xref: /linux/net/ipv4/fib_semantics.c (revision f788bb2fb915a79eae0a160580766ba52984d61f)
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 static inline 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 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
525 	if (err < 0) {
526 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
527 		WARN_ON(err == -EMSGSIZE);
528 		kfree_skb(skb);
529 		goto errout;
530 	}
531 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
532 		    info->nlh, GFP_KERNEL);
533 	return;
534 errout:
535 	if (err < 0)
536 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
537 }
538 
539 static int fib_detect_death(struct fib_info *fi, int order,
540 			    struct fib_info **last_resort, int *last_idx,
541 			    int dflt)
542 {
543 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
544 	struct neighbour *n;
545 	int state = NUD_NONE;
546 
547 	if (likely(nhc->nhc_gw_family == AF_INET))
548 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
549 	else if (nhc->nhc_gw_family == AF_INET6)
550 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
551 				 nhc->nhc_dev);
552 	else
553 		n = NULL;
554 
555 	if (n) {
556 		state = n->nud_state;
557 		neigh_release(n);
558 	} else {
559 		return 0;
560 	}
561 	if (state == NUD_REACHABLE)
562 		return 0;
563 	if ((state & NUD_VALID) && order != dflt)
564 		return 0;
565 	if ((state & NUD_VALID) ||
566 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
567 		*last_resort = fi;
568 		*last_idx = order;
569 	}
570 	return 1;
571 }
572 
573 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
574 		       struct nlattr *encap, u16 encap_type,
575 		       void *cfg, gfp_t gfp_flags,
576 		       struct netlink_ext_ack *extack)
577 {
578 	int err;
579 
580 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
581 						    gfp_flags);
582 	if (!nhc->nhc_pcpu_rth_output)
583 		return -ENOMEM;
584 
585 	if (encap) {
586 		struct lwtunnel_state *lwtstate;
587 
588 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
589 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
590 			err = -EINVAL;
591 			goto lwt_failure;
592 		}
593 		err = lwtunnel_build_state(net, encap_type, encap,
594 					   nhc->nhc_family, cfg, &lwtstate,
595 					   extack);
596 		if (err)
597 			goto lwt_failure;
598 
599 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
600 	}
601 
602 	return 0;
603 
604 lwt_failure:
605 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
606 	nhc->nhc_pcpu_rth_output = NULL;
607 	return err;
608 }
609 EXPORT_SYMBOL_GPL(fib_nh_common_init);
610 
611 int fib_nh_init(struct net *net, struct fib_nh *nh,
612 		struct fib_config *cfg, int nh_weight,
613 		struct netlink_ext_ack *extack)
614 {
615 	int err;
616 
617 	nh->fib_nh_family = AF_INET;
618 
619 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
620 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
621 	if (err)
622 		return err;
623 
624 	nh->fib_nh_oif = cfg->fc_oif;
625 	nh->fib_nh_gw_family = cfg->fc_gw_family;
626 	if (cfg->fc_gw_family == AF_INET)
627 		nh->fib_nh_gw4 = cfg->fc_gw4;
628 	else if (cfg->fc_gw_family == AF_INET6)
629 		nh->fib_nh_gw6 = cfg->fc_gw6;
630 
631 	nh->fib_nh_flags = cfg->fc_flags;
632 
633 #ifdef CONFIG_IP_ROUTE_CLASSID
634 	nh->nh_tclassid = cfg->fc_flow;
635 	if (nh->nh_tclassid)
636 		net->ipv4.fib_num_tclassid_users++;
637 #endif
638 #ifdef CONFIG_IP_ROUTE_MULTIPATH
639 	nh->fib_nh_weight = nh_weight;
640 #endif
641 	return 0;
642 }
643 
644 #ifdef CONFIG_IP_ROUTE_MULTIPATH
645 
646 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
647 			      struct netlink_ext_ack *extack)
648 {
649 	int nhs = 0;
650 
651 	while (rtnh_ok(rtnh, remaining)) {
652 		nhs++;
653 		rtnh = rtnh_next(rtnh, &remaining);
654 	}
655 
656 	/* leftover implies invalid nexthop configuration, discard it */
657 	if (remaining > 0) {
658 		NL_SET_ERR_MSG(extack,
659 			       "Invalid nexthop configuration - extra data after nexthops");
660 		nhs = 0;
661 	}
662 
663 	return nhs;
664 }
665 
666 /* only called when fib_nh is integrated into fib_info */
667 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
668 		       int remaining, struct fib_config *cfg,
669 		       struct netlink_ext_ack *extack)
670 {
671 	struct net *net = fi->fib_net;
672 	struct fib_config fib_cfg;
673 	struct fib_nh *nh;
674 	int ret;
675 
676 	change_nexthops(fi) {
677 		int attrlen;
678 
679 		memset(&fib_cfg, 0, sizeof(fib_cfg));
680 
681 		if (!rtnh_ok(rtnh, remaining)) {
682 			NL_SET_ERR_MSG(extack,
683 				       "Invalid nexthop configuration - extra data after nexthop");
684 			return -EINVAL;
685 		}
686 
687 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
688 			NL_SET_ERR_MSG(extack,
689 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
690 			return -EINVAL;
691 		}
692 
693 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
694 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
695 
696 		attrlen = rtnh_attrlen(rtnh);
697 		if (attrlen > 0) {
698 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
699 
