xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision b737eecd)
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 	if (nhc->nhc_dev)
212 		dev_put(nhc->nhc_dev);
213 
214 	lwtstate_put(nhc->nhc_lwtstate);
215 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
216 	rt_fibinfo_free(&nhc->nhc_rth_input);
217 	free_nh_exceptions(nhc);
218 }
219 EXPORT_SYMBOL_GPL(fib_nh_common_release);
220 
221 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
222 {
223 #ifdef CONFIG_IP_ROUTE_CLASSID
224 	if (fib_nh->nh_tclassid)
225 		net->ipv4.fib_num_tclassid_users--;
226 #endif
227 	fib_nh_common_release(&fib_nh->nh_common);
228 }
229 
230 /* Release a nexthop info record */
231 static void free_fib_info_rcu(struct rcu_head *head)
232 {
233 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
234 
235 	if (fi->nh) {
236 		nexthop_put(fi->nh);
237 	} else {
238 		change_nexthops(fi) {
239 			fib_nh_release(fi->fib_net, nexthop_nh);
240 		} endfor_nexthops(fi);
241 	}
242 
243 	ip_fib_metrics_put(fi->fib_metrics);
244 
245 	kfree(fi);
246 }
247 
248 void free_fib_info(struct fib_info *fi)
249 {
250 	if (fi->fib_dead == 0) {
251 		pr_warn("Freeing alive fib_info %p\n", fi);
252 		return;
253 	}
254 	fib_info_cnt--;
255 
256 	call_rcu(&fi->rcu, free_fib_info_rcu);
257 }
258 EXPORT_SYMBOL_GPL(free_fib_info);
259 
260 void fib_release_info(struct fib_info *fi)
261 {
262 	spin_lock_bh(&fib_info_lock);
263 	if (fi && --fi->fib_treeref == 0) {
264 		hlist_del(&fi->fib_hash);
265 		if (fi->fib_prefsrc)
266 			hlist_del(&fi->fib_lhash);
267 		if (fi->nh) {
268 			list_del(&fi->nh_list);
269 		} else {
270 			change_nexthops(fi) {
271 				if (!nexthop_nh->fib_nh_dev)
272 					continue;
273 				hlist_del(&nexthop_nh->nh_hash);
274 			} endfor_nexthops(fi)
275 		}
276 		fi->fib_dead = 1;
277 		fib_info_put(fi);
278 	}
279 	spin_unlock_bh(&fib_info_lock);
280 }
281 
282 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
283 {
284 	const struct fib_nh *onh;
285 
286 	if (fi->nh || ofi->nh)
287 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
288 
289 	if (ofi->fib_nhs == 0)
290 		return 0;
291 
292 	for_nexthops(fi) {
293 		onh = fib_info_nh(ofi, nhsel);
294 
295 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
296 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
297 		    nh->fib_nh_scope != onh->fib_nh_scope ||
298 #ifdef CONFIG_IP_ROUTE_MULTIPATH
299 		    nh->fib_nh_weight != onh->fib_nh_weight ||
300 #endif
301 #ifdef CONFIG_IP_ROUTE_CLASSID
302 		    nh->nh_tclassid != onh->nh_tclassid ||
303 #endif
304 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
305 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
306 			return -1;
307 
308 		if (nh->fib_nh_gw_family == AF_INET &&
309 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
310 			return -1;
311 
312 		if (nh->fib_nh_gw_family == AF_INET6 &&
313 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
314 			return -1;
315 	} endfor_nexthops(fi);
316 	return 0;
317 }
318 
319 static inline unsigned int fib_devindex_hashfn(unsigned int val)
320 {
321 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
322 
323 	return (val ^
324 		(val >> DEVINDEX_HASHBITS) ^
325 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
326 }
327 
328 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
329 				      u32 prefsrc, u32 priority)
330 {
331 	unsigned int val = init_val;
332 
333 	val ^= (protocol << 8) | scope;
334 	val ^= prefsrc;
335 	val ^= priority;
336 
337 	return val;
338 }
339 
340 static unsigned int fib_info_hashfn_result(unsigned int val)
341 {
342 	unsigned int mask = (fib_info_hash_size - 1);
343 
344 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
345 }
346 
347 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
348 {
349 	unsigned int val;
350 
351 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
352 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
353 				fi->fib_priority);
354 
355 	if (fi->nh) {
356 		val ^= fib_devindex_hashfn(fi->nh->id);
357 	} else {
358 		for_nexthops(fi) {
359 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
360 		} endfor_nexthops(fi)
361 	}
362 
363 	return fib_info_hashfn_result(val);
364 }
365 
366 /* no metrics, only nexthop id */
367 static struct fib_info *fib_find_info_nh(struct net *net,
368 					 const struct fib_config *cfg)
369 {
370 	struct hlist_head *head;
371 	struct fib_info *fi;
372 	unsigned int hash;
373 
374 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
375 				 cfg->fc_protocol, cfg->fc_scope,
376 				 (__force u32)cfg->fc_prefsrc,
377 				 cfg->fc_priority);
378 	hash = fib_info_hashfn_result(hash);
379 	head = &fib_info_hash[hash];
380 
381 	hlist_for_each_entry(fi, head, fib_hash) {
382 		if (!net_eq(fi->fib_net, net))
383 			continue;
384 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
385 			continue;
386 		if (cfg->fc_protocol == fi->fib_protocol &&
387 		    cfg->fc_scope == fi->fib_scope &&
388 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
389 		    cfg->fc_priority == fi->fib_priority &&
390 		    cfg->fc_type == fi->fib_type &&
391 		    cfg->fc_table == fi->fib_tb_id &&
392 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
393 			return fi;
394 	}
395 
396 	return NULL;
397 }
398 
399 static struct fib_info *fib_find_info(struct fib_info *nfi)
400 {
401 	struct hlist_head *head;
402 	struct fib_info *fi;
403 	unsigned int hash;
404 
405 	hash = fib_info_hashfn(nfi);
406 	head = &fib_info_hash[hash];
407 
408 	hlist_for_each_entry(fi, head, fib_hash) {
409 		if (!net_eq(fi->fib_net, nfi->fib_net))
410 			continue;
411 		if (fi->fib_nhs != nfi->fib_nhs)
412 			continue;
413 		if (nfi->fib_protocol == fi->fib_protocol &&
414 		    nfi->fib_scope == fi->fib_scope &&
415 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
416 		    nfi->fib_priority == fi->fib_priority &&
417 		    nfi->fib_type == fi->fib_type &&
418 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
419 			   sizeof(u32) * RTAX_MAX) == 0 &&
420 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
421 		    nh_comp(fi, nfi) == 0)
422 			return fi;
423 	}
424 
425 	return NULL;
426 }
427 
428 /* Check, that the gateway is already configured.
429  * Used only by redirect accept routine.
