xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 63705da3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: semantics.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/uaccess.h>
13 #include <linux/bitops.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
17 #include <linux/mm.h>
18 #include <linux/string.h>
19 #include <linux/socket.h>
20 #include <linux/sockios.h>
21 #include <linux/errno.h>
22 #include <linux/in.h>
23 #include <linux/inet.h>
24 #include <linux/inetdevice.h>
25 #include <linux/netdevice.h>
26 #include <linux/if_arp.h>
27 #include <linux/proc_fs.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/netlink.h>
32 
33 #include <net/arp.h>
34 #include <net/ip.h>
35 #include <net/protocol.h>
36 #include <net/route.h>
37 #include <net/tcp.h>
38 #include <net/sock.h>
39 #include <net/ip_fib.h>
40 #include <net/ip6_fib.h>
41 #include <net/nexthop.h>
42 #include <net/netlink.h>
43 #include <net/rtnh.h>
44 #include <net/lwtunnel.h>
45 #include <net/fib_notifier.h>
46 #include <net/addrconf.h>
47 
48 #include "fib_lookup.h"
49 
50 static DEFINE_SPINLOCK(fib_info_lock);
51 static struct hlist_head *fib_info_hash;
52 static struct hlist_head *fib_info_laddrhash;
53 static unsigned int fib_info_hash_size;
54 static unsigned int fib_info_cnt;
55 
56 #define DEVINDEX_HASHBITS 8
57 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
58 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
59 
60 /* for_nexthops and change_nexthops only used when nexthop object
61  * is not set in a fib_info. The logic within can reference fib_nh.
62  */
63 #ifdef CONFIG_IP_ROUTE_MULTIPATH
64 
65 #define for_nexthops(fi) {						\
66 	int nhsel; const struct fib_nh *nh;				\
67 	for (nhsel = 0, nh = (fi)->fib_nh;				\
68 	     nhsel < fib_info_num_path((fi));				\
69 	     nh++, nhsel++)
70 
71 #define change_nexthops(fi) {						\
72 	int nhsel; struct fib_nh *nexthop_nh;				\
73 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
74 	     nhsel < fib_info_num_path((fi));				\
75 	     nexthop_nh++, nhsel++)
76 
77 #else /* CONFIG_IP_ROUTE_MULTIPATH */
78 
79 /* Hope, that gcc will optimize it to get rid of dummy loop */
80 
81 #define for_nexthops(fi) {						\
82 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
83 	for (nhsel = 0; nhsel < 1; nhsel++)
84 
85 #define change_nexthops(fi) {						\
86 	int nhsel;							\
87 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
88 	for (nhsel = 0; nhsel < 1; nhsel++)
89 
90 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
91 
92 #define endfor_nexthops(fi) }
93 
94 
95 const struct fib_prop fib_props[RTN_MAX + 1] = {
96 	[RTN_UNSPEC] = {
97 		.error	= 0,
98 		.scope	= RT_SCOPE_NOWHERE,
99 	},
100 	[RTN_UNICAST] = {
101 		.error	= 0,
102 		.scope	= RT_SCOPE_UNIVERSE,
103 	},
104 	[RTN_LOCAL] = {
105 		.error	= 0,
106 		.scope	= RT_SCOPE_HOST,
107 	},
108 	[RTN_BROADCAST] = {
109 		.error	= 0,
110 		.scope	= RT_SCOPE_LINK,
111 	},
112 	[RTN_ANYCAST] = {
113 		.error	= 0,
114 		.scope	= RT_SCOPE_LINK,
115 	},
116 	[RTN_MULTICAST] = {
117 		.error	= 0,
118 		.scope	= RT_SCOPE_UNIVERSE,
119 	},
120 	[RTN_BLACKHOLE] = {
121 		.error	= -EINVAL,
122 		.scope	= RT_SCOPE_UNIVERSE,
123 	},
124 	[RTN_UNREACHABLE] = {
125 		.error	= -EHOSTUNREACH,
126 		.scope	= RT_SCOPE_UNIVERSE,
127 	},
128 	[RTN_PROHIBIT] = {
129 		.error	= -EACCES,
130 		.scope	= RT_SCOPE_UNIVERSE,
131 	},
132 	[RTN_THROW] = {
133 		.error	= -EAGAIN,
134 		.scope	= RT_SCOPE_UNIVERSE,
135 	},
136 	[RTN_NAT] = {
137 		.error	= -EINVAL,