700 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
701 			nlav = nla_find(attrs, attrlen, RTA_VIA);
702 			if (nla && nlav) {
703 				NL_SET_ERR_MSG(extack,
704 					       "Nexthop configuration can not contain both GATEWAY and VIA");
705 				return -EINVAL;
706 			}
707 			if (nla) {
708 				fib_cfg.fc_gw4 = nla_get_in_addr(nla);
709 				if (fib_cfg.fc_gw4)
710 					fib_cfg.fc_gw_family = AF_INET;
711 			} else if (nlav) {
712 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
713 				if (ret)
714 					goto errout;
715 			}
716 
717 			nla = nla_find(attrs, attrlen, RTA_FLOW);
718 			if (nla)
719 				fib_cfg.fc_flow = nla_get_u32(nla);
720 
721 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
722 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
723 			if (nla)
724 				fib_cfg.fc_encap_type = nla_get_u16(nla);
725 		}
726 
727 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
728 				  rtnh->rtnh_hops + 1, extack);
729 		if (ret)
730 			goto errout;
731 
732 		rtnh = rtnh_next(rtnh, &remaining);
733 	} endfor_nexthops(fi);
734 
735 	ret = -EINVAL;
736 	nh = fib_info_nh(fi, 0);
737 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
738 		NL_SET_ERR_MSG(extack,
739 			       "Nexthop device index does not match RTA_OIF");
740 		goto errout;
741 	}
742 	if (cfg->fc_gw_family) {
743 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
744 		    (cfg->fc_gw_family == AF_INET &&
745 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
746 		    (cfg->fc_gw_family == AF_INET6 &&
747 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
748 			NL_SET_ERR_MSG(extack,
749 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
750 			goto errout;
751 		}
752 	}
753 #ifdef CONFIG_IP_ROUTE_CLASSID
754 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
755 		NL_SET_ERR_MSG(extack,
756 			       "Nexthop class id does not match RTA_FLOW");
757 		goto errout;
758 	}
759 #endif
760 	ret = 0;
761 errout:
762 	return ret;
763 }
764 
765 /* only called when fib_nh is integrated into fib_info */
766 static void fib_rebalance(struct fib_info *fi)
767 {
768 	int total;
769 	int w;
770 
771 	if (fib_info_num_path(fi) < 2)
772 		return;
773 
774 	total = 0;
775 	for_nexthops(fi) {
776 		if (nh->fib_nh_flags & RTNH_F_DEAD)
777 			continue;
778 
779 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
780 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
781 			continue;
782 
783 		total += nh->fib_nh_weight;
784 	} endfor_nexthops(fi);
785 
786 	w = 0;
787 	change_nexthops(fi) {
788 		int upper_bound;
789 
790 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
791 			upper_bound = -1;
792 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
793 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
794 			upper_bound = -1;
795 		} else {
796 			w += nexthop_nh->fib_nh_weight;
797 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
798 							    total) - 1;
799 		}
800 
801 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
802 	} endfor_nexthops(fi);
803 }
804 #else /* CONFIG_IP_ROUTE_MULTIPATH */
805 
806 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
807 		       int remaining, struct fib_config *cfg,
808 		       struct netlink_ext_ack *extack)
809 {
810 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
811 
812 	return -EINVAL;
813 }
814 
815 #define fib_rebalance(fi) do { } while (0)
816 
817 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
818 
819 static int fib_encap_match(struct net *net, u16 encap_type,
820 			   struct nlattr *encap,
821 			   const struct fib_nh *nh,
822 			   const struct fib_config *cfg,
823 			   struct netlink_ext_ack *extack)
824 {
825 	struct lwtunnel_state *lwtstate;
826 	int ret, result = 0;
827 
828 	if (encap_type == LWTUNNEL_ENCAP_NONE)
829 		return 0;
830 
831 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
832 				   cfg, &lwtstate, extack);
833 	if (!ret) {
834 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
835 		lwtstate_free(lwtstate);
836 	}
837 
838 	return result;
839 }
840 
841 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
842 		 struct netlink_ext_ack *extack)
843 {
844 #ifdef CONFIG_IP_ROUTE_MULTIPATH
845 	struct rtnexthop *rtnh;
846 	int remaining;
847 #endif
848 
849 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
850 		return 1;
851 
852 	if (cfg->fc_nh_id) {
853 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
854 			return 0;
855 		return 1;
856 	}
857 
858 	if (cfg->fc_oif || cfg->fc_gw_family) {
859 		struct fib_nh *nh = fib_info_nh(fi, 0);
860 
861 		if (cfg->fc_encap) {
862 			if (fib_encap_match(net, cfg->fc_encap_type,
863 					    cfg->fc_encap, nh, cfg, extack))
864 				return 1;
865 		}
866 #ifdef CONFIG_IP_ROUTE_CLASSID
867 		if (cfg->fc_flow &&
868 		    cfg->fc_flow != nh->nh_tclassid)
869 			return 1;
870 #endif
871 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
872 		    (cfg->fc_gw_family &&
873 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
874 			return 1;
875 
876 		if (cfg->fc_gw_family == AF_INET &&
877 		    cfg->fc_gw4 != nh->fib_nh_gw4)
878 			return 1;
879 
