xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 664b9c4b7392ce723b013201843264bf95481ce5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: semantics.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/uaccess.h>
13 #include <linux/bitops.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
17 #include <linux/mm.h>
18 #include <linux/string.h>
19 #include <linux/socket.h>
20 #include <linux/sockios.h>
21 #include <linux/errno.h>
22 #include <linux/in.h>
23 #include <linux/inet.h>
24 #include <linux/inetdevice.h>
25 #include <linux/netdevice.h>
26 #include <linux/if_arp.h>
27 #include <linux/proc_fs.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/netlink.h>
32 
33 #include <net/arp.h>
34 #include <net/ip.h>
35 #include <net/protocol.h>
36 #include <net/route.h>
37 #include <net/tcp.h>
38 #include <net/sock.h>
39 #include <net/ip_fib.h>
40 #include <net/ip6_fib.h>
41 #include <net/nexthop.h>
42 #include <net/netlink.h>
43 #include <net/rtnh.h>
44 #include <net/lwtunnel.h>
45 #include <net/fib_notifier.h>
46 #include <net/addrconf.h>
47 
48 #include "fib_lookup.h"
49 
50 static DEFINE_SPINLOCK(fib_info_lock);
51 static struct hlist_head *fib_info_hash;
52 static struct hlist_head *fib_info_laddrhash;
53 static unsigned int fib_info_hash_size;
54 static unsigned int fib_info_cnt;
55 
56 #define DEVINDEX_HASHBITS 8
57 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
58 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
59 
60 /* for_nexthops and change_nexthops only used when nexthop object
61  * is not set in a fib_info. The logic within can reference fib_nh.
62  */
63 #ifdef CONFIG_IP_ROUTE_MULTIPATH
64 
65 #define for_nexthops(fi) {						\
66 	int nhsel; const struct fib_nh *nh;				\
67 	for (nhsel = 0, nh = (fi)->fib_nh;				\
68 	     nhsel < fib_info_num_path((fi));				\
69 	     nh++, nhsel++)
70 
71 #define change_nexthops(fi) {						\
72 	int nhsel; struct fib_nh *nexthop_nh;				\
73 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
74 	     nhsel < fib_info_num_path((fi));				\
75 	     nexthop_nh++, nhsel++)
76 
77 #else /* CONFIG_IP_ROUTE_MULTIPATH */
78 
79 /* Hope, that gcc will optimize it to get rid of dummy loop */
80 
81 #define for_nexthops(fi) {						\
82 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
83 	for (nhsel = 0; nhsel < 1; nhsel++)
84 
85 #define change_nexthops(fi) {						\
86 	int nhsel;							\
87 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
88 	for (nhsel = 0; nhsel < 1; nhsel++)
89 
90 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
91 
92 #define endfor_nexthops(fi) }
93 
94 
95 const struct fib_prop fib_props[RTN_MAX + 1] = {
96 	[RTN_UNSPEC] = {
97 		.error	= 0,
98 		.scope	= RT_SCOPE_NOWHERE,
99 	},
100 	[RTN_UNICAST] = {
101 		.error	= 0,
102 		.scope	= RT_SCOPE_UNIVERSE,
103 	},
104 	[RTN_LOCAL] = {
105 		.error	= 0,
106 		.scope	= RT_SCOPE_HOST,
107 	},
108 	[RTN_BROADCAST] = {
109 		.error	= 0,
110 		.scope	= RT_SCOPE_LINK,
111 	},
112 	[RTN_ANYCAST] = {
113 		.error	= 0,
114 		.scope	= RT_SCOPE_LINK,
115 	},
116 	[RTN_MULTICAST] = {
117 		.error	= 0,
118 		.scope	= RT_SCOPE_UNIVERSE,
119 	},
120 	[RTN_BLACKHOLE] = {
121 		.error	= -EINVAL,
122 		.scope	= RT_SCOPE_UNIVERSE,
123 	},
124 	[RTN_UNREACHABLE] = {
125 		.error	= -EHOSTUNREACH,
126 		.scope	= RT_SCOPE_UNIVERSE,
127 	},
128 	[RTN_PROHIBIT] = {
129 		.error	= -EACCES,
130 		.scope	= RT_SCOPE_UNIVERSE,
131 	},
132 	[RTN_THROW] = {
133 		.error	= -EAGAIN,
134 		.scope	= RT_SCOPE_UNIVERSE,
135 	},
136 	[RTN_NAT] = {
137 		.error	= -EINVAL,
138 		.scope	= RT_SCOPE_NOWHERE,
139 	},
140 	[RTN_XRESOLVE] = {
141 		.error	= -EINVAL,
142 		.scope	= RT_SCOPE_NOWHERE,
143 	},
144 };
145 
146 static void rt_fibinfo_free(struct rtable __rcu **rtp)
147 {
148 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
149 
150 	if (!rt)
151 		return;
152 
153 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
154 	 * because we waited an RCU grace period before calling
155 	 * free_fib_info_rcu()
156 	 */
157 
158 	dst_dev_put(&rt->dst);
159 	dst_release_immediate(&rt->dst);
160 }
161 
162 static void free_nh_exceptions(struct fib_nh_common *nhc)
163 {
164 	struct fnhe_hash_bucket *hash;
165 	int i;
166 
167 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
168 	if (!hash)
169 		return;
170 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
171 		struct fib_nh_exception *fnhe;
172 
173 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
174 		while (fnhe) {
175 			struct fib_nh_exception *next;
176 
177 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
178 
179 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
180 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
181 
182 			kfree(fnhe);
183 
184 			fnhe = next;
185 		}
186 	}
187 	kfree(hash);
188 }
189 
190 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
191 {
192 	int cpu;
193 
194 	if (!rtp)
195 		return;
196 
197 	for_each_possible_cpu(cpu) {
198 		struct rtable *rt;
199 
200 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
201 		if (rt) {
202 			dst_dev_put(&rt->dst);
203 			dst_release_immediate(&rt->dst);
204 		}
205 	}
206 	free_percpu(rtp);
207 }
208 
209 void fib_nh_common_release(struct fib_nh_common *nhc)
210 {
211 	dev_put(nhc->nhc_dev);
212 	lwtstate_put(nhc->nhc_lwtstate);
213 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
214 	rt_fibinfo_free(&nhc->nhc_rth_input);
215 	free_nh_exceptions(nhc);
216 }
217 EXPORT_SYMBOL_GPL(fib_nh_common_release);
218 
219 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
220 {
221 #ifdef CONFIG_IP_ROUTE_CLASSID
222 	if (fib_nh->nh_tclassid)
223 		atomic_dec(&net->ipv4.fib_num_tclassid_users);
224 #endif
225 	fib_nh_common_release(&fib_nh->nh_common);
226 }
227 
228 /* Release a nexthop info record */
229 static void free_fib_info_rcu(struct rcu_head *head)
230 {
231 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
232 
233 	if (fi->nh) {
234 		nexthop_put(fi->nh);
235 	} else {
236 		change_nexthops(fi) {
237 			fib_nh_release(fi->fib_net, nexthop_nh);
238 		} endfor_nexthops(fi);
239 	}
240 
241 	ip_fib_metrics_put(fi->fib_metrics);
242 
243 	kfree(fi);
244 }
245 
246 void free_fib_info(struct fib_info *fi)
247 {
248 	if (fi->fib_dead == 0) {
249 		pr_warn("Freeing alive fib_info %p\n", fi);
250 		return;
251 	}
252 	fib_info_cnt--;
253 
254 	call_rcu(&fi->rcu, free_fib_info_rcu);
255 }
256 EXPORT_SYMBOL_GPL(free_fib_info);
257 
258 void fib_release_info(struct fib_info *fi)
259 {
260 	spin_lock_bh(&fib_info_lock);
261 	if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
262 		hlist_del(&fi->fib_hash);
263 		if (fi->fib_prefsrc)
264 			hlist_del(&fi->fib_lhash);
265 		if (fi->nh) {
266 			list_del(&fi->nh_list);
267 		} else {
268 			change_nexthops(fi) {
269 				if (!nexthop_nh->fib_nh_dev)
270 					continue;
271 				hlist_del(&nexthop_nh->nh_hash);
272 			} endfor_nexthops(fi)
273 		}
274 		fi->fib_dead = 1;
275 		fib_info_put(fi);
276 	}
277 	spin_unlock_bh(&fib_info_lock);
278 }
279 
280 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
281 {
282 	const struct fib_nh *onh;
283 
284 	if (fi->nh || ofi->nh)
285 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
286 
287 	if (ofi->fib_nhs == 0)
288 		return 0;
289 
290 	for_nexthops(fi) {
291 		onh = fib_info_nh(ofi, nhsel);
292 
293 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
294 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
295 		    nh->fib_nh_scope != onh->fib_nh_scope ||
296 #ifdef CONFIG_IP_ROUTE_MULTIPATH
