xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 6724ed7f)
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 	struct dst_metrics *m;
212 
213 	change_nexthops(fi) {
214 		if (nexthop_nh->nh_dev)
215 			dev_put(nexthop_nh->nh_dev);
216 		lwtstate_put(nexthop_nh->nh_lwtstate);
217 		free_nh_exceptions(nexthop_nh);
218 		rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
219 		rt_fibinfo_free(&nexthop_nh->nh_rth_input);
220 	} endfor_nexthops(fi);
221 
222 	m = fi->fib_metrics;
223 	if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
224 		kfree(m);
225 	kfree(fi);
226 }
227 
228 void free_fib_info(struct fib_info *fi)
229 {
230 	if (fi->fib_dead == 0) {
231 		pr_warn("Freeing alive fib_info %p\n", fi);
232 		return;
233 	}
234 	fib_info_cnt--;
235 #ifdef CONFIG_IP_ROUTE_CLASSID
236 	change_nexthops(fi) {
237 		if (nexthop_nh->nh_tclassid)
238 			fi->fib_net->ipv4.fib_num_tclassid_users--;
239 	} endfor_nexthops(fi);
240 #endif
241 	call_rcu(&fi->rcu, free_fib_info_rcu);
242 }
243 EXPORT_SYMBOL_GPL(free_fib_info);
244 
245 void fib_release_info(struct fib_info *fi)
246 {
247 	spin_lock_bh(&fib_info_lock);
248 	if (fi && --fi->fib_treeref == 0) {
249 		hlist_del(&fi->fib_hash);
250 		if (fi->fib_prefsrc)
251 			hlist_del(&fi->fib_lhash);
252 		change_nexthops(fi) {
253 			if (!nexthop_nh->nh_dev)
254 				continue;
255 			hlist_del(&nexthop_nh->nh_hash);
256 		} endfor_nexthops(fi)
257 		fi->fib_dead = 1;
258 		fib_info_put(fi);
259 	}
260 	spin_unlock_bh(&fib_info_lock);
261 }
262 
263 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
264 {
265 	const struct fib_nh *onh = ofi->fib_nh;
266 
267 	for_nexthops(fi) {
268 		if (nh->nh_oif != onh->nh_oif ||
269 		    nh->nh_gw  != onh->nh_gw ||
270 		    nh->nh_scope != onh->nh_scope ||
271 #ifdef CONFIG_IP_ROUTE_MULTIPATH
272 		    nh->nh_weight != onh->nh_weight ||
273 #endif
274 #ifdef CONFIG_IP_ROUTE_CLASSID
275 		    nh->nh_tclassid != onh->nh_tclassid ||
276 #endif
277 		    lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
278 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
279 			return -1;
280 		onh++;
281 	} endfor_nexthops(fi);
282 	return 0;
283 }
284 
285 static inline unsigned int fib_devindex_hashfn(unsigned int val)
286 {
287 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
288 
289 	return (val ^
290 		(val >> DEVINDEX_HASHBITS) ^
291 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
292 }
293 
294 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
295 {
296 	unsigned int mask = (fib_info_hash_size - 1);
297 	unsigned int val = fi->fib_nhs;
298 
299 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
300 	val ^= (__force u32)fi->fib_prefsrc;
301 	val ^= fi->fib_priority;
302 	for_nexthops(fi) {
303 		val ^= fib_devindex_hashfn(nh->nh_oif);
304 	} endfor_nexthops(fi)
305 
306 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
307 }
308 
309 static struct fib_info *fib_find_info(const struct fib_info *nfi)
310 {
311 	struct hlist_head *head;
312 	struct fib_info *fi;
313 	unsigned int hash;
314 
315 	hash = fib_info_hashfn(nfi);
316 	head = &fib_info_hash[hash];
317 
318 	hlist_for_each_entry(fi, head, fib_hash) {
319 		if (!net_eq(fi->fib_net, nfi->fib_net))
320 			continue;
321 		if (fi->fib_nhs != nfi->fib_nhs)
322 			continue;
323 		if (nfi->fib_protocol == fi->fib_protocol &&
324 		    nfi->fib_scope == fi->fib_scope &&
325 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
326 		    nfi->fib_priority == fi->fib_priority &&
327 		    nfi->fib_type == fi->fib_type &&
328 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
329 			   sizeof(u32) * RTAX_MAX) == 0 &&
330 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
331 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
332 			return fi;
333 	}
334 
335 	return NULL;
336 }
337 
338 /* Check, that the gateway is already configured.
339  * Used only by redirect accept routine.