880 		if (cfg->fc_gw_family == AF_INET6 &&
881 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
882 			return 1;
883 
884 		return 0;
885 	}
886 
887 #ifdef CONFIG_IP_ROUTE_MULTIPATH
888 	if (!cfg->fc_mp)
889 		return 0;
890 
891 	rtnh = cfg->fc_mp;
892 	remaining = cfg->fc_mp_len;
893 
894 	for_nexthops(fi) {
895 		int attrlen;
896 
897 		if (!rtnh_ok(rtnh, remaining))
898 			return -EINVAL;
899 
900 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
901 			return 1;
902 
903 		attrlen = rtnh_attrlen(rtnh);
904 		if (attrlen > 0) {
905 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
906 
907 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
908 			nlav = nla_find(attrs, attrlen, RTA_VIA);
909 			if (nla && nlav) {
910 				NL_SET_ERR_MSG(extack,
911 					       "Nexthop configuration can not contain both GATEWAY and VIA");
912 				return -EINVAL;
913 			}
914 
915 			if (nla) {
916 				if (nh->fib_nh_gw_family != AF_INET ||
917 				    nla_get_in_addr(nla) != nh->fib_nh_gw4)
918 					return 1;
919 			} else if (nlav) {
920 				struct fib_config cfg2;
921 				int err;
922 
923 				err = fib_gw_from_via(&cfg2, nlav, extack);
924 				if (err)
925 					return err;
926 
927 				switch (nh->fib_nh_gw_family) {
928 				case AF_INET:
929 					if (cfg2.fc_gw_family != AF_INET ||
930 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
931 						return 1;
932 					break;
933 				case AF_INET6:
934 					if (cfg2.fc_gw_family != AF_INET6 ||
935 					    ipv6_addr_cmp(&cfg2.fc_gw6,
936 							  &nh->fib_nh_gw6))
937 						return 1;
938 					break;
939 				}
940 			}
941 
942 #ifdef CONFIG_IP_ROUTE_CLASSID
943 			nla = nla_find(attrs, attrlen, RTA_FLOW);
944 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
945 				return 1;
946 #endif
947 		}
948 
949 		rtnh = rtnh_next(rtnh, &remaining);
950 	} endfor_nexthops(fi);
951 #endif
952 	return 0;
953 }
954 
955 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
956 {
957 	struct nlattr *nla;
958 	int remaining;
959 
960 	if (!cfg->fc_mx)
961 		return true;
962 
963 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
964 		int type = nla_type(nla);
965 		u32 fi_val, val;
966 
967 		if (!type)
968 			continue;
969 		if (type > RTAX_MAX)
970 			return false;
971 
972 		if (type == RTAX_CC_ALGO) {
973 			char tmp[TCP_CA_NAME_MAX];
974 			bool ecn_ca = false;
975 
976 			nla_strlcpy(tmp, nla, sizeof(tmp));
977 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
978 		} else {
979 			if (nla_len(nla) != sizeof(u32))
980 				return false;
981 			val = nla_get_u32(nla);
982 		}
983 
984 		fi_val = fi->fib_metrics->metrics[type - 1];
985 		if (type == RTAX_FEATURES)
986 			fi_val &= ~DST_FEATURE_ECN_CA;
987 
988 		if (fi_val != val)
989 			return false;
990 	}
991 
992 	return true;
993 }
994 
995 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
996 			      u32 table, struct netlink_ext_ack *extack)
997 {
998 	struct fib6_config cfg = {
999 		.fc_table = table,
1000 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1001 		.fc_ifindex = nh->fib_nh_oif,
1002 		.fc_gateway = nh->fib_nh_gw6,
1003 	};
1004 	struct fib6_nh fib6_nh = {};
1005 	int err;
1006 
1007 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1008 	if (!err) {
1009 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1010 		dev_hold(nh->fib_nh_dev);
1011 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1012 		nh->fib_nh_scope = RT_SCOPE_LINK;
1013 
1014 		ipv6_stub->fib6_nh_release(&fib6_nh);
1015 	}
1016 
1017 	return err;
1018 }
1019 
1020 /*
1021  * Picture
1022  * -------
1023  *
1024  * Semantics of nexthop is very messy by historical reasons.
1025  * We have to take into account, that:
1026  * a) gateway can be actually local interface address,
1027  *    so that gatewayed route is direct.
1028  * b) gateway must be on-link address, possibly
1029  *    described not by an ifaddr, but also by a direct route.
1030  * c) If both gateway and interface are specified, they should not
1031  *    contradict.
1032  * d) If we use tunnel routes, gateway could be not on-link.
1033  *
1034  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1035  * results in pretty ugly and hairy code with obscure logic.
1036  *
1037  * I chose to generalized it instead, so that the size
1038  * of code does not increase practically, but it becomes
1039  * much more general.
1040  * Every prefix is assigned a "scope" value: "host" is local address,
1041  * "link" is direct route,
1042  * [ ... "site" ... "interior" ... ]
1043  * and "universe" is true gateway route with global meaning.
1044  *
1045  * Every prefix refers to a set of "nexthop"s (gw, oif),
1046  * where gw must have narrower scope. This recursion stops
1047  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1048  * which means that gw is forced to be on link.
1049  *
1050  * Code is still hairy, but now it is apparently logically
1051  * consistent and very flexible. F.e. as by-product it allows
1052  * to co-exists in peace independent exterior and interior
1053  * routing processes.
1054  *
1055  * Normally it looks as following.