297 		    nh->fib_nh_weight != onh->fib_nh_weight ||
298 #endif
299 #ifdef CONFIG_IP_ROUTE_CLASSID
300 		    nh->nh_tclassid != onh->nh_tclassid ||
301 #endif
302 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
303 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
304 			return -1;
305 
306 		if (nh->fib_nh_gw_family == AF_INET &&
307 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
308 			return -1;
309 
310 		if (nh->fib_nh_gw_family == AF_INET6 &&
311 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
312 			return -1;
313 	} endfor_nexthops(fi);
314 	return 0;
315 }
316 
317 static inline unsigned int fib_devindex_hashfn(unsigned int val)
318 {
319 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
320 
321 	return (val ^
322 		(val >> DEVINDEX_HASHBITS) ^
323 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
324 }
325 
326 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
327 				      u32 prefsrc, u32 priority)
328 {
329 	unsigned int val = init_val;
330 
331 	val ^= (protocol << 8) | scope;
332 	val ^= prefsrc;
333 	val ^= priority;
334 
335 	return val;
336 }
337 
338 static unsigned int fib_info_hashfn_result(unsigned int val)
339 {
340 	unsigned int mask = (fib_info_hash_size - 1);
341 
342 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
343 }
344 
345 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
346 {
347 	unsigned int val;
348 
349 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
350 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
351 				fi->fib_priority);
352 
353 	if (fi->nh) {
354 		val ^= fib_devindex_hashfn(fi->nh->id);
355 	} else {
356 		for_nexthops(fi) {
357 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
358 		} endfor_nexthops(fi)
359 	}
360 
361 	return fib_info_hashfn_result(val);
362 }
363 
364 /* no metrics, only nexthop id */
365 static struct fib_info *fib_find_info_nh(struct net *net,
366 					 const struct fib_config *cfg)
367 {
368 	struct hlist_head *head;
369 	struct fib_info *fi;
370 	unsigned int hash;
371 
372 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
373 				 cfg->fc_protocol, cfg->fc_scope,
374 				 (__force u32)cfg->fc_prefsrc,
375 				 cfg->fc_priority);
376 	hash = fib_info_hashfn_result(hash);
377 	head = &fib_info_hash[hash];
378 
379 	hlist_for_each_entry(fi, head, fib_hash) {
380 		if (!net_eq(fi->fib_net, net))
381 			continue;
382 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
383 			continue;
384 		if (cfg->fc_protocol == fi->fib_protocol &&
385 		    cfg->fc_scope == fi->fib_scope &&
386 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
387 		    cfg->fc_priority == fi->fib_priority &&
388 		    cfg->fc_type == fi->fib_type &&
389 		    cfg->fc_table == fi->fib_tb_id &&
390 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
391 			return fi;
392 	}
393 
394 	return NULL;
395 }
396 
397 static struct fib_info *fib_find_info(struct fib_info *nfi)
398 {
399 	struct hlist_head *head;
400 	struct fib_info *fi;
401 	unsigned int hash;
402 
403 	hash = fib_info_hashfn(nfi);
404 	head = &fib_info_hash[hash];
405 
406 	hlist_for_each_entry(fi, head, fib_hash) {
407 		if (!net_eq(fi->fib_net, nfi->fib_net))
408 			continue;
409 		if (fi->fib_nhs != nfi->fib_nhs)
410 			continue;
411 		if (nfi->fib_protocol == fi->fib_protocol &&
412 		    nfi->fib_scope == fi->fib_scope &&
413 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
414 		    nfi->fib_priority == fi->fib_priority &&
415 		    nfi->fib_type == fi->fib_type &&
416 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
417 			   sizeof(u32) * RTAX_MAX) == 0 &&
418 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
419 		    nh_comp(fi, nfi) == 0)
420 			return fi;
421 	}
422 
423 	return NULL;
424 }
425 
426 /* Check, that the gateway is already configured.
427  * Used only by redirect accept routine.
428  */
429 int ip_fib_check_default(__be32 gw, struct net_device *dev)
430 {
431 	struct hlist_head *head;
432 	struct fib_nh *nh;
433 	unsigned int hash;
434 
435 	spin_lock(&fib_info_lock);
436 
437 	hash = fib_devindex_hashfn(dev->ifindex);
438 	head = &fib_info_devhash[hash];
439 	hlist_for_each_entry(nh, head, nh_hash) {
440 		if (nh->fib_nh_dev == dev &&
441 		    nh->fib_nh_gw4 == gw &&
442 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
443 			spin_unlock(&fib_info_lock);
444 			return 0;
445 		}
446 	}
447 
448 	spin_unlock(&fib_info_lock);
449 
450 	return -1;
451 }
452 
453 size_t fib_nlmsg_size(struct fib_info *fi)
454 {
455 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
456 			 + nla_total_size(4) /* RTA_TABLE */
457 			 + nla_total_size(4) /* RTA_DST */
458 			 + nla_total_size(4) /* RTA_PRIORITY */
459 			 + nla_total_size(4) /* RTA_PREFSRC */
460 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
461 	unsigned int nhs = fib_info_num_path(fi);
462 
463 	/* space for nested metrics */
464 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
465 
466 	if (fi->nh)
467 		payload += nla_total_size(4); /* RTA_NH_ID */
468 
469 	if (nhs) {
470 		size_t nh_encapsize = 0;
471 		/* Also handles the special case nhs == 1 */
472 
473 		/* each nexthop is packed in an attribute */
474 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
475 		unsigned int i;
476 
477 		/* may contain flow and gateway attribute */
478 		nhsize += 2 * nla_total_size(4);
479 
480 		/* grab encap info */
481 		for (i = 0; i < fib_info_num_path(fi); i++) {
482 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
483 
484 			if (nhc->nhc_lwtstate) {
485 				/* RTA_ENCAP_TYPE */
486 				nh_encapsize += lwtunnel_get_encap_size(
487 						nhc->nhc_lwtstate);
488 				/* RTA_ENCAP */
489 				nh_encapsize +=  nla_total_size(2);
490 			}
491 		}
492 
493 		/* all nexthops are packed in a nested attribute */
494 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
495 
496 	}
497 
498 	return payload;
499 }
500 
501 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
502 	       int dst_len, u32 tb_id, const struct nl_info *info,
503 	       unsigned int nlm_flags)
504 {
505 	struct fib_rt_info fri;
506 	struct sk_buff *skb;
507 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
508 	int err = -ENOBUFS;
509 
510 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
511 	if (!skb)
512 		goto errout;
513 
514 	fri.fi = fa->fa_info;
515 	fri.tb_id = tb_id;
516 	fri.dst = key;
517 	fri.dst_len = dst_len;
518 	fri.tos = fa->fa_tos;
519 	fri.type = fa->fa_type;
520 	fri.offload = fa->offload;
521 	fri.trap = fa->trap;
522 	fri.offload_failed = fa->offload_failed;
523 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
524 	if (err < 0) {
525 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
526 		WARN_ON(err == -EMSGSIZE);
527 		kfree_skb(skb);
528 		goto errout;
529 	}
530 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
531 		    info->nlh, GFP_KERNEL);
532 	return;
533 errout:
534 	if (err < 0)
535 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
536 }
537 
538 static int fib_detect_death(struct fib_info *fi, int order,
539 			    struct fib_info **last_resort, int *last_idx,
540 			    int dflt)
541 {
542 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
543 	struct neighbour *n;
544 	int state = NUD_NONE;
545 
546 	if (likely(nhc->nhc_gw_family == AF_INET))
547 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
548 	else if (nhc->nhc_gw_family == AF_INET6)
549 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
550 				 nhc->nhc_dev);