340  */
341 int ip_fib_check_default(__be32 gw, struct net_device *dev)
342 {
343 	struct hlist_head *head;
344 	struct fib_nh *nh;
345 	unsigned int hash;
346 
347 	spin_lock(&fib_info_lock);
348 
349 	hash = fib_devindex_hashfn(dev->ifindex);
350 	head = &fib_info_devhash[hash];
351 	hlist_for_each_entry(nh, head, nh_hash) {
352 		if (nh->nh_dev == dev &&
353 		    nh->nh_gw == gw &&
354 		    !(nh->nh_flags & RTNH_F_DEAD)) {
355 			spin_unlock(&fib_info_lock);
356 			return 0;
357 		}
358 	}
359 
360 	spin_unlock(&fib_info_lock);
361 
362 	return -1;
363 }
364 
365 static inline size_t fib_nlmsg_size(struct fib_info *fi)
366 {
367 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
368 			 + nla_total_size(4) /* RTA_TABLE */
369 			 + nla_total_size(4) /* RTA_DST */
370 			 + nla_total_size(4) /* RTA_PRIORITY */
371 			 + nla_total_size(4) /* RTA_PREFSRC */
372 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
373 
374 	/* space for nested metrics */
375 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
376 
377 	if (fi->fib_nhs) {
378 		size_t nh_encapsize = 0;
379 		/* Also handles the special case fib_nhs == 1 */
380 
381 		/* each nexthop is packed in an attribute */
382 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
383 
384 		/* may contain flow and gateway attribute */
385 		nhsize += 2 * nla_total_size(4);
386 
387 		/* grab encap info */
388 		for_nexthops(fi) {
389 			if (nh->nh_lwtstate) {
390 				/* RTA_ENCAP_TYPE */
391 				nh_encapsize += lwtunnel_get_encap_size(
392 						nh->nh_lwtstate);
393 				/* RTA_ENCAP */
394 				nh_encapsize +=  nla_total_size(2);
395 			}
396 		} endfor_nexthops(fi);
397 
398 		/* all nexthops are packed in a nested attribute */
399 		payload += nla_total_size((fi->fib_nhs * nhsize) +
400 					  nh_encapsize);
401 
402 	}
403 
404 	return payload;
405 }
406 
407 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
408 	       int dst_len, u32 tb_id, const struct nl_info *info,
409 	       unsigned int nlm_flags)
410 {
411 	struct sk_buff *skb;
412 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
413 	int err = -ENOBUFS;
414 
415 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
416 	if (!skb)
417 		goto errout;
418 
419 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
420 			    fa->fa_type, key, dst_len,
421 			    fa->fa_tos, fa->fa_info, nlm_flags);
422 	if (err < 0) {
423 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
424 		WARN_ON(err == -EMSGSIZE);
425 		kfree_skb(skb);
426 		goto errout;
427 	}
428 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
429 		    info->nlh, GFP_KERNEL);
430 	return;
431 errout:
432 	if (err < 0)
433 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
434 }
435 
436 static int fib_detect_death(struct fib_info *fi, int order,
437 			    struct fib_info **last_resort, int *last_idx,
438 			    int dflt)
439 {
440 	struct neighbour *n;
441 	int state = NUD_NONE;
442 
443 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
444 	if (n) {
445 		state = n->nud_state;
446 		neigh_release(n);
447 	} else {
448 		return 0;
449 	}
450 	if (state == NUD_REACHABLE)
451 		return 0;
452 	if ((state & NUD_VALID) && order != dflt)
453 		return 0;
454 	if ((state & NUD_VALID) ||
455 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
456 		*last_resort = fi;
457 		*last_idx = order;
458 	}
459 	return 1;
460 }
461 
462 #ifdef CONFIG_IP_ROUTE_MULTIPATH
463 
464 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
465 			      struct netlink_ext_ack *extack)
466 {
467 	int nhs = 0;
468 
469 	while (rtnh_ok(rtnh, remaining)) {
470 		nhs++;
471 		rtnh = rtnh_next(rtnh, &remaining);
472 	}
473 
474 	/* leftover implies invalid nexthop configuration, discard it */
475 	if (remaining > 0) {
476 		NL_SET_ERR_MSG(extack,
477 			       "Invalid nexthop configuration - extra data after nexthops");
478 		nhs = 0;
479 	}
480 
481 	return nhs;
482 }
483 
484 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
485 		       int remaining, struct fib_config *cfg,
486 		       struct netlink_ext_ack *extack)
487 {
488 	int ret;
489 
490 	change_nexthops(fi) {
491 		int attrlen;
492 
493 		if (!rtnh_ok(rtnh, remaining)) {
494 			NL_SET_ERR_MSG(extack,
495 				       "Invalid nexthop configuration - extra data after nexthop");
496 			return -EINVAL;
497 		}
498 
499 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
500 			NL_SET_ERR_MSG(extack,
501 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
502 			return -EINVAL;
503 		}
504 
505 		nexthop_nh->nh_flags =
506 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
507 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
508 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
509 
510 		attrlen = rtnh_attrlen(rtnh);
511 		if (attrlen > 0) {
512 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
513 
514 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
515 			nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
516 #ifdef CONFIG_IP_ROUTE_CLASSID
517 			nla = nla_find(attrs, attrlen, RTA_FLOW);
518 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
519 			if (nexthop_nh->nh_tclassid)
520 				fi->fib_net->ipv4.fib_num_tclassid_users++;
521 #endif
522 			nla = nla_find(attrs, attrlen, RTA_ENCAP);
523 			if (nla) {
524 				struct lwtunnel_state *lwtstate;
525 				struct nlattr *nla_entype;
526 
527 				nla_entype = nla_find(attrs, attrlen,
528 						      RTA_ENCAP_TYPE);
529 				if (!nla_entype) {
530 					NL_SET_BAD_ATTR(extack, nla);
531 					NL_SET_ERR_MSG(extack,
532 						       "Encap type is missing");
533 					goto err_inval;
534 				}
535 
536 				ret = lwtunnel_build_state(nla_get_u16(
537 							   nla_entype),
538 							   nla,  AF_INET, cfg,
539 							   &lwtstate, extack);
540 				if (ret)
541 					goto errout;
542 				nexthop_nh->nh_lwtstate =
543 					lwtstate_get(lwtstate);
544 			}
545 		}
546 
547 		rtnh = rtnh_next(rtnh, &remaining);
548 	} endfor_nexthops(fi);
549 
550 	return 0;
551 
552 err_inval:
553 	ret = -EINVAL;
554 
555 errout:
556 	return ret;
557 }
558 
559 static void fib_rebalance(struct fib_info *fi)
560 {
561 	int total;
562 	int w;
563 	struct in_device *in_dev;
564 
565 	if (fi->fib_nhs < 2)
566 		return;
567 
568 	total = 0;
569 	for_nexthops(fi) {
570 		if (nh->nh_flags & RTNH_F_DEAD)
571 			continue;
572 
573 		in_dev = __in_dev_get_rtnl(nh->nh_dev);
574 