430  */
431 int ip_fib_check_default(__be32 gw, struct net_device *dev)
432 {
433 	struct hlist_head *head;
434 	struct fib_nh *nh;
435 	unsigned int hash;
436 
437 	spin_lock(&fib_info_lock);
438 
439 	hash = fib_devindex_hashfn(dev->ifindex);
440 	head = &fib_info_devhash[hash];
441 	hlist_for_each_entry(nh, head, nh_hash) {
442 		if (nh->fib_nh_dev == dev &&
443 		    nh->fib_nh_gw4 == gw &&
444 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
445 			spin_unlock(&fib_info_lock);
446 			return 0;
447 		}
448 	}
449 
450 	spin_unlock(&fib_info_lock);
451 
452 	return -1;
453 }
454 
455 size_t fib_nlmsg_size(struct fib_info *fi)
456 {
457 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
458 			 + nla_total_size(4) /* RTA_TABLE */
459 			 + nla_total_size(4) /* RTA_DST */
460 			 + nla_total_size(4) /* RTA_PRIORITY */
461 			 + nla_total_size(4) /* RTA_PREFSRC */
462 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
463 	unsigned int nhs = fib_info_num_path(fi);
464 
465 	/* space for nested metrics */
466 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
467 
468 	if (fi->nh)
469 		payload += nla_total_size(4); /* RTA_NH_ID */
470 
471 	if (nhs) {
472 		size_t nh_encapsize = 0;
473 		/* Also handles the special case nhs == 1 */
474 
475 		/* each nexthop is packed in an attribute */
476 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
477 		unsigned int i;
478 
479 		/* may contain flow and gateway attribute */
480 		nhsize += 2 * nla_total_size(4);
481 
482 		/* grab encap info */
483 		for (i = 0; i < fib_info_num_path(fi); i++) {
484 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
485 
486 			if (nhc->nhc_lwtstate) {
487 				/* RTA_ENCAP_TYPE */
488 				nh_encapsize += lwtunnel_get_encap_size(
489 						nhc->nhc_lwtstate);
490 				/* RTA_ENCAP */
491 				nh_encapsize +=  nla_total_size(2);
492 			}
493 		}
494 
495 		/* all nexthops are packed in a nested attribute */
496 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
497 
498 	}
499 
500 	return payload;
501 }
502 
503 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
504 	       int dst_len, u32 tb_id, const struct nl_info *info,
505 	       unsigned int nlm_flags)
506 {
507 	struct fib_rt_info fri;
508 	struct sk_buff *skb;
509 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
510 	int err = -ENOBUFS;
511 
512 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
513 	if (!skb)
514 		goto errout;
515 
516 	fri.fi = fa->fa_info;
517 	fri.tb_id = tb_id;
518 	fri.dst = key;
519 	fri.dst_len = dst_len;
520 	fri.tos = fa->fa_tos;
521 	fri.type = fa->fa_type;
522 	fri.offload = fa->offload;
523 	fri.trap = fa->trap;
524 	fri.offload_failed = fa->offload_failed;
525 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
526 	if (err < 0) {
527 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
528 		WARN_ON(err == -EMSGSIZE);
529 		kfree_skb(skb);
530 		goto errout;
531 	}
532 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
533 		    info->nlh, GFP_KERNEL);
534 	return;
535 errout:
536 	if (err < 0)
537 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
538 }
539 
540 static int fib_detect_death(struct fib_info *fi, int order,
541 			    struct fib_info **last_resort, int *last_idx,
542 			    int dflt)
543 {
544 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
545 	struct neighbour *n;
546 	int state = NUD_NONE;
547 
548 	if (likely(nhc->nhc_gw_family == AF_INET))
549 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
550 	else if (nhc->nhc_gw_family == AF_INET6)
551 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
552 				 nhc->nhc_dev);
553 	else
554 		n = NULL;
555 
556 	if (n) {
557 		state = n->nud_state;
558 		neigh_release(n);
559 	} else {
560 		return 0;
561 	}
562 	if (state == NUD_REACHABLE)
563 		return 0;
564 	if ((state & NUD_VALID) && order != dflt)
565 		return 0;
566 	if ((state & NUD_VALID) ||
567 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
568 		*last_resort = fi;
569 		*last_idx = order;
570 	}
571 	return 1;
572 }
573 
574 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
575 		       struct nlattr *encap, u16 encap_type,
576 		       void *cfg, gfp_t gfp_flags,
577 		       struct netlink_ext_ack *extack)
578 {
579 	int err;
580 
581 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
582 						    gfp_flags);
583 	if (!nhc->nhc_pcpu_rth_output)
584 		return -ENOMEM;
585 
586 	if (encap) {
587 		struct lwtunnel_state *lwtstate;
588 
589 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
590 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
591 			err = -EINVAL;
592 			goto lwt_failure;
593 		}
594 		err = lwtunnel_build_state(net, encap_type, encap,
595 					   nhc->nhc_family, cfg, &lwtstate,
596 					   extack);
597 		if (err)
598 			goto lwt_failure;
599 
600 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
601 	}
602 
603 	return 0;
604 
605 lwt_failure:
606 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
607 	nhc->nhc_pcpu_rth_output = NULL;
608 	return err;
609 }
610 EXPORT_SYMBOL_GPL(fib_nh_common_init);
611 
612 int fib_nh_init(struct net *net, struct fib_nh *nh,
613 		struct fib_config *cfg, int nh_weight,
614 		struct netlink_ext_ack *extack)
615 {
616 	int err;
617 
618 	nh->fib_nh_family = AF_INET;
619 
620 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
621 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
622 	if (err)
623 		return err;
624 
625 	nh->fib_nh_oif = cfg->fc_oif;
626 	nh->fib_nh_gw_family = cfg->fc_gw_family;
627 	if (cfg->fc_gw_family == AF_INET)
628 		nh->fib_nh_gw4 = cfg->fc_gw4;
629 	else if (cfg->fc_gw_family == AF_INET6)
630 		nh->fib_nh_gw6 = cfg->fc_gw6;
631 
632 	nh->fib_nh_flags = cfg->fc_flags;
633 
634 #ifdef CONFIG_IP_ROUTE_CLASSID
635 	nh->nh_tclassid = cfg->fc_flow;
636 	if (nh->nh_tclassid)
637 		net->ipv4.fib_num_tclassid_users++;
638 #endif
639 #ifdef CONFIG_IP_ROUTE_MULTIPATH
640 	nh->fib_nh_weight = nh_weight;
641 #endif
642 	return 0;
643 }
644 
645 #ifdef CONFIG_IP_ROUTE_MULTIPATH
646 
647 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
648 			      struct netlink_ext_ack *extack)
649 {
650 	int nhs = 0;
651 
652 	while (rtnh_ok(rtnh, remaining)) {
653 		nhs++;
654 		rtnh = rtnh_next(rtnh, &remaining);
655 	}
656 
657 	/* leftover implies invalid nexthop configuration, discard it */
658 	if (remaining > 0) {
659 		NL_SET_ERR_MSG(extack,
660 			       "Invalid nexthop configuration - extra data after nexthops");
661 		nhs = 0;
662 	}
663 
664 	return nhs;
665 }
666 
667 /* only called when fib_nh is integrated into fib_info */
668 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
669 		       int remaining, struct fib_config *cfg,
670 		       struct netlink_ext_ack *extack)
671 {
672 	struct net *net = fi->fib_net;
673 	struct fib_config fib_cfg;
674 	struct fib_nh *nh;
675 	int ret;
676 
677 	change_nexthops(fi) {
678 		int attrlen;
679 
680 		memset(&fib_cfg, 0, sizeof(fib_cfg));
681 
682 		if (!rtnh_ok(rtnh, remaining)) {
683 			NL_SET_ERR_MSG(extack,
684 				       "Invalid nexthop configuration - extra data after nexthop");
685 			return -EINVAL;
686 		}
687 
688 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
689 			NL_SET_ERR_MSG(extack,
690 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
691 			return -EINVAL;
692 		}
693 
694 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
695 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
696 
697 		attrlen = rtnh_attrlen(rtnh);
698 		if (attrlen > 0) {
699 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
700 
701 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
702 			nlav = nla_find(attrs, attrlen, RTA_VIA);
703 			if (nla && nlav) {
704 				NL_SET_ERR_MSG(extack,
705 					       "Nexthop configuration can not contain both GATEWAY and VIA");
706 				return -EINVAL;
707 			}
708 			if (nla) {
709 				fib_cfg.fc_gw4 = nla_get_in_addr(nla);
710 				if (fib_cfg.fc_gw4)
711 					fib_cfg.fc_gw_family = AF_INET;
712 			} else if (nlav) {
713 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
714 				if (ret)
715 					goto errout;
716 			}
717 
718 			nla = nla_find(attrs, attrlen, RTA_FLOW);