551 	else
552 		n = NULL;
553 
554 	if (n) {
555 		state = n->nud_state;
556 		neigh_release(n);
557 	} else {
558 		return 0;
559 	}
560 	if (state == NUD_REACHABLE)
561 		return 0;
562 	if ((state & NUD_VALID) && order != dflt)
563 		return 0;
564 	if ((state & NUD_VALID) ||
565 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
566 		*last_resort = fi;
567 		*last_idx = order;
568 	}
569 	return 1;
570 }
571 
572 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
573 		       struct nlattr *encap, u16 encap_type,
574 		       void *cfg, gfp_t gfp_flags,
575 		       struct netlink_ext_ack *extack)
576 {
577 	int err;
578 
579 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
580 						    gfp_flags);
581 	if (!nhc->nhc_pcpu_rth_output)
582 		return -ENOMEM;
583 
584 	if (encap) {
585 		struct lwtunnel_state *lwtstate;
586 
587 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
588 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
589 			err = -EINVAL;
590 			goto lwt_failure;
591 		}
592 		err = lwtunnel_build_state(net, encap_type, encap,
593 					   nhc->nhc_family, cfg, &lwtstate,
594 					   extack);
595 		if (err)
596 			goto lwt_failure;
597 
598 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
599 	}
600 
601 	return 0;
602 
603 lwt_failure:
604 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
605 	nhc->nhc_pcpu_rth_output = NULL;
606 	return err;
607 }
608 EXPORT_SYMBOL_GPL(fib_nh_common_init);
609 
610 int fib_nh_init(struct net *net, struct fib_nh *nh,
611 		struct fib_config *cfg, int nh_weight,
612 		struct netlink_ext_ack *extack)
613 {
614 	int err;
615 
616 	nh->fib_nh_family = AF_INET;
617 
618 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
619 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
620 	if (err)
621 		return err;
622 
623 	nh->fib_nh_oif = cfg->fc_oif;
624 	nh->fib_nh_gw_family = cfg->fc_gw_family;
625 	if (cfg->fc_gw_family == AF_INET)
626 		nh->fib_nh_gw4 = cfg->fc_gw4;
627 	else if (cfg->fc_gw_family == AF_INET6)
628 		nh->fib_nh_gw6 = cfg->fc_gw6;
629 
630 	nh->fib_nh_flags = cfg->fc_flags;
631 
632 #ifdef CONFIG_IP_ROUTE_CLASSID
633 	nh->nh_tclassid = cfg->fc_flow;
634 	if (nh->nh_tclassid)
635 		atomic_inc(&net->ipv4.fib_num_tclassid_users);
636 #endif
637 #ifdef CONFIG_IP_ROUTE_MULTIPATH
638 	nh->fib_nh_weight = nh_weight;
639 #endif
640 	return 0;
641 }
642 
643 #ifdef CONFIG_IP_ROUTE_MULTIPATH
644 
645 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
646 			      struct netlink_ext_ack *extack)
647 {
648 	int nhs = 0;
649 
650 	while (rtnh_ok(rtnh, remaining)) {
651 		nhs++;
652 		rtnh = rtnh_next(rtnh, &remaining);
653 	}
654 
655 	/* leftover implies invalid nexthop configuration, discard it */
656 	if (remaining > 0) {
657 		NL_SET_ERR_MSG(extack,
658 			       "Invalid nexthop configuration - extra data after nexthops");
659 		nhs = 0;
660 	}
661 
662 	return nhs;
663 }
664 
665 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
666 			    struct netlink_ext_ack *extack)
667 {
668 	if (nla_len(nla) < sizeof(*gw)) {
669 		NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
670 		return -EINVAL;
671 	}
672 
673 	*gw = nla_get_in_addr(nla);
674 
675 	return 0;
676 }
677 
678 /* only called when fib_nh is integrated into fib_info */
679 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
680 		       int remaining, struct fib_config *cfg,
681 		       struct netlink_ext_ack *extack)
682 {
683 	struct net *net = fi->fib_net;
684 	struct fib_config fib_cfg;
685 	struct fib_nh *nh;
686 	int ret;
687 
688 	change_nexthops(fi) {
689 		int attrlen;
690 
691 		memset(&fib_cfg, 0, sizeof(fib_cfg));
692 
693 		if (!rtnh_ok(rtnh, remaining)) {
694 			NL_SET_ERR_MSG(extack,
695 				       "Invalid nexthop configuration - extra data after nexthop");
696 			return -EINVAL;
697 		}
698 
699 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
700 			NL_SET_ERR_MSG(extack,
701 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
702 			return -EINVAL;
703 		}
704 
705 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
706 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
707 
708 		attrlen = rtnh_attrlen(rtnh);
709 		if (attrlen > 0) {
710 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
711 
712 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
713 			nlav = nla_find(attrs, attrlen, RTA_VIA);
714 			if (nla && nlav) {
715 				NL_SET_ERR_MSG(extack,
716 					       "Nexthop configuration can not contain both GATEWAY and VIA");
717 				return -EINVAL;
718 			}
719 			if (nla) {
720 				ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
721 						       extack);
722 				if (ret)
723 					goto errout;
724 
725 				if (fib_cfg.fc_gw4)
726 					fib_cfg.fc_gw_family = AF_INET;
727 			} else if (nlav) {
728 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
729 				if (ret)
730 					goto errout;
731 			}
732 
733 			nla = nla_find(attrs, attrlen, RTA_FLOW);
734 			if (nla) {
735 				if (nla_len(nla) < sizeof(u32)) {
736 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
737 					return -EINVAL;
738 				}
739 				fib_cfg.fc_flow = nla_get_u32(nla);
740 			}
741 
742 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
743 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
744 			if (nla)
745 				fib_cfg.fc_encap_type = nla_get_u16(nla);
746 		}
747 
748 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
749 				  rtnh->rtnh_hops + 1, extack);
750 		if (ret)
751 			goto errout;
752 
753 		rtnh = rtnh_next(rtnh, &remaining);
754 	} endfor_nexthops(fi);
755 
756 	ret = -EINVAL;
757 	nh = fib_info_nh(fi, 0);
758 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
759 		NL_SET_ERR_MSG(extack,
760 			       "Nexthop device index does not match RTA_OIF");
761 		goto errout;
762 	}
763 	if (cfg->fc_gw_family) {
764 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
765 		    (cfg->fc_gw_family == AF_INET &&
766 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
767 		    (cfg->fc_gw_family == AF_INET6 &&
768 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
769 			NL_SET_ERR_MSG(extack,
770 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
771 			goto errout;
772 		}
773 	}
774 #ifdef CONFIG_IP_ROUTE_CLASSID
775 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
776 		NL_SET_ERR_MSG(extack,
777 			       "Nexthop class id does not match RTA_FLOW");
778 		goto errout;
779 	}
780 #endif
781 	ret = 0;
782 errout:
783 	return ret;
784 }
785 
786 /* only called when fib_nh is integrated into fib_info */
787 static void fib_rebalance(struct fib_info *fi)
788 {
789 	int total;
790 	int w;
791 
792 	if (fib_info_num_path(fi) < 2)
793 		return;
794 
795 	total = 0;
796 	for_nexthops(fi) {
797 		if (nh->fib_nh_flags & RTNH_F_DEAD)
798 			continue;
799 
800 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
801 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
802 			continue;
803 
804 		total += nh->fib_nh_weight;
805 	} endfor_nexthops(fi);
806 
807 	w = 0;
808 	change_nexthops(fi) {
809 		int upper_bound;
810 
811 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
812 			upper_bound = -1;
813 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
814 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
815 			upper_bound = -1;
816 		} else {
817 			w += nexthop_nh->fib_nh_weight;
818 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
819 							    total) - 1;
820 		}
821 
822 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