575 		if (in_dev &&
576 		    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
577 		    nh->nh_flags & RTNH_F_LINKDOWN)
578 			continue;
579 
580 		total += nh->nh_weight;
581 	} endfor_nexthops(fi);
582 
583 	w = 0;
584 	change_nexthops(fi) {
585 		int upper_bound;
586 
587 		in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
588 
589 		if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
590 			upper_bound = -1;
591 		} else if (in_dev &&
592 			   IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
593 			   nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
594 			upper_bound = -1;
595 		} else {
596 			w += nexthop_nh->nh_weight;
597 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
598 							    total) - 1;
599 		}
600 
601 		atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
602 	} endfor_nexthops(fi);
603 }
604 #else /* CONFIG_IP_ROUTE_MULTIPATH */
605 
606 #define fib_rebalance(fi) do { } while (0)
607 
608 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
609 
610 static int fib_encap_match(u16 encap_type,
611 			   struct nlattr *encap,
612 			   const struct fib_nh *nh,
613 			   const struct fib_config *cfg,
614 			   struct netlink_ext_ack *extack)
615 {
616 	struct lwtunnel_state *lwtstate;
617 	int ret, result = 0;
618 
619 	if (encap_type == LWTUNNEL_ENCAP_NONE)
620 		return 0;
621 
622 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
623 				   cfg, &lwtstate, extack);
624 	if (!ret) {
625 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
626 		lwtstate_free(lwtstate);
627 	}
628 
629 	return result;
630 }
631 
632 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
633 		 struct netlink_ext_ack *extack)
634 {
635 #ifdef CONFIG_IP_ROUTE_MULTIPATH
636 	struct rtnexthop *rtnh;
637 	int remaining;
638 #endif
639 
640 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
641 		return 1;
642 
643 	if (cfg->fc_oif || cfg->fc_gw) {
644 		if (cfg->fc_encap) {
645 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
646 					    fi->fib_nh, cfg, extack))
647 				return 1;
648 		}
649 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
650 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
651 			return 0;
652 		return 1;
653 	}
654 
655 #ifdef CONFIG_IP_ROUTE_MULTIPATH
656 	if (!cfg->fc_mp)
657 		return 0;
658 
659 	rtnh = cfg->fc_mp;
660 	remaining = cfg->fc_mp_len;
661 
662 	for_nexthops(fi) {
663 		int attrlen;
664 
665 		if (!rtnh_ok(rtnh, remaining))
666 			return -EINVAL;
667 
668 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
669 			return 1;
670 
671 		attrlen = rtnh_attrlen(rtnh);
672 		if (attrlen > 0) {
673 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
674 
675 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
676 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
677 				return 1;
678 #ifdef CONFIG_IP_ROUTE_CLASSID
679 			nla = nla_find(attrs, attrlen, RTA_FLOW);
680 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
681 				return 1;
682 #endif
683 		}
684 
685 		rtnh = rtnh_next(rtnh, &remaining);
686 	} endfor_nexthops(fi);
687 #endif
688 	return 0;
689 }
690 
691 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
692 {
693 	struct nlattr *nla;
694 	int remaining;
695 
696 	if (!cfg->fc_mx)
697 		return true;
698 
699 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
700 		int type = nla_type(nla);
701 		u32 fi_val, val;
702 
703 		if (!type)
704 			continue;
705 		if (type > RTAX_MAX)
706 			return false;
707 
708 		if (type == RTAX_CC_ALGO) {
709 			char tmp[TCP_CA_NAME_MAX];
710 			bool ecn_ca = false;
711 
712 			nla_strlcpy(tmp, nla, sizeof(tmp));
713 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
714 		} else {
715 			val = nla_get_u32(nla);
716 		}
717 
718 		fi_val = fi->fib_metrics->metrics[type - 1];
719 		if (type == RTAX_FEATURES)
720 			fi_val &= ~DST_FEATURE_ECN_CA;
721 
722 		if (fi_val != val)
723 			return false;
724 	}
725 
726 	return true;
727 }
728 
729 
730 /*
731  * Picture
732  * -------
733  *
734  * Semantics of nexthop is very messy by historical reasons.
735  * We have to take into account, that:
736  * a) gateway can be actually local interface address,
737  *    so that gatewayed route is direct.
738  * b) gateway must be on-link address, possibly
739  *    described not by an ifaddr, but also by a direct route.
740  * c) If both gateway and interface are specified, they should not
741  *    contradict.
742  * d) If we use tunnel routes, gateway could be not on-link.
743  *
744  * Attempt to reconcile all of these (alas, self-contradictory) conditions
745  * results in pretty ugly and hairy code with obscure logic.
746  *
747  * I chose to generalized it instead, so that the size
748  * of code does not increase practically, but it becomes
749  * much more general.
750  * Every prefix is assigned a "scope" value: "host" is local address,
751  * "link" is direct route,
752  * [ ... "site" ... "interior" ... ]
753  * and "universe" is true gateway route with global meaning.
754  *
755  * Every prefix refers to a set of "nexthop"s (gw, oif),
756  * where gw must have narrower scope. This recursion stops
757  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
758  * which means that gw is forced to be on link.
759  *
760  * Code is still hairy, but now it is apparently logically
761  * consistent and very flexible. F.e. as by-product it allows
762  * to co-exists in peace independent exterior and interior
763  * routing processes.
764  *
765  * Normally it looks as following.
766  *
767  * {universe prefix}  -> (gw, oif) [scope link]
768  *		  |
769  *		  |-> {link prefix} -> (gw, oif) [scope local]
770  *					|
771  *					|-> {local prefix} (terminal node)
772  */
773 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
774 			struct netlink_ext_ack *extack)
775 {
776 	int err = 0;
777 	struct net *net;
778 	struct net_device *dev;
779 
780 	net = cfg->fc_nlinfo.nl_net;
781 	if (nh->nh_gw) {
782 		struct fib_result res;
783 
784 		if (nh->nh_flags & RTNH_F_ONLINK) {
785 			unsigned int addr_type;
786 
787 			if (cfg->fc_scope >= RT_SCOPE_LINK) {
788 				NL_SET_ERR_MSG(extack,
789 					       "Nexthop has invalid scope");
790 				return -EINVAL;
791 			}
792 			dev = __dev_get_by_index(net, nh->nh_oif);
793 			if (!dev)
794 				return -ENODEV;
795 			if (!(dev->flags & IFF_UP)) {
796 				NL_SET_ERR_MSG(extack,