1056  *
1057  * {universe prefix}  -> (gw, oif) [scope link]
1058  *		  |
1059  *		  |-> {link prefix} -> (gw, oif) [scope local]
1060  *					|
1061  *					|-> {local prefix} (terminal node)
1062  */
1063 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1064 			      u8 scope, struct netlink_ext_ack *extack)
1065 {
1066 	struct net_device *dev;
1067 	struct fib_result res;
1068 	int err = 0;
1069 
1070 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1071 		unsigned int addr_type;
1072 
1073 		if (scope >= RT_SCOPE_LINK) {
1074 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1075 			return -EINVAL;
1076 		}
1077 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1078 		if (!dev) {
1079 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1080 			return -ENODEV;
1081 		}
1082 		if (!(dev->flags & IFF_UP)) {
1083 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1084 			return -ENETDOWN;
1085 		}
1086 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1087 		if (addr_type != RTN_UNICAST) {
1088 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1089 			return -EINVAL;
1090 		}
1091 		if (!netif_carrier_ok(dev))
1092 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1093 		nh->fib_nh_dev = dev;
1094 		dev_hold(dev);
1095 		nh->fib_nh_scope = RT_SCOPE_LINK;
1096 		return 0;
1097 	}
1098 	rcu_read_lock();
1099 	{
1100 		struct fib_table *tbl = NULL;
1101 		struct flowi4 fl4 = {
1102 			.daddr = nh->fib_nh_gw4,
1103 			.flowi4_scope = scope + 1,
1104 			.flowi4_oif = nh->fib_nh_oif,
1105 			.flowi4_iif = LOOPBACK_IFINDEX,
1106 		};
1107 
1108 		/* It is not necessary, but requires a bit of thinking */
1109 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1110 			fl4.flowi4_scope = RT_SCOPE_LINK;
1111 
1112 		if (table && table != RT_TABLE_MAIN)
1113 			tbl = fib_get_table(net, table);
1114 
1115 		if (tbl)
1116 			err = fib_table_lookup(tbl, &fl4, &res,
1117 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1118 					       FIB_LOOKUP_NOREF);
1119 
1120 		/* on error or if no table given do full lookup. This
1121 		 * is needed for example when nexthops are in the local
1122 		 * table rather than the given table
1123 		 */
1124 		if (!tbl || err) {
1125 			err = fib_lookup(net, &fl4, &res,
1126 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1127 		}
1128 
1129 		if (err) {
1130 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1131 			goto out;
1132 		}
1133 	}
1134 
1135 	err = -EINVAL;
1136 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1137 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1138 		goto out;
1139 	}
1140 	nh->fib_nh_scope = res.scope;
1141 	nh->fib_nh_oif = FIB_RES_OIF(res);
1142 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1143 	if (!dev) {
1144 		NL_SET_ERR_MSG(extack,
1145 			       "No egress device for nexthop gateway");
1146 		goto out;
1147 	}
1148 	dev_hold(dev);
1149 	if (!netif_carrier_ok(dev))
1150 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1151 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1152 out:
1153 	rcu_read_unlock();
1154 	return err;
1155 }
1156 
1157 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1158 			      struct netlink_ext_ack *extack)
1159 {
1160 	struct in_device *in_dev;
1161 	int err;
1162 
1163 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1164 		NL_SET_ERR_MSG(extack,
1165 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1166 		return -EINVAL;
1167 	}
1168 
1169 	rcu_read_lock();
1170 
1171 	err = -ENODEV;
1172 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1173 	if (!in_dev)
1174 		goto out;
1175 	err = -ENETDOWN;
1176 	if (!(in_dev->dev->flags & IFF_UP)) {
1177 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1178 		goto out;
1179 	}
1180 
1181 	nh->fib_nh_dev = in_dev->dev;
1182 	dev_hold(nh->fib_nh_dev);
1183 	nh->fib_nh_scope = RT_SCOPE_HOST;
1184 	if (!netif_carrier_ok(nh->fib_nh_dev))
1185 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1186 	err = 0;
1187 out:
1188 	rcu_read_unlock();
1189 	return err;
1190 }
1191 
1192 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1193 		 struct netlink_ext_ack *extack)
1194 {
1195 	int err;
1196 
1197 	if (nh->fib_nh_gw_family == AF_INET)
1198 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1199 	else if (nh->fib_nh_gw_family == AF_INET6)
1200 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1201 	else
1202 		err = fib_check_nh_nongw(net, nh, extack);
1203 
1204 	return err;
1205 }
1206 
1207 static inline unsigned int fib_laddr_hashfn(__be32 val)
1208 {
1209 	unsigned int mask = (fib_info_hash_size - 1);
1210 
1211 	return ((__force u32)val ^
1212 		((__force u32)val >> 7) ^
1213 		((__force u32)val >> 14)) & mask;
1214 }
1215 
1216 static struct hlist_head *fib_info_hash_alloc(int bytes)
1217 {
1218 	if (bytes <= PAGE_SIZE)
1219 		return kzalloc(bytes, GFP_KERNEL);
1220 	else
1221 		return (struct hlist_head *)
1222 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1223 					 get_order(bytes));
1224 }
1225 
1226 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1227 {
1228 	if (!hash)
1229 		return;
1230 
1231 	if (bytes <= PAGE_SIZE)
1232 		kfree(hash);
1233 	else
1234 		free_pages((unsigned long) hash, get_order(bytes));
1235 }
1236 
1237 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1238 			       struct hlist_head *new_laddrhash,
1239 			       unsigned int new_size)
1240 {
1241 	struct hlist_head *old_info_hash, *old_laddrhash;
1242 	unsigned int old_size = fib_info_hash_size;
1243 	unsigned int i, bytes;
1244 
1245 	spin_lock_bh(&fib_info_lock);
1246 	old_info_hash = fib_info_hash;
1247 	old_laddrhash = fib_info_laddrhash;
1248 	fib_info_hash_size = new_size;
1249 
1250 	for (i = 0; i < old_size; i++) {
1251 		struct hlist_head *head = &fib_info_hash[i];
1252 		struct hlist_node *n;