823 	} endfor_nexthops(fi);
824 }
825 #else /* CONFIG_IP_ROUTE_MULTIPATH */
826 
827 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
828 		       int remaining, struct fib_config *cfg,
829 		       struct netlink_ext_ack *extack)
830 {
831 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
832 
833 	return -EINVAL;
834 }
835 
836 #define fib_rebalance(fi) do { } while (0)
837 
838 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
839 
840 static int fib_encap_match(struct net *net, u16 encap_type,
841 			   struct nlattr *encap,
842 			   const struct fib_nh *nh,
843 			   const struct fib_config *cfg,
844 			   struct netlink_ext_ack *extack)
845 {
846 	struct lwtunnel_state *lwtstate;
847 	int ret, result = 0;
848 
849 	if (encap_type == LWTUNNEL_ENCAP_NONE)
850 		return 0;
851 
852 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
853 				   cfg, &lwtstate, extack);
854 	if (!ret) {
855 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
856 		lwtstate_free(lwtstate);
857 	}
858 
859 	return result;
860 }
861 
862 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
863 		 struct netlink_ext_ack *extack)
864 {
865 #ifdef CONFIG_IP_ROUTE_MULTIPATH
866 	struct rtnexthop *rtnh;
867 	int remaining;
868 #endif
869 
870 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
871 		return 1;
872 
873 	if (cfg->fc_nh_id) {
874 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
875 			return 0;
876 		return 1;
877 	}
878 
879 	if (cfg->fc_oif || cfg->fc_gw_family) {
880 		struct fib_nh *nh = fib_info_nh(fi, 0);
881 
882 		if (cfg->fc_encap) {
883 			if (fib_encap_match(net, cfg->fc_encap_type,
884 					    cfg->fc_encap, nh, cfg, extack))
885 				return 1;
886 		}
887 #ifdef CONFIG_IP_ROUTE_CLASSID
888 		if (cfg->fc_flow &&
889 		    cfg->fc_flow != nh->nh_tclassid)
890 			return 1;
891 #endif
892 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
893 		    (cfg->fc_gw_family &&
894 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
895 			return 1;
896 
897 		if (cfg->fc_gw_family == AF_INET &&
898 		    cfg->fc_gw4 != nh->fib_nh_gw4)
899 			return 1;
900 
901 		if (cfg->fc_gw_family == AF_INET6 &&
902 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
903 			return 1;
904 
905 		return 0;
906 	}
907 
908 #ifdef CONFIG_IP_ROUTE_MULTIPATH
909 	if (!cfg->fc_mp)
910 		return 0;
911 
912 	rtnh = cfg->fc_mp;
913 	remaining = cfg->fc_mp_len;
914 
915 	for_nexthops(fi) {
916 		int attrlen;
917 
918 		if (!rtnh_ok(rtnh, remaining))
919 			return -EINVAL;
920 
921 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
922 			return 1;
923 
924 		attrlen = rtnh_attrlen(rtnh);
925 		if (attrlen > 0) {
926 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
927 			int err;
928 
929 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
930 			nlav = nla_find(attrs, attrlen, RTA_VIA);
931 			if (nla && nlav) {
932 				NL_SET_ERR_MSG(extack,
933 					       "Nexthop configuration can not contain both GATEWAY and VIA");
934 				return -EINVAL;
935 			}
936 
937 			if (nla) {
938 				__be32 gw;
939 
940 				err = fib_gw_from_attr(&gw, nla, extack);
941 				if (err)
942 					return err;
943 
944 				if (nh->fib_nh_gw_family != AF_INET ||
945 				    gw != nh->fib_nh_gw4)
946 					return 1;
947 			} else if (nlav) {
948 				struct fib_config cfg2;
949 
950 				err = fib_gw_from_via(&cfg2, nlav, extack);
951 				if (err)
952 					return err;
953 
954 				switch (nh->fib_nh_gw_family) {
955 				case AF_INET:
956 					if (cfg2.fc_gw_family != AF_INET ||
957 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
958 						return 1;
959 					break;
960 				case AF_INET6:
961 					if (cfg2.fc_gw_family != AF_INET6 ||
962 					    ipv6_addr_cmp(&cfg2.fc_gw6,
963 							  &nh->fib_nh_gw6))
964 						return 1;
965 					break;
966 				}
967 			}
968 
969 #ifdef CONFIG_IP_ROUTE_CLASSID
970 			nla = nla_find(attrs, attrlen, RTA_FLOW);
971 			if (nla) {
972 				if (nla_len(nla) < sizeof(u32)) {
973 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
974 					return -EINVAL;
975 				}
976 				if (nla_get_u32(nla) != nh->nh_tclassid)
977 					return 1;
978 			}
979 #endif
980 		}
981 
982 		rtnh = rtnh_next(rtnh, &remaining);
983 	} endfor_nexthops(fi);
984 #endif
985 	return 0;
986 }
987 
988 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
989 {
990 	struct nlattr *nla;
991 	int remaining;
992 
993 	if (!cfg->fc_mx)
994 		return true;
995 
996 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
997 		int type = nla_type(nla);
998 		u32 fi_val, val;
999 
1000 		if (!type)
1001 			continue;
1002 		if (type > RTAX_MAX)
1003 			return false;
1004 
1005 		if (type == RTAX_CC_ALGO) {
1006 			char tmp[TCP_CA_NAME_MAX];
1007 			bool ecn_ca = false;
1008 
1009 			nla_strscpy(tmp, nla, sizeof(tmp));
1010 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1011 		} else {
1012 			if (nla_len(nla) != sizeof(u32))
1013 				return false;
1014 			val = nla_get_u32(nla);
1015 		}
1016 
1017 		fi_val = fi->fib_metrics->metrics[type - 1];
1018 		if (type == RTAX_FEATURES)
1019 			fi_val &= ~DST_FEATURE_ECN_CA;
1020 
1021 		if (fi_val != val)
1022 			return false;
1023 	}
1024 
1025 	return true;
1026 }
1027 
1028 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1029 			      u32 table, struct netlink_ext_ack *extack)
1030 {
1031 	struct fib6_config cfg = {
1032 		.fc_table = table,
1033 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1034 		.fc_ifindex = nh->fib_nh_oif,
1035 		.fc_gateway = nh->fib_nh_gw6,
1036 	};
1037 	struct fib6_nh fib6_nh = {};
1038 	int err;
1039 
1040 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1041 	if (!err) {
1042 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1043 		dev_hold(nh->fib_nh_dev);
1044 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1045 		nh->fib_nh_scope = RT_SCOPE_LINK;
1046 
1047 		ipv6_stub->fib6_nh_release(&fib6_nh);
1048 	}
1049 
1050 	return err;
1051 }
1052 
1053 /*
1054  * Picture
1055  * -------
1056  *
1057  * Semantics of nexthop is very messy by historical reasons.
1058  * We have to take into account, that:
1059  * a) gateway can be actually local interface address,
1060  *    so that gatewayed route is direct.
1061  * b) gateway must be on-link address, possibly
1062  *    described not by an ifaddr, but also by a direct route.
1063  * c) If both gateway and interface are specified, they should not
1064  *    contradict.
1065  * d) If we use tunnel routes, gateway could be not on-link.
1066  *
1067  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1068  * results in pretty ugly and hairy code with obscure logic.
1069  *
1070  * I chose to generalized it instead, so that the size
1071  * of code does not increase practically, but it becomes
1072  * much more general.
1073  * Every prefix is assigned a "scope" value: "host" is local address,
1074  * "link" is direct route,
1075  * [ ... "site" ... "interior" ... ]
1076  * and "universe" is true gateway route with global meaning.
1077  *
1078  * Every prefix refers to a set of "nexthop"s (gw, oif),
1079  * where gw must have narrower scope. This recursion stops
1080  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1081  * which means that gw is forced to be on link.
1082  *
1083  * Code is still hairy, but now it is apparently logically
1084  * consistent and very flexible. F.e. as by-product it allows
1085  * to co-exists in peace independent exterior and interior
1086  * routing processes.
1087  *
1088  * Normally it looks as following.