797 					       "Nexthop device is not up");
798 				return -ENETDOWN;
799 			}
800 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
801 			if (addr_type != RTN_UNICAST) {
802 				NL_SET_ERR_MSG(extack,
803 					       "Nexthop has invalid gateway");
804 				return -EINVAL;
805 			}
806 			if (!netif_carrier_ok(dev))
807 				nh->nh_flags |= RTNH_F_LINKDOWN;
808 			nh->nh_dev = dev;
809 			dev_hold(dev);
810 			nh->nh_scope = RT_SCOPE_LINK;
811 			return 0;
812 		}
813 		rcu_read_lock();
814 		{
815 			struct fib_table *tbl = NULL;
816 			struct flowi4 fl4 = {
817 				.daddr = nh->nh_gw,
818 				.flowi4_scope = cfg->fc_scope + 1,
819 				.flowi4_oif = nh->nh_oif,
820 				.flowi4_iif = LOOPBACK_IFINDEX,
821 			};
822 
823 			/* It is not necessary, but requires a bit of thinking */
824 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
825 				fl4.flowi4_scope = RT_SCOPE_LINK;
826 
827 			if (cfg->fc_table)
828 				tbl = fib_get_table(net, cfg->fc_table);
829 
830 			if (tbl)
831 				err = fib_table_lookup(tbl, &fl4, &res,
832 						       FIB_LOOKUP_IGNORE_LINKSTATE |
833 						       FIB_LOOKUP_NOREF);
834 
835 			/* on error or if no table given do full lookup. This
836 			 * is needed for example when nexthops are in the local
837 			 * table rather than the given table
838 			 */
839 			if (!tbl || err) {
840 				err = fib_lookup(net, &fl4, &res,
841 						 FIB_LOOKUP_IGNORE_LINKSTATE);
842 			}
843 
844 			if (err) {
845 				NL_SET_ERR_MSG(extack,
846 					       "Nexthop has invalid gateway");
847 				rcu_read_unlock();
848 				return err;
849 			}
850 		}
851 		err = -EINVAL;
852 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
853 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
854 			goto out;
855 		}
856 		nh->nh_scope = res.scope;
857 		nh->nh_oif = FIB_RES_OIF(res);
858 		nh->nh_dev = dev = FIB_RES_DEV(res);
859 		if (!dev) {
860 			NL_SET_ERR_MSG(extack,
861 				       "No egress device for nexthop gateway");
862 			goto out;
863 		}
864 		dev_hold(dev);
865 		if (!netif_carrier_ok(dev))
866 			nh->nh_flags |= RTNH_F_LINKDOWN;
867 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
868 	} else {
869 		struct in_device *in_dev;
870 
871 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
872 			NL_SET_ERR_MSG(extack,
873 				       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
874 			return -EINVAL;
875 		}
876 		rcu_read_lock();
877 		err = -ENODEV;
878 		in_dev = inetdev_by_index(net, nh->nh_oif);
879 		if (!in_dev)
880 			goto out;
881 		err = -ENETDOWN;
882 		if (!(in_dev->dev->flags & IFF_UP)) {
883 			NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
884 			goto out;
885 		}
886 		nh->nh_dev = in_dev->dev;
887 		dev_hold(nh->nh_dev);
888 		nh->nh_scope = RT_SCOPE_HOST;
889 		if (!netif_carrier_ok(nh->nh_dev))
890 			nh->nh_flags |= RTNH_F_LINKDOWN;
891 		err = 0;
892 	}
893 out:
894 	rcu_read_unlock();
895 	return err;
896 }
897 
898 static inline unsigned int fib_laddr_hashfn(__be32 val)
899 {
900 	unsigned int mask = (fib_info_hash_size - 1);
901 
902 	return ((__force u32)val ^
903 		((__force u32)val >> 7) ^
904 		((__force u32)val >> 14)) & mask;
905 }
906 
907 static struct hlist_head *fib_info_hash_alloc(int bytes)
908 {
909 	if (bytes <= PAGE_SIZE)
910 		return kzalloc(bytes, GFP_KERNEL);
911 	else
912 		return (struct hlist_head *)
913 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
914 					 get_order(bytes));
915 }
916 
917 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
918 {
919 	if (!hash)
920 		return;
921 
922 	if (bytes <= PAGE_SIZE)
923 		kfree(hash);
924 	else
925 		free_pages((unsigned long) hash, get_order(bytes));
926 }
927 
928 static void fib_info_hash_move(struct hlist_head *new_info_hash,
929 			       struct hlist_head *new_laddrhash,
930 			       unsigned int new_size)
931 {
932 	struct hlist_head *old_info_hash, *old_laddrhash;
933 	unsigned int old_size = fib_info_hash_size;
934 	unsigned int i, bytes;
935 
936 	spin_lock_bh(&fib_info_lock);
937 	old_info_hash = fib_info_hash;
938 	old_laddrhash = fib_info_laddrhash;
939 	fib_info_hash_size = new_size;
940 
941 	for (i = 0; i < old_size; i++) {
942 		struct hlist_head *head = &fib_info_hash[i];
943 		struct hlist_node *n;
944 		struct fib_info *fi;
945 
946 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
947 			struct hlist_head *dest;
948 			unsigned int new_hash;
949 
950 			new_hash = fib_info_hashfn(fi);
951 			dest = &new_info_hash[new_hash];
952 			hlist_add_head(&fi->fib_hash, dest);
953 		}
954 	}
955 	fib_info_hash = new_info_hash;
956 
957 	for (i = 0; i < old_size; i++) {
958 		struct hlist_head *lhead = &fib_info_laddrhash[i];
959 		struct hlist_node *n;
960 		struct fib_info *fi;
961 
962 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
963 			struct hlist_head *ldest;
964 			unsigned int new_hash;
965 
966 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
967 			ldest = &new_laddrhash[new_hash];
968 			hlist_add_head(&fi->fib_lhash, ldest);
969 		}
970 	}
971 	fib_info_laddrhash = new_laddrhash;
972 
973 	spin_unlock_bh(&fib_info_lock);
974 
975 	bytes = old_size * sizeof(struct hlist_head *);
976 	fib_info_hash_free(old_info_hash, bytes);
977 	fib_info_hash_free(old_laddrhash, bytes);
978 }
979 
980 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
981 {
982 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
983 					nh->nh_gw,
984 					nh->nh_parent->fib_scope);
985 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
986 
987 	return nh->nh_saddr;
988 }
989 
990 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
991 {
992 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
993 	    fib_prefsrc != cfg->fc_dst) {
994 		u32 tb_id = cfg->fc_table;
995 		int rc;
996 
997 		if (tb_id == RT_TABLE_MAIN)
998 			tb_id = RT_TABLE_LOCAL;
999 
1000 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1001 					  fib_prefsrc, tb_id);
1002 
1003 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1004 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1005 						  fib_prefsrc, RT_TABLE_LOCAL);
1006 		}
1007 
1008 		if (rc != RTN_LOCAL)
1009 			return false;
1010 	}
1011 	return true;
1012 }
1013 
1014 static int
1015 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
1016 {
1017 	bool ecn_ca = false;
1018 	struct nlattr *nla;
1019 	int remaining;
1020 
1021 	if (!cfg->fc_mx)
1022 		return 0;