719 			if (nla)
720 				fib_cfg.fc_flow = nla_get_u32(nla);
721 
722 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
723 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
724 			if (nla)
725 				fib_cfg.fc_encap_type = nla_get_u16(nla);
726 		}
727 
728 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
729 				  rtnh->rtnh_hops + 1, extack);
730 		if (ret)
731 			goto errout;
732 
733 		rtnh = rtnh_next(rtnh, &remaining);
734 	} endfor_nexthops(fi);
735 
736 	ret = -EINVAL;
737 	nh = fib_info_nh(fi, 0);
738 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
739 		NL_SET_ERR_MSG(extack,
740 			       "Nexthop device index does not match RTA_OIF");
741 		goto errout;
742 	}
743 	if (cfg->fc_gw_family) {
744 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
745 		    (cfg->fc_gw_family == AF_INET &&
746 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
747 		    (cfg->fc_gw_family == AF_INET6 &&
748 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
749 			NL_SET_ERR_MSG(extack,
750 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
751 			goto errout;
752 		}
753 	}
754 #ifdef CONFIG_IP_ROUTE_CLASSID
755 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
756 		NL_SET_ERR_MSG(extack,
757 			       "Nexthop class id does not match RTA_FLOW");
758 		goto errout;
759 	}
760 #endif
761 	ret = 0;
762 errout:
763 	return ret;
764 }
765 
766 /* only called when fib_nh is integrated into fib_info */
767 static void fib_rebalance(struct fib_info *fi)
768 {
769 	int total;
770 	int w;
771 
772 	if (fib_info_num_path(fi) < 2)
773 		return;
774 
775 	total = 0;
776 	for_nexthops(fi) {
777 		if (nh->fib_nh_flags & RTNH_F_DEAD)
778 			continue;
779 
780 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
781 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
782 			continue;
783 
784 		total += nh->fib_nh_weight;
785 	} endfor_nexthops(fi);
786 
787 	w = 0;
788 	change_nexthops(fi) {
789 		int upper_bound;
790 
791 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
792 			upper_bound = -1;
793 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
794 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
795 			upper_bound = -1;
796 		} else {
797 			w += nexthop_nh->fib_nh_weight;
798 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
799 							    total) - 1;
800 		}
801 
802 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
803 	} endfor_nexthops(fi);
804 }
805 #else /* CONFIG_IP_ROUTE_MULTIPATH */
806 
807 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
808 		       int remaining, struct fib_config *cfg,
809 		       struct netlink_ext_ack *extack)
810 {
811 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
812 
813 	return -EINVAL;
814 }
815 
816 #define fib_rebalance(fi) do { } while (0)
817 
818 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
819 
820 static int fib_encap_match(struct net *net, u16 encap_type,
821 			   struct nlattr *encap,
822 			   const struct fib_nh *nh,
823 			   const struct fib_config *cfg,
824 			   struct netlink_ext_ack *extack)
825 {
826 	struct lwtunnel_state *lwtstate;
827 	int ret, result = 0;
828 
829 	if (encap_type == LWTUNNEL_ENCAP_NONE)
830 		return 0;
831 
832 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
833 				   cfg, &lwtstate, extack);
834 	if (!ret) {
835 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
836 		lwtstate_free(lwtstate);
837 	}
838 
839 	return result;
840 }
841 
842 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
843 		 struct netlink_ext_ack *extack)
844 {
845 #ifdef CONFIG_IP_ROUTE_MULTIPATH
846 	struct rtnexthop *rtnh;
847 	int remaining;
848 #endif
849 
850 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
851 		return 1;
852 
853 	if (cfg->fc_nh_id) {
854 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
855 			return 0;
856 		return 1;
857 	}
858 
859 	if (cfg->fc_oif || cfg->fc_gw_family) {
860 		struct fib_nh *nh = fib_info_nh(fi, 0);
861 
862 		if (cfg->fc_encap) {
863 			if (fib_encap_match(net, cfg->fc_encap_type,
864 					    cfg->fc_encap, nh, cfg, extack))
865 				return 1;
866 		}
867 #ifdef CONFIG_IP_ROUTE_CLASSID
868 		if (cfg->fc_flow &&
869 		    cfg->fc_flow != nh->nh_tclassid)
870 			return 1;
871 #endif
872 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
873 		    (cfg->fc_gw_family &&
874 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
875 			return 1;
876 
877 		if (cfg->fc_gw_family == AF_INET &&
878 		    cfg->fc_gw4 != nh->fib_nh_gw4)
879 			return 1;
880 
881 		if (cfg->fc_gw_family == AF_INET6 &&
882 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
883 			return 1;
884 
885 		return 0;
886 	}
887 
888 #ifdef CONFIG_IP_ROUTE_MULTIPATH
889 	if (!cfg->fc_mp)
890 		return 0;
891 
892 	rtnh = cfg->fc_mp;
893 	remaining = cfg->fc_mp_len;
894 
895 	for_nexthops(fi) {
896 		int attrlen;
897 
898 		if (!rtnh_ok(rtnh, remaining))
899 			return -EINVAL;
900 
901 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
902 			return 1;
903 
904 		attrlen = rtnh_attrlen(rtnh);
905 		if (attrlen > 0) {
906 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
907 
908 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
909 			nlav = nla_find(attrs, attrlen, RTA_VIA);
910 			if (nla && nlav) {
911 				NL_SET_ERR_MSG(extack,
912 					       "Nexthop configuration can not contain both GATEWAY and VIA");
913 				return -EINVAL;
914 			}
915 
916 			if (nla) {
917 				if (nh->fib_nh_gw_family != AF_INET ||
918 				    nla_get_in_addr(nla) != nh->fib_nh_gw4)
919 					return 1;
920 			} else if (nlav) {
921 				struct fib_config cfg2;
922 				int err;
923 
924 				err = fib_gw_from_via(&cfg2, nlav, extack);
925 				if (err)
926 					return err;
927 
928 				switch (nh->fib_nh_gw_family) {
929 				case AF_INET:
930 					if (cfg2.fc_gw_family != AF_INET ||
931 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
932 						return 1;
933 					break;
934 				case AF_INET6:
935 					if (cfg2.fc_gw_family != AF_INET6 ||
936 					    ipv6_addr_cmp(&cfg2.fc_gw6,
937 							  &nh->fib_nh_gw6))
938 						return 1;
939 					break;
940 				}
941 			}
942 
943 #ifdef CONFIG_IP_ROUTE_CLASSID
944 			nla = nla_find(attrs, attrlen, RTA_FLOW);
945 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
946 				return 1;
947 #endif
948 		}
949 
950 		rtnh = rtnh_next(rtnh, &remaining);
951 	} endfor_nexthops(fi);
952 #endif
953 	return 0;
954 }
955 
956 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
957 {
958 	struct nlattr *nla;
959 	int remaining;
960 
961 	if (!cfg->fc_mx)
962 		return true;
963 
964 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
965 		int type = nla_type(nla);
966 		u32 fi_val, val;
967 
968 		if (!type)
969 			continue;
970 		if (type > RTAX_MAX)
971 			return false;
972 
973 		if (type == RTAX_CC_ALGO) {
974 			char tmp[TCP_CA_NAME_MAX];
975 			bool ecn_ca = false;
976 
977 			nla_strscpy(tmp, nla, sizeof(tmp));
978 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
979 		} else {
980 			if (nla_len(nla) != sizeof(u32))
981 				return false;
982 			val = nla_get_u32(nla);
983 		}
984 
985 		fi_val = fi->fib_metrics->metrics[type - 1];
986 		if (type == RTAX_FEATURES)
987 			fi_val &= ~DST_FEATURE_ECN_CA;
988 
989 		if (fi_val != val)
990 			return false;
991 	}
992 
993 	return true;
994 }
995 
996 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
997 			      u32 table, struct netlink_ext_ack *extack)
998 {
999 	struct fib6_config cfg = {
1000 		.fc_table = table,
1001 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1002 		.fc_ifindex = nh->fib_nh_oif,
1003 		.fc_gateway = nh->fib_nh_gw6,
1004 	};
1005 	struct fib6_nh fib6_nh = {};
1006 	int err;
1007 
1008 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1009 	if (!err) {
1010 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1011 		dev_hold(nh->fib_nh_dev);
1012 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1013 		nh->fib_nh_scope = RT_SCOPE_LINK;
1014 
1015 		ipv6_stub->fib6_nh_release(&fib6_nh);
1016 	}
1017 
1018 	return err;
1019 }
1020 
1021 /*
1022  * Picture
1023  * -------
1024  *
1025  * Semantics of nexthop is very messy by historical reasons.