1089  *
1090  * {universe prefix}  -> (gw, oif) [scope link]
1091  *		  |
1092  *		  |-> {link prefix} -> (gw, oif) [scope local]
1093  *					|
1094  *					|-> {local prefix} (terminal node)
1095  */
1096 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1097 			      u8 scope, struct netlink_ext_ack *extack)
1098 {
1099 	struct net_device *dev;
1100 	struct fib_result res;
1101 	int err = 0;
1102 
1103 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1104 		unsigned int addr_type;
1105 
1106 		if (scope >= RT_SCOPE_LINK) {
1107 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1108 			return -EINVAL;
1109 		}
1110 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1111 		if (!dev) {
1112 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1113 			return -ENODEV;
1114 		}
1115 		if (!(dev->flags & IFF_UP)) {
1116 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1117 			return -ENETDOWN;
1118 		}
1119 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1120 		if (addr_type != RTN_UNICAST) {
1121 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1122 			return -EINVAL;
1123 		}
1124 		if (!netif_carrier_ok(dev))
1125 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1126 		nh->fib_nh_dev = dev;
1127 		dev_hold(dev);
1128 		nh->fib_nh_scope = RT_SCOPE_LINK;
1129 		return 0;
1130 	}
1131 	rcu_read_lock();
1132 	{
1133 		struct fib_table *tbl = NULL;
1134 		struct flowi4 fl4 = {
1135 			.daddr = nh->fib_nh_gw4,
1136 			.flowi4_scope = scope + 1,
1137 			.flowi4_oif = nh->fib_nh_oif,
1138 			.flowi4_iif = LOOPBACK_IFINDEX,
1139 		};
1140 
1141 		/* It is not necessary, but requires a bit of thinking */
1142 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1143 			fl4.flowi4_scope = RT_SCOPE_LINK;
1144 
1145 		if (table && table != RT_TABLE_MAIN)
1146 			tbl = fib_get_table(net, table);
1147 
1148 		if (tbl)
1149 			err = fib_table_lookup(tbl, &fl4, &res,
1150 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1151 					       FIB_LOOKUP_NOREF);
1152 
1153 		/* on error or if no table given do full lookup. This
1154 		 * is needed for example when nexthops are in the local
1155 		 * table rather than the given table
1156 		 */
1157 		if (!tbl || err) {
1158 			err = fib_lookup(net, &fl4, &res,
1159 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1160 		}
1161 
1162 		if (err) {
1163 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1164 			goto out;
1165 		}
1166 	}
1167 
1168 	err = -EINVAL;
1169 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1170 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1171 		goto out;
1172 	}
1173 	nh->fib_nh_scope = res.scope;
1174 	nh->fib_nh_oif = FIB_RES_OIF(res);
1175 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1176 	if (!dev) {
1177 		NL_SET_ERR_MSG(extack,
1178 			       "No egress device for nexthop gateway");
1179 		goto out;
1180 	}
1181 	dev_hold(dev);
1182 	if (!netif_carrier_ok(dev))
1183 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1184 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1185 out:
1186 	rcu_read_unlock();
1187 	return err;
1188 }
1189 
1190 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1191 			      struct netlink_ext_ack *extack)
1192 {
1193 	struct in_device *in_dev;
1194 	int err;
1195 
1196 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1197 		NL_SET_ERR_MSG(extack,
1198 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1199 		return -EINVAL;
1200 	}
1201 
1202 	rcu_read_lock();
1203 
1204 	err = -ENODEV;
1205 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1206 	if (!in_dev)
1207 		goto out;
1208 	err = -ENETDOWN;
1209 	if (!(in_dev->dev->flags & IFF_UP)) {
1210 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1211 		goto out;
1212 	}
1213 
1214 	nh->fib_nh_dev = in_dev->dev;
1215 	dev_hold(nh->fib_nh_dev);
1216 	nh->fib_nh_scope = RT_SCOPE_HOST;
1217 	if (!netif_carrier_ok(nh->fib_nh_dev))
1218 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1219 	err = 0;
1220 out:
1221 	rcu_read_unlock();
1222 	return err;
1223 }
1224 
1225 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1226 		 struct netlink_ext_ack *extack)
1227 {
1228 	int err;
1229 
1230 	if (nh->fib_nh_gw_family == AF_INET)
1231 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1232 	else if (nh->fib_nh_gw_family == AF_INET6)
1233 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1234 	else
1235 		err = fib_check_nh_nongw(net, nh, extack);
1236 
1237 	return err;
1238 }
1239 
1240 static inline unsigned int fib_laddr_hashfn(__be32 val)
1241 {
1242 	unsigned int mask = (fib_info_hash_size - 1);
1243 
1244 	return ((__force u32)val ^
1245 		((__force u32)val >> 7) ^
1246 		((__force u32)val >> 14)) & mask;
1247 }
1248 
1249 static struct hlist_head *fib_info_hash_alloc(int bytes)
1250 {
1251 	if (bytes <= PAGE_SIZE)
1252 		return kzalloc(bytes, GFP_KERNEL);
1253 	else
1254 		return (struct hlist_head *)
1255 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1256 					 get_order(bytes));
1257 }
1258 
1259 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1260 {
1261 	if (!hash)
1262 		return;
1263 
1264 	if (bytes <= PAGE_SIZE)
1265 		kfree(hash);
1266 	else
1267 		free_pages((unsigned long) hash, get_order(bytes));
1268 }
1269 
1270 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1271 			       struct hlist_head *new_laddrhash,
1272 			       unsigned int new_size)
1273 {
1274 	struct hlist_head *old_info_hash, *old_laddrhash;
1275 	unsigned int old_size = fib_info_hash_size;
1276 	unsigned int i, bytes;
1277 
1278 	spin_lock_bh(&fib_info_lock);
1279 	old_info_hash = fib_info_hash;
1280 	old_laddrhash = fib_info_laddrhash;
1281 	fib_info_hash_size = new_size;
1282 
1283 	for (i = 0; i < old_size; i++) {
1284 		struct hlist_head *head = &fib_info_hash[i];
1285 		struct hlist_node *n;
1286 		struct fib_info *fi;
1287 
1288 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1289 			struct hlist_head *dest;
1290 			unsigned int new_hash;
1291 
1292 			new_hash = fib_info_hashfn(fi);
1293 			dest = &new_info_hash[new_hash];
1294 			hlist_add_head(&fi->fib_hash, dest);
1295 		}
1296 	}
1297 	fib_info_hash = new_info_hash;
1298 
1299 	for (i = 0; i < old_size; i++) {
1300 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1301 		struct hlist_node *n;
1302 		struct fib_info *fi;
1303 
1304 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1305 			struct hlist_head *ldest;
1306 			unsigned int new_hash;
1307 
1308 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1309 			ldest = &new_laddrhash[new_hash];
1310 			hlist_add_head(&fi->fib_lhash, ldest);
1311 		}
1312 	}
1313 	fib_info_laddrhash = new_laddrhash;
1314 
1315 	spin_unlock_bh(&fib_info_lock);
1316 
1317 	bytes = old_size * sizeof(struct hlist_head *);
1318 	fib_info_hash_free(old_info_hash, bytes);
1319 	fib_info_hash_free(old_laddrhash, bytes);
1320 }
1321 
1322 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1323 				 unsigned char scope)
1324 {
1325 	struct fib_nh *nh;
1326 
1327 	if (nhc->nhc_family != AF_INET)
1328 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1329 
1330 	nh = container_of(nhc, struct fib_nh, nh_common);
1331 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1332 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1333 
1334 	return nh->nh_saddr;
1335 }
1336 
1337 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1338 {
1339 	struct fib_nh_common *nhc = res->nhc;
1340 
1341 	if (res->fi->fib_prefsrc)
1342 		return res->fi->fib_prefsrc;
1343 
1344 	if (nhc->nhc_family == AF_INET) {
1345 		struct fib_nh *nh;
1346 
1347 		nh = container_of(nhc, struct fib_nh, nh_common);
1348 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1349 			return nh->nh_saddr;
1350 	}
1351 
1352 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1353 }
1354 
1355 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1356 {
1357 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1358 	    fib_prefsrc != cfg->fc_dst) {
1359 		u32 tb_id = cfg->fc_table;
1360 		int rc;
1361 
1362 		if (tb_id == RT_TABLE_MAIN)
1363 			tb_id = RT_TABLE_LOCAL;
1364 
1365 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1366 					  fib_prefsrc, tb_id);
1367 
1368 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1369 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1370 						  fib_prefsrc, RT_TABLE_LOCAL);
1371 		}
1372 
1373 		if (rc != RTN_LOCAL)
1374 			return false;
1375 	}