1023 
1024 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1025 		int type = nla_type(nla);
1026 		u32 val;
1027 
1028 		if (!type)
1029 			continue;
1030 		if (type > RTAX_MAX)
1031 			return -EINVAL;
1032 
1033 		if (type == RTAX_CC_ALGO) {
1034 			char tmp[TCP_CA_NAME_MAX];
1035 
1036 			nla_strlcpy(tmp, nla, sizeof(tmp));
1037 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1038 			if (val == TCP_CA_UNSPEC)
1039 				return -EINVAL;
1040 		} else {
1041 			val = nla_get_u32(nla);
1042 		}
1043 		if (type == RTAX_ADVMSS && val > 65535 - 40)
1044 			val = 65535 - 40;
1045 		if (type == RTAX_MTU && val > 65535 - 15)
1046 			val = 65535 - 15;
1047 		if (type == RTAX_HOPLIMIT && val > 255)
1048 			val = 255;
1049 		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
1050 			return -EINVAL;
1051 		fi->fib_metrics->metrics[type - 1] = val;
1052 	}
1053 
1054 	if (ecn_ca)
1055 		fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
1056 
1057 	return 0;
1058 }
1059 
1060 struct fib_info *fib_create_info(struct fib_config *cfg,
1061 				 struct netlink_ext_ack *extack)
1062 {
1063 	int err;
1064 	struct fib_info *fi = NULL;
1065 	struct fib_info *ofi;
1066 	int nhs = 1;
1067 	struct net *net = cfg->fc_nlinfo.nl_net;
1068 
1069 	if (cfg->fc_type > RTN_MAX)
1070 		goto err_inval;
1071 
1072 	/* Fast check to catch the most weird cases */
1073 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1074 		NL_SET_ERR_MSG(extack, "Invalid scope");
1075 		goto err_inval;
1076 	}
1077 
1078 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1079 		NL_SET_ERR_MSG(extack,
1080 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1081 		goto err_inval;
1082 	}
1083 
1084 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1085 	if (cfg->fc_mp) {
1086 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1087 		if (nhs == 0)
1088 			goto err_inval;
1089 	}
1090 #endif
1091 
1092 	err = -ENOBUFS;
1093 	if (fib_info_cnt >= fib_info_hash_size) {
1094 		unsigned int new_size = fib_info_hash_size << 1;
1095 		struct hlist_head *new_info_hash;
1096 		struct hlist_head *new_laddrhash;
1097 		unsigned int bytes;
1098 
1099 		if (!new_size)
1100 			new_size = 16;
1101 		bytes = new_size * sizeof(struct hlist_head *);
1102 		new_info_hash = fib_info_hash_alloc(bytes);
1103 		new_laddrhash = fib_info_hash_alloc(bytes);
1104 		if (!new_info_hash || !new_laddrhash) {
1105 			fib_info_hash_free(new_info_hash, bytes);
1106 			fib_info_hash_free(new_laddrhash, bytes);
1107 		} else
1108 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1109 
1110 		if (!fib_info_hash_size)
1111 			goto failure;
1112 	}
1113 
1114 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1115 	if (!fi)
1116 		goto failure;
1117 	if (cfg->fc_mx) {
1118 		fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1119 		if (unlikely(!fi->fib_metrics)) {
1120 			kfree(fi);
1121 			return ERR_PTR(err);
1122 		}
1123 		refcount_set(&fi->fib_metrics->refcnt, 1);
1124 	} else {
1125 		fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1126 	}
1127 	fib_info_cnt++;
1128 	fi->fib_net = net;
1129 	fi->fib_protocol = cfg->fc_protocol;
1130 	fi->fib_scope = cfg->fc_scope;
1131 	fi->fib_flags = cfg->fc_flags;
1132 	fi->fib_priority = cfg->fc_priority;
1133 	fi->fib_prefsrc = cfg->fc_prefsrc;
1134 	fi->fib_type = cfg->fc_type;
1135 	fi->fib_tb_id = cfg->fc_table;
1136 
1137 	fi->fib_nhs = nhs;
1138 	change_nexthops(fi) {
1139 		nexthop_nh->nh_parent = fi;
1140 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1141 		if (!nexthop_nh->nh_pcpu_rth_output)
1142 			goto failure;
1143 	} endfor_nexthops(fi)
1144 
1145 	err = fib_convert_metrics(fi, cfg);
1146 	if (err)
1147 		goto failure;
1148 
1149 	if (cfg->fc_mp) {
1150 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1151 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1152 		if (err != 0)
1153 			goto failure;
1154 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1155 			NL_SET_ERR_MSG(extack,
1156 				       "Nexthop device index does not match RTA_OIF");
1157 			goto err_inval;
1158 		}
1159 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1160 			NL_SET_ERR_MSG(extack,
1161 				       "Nexthop gateway does not match RTA_GATEWAY");
1162 			goto err_inval;
1163 		}
1164 #ifdef CONFIG_IP_ROUTE_CLASSID
1165 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1166 			NL_SET_ERR_MSG(extack,
1167 				       "Nexthop class id does not match RTA_FLOW");
1168 			goto err_inval;
1169 		}
1170 #endif
1171 #else
1172 		NL_SET_ERR_MSG(extack,
1173 			       "Multipath support not enabled in kernel");
1174 		goto err_inval;
1175 #endif
1176 	} else {
1177 		struct fib_nh *nh = fi->fib_nh;
1178 
1179 		if (cfg->fc_encap) {
1180 			struct lwtunnel_state *lwtstate;
1181 
1182 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1183 				NL_SET_ERR_MSG(extack,
1184 					       "LWT encap type not specified");
1185 				goto err_inval;
1186 			}
1187 			err = lwtunnel_build_state(cfg->fc_encap_type,
1188 						   cfg->fc_encap, AF_INET, cfg,
1189 						   &lwtstate, extack);
1190 			if (err)
1191 				goto failure;
1192 
1193 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1194 		}
1195 		nh->nh_oif = cfg->fc_oif;
1196 		nh->nh_gw = cfg->fc_gw;
1197 		nh->nh_flags = cfg->fc_flags;
1198 #ifdef CONFIG_IP_ROUTE_CLASSID
1199 		nh->nh_tclassid = cfg->fc_flow;
1200 		if (nh->nh_tclassid)
1201 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1202 #endif
1203 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1204 		nh->nh_weight = 1;
1205 #endif
1206 	}
1207 
1208 	if (fib_props[cfg->fc_type].error) {
1209 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1210 			NL_SET_ERR_MSG(extack,
1211 				       "Gateway, device and multipath can not be specified for this route type");
1212 			goto err_inval;
1213 		}
1214 		goto link_it;
1215 	} else {
1216 		switch (cfg->fc_type) {
1217 		case RTN_UNICAST:
1218 		case RTN_LOCAL:
1219 		case RTN_BROADCAST:
1220 		case RTN_ANYCAST:
1221 		case RTN_MULTICAST:
1222 			break;
1223 		default:
1224 			NL_SET_ERR_MSG(extack, "Invalid route type");
1225 			goto err_inval;
1226 		}
1227 	}
1228 
1229 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1230 		NL_SET_ERR_MSG(extack, "Invalid scope");
1231 		goto err_inval;
1232 	}
1233 
1234 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1235 		struct fib_nh *nh = fi->fib_nh;