138 		.scope	= RT_SCOPE_NOWHERE,
139 	},
140 	[RTN_XRESOLVE] = {
141 		.error	= -EINVAL,
142 		.scope	= RT_SCOPE_NOWHERE,
143 	},
144 };
145 
146 static void rt_fibinfo_free(struct rtable __rcu **rtp)
147 {
148 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
149 
150 	if (!rt)
151 		return;
152 
153 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
154 	 * because we waited an RCU grace period before calling
155 	 * free_fib_info_rcu()
156 	 */
157 
158 	dst_dev_put(&rt->dst);
159 	dst_release_immediate(&rt->dst);
160 }
161 
162 static void free_nh_exceptions(struct fib_nh_common *nhc)
163 {
164 	struct fnhe_hash_bucket *hash;
165 	int i;
166 
167 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
168 	if (!hash)
169 		return;
170 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
171 		struct fib_nh_exception *fnhe;
172 
173 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
174 		while (fnhe) {
175 			struct fib_nh_exception *next;
176 
177 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
178 
179 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
180 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
181 
182 			kfree(fnhe);
183 
184 			fnhe = next;
185 		}
186 	}
187 	kfree(hash);
188 }
189 
190 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
191 {
192 	int cpu;
193 
194 	if (!rtp)
195 		return;
196 
197 	for_each_possible_cpu(cpu) {
198 		struct rtable *rt;
199 
200 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
201 		if (rt) {
202 			dst_dev_put(&rt->dst);
203 			dst_release_immediate(&rt->dst);
204 		}
205 	}
206 	free_percpu(rtp);
207 }
208 
209 void fib_nh_common_release(struct fib_nh_common *nhc)
210 {
211 	dev_put(nhc->nhc_dev);
212 	lwtstate_put(nhc->nhc_lwtstate);
213 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
214 	rt_fibinfo_free(&nhc->nhc_rth_input);
215 	free_nh_exceptions(nhc);
216 }
217 EXPORT_SYMBOL_GPL(fib_nh_common_release);
218 
219 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
220 {
221 #ifdef CONFIG_IP_ROUTE_CLASSID
222 	if (fib_nh->nh_tclassid)
223 		net->ipv4.fib_num_tclassid_users--;
224 #endif
225 	fib_nh_common_release(&fib_nh->nh_common);
226 }
227 
228 /* Release a nexthop info record */
229 static void free_fib_info_rcu(struct rcu_head *head)
230 {
231 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
232 
233 	if (fi->nh) {
234 		nexthop_put(fi->nh);
235 	} else {
236 		change_nexthops(fi) {
237 			fib_nh_release(fi->fib_net, nexthop_nh);
238 		} endfor_nexthops(fi);
239 	}
240 
241 	ip_fib_metrics_put(fi->fib_metrics);
242 
243 	kfree(fi);
244 }
245 
246 void free_fib_info(struct fib_info *fi)
247 {
248 	if (fi->fib_dead == 0) {
249 		pr_warn("Freeing alive fib_info %p\n", fi);
250 		return;
251 	}
252 	fib_info_cnt--;
253 
254 	call_rcu(&fi->rcu, free_fib_info_rcu);
255 }
256 EXPORT_SYMBOL_GPL(free_fib_info);
257 
258 void fib_release_info(struct fib_info *fi)
259 {
260 	spin_lock_bh(&fib_info_lock);
261 	if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
262 		hlist_del(&fi->fib_hash);
263 		if (fi->fib_prefsrc)
264 			hlist_del(&fi->fib_lhash);
265 		if (fi->nh) {
266 			list_del(&fi->nh_list);
267 		} else {
268 			change_nexthops(fi) {
269 				if (!nexthop_nh->fib_nh_dev)
270 					continue;
271 				hlist_del(&nexthop_nh->nh_hash);
272 			} endfor_nexthops(fi)
273 		}
274 		fi->fib_dead = 1;
275 		fib_info_put(fi);
276 	}
277 	spin_unlock_bh(&fib_info_lock);
278 }
279 
280 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
281 {
282 	const struct fib_nh *onh;