1376 	return true;
1377 }
1378 
1379 struct fib_info *fib_create_info(struct fib_config *cfg,
1380 				 struct netlink_ext_ack *extack)
1381 {
1382 	int err;
1383 	struct fib_info *fi = NULL;
1384 	struct nexthop *nh = NULL;
1385 	struct fib_info *ofi;
1386 	int nhs = 1;
1387 	struct net *net = cfg->fc_nlinfo.nl_net;
1388 
1389 	if (cfg->fc_type > RTN_MAX)
1390 		goto err_inval;
1391 
1392 	/* Fast check to catch the most weird cases */
1393 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1394 		NL_SET_ERR_MSG(extack, "Invalid scope");
1395 		goto err_inval;
1396 	}
1397 
1398 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1399 		NL_SET_ERR_MSG(extack,
1400 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1401 		goto err_inval;
1402 	}
1403 
1404 	if (cfg->fc_nh_id) {
1405 		if (!cfg->fc_mx) {
1406 			fi = fib_find_info_nh(net, cfg);
1407 			if (fi) {
1408 				refcount_inc(&fi->fib_treeref);
1409 				return fi;
1410 			}
1411 		}
1412 
1413 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1414 		if (!nh) {
1415 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1416 			goto err_inval;
1417 		}
1418 		nhs = 0;
1419 	}
1420 
1421 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1422 	if (cfg->fc_mp) {
1423 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1424 		if (nhs == 0)
1425 			goto err_inval;
1426 	}
1427 #endif
1428 
1429 	err = -ENOBUFS;
1430 	if (fib_info_cnt >= fib_info_hash_size) {
1431 		unsigned int new_size = fib_info_hash_size << 1;
1432 		struct hlist_head *new_info_hash;
1433 		struct hlist_head *new_laddrhash;
1434 		unsigned int bytes;
1435 
1436 		if (!new_size)
1437 			new_size = 16;
1438 		bytes = new_size * sizeof(struct hlist_head *);
1439 		new_info_hash = fib_info_hash_alloc(bytes);
1440 		new_laddrhash = fib_info_hash_alloc(bytes);
1441 		if (!new_info_hash || !new_laddrhash) {
1442 			fib_info_hash_free(new_info_hash, bytes);
1443 			fib_info_hash_free(new_laddrhash, bytes);
1444 		} else
1445 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1446 
1447 		if (!fib_info_hash_size)
1448 			goto failure;
1449 	}
1450 
1451 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1452 	if (!fi)
1453 		goto failure;
1454 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1455 					      cfg->fc_mx_len, extack);
1456 	if (IS_ERR(fi->fib_metrics)) {
1457 		err = PTR_ERR(fi->fib_metrics);
1458 		kfree(fi);
1459 		return ERR_PTR(err);
1460 	}
1461 
1462 	fib_info_cnt++;
1463 	fi->fib_net = net;
1464 	fi->fib_protocol = cfg->fc_protocol;
1465 	fi->fib_scope = cfg->fc_scope;
1466 	fi->fib_flags = cfg->fc_flags;
1467 	fi->fib_priority = cfg->fc_priority;
1468 	fi->fib_prefsrc = cfg->fc_prefsrc;
1469 	fi->fib_type = cfg->fc_type;
1470 	fi->fib_tb_id = cfg->fc_table;
1471 
1472 	fi->fib_nhs = nhs;
1473 	if (nh) {
1474 		if (!nexthop_get(nh)) {
1475 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1476 			err = -EINVAL;
1477 		} else {
1478 			err = 0;
1479 			fi->nh = nh;
1480 		}
1481 	} else {
1482 		change_nexthops(fi) {
1483 			nexthop_nh->nh_parent = fi;
1484 		} endfor_nexthops(fi)
1485 
1486 		if (cfg->fc_mp)
1487 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1488 					  extack);
1489 		else
1490 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1491 	}
1492 
1493 	if (err != 0)
1494 		goto failure;
1495 
1496 	if (fib_props[cfg->fc_type].error) {
1497 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1498 			NL_SET_ERR_MSG(extack,
1499 				       "Gateway, device and multipath can not be specified for this route type");
1500 			goto err_inval;
1501 		}
1502 		goto link_it;
1503 	} else {
1504 		switch (cfg->fc_type) {
1505 		case RTN_UNICAST:
1506 		case RTN_LOCAL:
1507 		case RTN_BROADCAST:
1508 		case RTN_ANYCAST:
1509 		case RTN_MULTICAST:
1510 			break;
1511 		default:
1512 			NL_SET_ERR_MSG(extack, "Invalid route type");
1513 			goto err_inval;
1514 		}
1515 	}
1516 
1517 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1518 		NL_SET_ERR_MSG(extack, "Invalid scope");
1519 		goto err_inval;
1520 	}
1521 
1522 	if (fi->nh) {
1523 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1524 		if (err)
1525 			goto failure;
1526 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1527 		struct fib_nh *nh = fi->fib_nh;
1528 
1529 		/* Local address is added. */
1530 		if (nhs != 1) {
1531 			NL_SET_ERR_MSG(extack,
1532 				       "Route with host scope can not have multiple nexthops");
1533 			goto err_inval;
1534 		}
1535 		if (nh->fib_nh_gw_family) {
1536 			NL_SET_ERR_MSG(extack,
1537 				       "Route with host scope can not have a gateway");
1538 			goto err_inval;
1539 		}
1540 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1541 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1542 		err = -ENODEV;
1543 		if (!nh->fib_nh_dev)
1544 			goto failure;
1545 	} else {
1546 		int linkdown = 0;
1547 
1548 		change_nexthops(fi) {
1549 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1550 					   cfg->fc_table, cfg->fc_scope,
1551 					   extack);
1552 			if (err != 0)
1553 				goto failure;
1554 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1555 				linkdown++;
1556 		} endfor_nexthops(fi)
1557 		if (linkdown == fi->fib_nhs)
1558 			fi->fib_flags |= RTNH_F_LINKDOWN;
1559 	}
1560 
1561 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1562 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1563 		goto err_inval;
1564 	}
1565 
1566 	if (!fi->nh) {
1567 		change_nexthops(fi) {
1568 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1569 						  fi->fib_scope);
1570 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1571 				fi->fib_nh_is_v6 = true;
1572 		} endfor_nexthops(fi)
1573 
1574 		fib_rebalance(fi);
1575 	}
1576 
1577 link_it:
1578 	ofi = fib_find_info(fi);
1579 	if (ofi) {
1580 		fi->fib_dead = 1;
1581 		free_fib_info(fi);
1582 		refcount_inc(&ofi->fib_treeref);
1583 		return ofi;
1584 	}
1585 
1586 	refcount_set(&fi->fib_treeref, 1);
1587 	refcount_set(&fi->fib_clntref, 1);
1588 	spin_lock_bh(&fib_info_lock);
1589 	hlist_add_head(&fi->fib_hash,
1590 		       &fib_info_hash[fib_info_hashfn(fi)]);
1591 	if (fi->fib_prefsrc) {
1592 		struct hlist_head *head;
1593 
1594 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1595 		hlist_add_head(&fi->fib_lhash, head);
1596 	}
1597 	if (fi->nh) {
1598 		list_add(&fi->nh_list, &nh->fi_list);
1599 	} else {
1600 		change_nexthops(fi) {
1601 			struct hlist_head *head;
1602 			unsigned int hash;
1603 
1604 			if (!nexthop_nh->fib_nh_dev)
1605 				continue;
1606 			hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1607 			head = &fib_info_devhash[hash];
1608 			hlist_add_head(&nexthop_nh->nh_hash, head);
1609 		} endfor_nexthops(fi)
1610 	}
1611 	spin_unlock_bh(&fib_info_lock);
1612 	return fi;
1613 
1614 err_inval:
1615 	err = -EINVAL;
1616 
1617 failure:
1618 	if (fi) {
1619 		fi->fib_dead = 1;
1620 		free_fib_info(fi);
1621 	}
1622 
1623 	return ERR_PTR(err);
1624 }
1625 
1626 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1627 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1628 {
1629 	if (nhc->nhc_flags & RTNH_F_DEAD)
1630 		*flags |= RTNH_F_DEAD;
1631 
1632 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1633 		*flags |= RTNH_F_LINKDOWN;
1634 
1635 		rcu_read_lock();
1636 		switch (nhc->nhc_family) {
1637 		case AF_INET:
1638 			if (ip_ignore_linkdown(nhc->nhc_dev))
1639 				*flags |= RTNH_F_DEAD;
1640 			break;
1641 		case AF_INET6:
1642 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1643 				*flags |= RTNH_F_DEAD;
1644 			break;
1645 		}
1646 		rcu_read_unlock();
1647 	}
1648 
1649 	switch (nhc->nhc_gw_family) {
1650 	case AF_INET:
1651 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1652 			goto nla_put_failure;
1653 		break;
1654 	case AF_INET6:
1655 		/* if gateway family does not match nexthop family
1656 		 * gateway is encoded as RTA_VIA
1657 		 */
1658 		if (rt_family != nhc->nhc_gw_family) {
1659 			int alen = sizeof(struct in6_addr);
1660 			struct nlattr *nla;
1661 			struct rtvia *via;
1662 
1663 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1664 			if (!nla)
1665 				goto nla_put_failure;
1666 
1667 			via = nla_data(nla);
1668 			via->rtvia_family = AF_INET6;
1669 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1670 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1671 					    &nhc->nhc_gw.ipv6) < 0) {
1672 			goto nla_put_failure;
1673 		}
1674 		break;
1675 	}
1676 