1253 		struct fib_info *fi;
1254 
1255 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1256 			struct hlist_head *dest;
1257 			unsigned int new_hash;
1258 
1259 			new_hash = fib_info_hashfn(fi);
1260 			dest = &new_info_hash[new_hash];
1261 			hlist_add_head(&fi->fib_hash, dest);
1262 		}
1263 	}
1264 	fib_info_hash = new_info_hash;
1265 
1266 	for (i = 0; i < old_size; i++) {
1267 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1268 		struct hlist_node *n;
1269 		struct fib_info *fi;
1270 
1271 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1272 			struct hlist_head *ldest;
1273 			unsigned int new_hash;
1274 
1275 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1276 			ldest = &new_laddrhash[new_hash];
1277 			hlist_add_head(&fi->fib_lhash, ldest);
1278 		}
1279 	}
1280 	fib_info_laddrhash = new_laddrhash;
1281 
1282 	spin_unlock_bh(&fib_info_lock);
1283 
1284 	bytes = old_size * sizeof(struct hlist_head *);
1285 	fib_info_hash_free(old_info_hash, bytes);
1286 	fib_info_hash_free(old_laddrhash, bytes);
1287 }
1288 
1289 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1290 				 unsigned char scope)
1291 {
1292 	struct fib_nh *nh;
1293 
1294 	if (nhc->nhc_family != AF_INET)
1295 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1296 
1297 	nh = container_of(nhc, struct fib_nh, nh_common);
1298 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1299 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1300 
1301 	return nh->nh_saddr;
1302 }
1303 
1304 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1305 {
1306 	struct fib_nh_common *nhc = res->nhc;
1307 
1308 	if (res->fi->fib_prefsrc)
1309 		return res->fi->fib_prefsrc;
1310 
1311 	if (nhc->nhc_family == AF_INET) {
1312 		struct fib_nh *nh;
1313 
1314 		nh = container_of(nhc, struct fib_nh, nh_common);
1315 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1316 			return nh->nh_saddr;
1317 	}
1318 
1319 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1320 }
1321 
1322 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1323 {
1324 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1325 	    fib_prefsrc != cfg->fc_dst) {
1326 		u32 tb_id = cfg->fc_table;
1327 		int rc;
1328 
1329 		if (tb_id == RT_TABLE_MAIN)
1330 			tb_id = RT_TABLE_LOCAL;
1331 
1332 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1333 					  fib_prefsrc, tb_id);
1334 
1335 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1336 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1337 						  fib_prefsrc, RT_TABLE_LOCAL);
1338 		}
1339 
1340 		if (rc != RTN_LOCAL)
1341 			return false;
1342 	}
1343 	return true;
1344 }
1345 
1346 struct fib_info *fib_create_info(struct fib_config *cfg,
1347 				 struct netlink_ext_ack *extack)
1348 {
1349 	int err;
1350 	struct fib_info *fi = NULL;
1351 	struct nexthop *nh = NULL;
1352 	struct fib_info *ofi;
1353 	int nhs = 1;
1354 	struct net *net = cfg->fc_nlinfo.nl_net;
1355 
1356 	if (cfg->fc_type > RTN_MAX)
1357 		goto err_inval;
1358 
1359 	/* Fast check to catch the most weird cases */
1360 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1361 		NL_SET_ERR_MSG(extack, "Invalid scope");
1362 		goto err_inval;
1363 	}
1364 
1365 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1366 		NL_SET_ERR_MSG(extack,
1367 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1368 		goto err_inval;
1369 	}
1370 
1371 	if (cfg->fc_nh_id) {
1372 		if (!cfg->fc_mx) {
1373 			fi = fib_find_info_nh(net, cfg);
1374 			if (fi) {
1375 				fi->fib_treeref++;
1376 				return fi;
1377 			}
1378 		}
1379 
1380 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1381 		if (!nh) {
1382 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1383 			goto err_inval;
1384 		}
1385 		nhs = 0;
1386 	}
1387 
1388 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1389 	if (cfg->fc_mp) {
1390 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1391 		if (nhs == 0)
1392 			goto err_inval;
1393 	}
1394 #endif
1395 
1396 	err = -ENOBUFS;
1397 	if (fib_info_cnt >= fib_info_hash_size) {
1398 		unsigned int new_size = fib_info_hash_size << 1;
1399 		struct hlist_head *new_info_hash;
1400 		struct hlist_head *new_laddrhash;
1401 		unsigned int bytes;
1402 
1403 		if (!new_size)
1404 			new_size = 16;
1405 		bytes = new_size * sizeof(struct hlist_head *);
1406 		new_info_hash = fib_info_hash_alloc(bytes);
1407 		new_laddrhash = fib_info_hash_alloc(bytes);
1408 		if (!new_info_hash || !new_laddrhash) {
1409 			fib_info_hash_free(new_info_hash, bytes);
1410 			fib_info_hash_free(new_laddrhash, bytes);
1411 		} else
1412 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1413 
1414 		if (!fib_info_hash_size)
1415 			goto failure;
1416 	}
1417 
1418 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1419 	if (!fi)
1420 		goto failure;
1421 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1422 					      cfg->fc_mx_len, extack);
1423 	if (IS_ERR(fi->fib_metrics)) {
1424 		err = PTR_ERR(fi->fib_metrics);
1425 		kfree(fi);
1426 		return ERR_PTR(err);
1427 	}
1428 
1429 	fib_info_cnt++;
1430 	fi->fib_net = net;
1431 	fi->fib_protocol = cfg->fc_protocol;
1432 	fi->fib_scope = cfg->fc_scope;
1433 	fi->fib_flags = cfg->fc_flags;
1434 	fi->fib_priority = cfg->fc_priority;
1435 	fi->fib_prefsrc = cfg->fc_prefsrc;
1436 	fi->fib_type = cfg->fc_type;
1437 	fi->fib_tb_id = cfg->fc_table;
1438 
1439 	fi->fib_nhs = nhs;
1440 	if (nh) {
1441 		if (!nexthop_get(nh)) {
1442 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1443 			err = -EINVAL;
1444 		} else {
1445 			err = 0;
1446 			fi->nh = nh;
1447 		}
1448 	} else {
1449 		change_nexthops(fi) {
1450 			nexthop_nh->nh_parent = fi;
1451 		} endfor_nexthops(fi)
1452 
1453 		if (cfg->fc_mp)
1454 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1455 					  extack);
1456 		else