1236 
1237 		/* Local address is added. */
1238 		if (nhs != 1) {
1239 			NL_SET_ERR_MSG(extack,
1240 				       "Route with host scope can not have multiple nexthops");
1241 			goto err_inval;
1242 		}
1243 		if (nh->nh_gw) {
1244 			NL_SET_ERR_MSG(extack,
1245 				       "Route with host scope can not have a gateway");
1246 			goto err_inval;
1247 		}
1248 		nh->nh_scope = RT_SCOPE_NOWHERE;
1249 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1250 		err = -ENODEV;
1251 		if (!nh->nh_dev)
1252 			goto failure;
1253 	} else {
1254 		int linkdown = 0;
1255 
1256 		change_nexthops(fi) {
1257 			err = fib_check_nh(cfg, nexthop_nh, extack);
1258 			if (err != 0)
1259 				goto failure;
1260 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1261 				linkdown++;
1262 		} endfor_nexthops(fi)
1263 		if (linkdown == fi->fib_nhs)
1264 			fi->fib_flags |= RTNH_F_LINKDOWN;
1265 	}
1266 
1267 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1268 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1269 		goto err_inval;
1270 	}
1271 
1272 	change_nexthops(fi) {
1273 		fib_info_update_nh_saddr(net, nexthop_nh);
1274 	} endfor_nexthops(fi)
1275 
1276 	fib_rebalance(fi);
1277 
1278 link_it:
1279 	ofi = fib_find_info(fi);
1280 	if (ofi) {
1281 		fi->fib_dead = 1;
1282 		free_fib_info(fi);
1283 		ofi->fib_treeref++;
1284 		return ofi;
1285 	}
1286 
1287 	fi->fib_treeref++;
1288 	refcount_set(&fi->fib_clntref, 1);
1289 	spin_lock_bh(&fib_info_lock);
1290 	hlist_add_head(&fi->fib_hash,
1291 		       &fib_info_hash[fib_info_hashfn(fi)]);
1292 	if (fi->fib_prefsrc) {
1293 		struct hlist_head *head;
1294 
1295 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1296 		hlist_add_head(&fi->fib_lhash, head);
1297 	}
1298 	change_nexthops(fi) {
1299 		struct hlist_head *head;
1300 		unsigned int hash;
1301 
1302 		if (!nexthop_nh->nh_dev)
1303 			continue;
1304 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1305 		head = &fib_info_devhash[hash];
1306 		hlist_add_head(&nexthop_nh->nh_hash, head);
1307 	} endfor_nexthops(fi)
1308 	spin_unlock_bh(&fib_info_lock);
1309 	return fi;
1310 
1311 err_inval:
1312 	err = -EINVAL;
1313 
1314 failure:
1315 	if (fi) {
1316 		fi->fib_dead = 1;
1317 		free_fib_info(fi);
1318 	}
1319 
1320 	return ERR_PTR(err);
1321 }
1322 
1323 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1324 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1325 		  struct fib_info *fi, unsigned int flags)
1326 {
1327 	struct nlmsghdr *nlh;
1328 	struct rtmsg *rtm;
1329 
1330 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1331 	if (!nlh)
1332 		return -EMSGSIZE;
1333 
1334 	rtm = nlmsg_data(nlh);
1335 	rtm->rtm_family = AF_INET;
1336 	rtm->rtm_dst_len = dst_len;
1337 	rtm->rtm_src_len = 0;
1338 	rtm->rtm_tos = tos;
1339 	if (tb_id < 256)
1340 		rtm->rtm_table = tb_id;
1341 	else
1342 		rtm->rtm_table = RT_TABLE_COMPAT;
1343 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1344 		goto nla_put_failure;
1345 	rtm->rtm_type = type;
1346 	rtm->rtm_flags = fi->fib_flags;
1347 	rtm->rtm_scope = fi->fib_scope;
1348 	rtm->rtm_protocol = fi->fib_protocol;
1349 
1350 	if (rtm->rtm_dst_len &&
1351 	    nla_put_in_addr(skb, RTA_DST, dst))
1352 		goto nla_put_failure;
1353 	if (fi->fib_priority &&
1354 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1355 		goto nla_put_failure;
1356 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1357 		goto nla_put_failure;
1358 
1359 	if (fi->fib_prefsrc &&
1360 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1361 		goto nla_put_failure;
1362 	if (fi->fib_nhs == 1) {
1363 		if (fi->fib_nh->nh_gw &&
1364 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1365 			goto nla_put_failure;
1366 		if (fi->fib_nh->nh_oif &&
1367 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1368 			goto nla_put_failure;
1369 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1370 			struct in_device *in_dev;
1371 
1372 			rcu_read_lock();
1373 			in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev);
1374 			if (in_dev &&
1375 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1376 				rtm->rtm_flags |= RTNH_F_DEAD;
1377 			rcu_read_unlock();
1378 		}
1379 		if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1380 			rtm->rtm_flags |= RTNH_F_OFFLOAD;
1381 #ifdef CONFIG_IP_ROUTE_CLASSID
1382 		if (fi->fib_nh[0].nh_tclassid &&
1383 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1384 			goto nla_put_failure;
1385 #endif
1386 		if (fi->fib_nh->nh_lwtstate &&
1387 		    lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1388 			goto nla_put_failure;
1389 	}
1390 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1391 	if (fi->fib_nhs > 1) {
1392 		struct rtnexthop *rtnh;
1393 		struct nlattr *mp;
1394 
1395 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1396 		if (!mp)
1397 			goto nla_put_failure;
1398 
1399 		for_nexthops(fi) {
1400 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1401 			if (!rtnh)
1402 				goto nla_put_failure;
1403 
1404 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1405 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1406 				struct in_device *in_dev;
1407 
1408 				rcu_read_lock();
1409 				in_dev = __in_dev_get_rcu(nh->nh_dev);
1410 				if (in_dev &&
1411 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1412 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1413 				rcu_read_unlock();
1414 			}
1415 			rtnh->rtnh_hops = nh->nh_weight - 1;
1416 			rtnh->rtnh_ifindex = nh->nh_oif;
1417 
1418 			if (nh->nh_gw &&
1419 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1420 				goto nla_put_failure;
1421 #ifdef CONFIG_IP_ROUTE_CLASSID
1422 			if (nh->nh_tclassid &&
1423 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1424 				goto nla_put_failure;
1425 #endif
1426 			if (nh->nh_lwtstate &&
1427 			    lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1428 				goto nla_put_failure;
1429 
1430 			/* length of rtnetlink header + attributes */
1431 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1432 		} endfor_nexthops(fi);
1433 
1434 		nla_nest_end(skb, mp);
1435 	}
1436 #endif
1437 	nlmsg_end(skb, nlh);
1438 	return 0;
1439 
1440 nla_put_failure:
1441 	nlmsg_cancel(skb, nlh);
1442 	return -EMSGSIZE;
1443 }
1444 
1445 /*
1446  * Update FIB if:
1447  * - local address disappeared -> we must delete all the entries
1448  *   referring to it.