1026  * We have to take into account, that:
1027  * a) gateway can be actually local interface address,
1028  *    so that gatewayed route is direct.
1029  * b) gateway must be on-link address, possibly
1030  *    described not by an ifaddr, but also by a direct route.
1031  * c) If both gateway and interface are specified, they should not
1032  *    contradict.
1033  * d) If we use tunnel routes, gateway could be not on-link.
1034  *
1035  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1036  * results in pretty ugly and hairy code with obscure logic.
1037  *
1038  * I chose to generalized it instead, so that the size
1039  * of code does not increase practically, but it becomes
1040  * much more general.
1041  * Every prefix is assigned a "scope" value: "host" is local address,
1042  * "link" is direct route,
1043  * [ ... "site" ... "interior" ... ]
1044  * and "universe" is true gateway route with global meaning.
1045  *
1046  * Every prefix refers to a set of "nexthop"s (gw, oif),
1047  * where gw must have narrower scope. This recursion stops
1048  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1049  * which means that gw is forced to be on link.
1050  *
1051  * Code is still hairy, but now it is apparently logically
1052  * consistent and very flexible. F.e. as by-product it allows
1053  * to co-exists in peace independent exterior and interior
1054  * routing processes.
1055  *
1056  * Normally it looks as following.
1057  *
1058  * {universe prefix}  -> (gw, oif) [scope link]
1059  *		  |
1060  *		  |-> {link prefix} -> (gw, oif) [scope local]
1061  *					|
1062  *					|-> {local prefix} (terminal node)
1063  */
1064 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1065 			      u8 scope, struct netlink_ext_ack *extack)
1066 {
1067 	struct net_device *dev;
1068 	struct fib_result res;
1069 	int err = 0;
1070 
1071 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1072 		unsigned int addr_type;
1073 
1074 		if (scope >= RT_SCOPE_LINK) {
1075 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1076 			return -EINVAL;
1077 		}
1078 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1079 		if (!dev) {
1080 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1081 			return -ENODEV;
1082 		}
1083 		if (!(dev->flags & IFF_UP)) {
1084 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1085 			return -ENETDOWN;
1086 		}
1087 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1088 		if (addr_type != RTN_UNICAST) {
1089 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1090 			return -EINVAL;
1091 		}
1092 		if (!netif_carrier_ok(dev))
1093 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1094 		nh->fib_nh_dev = dev;
1095 		dev_hold(dev);
1096 		nh->fib_nh_scope = RT_SCOPE_LINK;
1097 		return 0;
1098 	}
1099 	rcu_read_lock();
1100 	{
1101 		struct fib_table *tbl = NULL;
1102 		struct flowi4 fl4 = {
1103 			.daddr = nh->fib_nh_gw4,
1104 			.flowi4_scope = scope + 1,
1105 			.flowi4_oif = nh->fib_nh_oif,
1106 			.flowi4_iif = LOOPBACK_IFINDEX,
1107 		};
1108 
1109 		/* It is not necessary, but requires a bit of thinking */
1110 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1111 			fl4.flowi4_scope = RT_SCOPE_LINK;
1112 
1113 		if (table && table != RT_TABLE_MAIN)
1114 			tbl = fib_get_table(net, table);
1115 
1116 		if (tbl)
1117 			err = fib_table_lookup(tbl, &fl4, &res,
1118 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1119 					       FIB_LOOKUP_NOREF);
1120 
1121 		/* on error or if no table given do full lookup. This
1122 		 * is needed for example when nexthops are in the local
1123 		 * table rather than the given table
1124 		 */
1125 		if (!tbl || err) {
1126 			err = fib_lookup(net, &fl4, &res,
1127 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1128 		}
1129 
1130 		if (err) {
1131 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1132 			goto out;
1133 		}
1134 	}
1135 
1136 	err = -EINVAL;
1137 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1138 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1139 		goto out;
1140 	}
1141 	nh->fib_nh_scope = res.scope;
1142 	nh->fib_nh_oif = FIB_RES_OIF(res);
1143 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1144 	if (!dev) {
1145 		NL_SET_ERR_MSG(extack,
1146 			       "No egress device for nexthop gateway");
1147 		goto out;
1148 	}
1149 	dev_hold(dev);
1150 	if (!netif_carrier_ok(dev))
1151 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1152 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1153 out:
1154 	rcu_read_unlock();
1155 	return err;
1156 }
1157 
1158 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1159 			      struct netlink_ext_ack *extack)
1160 {
1161 	struct in_device *in_dev;
1162 	int err;
1163 
1164 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1165 		NL_SET_ERR_MSG(extack,
1166 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1167 		return -EINVAL;
1168 	}
1169 
1170 	rcu_read_lock();
1171 
1172 	err = -ENODEV;
1173 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1174 	if (!in_dev)
1175 		goto out;
1176 	err = -ENETDOWN;
1177 	if (!(in_dev->dev->flags & IFF_UP)) {
1178 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1179 		goto out;
1180 	}
1181 
1182 	nh->fib_nh_dev = in_dev->dev;
1183 	dev_hold(nh->fib_nh_dev);
1184 	nh->fib_nh_scope = RT_SCOPE_HOST;
1185 	if (!netif_carrier_ok(nh->fib_nh_dev))
1186 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1187 	err = 0;
1188 out:
1189 	rcu_read_unlock();
1190 	return err;
1191 }
1192 
1193 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1194 		 struct netlink_ext_ack *extack)
1195 {
1196 	int err;
1197 
1198 	if (nh->fib_nh_gw_family == AF_INET)
1199 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1200 	else if (nh->fib_nh_gw_family == AF_INET6)
1201 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1202 	else
1203 		err = fib_check_nh_nongw(net, nh, extack);
1204 
1205 	return err;
1206 }
1207 
1208 static inline unsigned int fib_laddr_hashfn(__be32 val)
1209 {
1210 	unsigned int mask = (fib_info_hash_size - 1);
1211 
1212 	return ((__force u32)val ^
1213 		((__force u32)val >> 7) ^
1214 		((__force u32)val >> 14)) & mask;
1215 }
1216 
1217 static struct hlist_head *fib_info_hash_alloc(int bytes)
1218 {
1219 	if (bytes <= PAGE_SIZE)
1220 		return kzalloc(bytes, GFP_KERNEL);
1221 	else
1222 		return (struct hlist_head *)
1223 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1224 					 get_order(bytes));
1225 }
1226 
1227 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1228 {
1229 	if (!hash)
1230 		return;
1231 
1232 	if (bytes <= PAGE_SIZE)
1233 		kfree(hash);
1234 	else
1235 		free_pages((unsigned long) hash, get_order(bytes));
1236 }
1237 
1238 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1239 			       struct hlist_head *new_laddrhash,
1240 			       unsigned int new_size)
1241 {
1242 	struct hlist_head *old_info_hash, *old_laddrhash;
1243 	unsigned int old_size = fib_info_hash_size;
1244 	unsigned int i, bytes;
1245 
1246 	spin_lock_bh(&fib_info_lock);