1457 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1458 	}
1459 
1460 	if (err != 0)
1461 		goto failure;
1462 
1463 	if (fib_props[cfg->fc_type].error) {
1464 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1465 			NL_SET_ERR_MSG(extack,
1466 				       "Gateway, device and multipath can not be specified for this route type");
1467 			goto err_inval;
1468 		}
1469 		goto link_it;
1470 	} else {
1471 		switch (cfg->fc_type) {
1472 		case RTN_UNICAST:
1473 		case RTN_LOCAL:
1474 		case RTN_BROADCAST:
1475 		case RTN_ANYCAST:
1476 		case RTN_MULTICAST:
1477 			break;
1478 		default:
1479 			NL_SET_ERR_MSG(extack, "Invalid route type");
1480 			goto err_inval;
1481 		}
1482 	}
1483 
1484 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1485 		NL_SET_ERR_MSG(extack, "Invalid scope");
1486 		goto err_inval;
1487 	}
1488 
1489 	if (fi->nh) {
1490 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1491 		if (err)
1492 			goto failure;
1493 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1494 		struct fib_nh *nh = fi->fib_nh;
1495 
1496 		/* Local address is added. */
1497 		if (nhs != 1) {
1498 			NL_SET_ERR_MSG(extack,
1499 				       "Route with host scope can not have multiple nexthops");
1500 			goto err_inval;
1501 		}
1502 		if (nh->fib_nh_gw_family) {
1503 			NL_SET_ERR_MSG(extack,
1504 				       "Route with host scope can not have a gateway");
1505 			goto err_inval;
1506 		}
1507 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1508 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1509 		err = -ENODEV;
1510 		if (!nh->fib_nh_dev)
1511 			goto failure;
1512 	} else {
1513 		int linkdown = 0;
1514 
1515 		change_nexthops(fi) {
1516 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1517 					   cfg->fc_table, cfg->fc_scope,
1518 					   extack);
1519 			if (err != 0)
1520 				goto failure;
1521 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1522 				linkdown++;
1523 		} endfor_nexthops(fi)
1524 		if (linkdown == fi->fib_nhs)
1525 			fi->fib_flags |= RTNH_F_LINKDOWN;
1526 	}
1527 
1528 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1529 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1530 		goto err_inval;
1531 	}
1532 
1533 	if (!fi->nh) {
1534 		change_nexthops(fi) {
1535 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1536 						  fi->fib_scope);
1537 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1538 				fi->fib_nh_is_v6 = true;
1539 		} endfor_nexthops(fi)
1540 
1541 		fib_rebalance(fi);
1542 	}
1543 
1544 link_it:
1545 	ofi = fib_find_info(fi);
1546 	if (ofi) {
1547 		fi->fib_dead = 1;
1548 		free_fib_info(fi);
1549 		ofi->fib_treeref++;
1550 		return ofi;
1551 	}
1552 
1553 	fi->fib_treeref++;
1554 	refcount_set(&fi->fib_clntref, 1);
1555 	spin_lock_bh(&fib_info_lock);
1556 	hlist_add_head(&fi->fib_hash,
1557 		       &fib_info_hash[fib_info_hashfn(fi)]);
1558 	if (fi->fib_prefsrc) {
1559 		struct hlist_head *head;
1560 
1561 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1562 		hlist_add_head(&fi->fib_lhash, head);
1563 	}
1564 	if (fi->nh) {
1565 		list_add(&fi->nh_list, &nh->fi_list);
1566 	} else {
1567 		change_nexthops(fi) {
1568 			struct hlist_head *head;
1569 			unsigned int hash;
1570 
1571 			if (!nexthop_nh->fib_nh_dev)
1572 				continue;
1573 			hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1574 			head = &fib_info_devhash[hash];
1575 			hlist_add_head(&nexthop_nh->nh_hash, head);
1576 		} endfor_nexthops(fi)
1577 	}
1578 	spin_unlock_bh(&fib_info_lock);
1579 	return fi;
1580 
1581 err_inval:
1582 	err = -EINVAL;
1583 
1584 failure:
1585 	if (fi) {
1586 		fi->fib_dead = 1;
1587 		free_fib_info(fi);
1588 	}
1589 
1590 	return ERR_PTR(err);
1591 }
1592 
1593 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1594 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1595 {
1596 	if (nhc->nhc_flags & RTNH_F_DEAD)
1597 		*flags |= RTNH_F_DEAD;
1598 
1599 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1600 		*flags |= RTNH_F_LINKDOWN;
1601 
1602 		rcu_read_lock();
1603 		switch (nhc->nhc_family) {
1604 		case AF_INET:
1605 			if (ip_ignore_linkdown(nhc->nhc_dev))
1606 				*flags |= RTNH_F_DEAD;
1607 			break;
1608 		case AF_INET6:
1609 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1610 				*flags |= RTNH_F_DEAD;
1611 			break;
1612 		}
1613 		rcu_read_unlock();
1614 	}
1615 
1616 	switch (nhc->nhc_gw_family) {
1617 	case AF_INET:
1618 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1619 			goto nla_put_failure;
1620 		break;
1621 	case AF_INET6:
1622 		/* if gateway family does not match nexthop family
1623 		 * gateway is encoded as RTA_VIA
1624 		 */
1625 		if (rt_family != nhc->nhc_gw_family) {
1626 			int alen = sizeof(struct in6_addr);
1627 			struct nlattr *nla;
1628 			struct rtvia *via;
1629 
1630 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1631 			if (!nla)
1632 				goto nla_put_failure;
1633 
1634 			via = nla_data(nla);
1635 			via->rtvia_family = AF_INET6;
1636 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1637 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1638 					    &nhc->nhc_gw.ipv6) < 0) {
1639 			goto nla_put_failure;
1640 		}
1641 		break;
1642 	}
1643 
1644 	*flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1645 	if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1646 		*flags |= RTNH_F_OFFLOAD;
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 		  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 
1816 	nlmsg_end(skb, nlh);
1817 	return 0;
1818 
1819 nla_put_failure:
1820 	nlmsg_cancel(skb, nlh);
1821 	return -EMSGSIZE;
1822 }
1823 
1824 /*
1825  * Update FIB if:
1826  * - local address disappeared -> we must delete all the entries
1827  *   referring to it.
1828  * - device went down -> we must shutdown all nexthops going via it.