1449  * - device went down -> we must shutdown all nexthops going via it.
1450  */
1451 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1452 {
1453 	int ret = 0;
1454 	unsigned int hash = fib_laddr_hashfn(local);
1455 	struct hlist_head *head = &fib_info_laddrhash[hash];
1456 	struct net *net = dev_net(dev);
1457 	int tb_id = l3mdev_fib_table(dev);
1458 	struct fib_info *fi;
1459 
1460 	if (!fib_info_laddrhash || local == 0)
1461 		return 0;
1462 
1463 	hlist_for_each_entry(fi, head, fib_lhash) {
1464 		if (!net_eq(fi->fib_net, net) ||
1465 		    fi->fib_tb_id != tb_id)
1466 			continue;
1467 		if (fi->fib_prefsrc == local) {
1468 			fi->fib_flags |= RTNH_F_DEAD;
1469 			ret++;
1470 		}
1471 	}
1472 	return ret;
1473 }
1474 
1475 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1476 				 enum fib_event_type event_type)
1477 {
1478 	struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1479 	struct fib_nh_notifier_info info = {
1480 		.fib_nh = fib_nh,
1481 	};
1482 
1483 	switch (event_type) {
1484 	case FIB_EVENT_NH_ADD:
1485 		if (fib_nh->nh_flags & RTNH_F_DEAD)
1486 			break;
1487 		if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1488 		    fib_nh->nh_flags & RTNH_F_LINKDOWN)
1489 			break;
1490 		return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1491 					   &info.info);
1492 	case FIB_EVENT_NH_DEL:
1493 		if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1494 		     fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1495 		    (fib_nh->nh_flags & RTNH_F_DEAD))
1496 			return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1497 						   event_type, &info.info);
1498 	default:
1499 		break;
1500 	}
1501 
1502 	return NOTIFY_DONE;
1503 }
1504 
1505 /* Event              force Flags           Description
1506  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1507  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1508  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1509  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1510  */
1511 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1512 {
1513 	int ret = 0;
1514 	int scope = RT_SCOPE_NOWHERE;
1515 	struct fib_info *prev_fi = NULL;
1516 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1517 	struct hlist_head *head = &fib_info_devhash[hash];
1518 	struct fib_nh *nh;
1519 
1520 	if (force)
1521 		scope = -1;
1522 
1523 	hlist_for_each_entry(nh, head, nh_hash) {
1524 		struct fib_info *fi = nh->nh_parent;
1525 		int dead;
1526 
1527 		BUG_ON(!fi->fib_nhs);
1528 		if (nh->nh_dev != dev || fi == prev_fi)
1529 			continue;
1530 		prev_fi = fi;
1531 		dead = 0;
1532 		change_nexthops(fi) {
1533 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1534 				dead++;
1535 			else if (nexthop_nh->nh_dev == dev &&
1536 				 nexthop_nh->nh_scope != scope) {
1537 				switch (event) {
1538 				case NETDEV_DOWN:
1539 				case NETDEV_UNREGISTER:
1540 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1541 					/* fall through */
1542 				case NETDEV_CHANGE:
1543 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1544 					break;
1545 				}
1546 				call_fib_nh_notifiers(nexthop_nh,
1547 						      FIB_EVENT_NH_DEL);
1548 				dead++;
1549 			}
1550 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1551 			if (event == NETDEV_UNREGISTER &&
1552 			    nexthop_nh->nh_dev == dev) {
1553 				dead = fi->fib_nhs;
1554 				break;
1555 			}
1556 #endif
1557 		} endfor_nexthops(fi)
1558 		if (dead == fi->fib_nhs) {
1559 			switch (event) {
1560 			case NETDEV_DOWN:
1561 			case NETDEV_UNREGISTER:
1562 				fi->fib_flags |= RTNH_F_DEAD;
1563 				/* fall through */
1564 			case NETDEV_CHANGE:
1565 				fi->fib_flags |= RTNH_F_LINKDOWN;
1566 				break;
1567 			}
1568 			ret++;
1569 		}
1570 
1571 		fib_rebalance(fi);
1572 	}
1573 
1574 	return ret;
1575 }
1576 
1577 /* Must be invoked inside of an RCU protected region.  */
1578 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1579 {
1580 	struct fib_info *fi = NULL, *last_resort = NULL;
1581 	struct hlist_head *fa_head = res->fa_head;
1582 	struct fib_table *tb = res->table;
1583 	u8 slen = 32 - res->prefixlen;
1584 	int order = -1, last_idx = -1;
1585 	struct fib_alias *fa, *fa1 = NULL;
1586 	u32 last_prio = res->fi->fib_priority;
1587 	u8 last_tos = 0;
1588 
1589 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1590 		struct fib_info *next_fi = fa->fa_info;
1591 
1592 		if (fa->fa_slen != slen)
1593 			continue;
1594 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1595 			continue;
1596 		if (fa->tb_id != tb->tb_id)
1597 			continue;
1598 		if (next_fi->fib_priority > last_prio &&
1599 		    fa->fa_tos == last_tos) {
1600 			if (last_tos)
1601 				continue;