1247 	old_info_hash = fib_info_hash;
1248 	old_laddrhash = fib_info_laddrhash;
1249 	fib_info_hash_size = new_size;
1250 
1251 	for (i = 0; i < old_size; i++) {
1252 		struct hlist_head *head = &fib_info_hash[i];
1253 		struct hlist_node *n;
1254 		struct fib_info *fi;
1255 
1256 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1257 			struct hlist_head *dest;
1258 			unsigned int new_hash;
1259 
1260 			new_hash = fib_info_hashfn(fi);
1261 			dest = &new_info_hash[new_hash];
1262 			hlist_add_head(&fi->fib_hash, dest);
1263 		}
1264 	}
1265 	fib_info_hash = new_info_hash;
1266 
1267 	for (i = 0; i < old_size; i++) {
1268 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1269 		struct hlist_node *n;
1270 		struct fib_info *fi;
1271 
1272 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1273 			struct hlist_head *ldest;
1274 			unsigned int new_hash;
1275 
1276 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1277 			ldest = &new_laddrhash[new_hash];
1278 			hlist_add_head(&fi->fib_lhash, ldest);
1279 		}
1280 	}
1281 	fib_info_laddrhash = new_laddrhash;
1282 
1283 	spin_unlock_bh(&fib_info_lock);
1284 
1285 	bytes = old_size * sizeof(struct hlist_head *);
1286 	fib_info_hash_free(old_info_hash, bytes);
1287 	fib_info_hash_free(old_laddrhash, bytes);
1288 }
1289 
1290 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1291 				 unsigned char scope)
1292 {
1293 	struct fib_nh *nh;
1294 
1295 	if (nhc->nhc_family != AF_INET)
1296 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1297 
1298 	nh = container_of(nhc, struct fib_nh, nh_common);
1299 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1300 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1301 
1302 	return nh->nh_saddr;
1303 }
1304 
1305 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1306 {
1307 	struct fib_nh_common *nhc = res->nhc;
1308 
1309 	if (res->fi->fib_prefsrc)
1310 		return res->fi->fib_prefsrc;
1311 
1312 	if (nhc->nhc_family == AF_INET) {
1313 		struct fib_nh *nh;
1314 
1315 		nh = container_of(nhc, struct fib_nh, nh_common);
1316 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1317 			return nh->nh_saddr;
1318 	}
1319 
1320 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1321 }
1322 
1323 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1324 {
1325 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1326 	    fib_prefsrc != cfg->fc_dst) {
1327 		u32 tb_id = cfg->fc_table;
1328 		int rc;
1329 
1330 		if (tb_id == RT_TABLE_MAIN)
1331 			tb_id = RT_TABLE_LOCAL;
1332 
1333 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1334 					  fib_prefsrc, tb_id);
1335 
1336 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1337 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1338 						  fib_prefsrc, RT_TABLE_LOCAL);
1339 		}
1340 
1341 		if (rc != RTN_LOCAL)
1342 			return false;
1343 	}
1344 	return true;
1345 }
1346 
1347 struct fib_info *fib_create_info(struct fib_config *cfg,
1348 				 struct netlink_ext_ack *extack)
1349 {
1350 	int err;
1351 	struct fib_info *fi = NULL;
1352 	struct nexthop *nh = NULL;
1353 	struct fib_info *ofi;
1354 	int nhs = 1;
1355 	struct net *net = cfg->fc_nlinfo.nl_net;
1356 
1357 	if (cfg->fc_type > RTN_MAX)
1358 		goto err_inval;
1359 
1360 	/* Fast check to catch the most weird cases */
1361 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1362 		NL_SET_ERR_MSG(extack, "Invalid scope");
1363 		goto err_inval;
1364 	}
1365 
1366 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1367 		NL_SET_ERR_MSG(extack,
1368 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1369 		goto err_inval;
1370 	}
1371 
1372 	if (cfg->fc_nh_id) {
1373 		if (!cfg->fc_mx) {
1374 			fi = fib_find_info_nh(net, cfg);
1375 			if (fi) {
1376 				fi->fib_treeref++;
1377 				return fi;
1378 			}
1379 		}
1380 
1381 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1382 		if (!nh) {
1383 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1384 			goto err_inval;
1385 		}
1386 		nhs = 0;
1387 	}
1388 
1389 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1390 	if (cfg->fc_mp) {
1391 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1392 		if (nhs == 0)
1393 			goto err_inval;
1394 	}
1395 #endif
1396 
1397 	err = -ENOBUFS;
1398 	if (fib_info_cnt >= fib_info_hash_size) {
1399 		unsigned int new_size = fib_info_hash_size << 1;
1400 		struct hlist_head *new_info_hash;
1401 		struct hlist_head *new_laddrhash;
1402 		unsigned int bytes;
1403 
1404 		if (!new_size)
1405 			new_size = 16;
1406 		bytes = new_size * sizeof(struct hlist_head *);
1407 		new_info_hash = fib_info_hash_alloc(bytes);
1408 		new_laddrhash = fib_info_hash_alloc(bytes);
1409 		if (!new_info_hash || !new_laddrhash) {
1410 			fib_info_hash_free(new_info_hash, bytes);
1411 			fib_info_hash_free(new_laddrhash, bytes);
1412 		} else
1413 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1414 
1415 		if (!fib_info_hash_size)
1416 			goto failure;
1417 	}
1418 
1419 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1420 	if (!fi)
1421 		goto failure;
1422 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1423 					      cfg->fc_mx_len, extack);
1424 	if (IS_ERR(fi->fib_metrics)) {
1425 		err = PTR_ERR(fi->fib_metrics);
1426 		kfree(fi);
1427 		return ERR_PTR(err);
1428 	}
1429 
1430 	fib_info_cnt++;
1431 	fi->fib_net = net;
1432 	fi->fib_protocol = cfg->fc_protocol;
1433 	fi->fib_scope = cfg->fc_scope;
1434 	fi->fib_flags = cfg->fc_flags;
1435 	fi->fib_priority = cfg->fc_priority;
1436 	fi->fib_prefsrc = cfg->fc_prefsrc;
1437 	fi->fib_type = cfg->fc_type;
1438 	fi->fib_tb_id = cfg->fc_table;
1439 
1440 	fi->fib_nhs = nhs;
1441 	if (nh) {
1442 		if (!nexthop_get(nh)) {
1443 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1444 			err = -EINVAL;
1445 		} else {
1446 			err = 0;
1447 			fi->nh = nh;
1448 		}
1449 	} else {
1450 		change_nexthops(fi) {
1451 			nexthop_nh->nh_parent = fi;
1452 		} endfor_nexthops(fi)
1453 
1454 		if (cfg->fc_mp)
1455 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1456 					  extack);
1457 		else
1458 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1459 	}
1460 
1461 	if (err != 0)
1462 		goto failure;
1463 
1464 	if (fib_props[cfg->fc_type].error) {
1465 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1466 			NL_SET_ERR_MSG(extack,
1467 				       "Gateway, device and multipath can not be specified for this route type");
1468 			goto err_inval;
1469 		}
1470 		goto link_it;
1471 	} else {
1472 		switch (cfg->fc_type) {
1473 		case RTN_UNICAST:
1474 		case RTN_LOCAL:
1475 		case RTN_BROADCAST:
1476 		case RTN_ANYCAST:
1477 		case RTN_MULTICAST:
1478 			break;
1479 		default:
1480 			NL_SET_ERR_MSG(extack, "Invalid route type");
1481 			goto err_inval;
1482 		}
1483 	}
1484 
1485 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1486 		NL_SET_ERR_MSG(extack, "Invalid scope");
1487 		goto err_inval;
1488 	}
1489 
1490 	if (fi->nh) {
1491 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1492 		if (err)
1493 			goto failure;
