1461  * - the new MTU of the first hop becomes smaller than the PMTU
1462  * - the old MTU was the same as the PMTU, and it limited discovery of
1463  *   larger MTUs on the path. With that limit raised, we can now
1464  *   discover larger MTUs
1465  * A special case is locked exceptions, for which the PMTU is smaller
1466  * than the minimal accepted PMTU:
1467  * - if the new MTU is greater than the PMTU, don't make any change
1468  * - otherwise, unlock and set PMTU
1469  */
1470 static void nh_update_mtu(struct fib_nh *nh, u32 new, u32 orig)
1471 {
1472 	struct fnhe_hash_bucket *bucket;
1473 	int i;
1474 
1475 	bucket = rcu_dereference_protected(nh->nh_exceptions, 1);
1476 	if (!bucket)
1477 		return;
1478 
1479 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1480 		struct fib_nh_exception *fnhe;
1481 
1482 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1483 		     fnhe;
1484 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1485 			if (fnhe->fnhe_mtu_locked) {
1486 				if (new <= fnhe->fnhe_pmtu) {
1487 					fnhe->fnhe_pmtu = new;
1488 					fnhe->fnhe_mtu_locked = false;
1489 				}
1490 			} else if (new < fnhe->fnhe_pmtu ||
1491 				   orig == fnhe->fnhe_pmtu) {
1492 				fnhe->fnhe_pmtu = new;
1493 			}
1494 		}
1495 	}
1496 }
1497 
1498 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1499 {
1500 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1501 	struct hlist_head *head = &fib_info_devhash[hash];
1502 	struct fib_nh *nh;
1503 
1504 	hlist_for_each_entry(nh, head, nh_hash) {
1505 		if (nh->nh_dev == dev)
1506 			nh_update_mtu(nh, dev->mtu, orig_mtu);
1507 	}
1508 }
1509 
1510 /* Event              force Flags           Description
1511  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1512  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1513  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1514  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1515  */
1516 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1517 {
1518 	int ret = 0;
1519 	int scope = RT_SCOPE_NOWHERE;
1520 	struct fib_info *prev_fi = NULL;
1521 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1522 	struct hlist_head *head = &fib_info_devhash[hash];
1523 	struct fib_nh *nh;
1524 
1525 	if (force)
1526 		scope = -1;
1527 
1528 	hlist_for_each_entry(nh, head, nh_hash) {
1529 		struct fib_info *fi = nh->nh_parent;
1530 		int dead;
1531 
1532 		BUG_ON(!fi->fib_nhs);
1533 		if (nh->nh_dev != dev || fi == prev_fi)
1534 			continue;
1535 		prev_fi = fi;
1536 		dead = 0;
1537 		change_nexthops(fi) {
1538 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1539 				dead++;
1540 			else if (nexthop_nh->nh_dev == dev &&
1541 				 nexthop_nh->nh_scope != scope) {
1542 				switch (event) {
1543 				case NETDEV_DOWN:
1544 				case NETDEV_UNREGISTER:
1545 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1546 					/* fall through */
1547 				case NETDEV_CHANGE:
1548 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1549 					break;
1550 				}
1551 				call_fib_nh_notifiers(nexthop_nh,
1552 						      FIB_EVENT_NH_DEL);
1553 				dead++;
1554 			}
1555 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1556 			if (event == NETDEV_UNREGISTER &&
1557 			    nexthop_nh->nh_dev == dev) {
1558 				dead = fi->fib_nhs;
1559 				break;
1560 			}
1561 #endif
1562 		} endfor_nexthops(fi)
1563 		if (dead == fi->fib_nhs) {
1564 			switch (event) {
1565 			case NETDEV_DOWN:
1566 			case NETDEV_UNREGISTER:
1567 				fi->fib_flags |= RTNH_F_DEAD;
1568 				/* fall through */
1569 			case NETDEV_CHANGE:
1570 				fi->fib_flags |= RTNH_F_LINKDOWN;
1571 				break;
1572 			}
1573 			ret++;
1574 		}
1575 
1576 		fib_rebalance(fi);
1577 	}
1578 
1579 	return ret;
1580 }
1581 
1582 /* Must be invoked inside of an RCU protected region.  */
1583 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1584 {
1585 	struct fib_info *fi = NULL, *last_resort = NULL;
1586 	struct hlist_head *fa_head = res->fa_head;
1587 	struct fib_table *tb = res->table;
1588 	u8 slen = 32 - res->prefixlen;
1589 	int order = -1, last_idx = -1;
1590 	struct fib_alias *fa, *fa1 = NULL;
1591 	u32 last_prio = res->fi->fib_priority;
1592 	u8 last_tos = 0;
1593 
1594 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1595 		struct fib_info *next_fi = fa->fa_info;
1596 
1597 		if (fa->fa_slen != slen)
1598 			continue;
1599 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1600 			continue;
1601 		if (fa->tb_id != tb->tb_id)
1602 			continue;
1603 		if (next_fi->fib_priority > last_prio &&
1604 		    fa->fa_tos == last_tos) {
1605 			if (last_tos)
1606 				continue;
1607 			break;
1608 		}
1609 		if (next_fi->fib_flags & RTNH_F_DEAD)
1610 			continue;
1611 		last_tos = fa->fa_tos;
1612 		last_prio = next_fi->fib_priority;
1613 
1614 		if (next_fi->fib_scope != res->scope ||
1615 		    fa->fa_type != RTN_UNICAST)
1616 			continue;
1617 		if (!next_fi->fib_nh[0].nh_gw ||
1618 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1619 			continue;
1620 
1621 		fib_alias_accessed(fa);
1622 
1623 		if (!fi) {
1624 			if (next_fi != res->fi)
1625 				break;
1626 			fa1 = fa;
1627 		} else if (!fib_detect_death(fi, order, &last_resort,
1628 					     &last_idx, fa1->fa_default)) {
1629 			fib_result_assign(res, fi);
1630 			fa1->fa_default = order;
1631 			goto out;
1632 		}
1633 		fi = next_fi;
1634 		order++;
1635 	}
1636 
1637 	if (order <= 0 || !fi) {
1638 		if (fa1)
1639 			fa1->fa_default = -1;
1640 		goto out;
1641 	}
1642 
1643 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1644 			      fa1->fa_default)) {
1645 		fib_result_assign(res, fi);
1646 		fa1->fa_default = order;
1647 		goto out;
1648 	}
1649 
1650 	if (last_idx >= 0)
1651 		fib_result_assign(res, last_resort);
1652 	fa1->fa_default = last_idx;
1653 out:
1654 	return;
1655 }
1656 
1657 /*
1658  * Dead device goes up. We wake up dead nexthops.