1602 			break;
1603 		}
1604 		if (next_fi->fib_flags & RTNH_F_DEAD)
1605 			continue;
1606 		last_tos = fa->fa_tos;
1607 		last_prio = next_fi->fib_priority;
1608 
1609 		if (next_fi->fib_scope != res->scope ||
1610 		    fa->fa_type != RTN_UNICAST)
1611 			continue;
1612 		if (!next_fi->fib_nh[0].nh_gw ||
1613 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1614 			continue;
1615 
1616 		fib_alias_accessed(fa);
1617 
1618 		if (!fi) {
1619 			if (next_fi != res->fi)
1620 				break;
1621 			fa1 = fa;
1622 		} else if (!fib_detect_death(fi, order, &last_resort,
1623 					     &last_idx, fa1->fa_default)) {
1624 			fib_result_assign(res, fi);
1625 			fa1->fa_default = order;
1626 			goto out;
1627 		}
1628 		fi = next_fi;
1629 		order++;
1630 	}
1631 
1632 	if (order <= 0 || !fi) {
1633 		if (fa1)
1634 			fa1->fa_default = -1;
1635 		goto out;
1636 	}
1637 
1638 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1639 			      fa1->fa_default)) {
1640 		fib_result_assign(res, fi);
1641 		fa1->fa_default = order;
1642 		goto out;
1643 	}
1644 
1645 	if (last_idx >= 0)
1646 		fib_result_assign(res, last_resort);
1647 	fa1->fa_default = last_idx;
1648 out:
1649 	return;
1650 }
1651 
1652 /*
1653  * Dead device goes up. We wake up dead nexthops.
1654  * It takes sense only on multipath routes.
1655  */
1656 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1657 {
1658 	struct fib_info *prev_fi;
1659 	unsigned int hash;
1660 	struct hlist_head *head;
1661 	struct fib_nh *nh;
1662 	int ret;
1663 
1664 	if (!(dev->flags & IFF_UP))
1665 		return 0;
1666 
1667 	if (nh_flags & RTNH_F_DEAD) {
1668 		unsigned int flags = dev_get_flags(dev);
1669 
1670 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1671 			nh_flags |= RTNH_F_LINKDOWN;
1672 	}
1673 
1674 	prev_fi = NULL;
1675 	hash = fib_devindex_hashfn(dev->ifindex);
1676 	head = &fib_info_devhash[hash];
1677 	ret = 0;
1678 
1679 	hlist_for_each_entry(nh, head, nh_hash) {
1680 		struct fib_info *fi = nh->nh_parent;
1681 		int alive;
1682 
1683 		BUG_ON(!fi->fib_nhs);
1684 		if (nh->nh_dev != dev || fi == prev_fi)
1685 			continue;
1686 
1687 		prev_fi = fi;
1688 		alive = 0;
1689 		change_nexthops(fi) {
1690 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1691 				alive++;
1692 				continue;
1693 			}
1694 			if (!nexthop_nh->nh_dev ||
1695 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1696 				continue;
1697 			if (nexthop_nh->nh_dev != dev ||
1698 			    !__in_dev_get_rtnl(dev))
1699 				continue;
1700 			alive++;
1701 			nexthop_nh->nh_flags &= ~nh_flags;
1702 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1703 		} endfor_nexthops(fi)
1704 
1705 		if (alive > 0) {
1706 			fi->fib_flags &= ~nh_flags;
1707 			ret++;
1708 		}
1709 
1710 		fib_rebalance(fi);
1711 	}
1712 
1713 	return ret;
1714 }
1715 
1716 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1717 static bool fib_good_nh(const struct fib_nh *nh)
1718 {
1719 	int state = NUD_REACHABLE;
1720 
1721 	if (nh->nh_scope == RT_SCOPE_LINK) {
1722 		struct neighbour *n;
1723 
1724 		rcu_read_lock_bh();
1725 
1726 		n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1727 					      (__force u32)nh->nh_gw);
1728 		if (n)
1729 			state = n->nud_state;
1730 
1731 		rcu_read_unlock_bh();
1732 	}
1733 
1734 	return !!(state & NUD_VALID);
1735 }
1736 
1737 void fib_select_multipath(struct fib_result *res, int hash)
1738 {
1739 	struct fib_info *fi = res->fi;
1740 	struct net *net = fi->fib_net;
1741 	bool first = false;
1742 
1743 	for_nexthops(fi) {
1744 		if (hash > atomic_read(&nh->nh_upper_bound))
1745 			continue;
1746 
1747 		if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1748 		    fib_good_nh(nh)) {
1749 			res->nh_sel = nhsel;
1750 			return;
1751 		}
1752 		if (!first) {
1753 			res->nh_sel = nhsel;
1754 			first = true;
1755 		}
1756 	} endfor_nexthops(fi);
1757 }
1758 #endif
1759 
1760 void fib_select_path(struct net *net, struct fib_result *res,
1761 		     struct flowi4 *fl4, const struct sk_buff *skb)
1762 {
1763 	bool oif_check;
1764 
1765 	oif_check = (fl4->flowi4_oif == 0 ||
1766 		     fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1767 
1768 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1769 	if (res->fi->fib_nhs > 1 && oif_check) {
1770 		int h = fib_multipath_hash(res->fi, fl4, skb);
1771 
1772 		fib_select_multipath(res, h);
1773 	}
1774 	else
1775 #endif
1776 	if (!res->prefixlen &&
1777 	    res->table->tb_num_default > 1 &&
1778 	    res->type == RTN_UNICAST && oif_check)
1779 		fib_select_default(fl4, res);
1780 
1781 	if (!fl4->saddr)
1782 		fl4->saddr = FIB_RES_PREFSRC(net, *res);
1783 }
1784 EXPORT_SYMBOL_GPL(fib_select_path);
1785