1494 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1495 		struct fib_nh *nh = fi->fib_nh;
1496 
1497 		/* Local address is added. */
1498 		if (nhs != 1) {
1499 			NL_SET_ERR_MSG(extack,
1500 				       "Route with host scope can not have multiple nexthops");
1501 			goto err_inval;
1502 		}
1503 		if (nh->fib_nh_gw_family) {
1504 			NL_SET_ERR_MSG(extack,
1505 				       "Route with host scope can not have a gateway");
1506 			goto err_inval;
1507 		}
1508 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1509 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1510 		err = -ENODEV;
1511 		if (!nh->fib_nh_dev)
1512 			goto failure;
1513 	} else {
1514 		int linkdown = 0;
1515 
1516 		change_nexthops(fi) {
1517 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1518 					   cfg->fc_table, cfg->fc_scope,
1519 					   extack);
1520 			if (err != 0)
1521 				goto failure;
1522 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1523 				linkdown++;
1524 		} endfor_nexthops(fi)
1525 		if (linkdown == fi->fib_nhs)
1526 			fi->fib_flags |= RTNH_F_LINKDOWN;
1527 	}
1528 
1529 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1530 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1531 		goto err_inval;
1532 	}
1533 
1534 	if (!fi->nh) {
1535 		change_nexthops(fi) {
1536 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1537 						  fi->fib_scope);
1538 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1539 				fi->fib_nh_is_v6 = true;
1540 		} endfor_nexthops(fi)
1541 
1542 		fib_rebalance(fi);
1543 	}
1544 
1545 link_it:
1546 	ofi = fib_find_info(fi);
1547 	if (ofi) {
1548 		fi->fib_dead = 1;
1549 		free_fib_info(fi);
1550 		ofi->fib_treeref++;
1551 		return ofi;
1552 	}
1553 
1554 	fi->fib_treeref++;
1555 	refcount_set(&fi->fib_clntref, 1);
1556 	spin_lock_bh(&fib_info_lock);
1557 	hlist_add_head(&fi->fib_hash,
1558 		       &fib_info_hash[fib_info_hashfn(fi)]);
1559 	if (fi->fib_prefsrc) {
1560 		struct hlist_head *head;
1561 
1562 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1563 		hlist_add_head(&fi->fib_lhash, head);
1564 	}
1565 	if (fi->nh) {
1566 		list_add(&fi->nh_list, &nh->fi_list);
1567 	} else {
1568 		change_nexthops(fi) {
1569 			struct hlist_head *head;
1570 			unsigned int hash;
1571 
1572 			if (!nexthop_nh->fib_nh_dev)
1573 				continue;
1574 			hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1575 			head = &fib_info_devhash[hash];
1576 			hlist_add_head(&nexthop_nh->nh_hash, head);
1577 		} endfor_nexthops(fi)
1578 	}
1579 	spin_unlock_bh(&fib_info_lock);
1580 	return fi;
1581 
1582 err_inval:
1583 	err = -EINVAL;
1584 
1585 failure:
1586 	if (fi) {
1587 		fi->fib_dead = 1;
1588 		free_fib_info(fi);
1589 	}
1590 
1591 	return ERR_PTR(err);
1592 }
1593 
1594 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1595 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1596 {
1597 	if (nhc->nhc_flags & RTNH_F_DEAD)
1598 		*flags |= RTNH_F_DEAD;
1599 
1600 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1601 		*flags |= RTNH_F_LINKDOWN;
1602 
1603 		rcu_read_lock();
1604 		switch (nhc->nhc_family) {
1605 		case AF_INET:
1606 			if (ip_ignore_linkdown(nhc->nhc_dev))
1607 				*flags |= RTNH_F_DEAD;
1608 			break;
1609 		case AF_INET6:
1610 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1611 				*flags |= RTNH_F_DEAD;
1612 			break;
1613 		}
1614 		rcu_read_unlock();
1615 	}
1616 
1617 	switch (nhc->nhc_gw_family) {
1618 	case AF_INET:
1619 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1620 			goto nla_put_failure;
1621 		break;
1622 	case AF_INET6:
1623 		/* if gateway family does not match nexthop family
1624 		 * gateway is encoded as RTA_VIA
1625 		 */
1626 		if (rt_family != nhc->nhc_gw_family) {
1627 			int alen = sizeof(struct in6_addr);
1628 			struct nlattr *nla;
1629 			struct rtvia *via;
1630 
1631 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1632 			if (!nla)
1633 				goto nla_put_failure;
1634 
1635 			via = nla_data(nla);
1636 			via->rtvia_family = AF_INET6;
1637 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1638 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1639 					    &nhc->nhc_gw.ipv6) < 0) {
1640 			goto nla_put_failure;
1641 		}
1642 		break;
1643 	}
1644 
1645 	*flags |= (nhc->nhc_flags &
1646 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1647 
1648 	if (!skip_oif && nhc->nhc_dev &&
1649 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1650 		goto nla_put_failure;
1651 
1652 	if (nhc->nhc_lwtstate &&
1653 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1654 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1655 		goto nla_put_failure;
1656 
1657 	return 0;
1658 
1659 nla_put_failure:
1660 	return -EMSGSIZE;
1661 }
1662 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1663 
1664 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1665 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1666 		    int nh_weight, u8 rt_family)
1667 {
1668 	const struct net_device *dev = nhc->nhc_dev;
1669 	struct rtnexthop *rtnh;
1670 	unsigned char flags = 0;
1671 
1672 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1673 	if (!rtnh)
1674 		goto nla_put_failure;
1675 
1676 	rtnh->rtnh_hops = nh_weight - 1;
1677 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1678 
1679 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1680 		goto nla_put_failure;
1681 
1682 	rtnh->rtnh_flags = flags;
1683 
1684 	/* length of rtnetlink header + attributes */
1685 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1686 
1687 	return 0;
1688 
1689 nla_put_failure:
1690 	return -EMSGSIZE;
1691 }
1692 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1693 #endif
1694 
1695 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1696 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1697 {
1698 	struct nlattr *mp;
1699 
1700 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1701 	if (!mp)
1702 		goto nla_put_failure;
1703 
1704 	if (unlikely(fi->nh)) {
1705 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1706 			goto nla_put_failure;
1707 		goto mp_end;
1708 	}
1709 
1710 	for_nexthops(fi) {
1711 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1712 				    AF_INET) < 0)
1713 			goto nla_put_failure;
1714 #ifdef CONFIG_IP_ROUTE_CLASSID
1715 		if (nh->nh_tclassid &&
1716 		    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1717 			goto nla_put_failure;
1718 #endif
1719 	} endfor_nexthops(fi);
1720 
1721 mp_end:
1722 	nla_nest_end(skb, mp);
1723 
1724 	return 0;
1725 
1726 nla_put_failure:
1727 	return -EMSGSIZE;
1728 }
1729 #else
1730 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1731 {
1732 	return 0;
1733 }
1734 #endif
1735 
1736 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1737 		  const struct fib_rt_info *fri, unsigned int flags)
1738 {
1739 	unsigned int nhs = fib_info_num_path(fri->fi);
1740 	struct fib_info *fi = fri->fi;
1741 	u32 tb_id = fri->tb_id;
1742 	struct nlmsghdr *nlh;
1743 	struct rtmsg *rtm;
1744 