1659  * It takes sense only on multipath routes.
1660  */
1661 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1662 {
1663 	struct fib_info *prev_fi;
1664 	unsigned int hash;
1665 	struct hlist_head *head;
1666 	struct fib_nh *nh;
1667 	int ret;
1668 
1669 	if (!(dev->flags & IFF_UP))
1670 		return 0;
1671 
1672 	if (nh_flags & RTNH_F_DEAD) {
1673 		unsigned int flags = dev_get_flags(dev);
1674 
1675 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1676 			nh_flags |= RTNH_F_LINKDOWN;
1677 	}
1678 
1679 	prev_fi = NULL;
1680 	hash = fib_devindex_hashfn(dev->ifindex);
1681 	head = &fib_info_devhash[hash];
1682 	ret = 0;
1683 
1684 	hlist_for_each_entry(nh, head, nh_hash) {
1685 		struct fib_info *fi = nh->nh_parent;
1686 		int alive;
1687 
1688 		BUG_ON(!fi->fib_nhs);
1689 		if (nh->nh_dev != dev || fi == prev_fi)
1690 			continue;
1691 
1692 		prev_fi = fi;
1693 		alive = 0;
1694 		change_nexthops(fi) {
1695 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1696 				alive++;
1697 				continue;
1698 			}
1699 			if (!nexthop_nh->nh_dev ||
1700 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1701 				continue;
1702 			if (nexthop_nh->nh_dev != dev ||
1703 			    !__in_dev_get_rtnl(dev))
1704 				continue;
1705 			alive++;
1706 			nexthop_nh->nh_flags &= ~nh_flags;
1707 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1708 		} endfor_nexthops(fi)
1709 
1710 		if (alive > 0) {
1711 			fi->fib_flags &= ~nh_flags;
1712 			ret++;
1713 		}
1714 
1715 		fib_rebalance(fi);
1716 	}
1717 
1718 	return ret;
1719 }
1720 
1721 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1722 static bool fib_good_nh(const struct fib_nh *nh)
1723 {
1724 	int state = NUD_REACHABLE;
1725 
1726 	if (nh->nh_scope == RT_SCOPE_LINK) {
1727 		struct neighbour *n;
1728 
1729 		rcu_read_lock_bh();
1730 
1731 		n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1732 					      (__force u32)nh->nh_gw);
1733 		if (n)
1734 			state = n->nud_state;
1735 
1736 		rcu_read_unlock_bh();
1737 	}
1738 
1739 	return !!(state & NUD_VALID);
1740 }
1741 
1742 void fib_select_multipath(struct fib_result *res, int hash)
1743 {
1744 	struct fib_info *fi = res->fi;
1745 	struct net *net = fi->fib_net;
1746 	bool first = false;
1747 
1748 	for_nexthops(fi) {
1749 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
1750 			if (!fib_good_nh(nh))
1751 				continue;
1752 			if (!first) {
1753 				res->nh_sel = nhsel;
1754 				first = true;
1755 			}
1756 		}
1757 
1758 		if (hash > atomic_read(&nh->nh_upper_bound))
1759 			continue;
1760 
1761 		res->nh_sel = nhsel;
1762 		return;
1763 	} endfor_nexthops(fi);
1764 }
1765 #endif
1766 
1767 void fib_select_path(struct net *net, struct fib_result *res,
1768 		     struct flowi4 *fl4, const struct sk_buff *skb)
1769 {
1770 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
1771 		goto check_saddr;
1772 
1773 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1774 	if (res->fi->fib_nhs > 1) {
1775 		int h = fib_multipath_hash(net, fl4, skb, NULL);
1776 
1777 		fib_select_multipath(res, h);
1778 	}
1779 	else
1780 #endif
1781 	if (!res->prefixlen &&
1782 	    res->table->tb_num_default > 1 &&
1783 	    res->type == RTN_UNICAST)
1784 		fib_select_default(fl4, res);
1785 
1786 check_saddr:
1787 	if (!fl4->saddr)
1788 		fl4->saddr = FIB_RES_PREFSRC(net, *res);
1789 }
1790