1677 	*flags |= (nhc->nhc_flags &
1678 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1679 
1680 	if (!skip_oif && nhc->nhc_dev &&
1681 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1682 		goto nla_put_failure;
1683 
1684 	if (nhc->nhc_lwtstate &&
1685 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1686 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1687 		goto nla_put_failure;
1688 
1689 	return 0;
1690 
1691 nla_put_failure:
1692 	return -EMSGSIZE;
1693 }
1694 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1695 
1696 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1697 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1698 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1699 {
1700 	const struct net_device *dev = nhc->nhc_dev;
1701 	struct rtnexthop *rtnh;
1702 	unsigned char flags = 0;
1703 
1704 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1705 	if (!rtnh)
1706 		goto nla_put_failure;
1707 
1708 	rtnh->rtnh_hops = nh_weight - 1;
1709 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1710 
1711 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1712 		goto nla_put_failure;
1713 
1714 	rtnh->rtnh_flags = flags;
1715 
1716 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1717 		goto nla_put_failure;
1718 
1719 	/* length of rtnetlink header + attributes */
1720 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1721 
1722 	return 0;
1723 
1724 nla_put_failure:
1725 	return -EMSGSIZE;
1726 }
1727 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1728 #endif
1729 
1730 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1731 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1732 {
1733 	struct nlattr *mp;
1734 
1735 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1736 	if (!mp)
1737 		goto nla_put_failure;
1738 
1739 	if (unlikely(fi->nh)) {
1740 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1741 			goto nla_put_failure;
1742 		goto mp_end;
1743 	}
1744 
1745 	for_nexthops(fi) {
1746 		u32 nh_tclassid = 0;
1747 #ifdef CONFIG_IP_ROUTE_CLASSID
1748 		nh_tclassid = nh->nh_tclassid;
1749 #endif
1750 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1751 				    AF_INET, nh_tclassid) < 0)
1752 			goto nla_put_failure;
1753 	} endfor_nexthops(fi);
1754 
1755 mp_end:
1756 	nla_nest_end(skb, mp);
1757 
1758 	return 0;
1759 
1760 nla_put_failure:
1761 	return -EMSGSIZE;
1762 }
1763 #else
1764 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1765 {
1766 	return 0;
1767 }
1768 #endif
1769 
1770 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1771 		  const struct fib_rt_info *fri, unsigned int flags)
1772 {
1773 	unsigned int nhs = fib_info_num_path(fri->fi);
1774 	struct fib_info *fi = fri->fi;
1775 	u32 tb_id = fri->tb_id;
1776 	struct nlmsghdr *nlh;
1777 	struct rtmsg *rtm;
1778 
1779 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1780 	if (!nlh)
1781 		return -EMSGSIZE;
1782 
1783 	rtm = nlmsg_data(nlh);
1784 	rtm->rtm_family = AF_INET;
1785 	rtm->rtm_dst_len = fri->dst_len;
1786 	rtm->rtm_src_len = 0;
1787 	rtm->rtm_tos = fri->tos;
1788 	if (tb_id < 256)
1789 		rtm->rtm_table = tb_id;
1790 	else
1791 		rtm->rtm_table = RT_TABLE_COMPAT;
1792 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1793 		goto nla_put_failure;
1794 	rtm->rtm_type = fri->type;
1795 	rtm->rtm_flags = fi->fib_flags;
1796 	rtm->rtm_scope = fi->fib_scope;
1797 	rtm->rtm_protocol = fi->fib_protocol;
1798 
1799 	if (rtm->rtm_dst_len &&
1800 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1801 		goto nla_put_failure;
1802 	if (fi->fib_priority &&
1803 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1804 		goto nla_put_failure;
1805 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1806 		goto nla_put_failure;
1807 
1808 	if (fi->fib_prefsrc &&
1809 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1810 		goto nla_put_failure;
1811 
1812 	if (fi->nh) {
1813 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1814 			goto nla_put_failure;
1815 		if (nexthop_is_blackhole(fi->nh))
1816 			rtm->rtm_type = RTN_BLACKHOLE;
1817 		if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode)
1818 			goto offload;
1819 	}
1820 
1821 	if (nhs == 1) {
1822 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1823 		unsigned char flags = 0;
1824 
1825 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1826 			goto nla_put_failure;
1827 
1828 		rtm->rtm_flags = flags;
1829 #ifdef CONFIG_IP_ROUTE_CLASSID
1830 		if (nhc->nhc_family == AF_INET) {
1831 			struct fib_nh *nh;
1832 
1833 			nh = container_of(nhc, struct fib_nh, nh_common);
1834 			if (nh->nh_tclassid &&
1835 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1836 				goto nla_put_failure;
1837 		}
1838 #endif
1839 	} else {
1840 		if (fib_add_multipath(skb, fi) < 0)
1841 			goto nla_put_failure;
1842 	}
1843 
1844 offload:
1845 	if (fri->offload)
1846 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1847 	if (fri->trap)
1848 		rtm->rtm_flags |= RTM_F_TRAP;
1849 	if (fri->offload_failed)
1850 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1851 
1852 	nlmsg_end(skb, nlh);
1853 	return 0;
1854 
1855 nla_put_failure:
1856 	nlmsg_cancel(skb, nlh);
1857 	return -EMSGSIZE;
1858 }
1859 
1860 /*
1861  * Update FIB if:
1862  * - local address disappeared -> we must delete all the entries
1863  *   referring to it.
1864  * - device went down -> we must shutdown all nexthops going via it.
1865  */
1866 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1867 {
1868 	int ret = 0;
1869 	unsigned int hash = fib_laddr_hashfn(local);
1870 	struct hlist_head *head = &fib_info_laddrhash[hash];
1871 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1872 	struct net *net = dev_net(dev);
1873 	struct fib_info *fi;
1874 
1875 	if (!fib_info_laddrhash || local == 0)
1876 		return 0;
1877 
1878 	hlist_for_each_entry(fi, head, fib_lhash) {
1879 		if (!net_eq(fi->fib_net, net) ||
1880 		    fi->fib_tb_id != tb_id)
1881 			continue;
1882 		if (fi->fib_prefsrc == local) {
1883 			fi->fib_flags |= RTNH_F_DEAD;
1884 			ret++;
1885 		}
1886 	}
1887 	return ret;
1888 }
1889 
1890 static int call_fib_nh_notifiers(struct fib_nh *nh,
1891 				 enum fib_event_type event_type)
1892 {
1893 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1894 	struct fib_nh_notifier_info info = {
1895 		.fib_nh = nh,
1896 	};
1897 
1898 	switch (event_type) {
1899 	case FIB_EVENT_NH_ADD:
1900 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1901 			break;
1902 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1903 			break;
1904 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1905 					   &info.info);
1906 	case FIB_EVENT_NH_DEL:
1907 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1908 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1909 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1910 						   event_type, &info.info);
1911 		break;
1912 	default:
1913 		break;
1914 	}
1915 
1916 	return NOTIFY_DONE;
1917 }
1918 
1919 /* Update the PMTU of exceptions when:
1920  * - the new MTU of the first hop becomes smaller than the PMTU
1921  * - the old MTU was the same as the PMTU, and it limited discovery of
1922  *   larger MTUs on the path. With that limit raised, we can now
1923  *   discover larger MTUs
1924  * A special case is locked exceptions, for which the PMTU is smaller
1925  * than the minimal accepted PMTU:
1926  * - if the new MTU is greater than the PMTU, don't make any change
1927  * - otherwise, unlock and set PMTU
1928  */
1929 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1930 {
1931 	struct fnhe_hash_bucket *bucket;
1932 	int i;
1933 
1934 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1935 	if (!bucket)
1936 		return;
1937 
1938 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1939 		struct fib_nh_exception *fnhe;
1940 
1941 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1942 		     fnhe;
1943 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1944 			if (fnhe->fnhe_mtu_locked) {
1945 				if (new <= fnhe->fnhe_pmtu) {
1946 					fnhe->fnhe_pmtu = new;
1947 					fnhe->fnhe_mtu_locked = false;
1948 				}
1949 			} else if (new < fnhe->fnhe_pmtu ||
1950 				   orig == fnhe->fnhe_pmtu) {
1951 				fnhe->fnhe_pmtu = new;
1952 			}
1953 		}
1954 	}
1955 }
1956 
1957 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1958 {
1959 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1960 	struct hlist_head *head = &fib_info_devhash[hash];