1745 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1746 	if (!nlh)
1747 		return -EMSGSIZE;
1748 
1749 	rtm = nlmsg_data(nlh);
1750 	rtm->rtm_family = AF_INET;
1751 	rtm->rtm_dst_len = fri->dst_len;
1752 	rtm->rtm_src_len = 0;
1753 	rtm->rtm_tos = fri->tos;
1754 	if (tb_id < 256)
1755 		rtm->rtm_table = tb_id;
1756 	else
1757 		rtm->rtm_table = RT_TABLE_COMPAT;
1758 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1759 		goto nla_put_failure;
1760 	rtm->rtm_type = fri->type;
1761 	rtm->rtm_flags = fi->fib_flags;
1762 	rtm->rtm_scope = fi->fib_scope;
1763 	rtm->rtm_protocol = fi->fib_protocol;
1764 
1765 	if (rtm->rtm_dst_len &&
1766 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1767 		goto nla_put_failure;
1768 	if (fi->fib_priority &&
1769 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1770 		goto nla_put_failure;
1771 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1772 		goto nla_put_failure;
1773 
1774 	if (fi->fib_prefsrc &&
1775 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1776 		goto nla_put_failure;
1777 
1778 	if (fi->nh) {
1779 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1780 			goto nla_put_failure;
1781 		if (nexthop_is_blackhole(fi->nh))
1782 			rtm->rtm_type = RTN_BLACKHOLE;
1783 		if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode)
1784 			goto offload;
1785 	}
1786 
1787 	if (nhs == 1) {
1788 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1789 		unsigned char flags = 0;
1790 
1791 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1792 			goto nla_put_failure;
1793 
1794 		rtm->rtm_flags = flags;
1795 #ifdef CONFIG_IP_ROUTE_CLASSID
1796 		if (nhc->nhc_family == AF_INET) {
1797 			struct fib_nh *nh;
1798 
1799 			nh = container_of(nhc, struct fib_nh, nh_common);
1800 			if (nh->nh_tclassid &&
1801 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1802 				goto nla_put_failure;
1803 		}
1804 #endif
1805 	} else {
1806 		if (fib_add_multipath(skb, fi) < 0)
1807 			goto nla_put_failure;
1808 	}
1809 
1810 offload:
1811 	if (fri->offload)
1812 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1813 	if (fri->trap)
1814 		rtm->rtm_flags |= RTM_F_TRAP;
1815 	if (fri->offload_failed)
1816 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1817 
1818 	nlmsg_end(skb, nlh);
1819 	return 0;
1820 
1821 nla_put_failure:
1822 	nlmsg_cancel(skb, nlh);
1823 	return -EMSGSIZE;
1824 }
1825 
1826 /*
1827  * Update FIB if:
1828  * - local address disappeared -> we must delete all the entries
1829  *   referring to it.
1830  * - device went down -> we must shutdown all nexthops going via it.
1831  */
1832 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1833 {
1834 	int ret = 0;
1835 	unsigned int hash = fib_laddr_hashfn(local);
1836 	struct hlist_head *head = &fib_info_laddrhash[hash];
1837 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1838 	struct net *net = dev_net(dev);
1839 	struct fib_info *fi;
1840 
1841 	if (!fib_info_laddrhash || local == 0)
1842 		return 0;
1843 
1844 	hlist_for_each_entry(fi, head, fib_lhash) {
1845 		if (!net_eq(fi->fib_net, net) ||
1846 		    fi->fib_tb_id != tb_id)
1847 			continue;
1848 		if (fi->fib_prefsrc == local) {
1849 			fi->fib_flags |= RTNH_F_DEAD;
1850 			ret++;
1851 		}
1852 	}
1853 	return ret;
1854 }
1855 
1856 static int call_fib_nh_notifiers(struct fib_nh *nh,
1857 				 enum fib_event_type event_type)
1858 {
1859 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1860 	struct fib_nh_notifier_info info = {
1861 		.fib_nh = nh,
1862 	};
1863 
1864 	switch (event_type) {
1865 	case FIB_EVENT_NH_ADD:
1866 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1867 			break;
1868 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1869 			break;
1870 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1871 					   &info.info);
1872 	case FIB_EVENT_NH_DEL:
1873 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1874 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1875 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1876 						   event_type, &info.info);
1877 	default:
1878 		break;
1879 	}
1880 
1881 	return NOTIFY_DONE;
1882 }
1883 
1884 /* Update the PMTU of exceptions when:
1885  * - the new MTU of the first hop becomes smaller than the PMTU
1886  * - the old MTU was the same as the PMTU, and it limited discovery of
1887  *   larger MTUs on the path. With that limit raised, we can now
1888  *   discover larger MTUs
1889  * A special case is locked exceptions, for which the PMTU is smaller
1890  * than the minimal accepted PMTU:
1891  * - if the new MTU is greater than the PMTU, don't make any change
1892  * - otherwise, unlock and set PMTU
1893  */
1894 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1895 {
1896 	struct fnhe_hash_bucket *bucket;
1897 	int i;
1898 
1899 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1900 	if (!bucket)
1901 		return;
1902 
1903 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1904 		struct fib_nh_exception *fnhe;
1905 
1906 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1907 		     fnhe;
1908 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1909 			if (fnhe->fnhe_mtu_locked) {
1910 				if (new <= fnhe->fnhe_pmtu) {
1911 					fnhe->fnhe_pmtu = new;
1912 					fnhe->fnhe_mtu_locked = false;
1913 				}
1914 			} else if (new < fnhe->fnhe_pmtu ||
1915 				   orig == fnhe->fnhe_pmtu) {
1916 				fnhe->fnhe_pmtu = new;
1917 			}
1918 		}
1919 	}
1920 }
1921 
1922 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1923 {
1924 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1925 	struct hlist_head *head = &fib_info_devhash[hash];
1926 	struct fib_nh *nh;
1927 
1928 	hlist_for_each_entry(nh, head, nh_hash) {
1929 		if (nh->fib_nh_dev == dev)
1930 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1931 	}
1932 }
1933 
1934 /* Event              force Flags           Description
1935  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1936  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1937  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1938  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1939  *
1940  * only used when fib_nh is built into fib_info
1941  */
1942 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1943 {
1944 	int ret = 0;
1945 	int scope = RT_SCOPE_NOWHERE;
1946 	struct fib_info *prev_fi = NULL;
1947 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1948 	struct hlist_head *head = &fib_info_devhash[hash];
1949 	struct fib_nh *nh;
1950 
1951 	if (force)
1952 		scope = -1;
1953 
1954 	hlist_for_each_entry(nh, head, nh_hash) {
1955 		struct fib_info *fi = nh->nh_parent;
1956 		int dead;
1957 
1958 		BUG_ON(!fi->fib_nhs);
1959 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1960 			continue;
1961 		prev_fi = fi;
1962 		dead = 0;
1963 		change_nexthops(fi) {
1964 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1965 				dead++;
1966 			else if (nexthop_nh->fib_nh_dev == dev &&
1967 				 nexthop_nh->fib_nh_scope != scope) {
1968 				switch (event) {
1969 				case NETDEV_DOWN:
1970 				case NETDEV_UNREGISTER:
1971 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1972 					fallthrough;
1973 				case NETDEV_CHANGE:
1974 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1975 					break;
1976 				}
1977 				call_fib_nh_notifiers(nexthop_nh,
1978 						      FIB_EVENT_NH_DEL);
1979 				dead++;
1980 			}
1981 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1982 			if (event == NETDEV_UNREGISTER &&
1983 			    nexthop_nh->fib_nh_dev == dev) {
1984 				dead = fi->fib_nhs;
1985 				break;
1986 			}
1987 #endif
1988 		} endfor_nexthops(fi)
1989 		if (dead == fi->fib_nhs) {
1990 			switch (event) {
1991 			case NETDEV_DOWN:
1992 			case NETDEV_UNREGISTER:
1993 				fi->fib_flags |= RTNH_F_DEAD;
1994 				fallthrough;
1995 			case NETDEV_CHANGE:
1996 				fi->fib_flags |= RTNH_F_LINKDOWN;
1997 				break;
1998 			}
1999 			ret++;
2000 		}
2001 
2002 		fib_rebalance(fi);
2003 	}
2004 
2005 	return ret;
2006 }
2007 
2008 /* Must be invoked inside of an RCU protected region.  */
2009 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2010 {
2011 	struct fib_info *fi = NULL, *last_resort = NULL;
2012 	struct hlist_head *fa_head = res->fa_head;
2013 	struct fib_table *tb = res->table;
2014 	u8 slen = 32 - res->prefixlen;
2015 	int order = -1, last_idx = -1;
2016 	struct fib_alias *fa, *fa1 = NULL;
2017 	u32 last_prio = res->fi->fib_priority;
2018 	u8 last_tos = 0;
2019 
2020 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2021 		struct fib_info *next_fi = fa->fa_info;
2022 		struct fib_nh_common *nhc;
2023 
2024 		if (fa->fa_slen != slen)
2025 			continue;
2026 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2027 			continue;
2028 		if (fa->tb_id != tb->tb_id)
2029 			continue;
2030 		if (next_fi->fib_priority > last_prio &&
2031 		    fa->fa_tos == last_tos) {
2032 			if (last_tos)
2033 				continue;
2034 			break;
2035 		}
2036 		if (next_fi->fib_flags & RTNH_F_DEAD)
2037 			continue;
2038 		last_tos = fa->fa_tos;
2039 		last_prio = next_fi->fib_priority;
2040 
2041 		if (next_fi->fib_scope != res->scope ||
2042 		    fa->fa_type != RTN_UNICAST)
2043 			continue;
2044 
2045 		nhc = fib_info_nhc(next_fi, 0);
2046 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2047 			continue;
2048 
2049 		fib_alias_accessed(fa);
2050 
2051 		if (!fi) {
2052 			if (next_fi != res->fi)
2053 				break;
2054 			fa1 = fa;
2055 		} else if (!fib_detect_death(fi, order, &last_resort,
2056 					     &last_idx, fa1->fa_default)) {
2057 			fib_result_assign(res, fi);
2058 			fa1->fa_default = order;
2059 			goto out;
2060 		}
2061 		fi = next_fi;
2062 		order++;
2063 	}
2064 
2065 	if (order <= 0 || !fi) {
2066 		if (fa1)
2067 			fa1->fa_default = -1;
2068 		goto out;
2069 	}
2070 
2071 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2072 			      fa1->fa_default)) {
2073 		fib_result_assign(res, fi);
2074 		fa1->fa_default = order;
2075 		goto out;
2076 	}
2077 
2078 	if (last_idx >= 0)
2079 		fib_result_assign(res, last_resort);
2080 	fa1->fa_default = last_idx;
2081 out:
2082 	return;
2083 }
2084 
2085 /*
2086  * Dead device goes up. We wake up dead nexthops.
2087  * It takes sense only on multipath routes.
2088  *
2089  * only used when fib_nh is built into fib_info
2090  */
2091 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2092 {
2093 	struct fib_info *prev_fi;
2094 	unsigned int hash;
2095 	struct hlist_head *head;
2096 	struct fib_nh *nh;
2097 	int ret;
2098 
2099 	if (!(dev->flags & IFF_UP))
2100 		return 0;
2101 
2102 	if (nh_flags & RTNH_F_DEAD) {
2103 		unsigned int flags = dev_get_flags(dev);
2104 
2105 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2106 			nh_flags |= RTNH_F_LINKDOWN;
2107 	}
2108 
2109 	prev_fi = NULL;
2110 	hash = fib_devindex_hashfn(dev->ifindex);
2111 	head = &fib_info_devhash[hash];
2112 	ret = 0;
2113 
2114 	hlist_for_each_entry(nh, head, nh_hash) {
2115 		struct fib_info *fi = nh->nh_parent;
2116 		int alive;
2117 
2118 		BUG_ON(!fi->fib_nhs);
2119 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2120 			continue;
2121 
2122 		prev_fi = fi;
2123 		alive = 0;
2124 		change_nexthops(fi) {
2125 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2126 				alive++;
2127 				continue;
2128 			}
2129 			if (!nexthop_nh->fib_nh_dev ||
2130 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2131 				continue;
2132 			if (nexthop_nh->fib_nh_dev != dev ||
2133 			    !__in_dev_get_rtnl(dev))
2134 				continue;
2135 			alive++;
2136 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2137 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2138 		} endfor_nexthops(fi)
2139 
2140 		if (alive > 0) {
2141 			fi->fib_flags &= ~nh_flags;
2142 			ret++;
2143 		}
2144 
2145 		fib_rebalance(fi);
2146 	}
2147 
2148 	return ret;
2149 }
2150 
2151 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2152 static bool fib_good_nh(const struct fib_nh *nh)
2153 {
2154 	int state = NUD_REACHABLE;
2155 
2156 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2157 		struct neighbour *n;
2158 
2159 		rcu_read_lock_bh();
2160 
2161 		if (likely(nh->fib_nh_gw_family == AF_INET))
2162 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2163 						   (__force u32)nh->fib_nh_gw4);
2164 		else if (nh->fib_nh_gw_family == AF_INET6)
2165 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2166 							   &nh->fib_nh_gw6);
2167 		else
2168 			n = NULL;
2169 		if (n)
2170 			state = n->nud_state;
2171 
2172 		rcu_read_unlock_bh();
2173 	}
2174 
2175 	return !!(state & NUD_VALID);
2176 }
2177 
2178 void fib_select_multipath(struct fib_result *res, int hash)
2179 {
2180 	struct fib_info *fi = res->fi;
2181 	struct net *net = fi->fib_net;
2182 	bool first = false;
2183 
2184 	if (unlikely(res->fi->nh)) {
2185 		nexthop_path_fib_result(res, hash);
2186 		return;
2187 	}
2188 
2189 	change_nexthops(fi) {
2190 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2191 			if (!fib_good_nh(nexthop_nh))
2192 				continue;
2193 			if (!first) {
2194 				res->nh_sel = nhsel;
2195 				res->nhc = &nexthop_nh->nh_common;
2196 				first = true;
2197 			}
2198 		}
2199 
2200 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2201 			continue;
2202 
2203 		res->nh_sel = nhsel;
2204 		res->nhc = &nexthop_nh->nh_common;
2205 		return;
2206 	} endfor_nexthops(fi);
2207 }
2208 #endif
2209 
2210 void fib_select_path(struct net *net, struct fib_result *res,
2211 		     struct flowi4 *fl4, const struct sk_buff *skb)
2212 {
2213 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2214 		goto check_saddr;
2215 
2216 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2217 	if (fib_info_num_path(res->fi) > 1) {
2218 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2219 
2220 		fib_select_multipath(res, h);
2221 	}
2222 	else
2223 #endif
2224 	if (!res->prefixlen &&
2225 	    res->table->tb_num_default > 1 &&
2226 	    res->type == RTN_UNICAST)
2227 		fib_select_default(fl4, res);
2228 
2229 check_saddr:
2230 	if (!fl4->saddr)
2231 		fl4->saddr = fib_result_prefsrc(net, res);
2232 }
2233