1829  */
1830 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1831 {
1832 	int ret = 0;
1833 	unsigned int hash = fib_laddr_hashfn(local);
1834 	struct hlist_head *head = &fib_info_laddrhash[hash];
1835 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1836 	struct net *net = dev_net(dev);
1837 	struct fib_info *fi;
1838 
1839 	if (!fib_info_laddrhash || local == 0)
1840 		return 0;
1841 
1842 	hlist_for_each_entry(fi, head, fib_lhash) {
1843 		if (!net_eq(fi->fib_net, net) ||
1844 		    fi->fib_tb_id != tb_id)
1845 			continue;
1846 		if (fi->fib_prefsrc == local) {
1847 			fi->fib_flags |= RTNH_F_DEAD;
1848 			ret++;
1849 		}
1850 	}
1851 	return ret;
1852 }
1853 
1854 static int call_fib_nh_notifiers(struct fib_nh *nh,
1855 				 enum fib_event_type event_type)
1856 {
1857 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1858 	struct fib_nh_notifier_info info = {
1859 		.fib_nh = nh,
1860 	};
1861 
1862 	switch (event_type) {
1863 	case FIB_EVENT_NH_ADD:
1864 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1865 			break;
1866 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1867 			break;
1868 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1869 					   &info.info);
1870 	case FIB_EVENT_NH_DEL:
1871 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1872 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1873 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1874 						   event_type, &info.info);
1875 	default:
1876 		break;
1877 	}
1878 
1879 	return NOTIFY_DONE;
1880 }
1881 
1882 /* Update the PMTU of exceptions when:
1883  * - the new MTU of the first hop becomes smaller than the PMTU
1884  * - the old MTU was the same as the PMTU, and it limited discovery of
1885  *   larger MTUs on the path. With that limit raised, we can now
1886  *   discover larger MTUs
1887  * A special case is locked exceptions, for which the PMTU is smaller
1888  * than the minimal accepted PMTU:
1889  * - if the new MTU is greater than the PMTU, don't make any change
1890  * - otherwise, unlock and set PMTU
1891  */
1892 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1893 {
1894 	struct fnhe_hash_bucket *bucket;
1895 	int i;
1896 
1897 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1898 	if (!bucket)
1899 		return;
1900 
1901 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1902 		struct fib_nh_exception *fnhe;
1903 
1904 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1905 		     fnhe;
1906 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1907 			if (fnhe->fnhe_mtu_locked) {
1908 				if (new <= fnhe->fnhe_pmtu) {
1909 					fnhe->fnhe_pmtu = new;
1910 					fnhe->fnhe_mtu_locked = false;
1911 				}
1912 			} else if (new < fnhe->fnhe_pmtu ||
1913 				   orig == fnhe->fnhe_pmtu) {
1914 				fnhe->fnhe_pmtu = new;
1915 			}
1916 		}
1917 	}
1918 }
1919 
1920 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1921 {
1922 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1923 	struct hlist_head *head = &fib_info_devhash[hash];
1924 	struct fib_nh *nh;
1925 
1926 	hlist_for_each_entry(nh, head, nh_hash) {
1927 		if (nh->fib_nh_dev == dev)
1928 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1929 	}
1930 }
1931 
1932 /* Event              force Flags           Description
1933  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1934  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1935  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1936  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1937  *
1938  * only used when fib_nh is built into fib_info
1939  */
1940 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1941 {
1942 	int ret = 0;
1943 	int scope = RT_SCOPE_NOWHERE;
1944 	struct fib_info *prev_fi = NULL;
1945 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1946 	struct hlist_head *head = &fib_info_devhash[hash];
1947 	struct fib_nh *nh;
1948 
1949 	if (force)
1950 		scope = -1;
1951 
1952 	hlist_for_each_entry(nh, head, nh_hash) {
1953 		struct fib_info *fi = nh->nh_parent;
1954 		int dead;
1955 
1956 		BUG_ON(!fi->fib_nhs);
1957 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1958 			continue;
1959 		prev_fi = fi;
1960 		dead = 0;
1961 		change_nexthops(fi) {
1962 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1963 				dead++;
1964 			else if (nexthop_nh->fib_nh_dev == dev &&
1965 				 nexthop_nh->fib_nh_scope != scope) {
1966 				switch (event) {
1967 				case NETDEV_DOWN:
1968 				case NETDEV_UNREGISTER:
1969 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1970 					fallthrough;
1971 				case NETDEV_CHANGE:
1972 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1973 					break;
1974 				}
1975 				call_fib_nh_notifiers(nexthop_nh,
1976 						      FIB_EVENT_NH_DEL);
1977 				dead++;
1978 			}
1979 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1980 			if (event == NETDEV_UNREGISTER &&
1981 			    nexthop_nh->fib_nh_dev == dev) {
1982 				dead = fi->fib_nhs;
1983 				break;
1984 			}
1985 #endif
1986 		} endfor_nexthops(fi)
1987 		if (dead == fi->fib_nhs) {
1988 			switch (event) {
1989 			case NETDEV_DOWN:
1990 			case NETDEV_UNREGISTER:
1991 				fi->fib_flags |= RTNH_F_DEAD;
1992 				fallthrough;
1993 			case NETDEV_CHANGE:
1994 				fi->fib_flags |= RTNH_F_LINKDOWN;
1995 				break;
1996 			}
1997 			ret++;
1998 		}
1999 
2000 		fib_rebalance(fi);
2001 	}
2002 
2003 	return ret;
2004 }
2005 
2006 /* Must be invoked inside of an RCU protected region.  */
2007 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2008 {
2009 	struct fib_info *fi = NULL, *last_resort = NULL;
2010 	struct hlist_head *fa_head = res->fa_head;
2011 	struct fib_table *tb = res->table;
2012 	u8 slen = 32 - res->prefixlen;
2013 	int order = -1, last_idx = -1;
2014 	struct fib_alias *fa, *fa1 = NULL;
2015 	u32 last_prio = res->fi->fib_priority;
2016 	u8 last_tos = 0;
2017 
2018 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2019 		struct fib_info *next_fi = fa->fa_info;
2020 		struct fib_nh_common *nhc;
2021 
2022 		if (fa->fa_slen != slen)
2023 			continue;
2024 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2025 			continue;
2026 		if (fa->tb_id != tb->tb_id)
2027 			continue;
2028 		if (next_fi->fib_priority > last_prio &&
2029 		    fa->fa_tos == last_tos) {
2030 			if (last_tos)
2031 				continue;
2032 			break;
2033 		}
2034 		if (next_fi->fib_flags & RTNH_F_DEAD)
2035 			continue;
2036 		last_tos = fa->fa_tos;
2037 		last_prio = next_fi->fib_priority;
2038 
2039 		if (next_fi->fib_scope != res->scope ||
2040 		    fa->fa_type != RTN_UNICAST)
2041 			continue;
2042 
2043 		nhc = fib_info_nhc(next_fi, 0);
2044 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2045 			continue;
2046 
2047 		fib_alias_accessed(fa);
2048 
2049 		if (!fi) {
2050 			if (next_fi != res->fi)
2051 				break;
2052 			fa1 = fa;
2053 		} else if (!fib_detect_death(fi, order, &last_resort,
2054 					     &last_idx, fa1->fa_default)) {
2055 			fib_result_assign(res, fi);
2056 			fa1->fa_default = order;
2057 			goto out;
2058 		}
2059 		fi = next_fi;
2060 		order++;
2061 	}
2062 
2063 	if (order <= 0 || !fi) {
2064 		if (fa1)
2065 			fa1->fa_default = -1;
2066 		goto out;
2067 	}
2068 
2069 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2070 			      fa1->fa_default)) {
2071 		fib_result_assign(res, fi);
2072 		fa1->fa_default = order;
2073 		goto out;
2074 	}
2075 
2076 	if (last_idx >= 0)
2077 		fib_result_assign(res, last_resort);
2078 	fa1->fa_default = last_idx;
2079 out:
2080 	return;
2081 }
2082 
2083 /*
2084  * Dead device goes up. We wake up dead nexthops.