1961 	struct fib_nh *nh;
1962 
1963 	hlist_for_each_entry(nh, head, nh_hash) {
1964 		if (nh->fib_nh_dev == dev)
1965 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1966 	}
1967 }
1968 
1969 /* Event              force Flags           Description
1970  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1971  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1972  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1973  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1974  *
1975  * only used when fib_nh is built into fib_info
1976  */
1977 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1978 {
1979 	int ret = 0;
1980 	int scope = RT_SCOPE_NOWHERE;
1981 	struct fib_info *prev_fi = NULL;
1982 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1983 	struct hlist_head *head = &fib_info_devhash[hash];
1984 	struct fib_nh *nh;
1985 
1986 	if (force)
1987 		scope = -1;
1988 
1989 	hlist_for_each_entry(nh, head, nh_hash) {
1990 		struct fib_info *fi = nh->nh_parent;
1991 		int dead;
1992 
1993 		BUG_ON(!fi->fib_nhs);
1994 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1995 			continue;
1996 		prev_fi = fi;
1997 		dead = 0;
1998 		change_nexthops(fi) {
1999 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2000 				dead++;
2001 			else if (nexthop_nh->fib_nh_dev == dev &&
2002 				 nexthop_nh->fib_nh_scope != scope) {
2003 				switch (event) {
2004 				case NETDEV_DOWN:
2005 				case NETDEV_UNREGISTER:
2006 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2007 					fallthrough;
2008 				case NETDEV_CHANGE:
2009 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2010 					break;
2011 				}
2012 				call_fib_nh_notifiers(nexthop_nh,
2013 						      FIB_EVENT_NH_DEL);
2014 				dead++;
2015 			}
2016 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2017 			if (event == NETDEV_UNREGISTER &&
2018 			    nexthop_nh->fib_nh_dev == dev) {
2019 				dead = fi->fib_nhs;
2020 				break;
2021 			}
2022 #endif
2023 		} endfor_nexthops(fi)
2024 		if (dead == fi->fib_nhs) {
2025 			switch (event) {
2026 			case NETDEV_DOWN:
2027 			case NETDEV_UNREGISTER:
2028 				fi->fib_flags |= RTNH_F_DEAD;
2029 				fallthrough;
2030 			case NETDEV_CHANGE:
2031 				fi->fib_flags |= RTNH_F_LINKDOWN;
2032 				break;
2033 			}
2034 			ret++;
2035 		}
2036 
2037 		fib_rebalance(fi);
2038 	}
2039 
2040 	return ret;
2041 }
2042 
2043 /* Must be invoked inside of an RCU protected region.  */
2044 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2045 {
2046 	struct fib_info *fi = NULL, *last_resort = NULL;
2047 	struct hlist_head *fa_head = res->fa_head;
2048 	struct fib_table *tb = res->table;
2049 	u8 slen = 32 - res->prefixlen;
2050 	int order = -1, last_idx = -1;
2051 	struct fib_alias *fa, *fa1 = NULL;
2052 	u32 last_prio = res->fi->fib_priority;
2053 	u8 last_tos = 0;
2054 
2055 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2056 		struct fib_info *next_fi = fa->fa_info;
2057 		struct fib_nh_common *nhc;
2058 
2059 		if (fa->fa_slen != slen)
2060 			continue;
2061 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2062 			continue;
2063 		if (fa->tb_id != tb->tb_id)
2064 			continue;
2065 		if (next_fi->fib_priority > last_prio &&
2066 		    fa->fa_tos == last_tos) {
2067 			if (last_tos)
2068 				continue;
2069 			break;
2070 		}
2071 		if (next_fi->fib_flags & RTNH_F_DEAD)
2072 			continue;
2073 		last_tos = fa->fa_tos;
2074 		last_prio = next_fi->fib_priority;
2075 
2076 		if (next_fi->fib_scope != res->scope ||
2077 		    fa->fa_type != RTN_UNICAST)
2078 			continue;
2079 
2080 		nhc = fib_info_nhc(next_fi, 0);
2081 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2082 			continue;
2083 
2084 		fib_alias_accessed(fa);
2085 
2086 		if (!fi) {
2087 			if (next_fi != res->fi)
2088 				break;
2089 			fa1 = fa;
2090 		} else if (!fib_detect_death(fi, order, &last_resort,
2091 					     &last_idx, fa1->fa_default)) {
2092 			fib_result_assign(res, fi);
2093 			fa1->fa_default = order;
2094 			goto out;
2095 		}
2096 		fi = next_fi;
2097 		order++;
2098 	}
2099 
2100 	if (order <= 0 || !fi) {
2101 		if (fa1)
2102 			fa1->fa_default = -1;
2103 		goto out;
2104 	}
2105 
2106 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2107 			      fa1->fa_default)) {
2108 		fib_result_assign(res, fi);
2109 		fa1->fa_default = order;
2110 		goto out;
2111 	}
2112 
2113 	if (last_idx >= 0)
2114 		fib_result_assign(res, last_resort);
2115 	fa1->fa_default = last_idx;
2116 out:
2117 	return;
2118 }
2119 
2120 /*
2121  * Dead device goes up. We wake up dead nexthops.
2122  * It takes sense only on multipath routes.
2123  *
2124  * only used when fib_nh is built into fib_info
2125  */
2126 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2127 {
2128 	struct fib_info *prev_fi;
2129 	unsigned int hash;
2130 	struct hlist_head *head;
2131 	struct fib_nh *nh;
2132 	int ret;
2133 
2134 	if (!(dev->flags & IFF_UP))
2135 		return 0;
2136 
2137 	if (nh_flags & RTNH_F_DEAD) {
2138 		unsigned int flags = dev_get_flags(dev);
2139 
2140 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2141 			nh_flags |= RTNH_F_LINKDOWN;
2142 	}
2143 
2144 	prev_fi = NULL;
2145 	hash = fib_devindex_hashfn(dev->ifindex);
2146 	head = &fib_info_devhash[hash];
2147 	ret = 0;
2148 
2149 	hlist_for_each_entry(nh, head, nh_hash) {
2150 		struct fib_info *fi = nh->nh_parent;
2151 		int alive;
2152 
2153 		BUG_ON(!fi->fib_nhs);
2154 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2155 			continue;
2156 
2157 		prev_fi = fi;
2158 		alive = 0;
2159 		change_nexthops(fi) {
2160 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2161 				alive++;
2162 				continue;
2163 			}
2164 			if (!nexthop_nh->fib_nh_dev ||
2165 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2166 				continue;
2167 			if (nexthop_nh->fib_nh_dev != dev ||
2168 			    !__in_dev_get_rtnl(dev))
2169 				continue;
2170 			alive++;
2171 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2172 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2173 		} endfor_nexthops(fi)
2174 
2175 		if (alive > 0) {
2176 			fi->fib_flags &= ~nh_flags;
2177 			ret++;
2178 		}
2179 
2180 		fib_rebalance(fi);
2181 	}
2182 
2183 	return ret;
2184 }
2185 
2186 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2187 static bool fib_good_nh(const struct fib_nh *nh)
2188 {
2189 	int state = NUD_REACHABLE;
2190 
2191 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2192 		struct neighbour *n;
2193 
2194 		rcu_read_lock_bh();
2195 
2196 		if (likely(nh->fib_nh_gw_family == AF_INET))
2197 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2198 						   (__force u32)nh->fib_nh_gw4);
2199 		else if (nh->fib_nh_gw_family == AF_INET6)
2200 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2201 							   &nh->fib_nh_gw6);
2202 		else
2203 			n = NULL;
2204 		if (n)
2205 			state = n->nud_state;
2206 
2207 		rcu_read_unlock_bh();
2208 	}
2209 
2210 	return !!(state & NUD_VALID);
2211 }
2212 
2213 void fib_select_multipath(struct fib_result *res, int hash)
2214 {
2215 	struct fib_info *fi = res->fi;
2216 	struct net *net = fi->fib_net;
2217 	bool first = false;
2218 
2219 	if (unlikely(res->fi->nh)) {
2220 		nexthop_path_fib_result(res, hash);
2221 		return;
2222 	}
2223 
2224 	change_nexthops(fi) {
2225 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2226 			if (!fib_good_nh(nexthop_nh))
2227 				continue;
2228 			if (!first) {
2229 				res->nh_sel = nhsel;
2230 				res->nhc = &nexthop_nh->nh_common;
2231 				first = true;
2232 			}
2233 		}
2234 
2235 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2236 			continue;
2237 
2238 		res->nh_sel = nhsel;
2239 		res->nhc = &nexthop_nh->nh_common;
2240 		return;
2241 	} endfor_nexthops(fi);
2242 }
2243 #endif
2244 
2245 void fib_select_path(struct net *net, struct fib_result *res,
2246 		     struct flowi4 *fl4, const struct sk_buff *skb)
2247 {
2248 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2249 		goto check_saddr;
2250 
2251 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2252 	if (fib_info_num_path(res->fi) > 1) {
2253 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2254 
2255 		fib_select_multipath(res, h);
2256 	}
2257 	else
2258 #endif
2259 	if (!res->prefixlen &&
2260 	    res->table->tb_num_default > 1 &&
2261 	    res->type == RTN_UNICAST)
2262 		fib_select_default(fl4, res);
2263 
2264 check_saddr:
2265 	if (!fl4->saddr)
2266 		fl4->saddr = fib_result_prefsrc(net, res);
2267 }
2268