2085  * It takes sense only on multipath routes.
2086  *
2087  * only used when fib_nh is built into fib_info
2088  */
2089 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2090 {
2091 	struct fib_info *prev_fi;
2092 	unsigned int hash;
2093 	struct hlist_head *head;
2094 	struct fib_nh *nh;
2095 	int ret;
2096 
2097 	if (!(dev->flags & IFF_UP))
2098 		return 0;
2099 
2100 	if (nh_flags & RTNH_F_DEAD) {
2101 		unsigned int flags = dev_get_flags(dev);
2102 
2103 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2104 			nh_flags |= RTNH_F_LINKDOWN;
2105 	}
2106 
2107 	prev_fi = NULL;
2108 	hash = fib_devindex_hashfn(dev->ifindex);
2109 	head = &fib_info_devhash[hash];
2110 	ret = 0;
2111 
2112 	hlist_for_each_entry(nh, head, nh_hash) {
2113 		struct fib_info *fi = nh->nh_parent;
2114 		int alive;
2115 
2116 		BUG_ON(!fi->fib_nhs);
2117 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2118 			continue;
2119 
2120 		prev_fi = fi;
2121 		alive = 0;
2122 		change_nexthops(fi) {
2123 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2124 				alive++;
2125 				continue;
2126 			}
2127 			if (!nexthop_nh->fib_nh_dev ||
2128 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2129 				continue;
2130 			if (nexthop_nh->fib_nh_dev != dev ||
2131 			    !__in_dev_get_rtnl(dev))
2132 				continue;
2133 			alive++;
2134 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2135 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2136 		} endfor_nexthops(fi)
2137 
2138 		if (alive > 0) {
2139 			fi->fib_flags &= ~nh_flags;
2140 			ret++;
2141 		}
2142 
2143 		fib_rebalance(fi);
2144 	}
2145 
2146 	return ret;
2147 }
2148 
2149 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2150 static bool fib_good_nh(const struct fib_nh *nh)
2151 {
2152 	int state = NUD_REACHABLE;
2153 
2154 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2155 		struct neighbour *n;
2156 
2157 		rcu_read_lock_bh();
2158 
2159 		if (likely(nh->fib_nh_gw_family == AF_INET))
2160 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2161 						   (__force u32)nh->fib_nh_gw4);
2162 		else if (nh->fib_nh_gw_family == AF_INET6)
2163 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2164 							   &nh->fib_nh_gw6);
2165 		else
2166 			n = NULL;
2167 		if (n)
2168 			state = n->nud_state;
2169 
2170 		rcu_read_unlock_bh();
2171 	}
2172 
2173 	return !!(state & NUD_VALID);
2174 }
2175 
2176 void fib_select_multipath(struct fib_result *res, int hash)
2177 {
2178 	struct fib_info *fi = res->fi;
2179 	struct net *net = fi->fib_net;
2180 	bool first = false;
2181 
2182 	if (unlikely(res->fi->nh)) {
2183 		nexthop_path_fib_result(res, hash);
2184 		return;
2185 	}
2186 
2187 	change_nexthops(fi) {
2188 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2189 			if (!fib_good_nh(nexthop_nh))
2190 				continue;
2191 			if (!first) {
2192 				res->nh_sel = nhsel;
2193 				res->nhc = &nexthop_nh->nh_common;
2194 				first = true;
2195 			}
2196 		}
2197 
2198 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2199 			continue;
2200 
2201 		res->nh_sel = nhsel;
2202 		res->nhc = &nexthop_nh->nh_common;
2203 		return;
2204 	} endfor_nexthops(fi);
2205 }
2206 #endif
2207 
2208 void fib_select_path(struct net *net, struct fib_result *res,
2209 		     struct flowi4 *fl4, const struct sk_buff *skb)
2210 {
2211 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2212 		goto check_saddr;
2213 
2214 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2215 	if (fib_info_num_path(res->fi) > 1) {
2216 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2217 
2218 		fib_select_multipath(res, h);
2219 	}
2220 	else
2221 #endif
2222 	if (!res->prefixlen &&
2223 	    res->table->tb_num_default > 1 &&
2224 	    res->type == RTN_UNICAST)
2225 		fib_select_default(fl4, res);
2226 
2227 check_saddr:
2228 	if (!fl4->saddr)
2229 		fl4->saddr = fib_result_prefsrc(net, res);
2230 }
2231