283 
284 	if (fi->nh || ofi->nh)
285 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
286 
287 	if (ofi->fib_nhs == 0)
288 		return 0;
289 
290 	for_nexthops(fi) {
291 		onh = fib_info_nh(ofi, nhsel);
292 
293 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
294 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
295 		    nh->fib_nh_scope != onh->fib_nh_scope ||
296 #ifdef CONFIG_IP_ROUTE_MULTIPATH
297 		    nh->fib_nh_weight != onh->fib_nh_weight ||
298 #endif
299 #ifdef CONFIG_IP_ROUTE_CLASSID
300 		    nh->nh_tclassid != onh->nh_tclassid ||
301 #endif
302 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
303 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
304 			return -1;
305 
306 		if (nh->fib_nh_gw_family == AF_INET &&
307 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
308 			return -1;
309 
310 		if (nh->fib_nh_gw_family == AF_INET6 &&
311 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
312 			return -1;
313 	} endfor_nexthops(fi);
314 	return 0;
315 }
316 
317 static inline unsigned int fib_devindex_hashfn(unsigned int val)
318 {
319 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
320 
321 	return (val ^
322 		(val >> DEVINDEX_HASHBITS) ^
323 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
324 }
325 
326 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
327 				      u32 prefsrc, u32 priority)
328 {
329 	unsigned int val = init_val;
330 
331 	val ^= (protocol << 8) | scope;
332 	val ^= prefsrc;
333 	val ^= priority;
334 
335 	return val;
336 }
337 
338 static unsigned int fib_info_hashfn_result(unsigned int val)
339 {
340 	unsigned int mask = (fib_info_hash_size - 1);
341 
342 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
343 }
344 
345 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
346 {
347 	unsigned int val;
348 
349 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
350 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
351 				fi->fib_priority);
352 
353 	if (fi->nh) {
354 		val ^= fib_devindex_hashfn(fi->nh->id);
355 	} else {
356 		for_nexthops(fi) {
357 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
358 		} endfor_nexthops(fi)
359 	}
360 
361 	return fib_info_hashfn_result(val);
362 }
363 
364 /* no metrics, only nexthop id */
365 static struct fib_info *fib_find_info_nh(struct net *net,
366 					 const struct fib_config *cfg)
367 {
368 	struct hlist_head *head;
369 	struct fib_info *fi;
370 	unsigned int hash;
371 
372 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
373 				 cfg->fc_protocol, cfg->fc_scope,
374 				 (__force u32)cfg->fc_prefsrc,
375 				 cfg->fc_priority);
376 	hash = fib_info_hashfn_result(hash);
377 	head = &fib_info_hash[hash];
378 
379 	hlist_for_each_entry(fi, head, fib_hash) {
380 		if (!net_eq(fi->fib_net, net))
381 			continue;
382 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
383 			continue;
384 		if (cfg->fc_protocol == fi->fib_protocol &&
385 		    cfg->fc_scope == fi->fib_scope &&
386 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
387 		    cfg->fc_priority == fi->fib_priority &&
388 		    cfg->fc_type == fi->fib_type &&
389 		    cfg->fc_table == fi->fib_tb_id &&
390 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
391 			return fi;
392 	}
393 
394 	return NULL;
395 }
396 
397 static struct fib_info *fib_find_info(struct fib_info *nfi)
398 {
399 	struct hlist_head *head;
400 	struct fib_info *fi;
401 	unsigned int hash;
402 
403 	hash = fib_info_hashfn(nfi);
404 	head = &fib_info_hash[hash];
405 
406 	hlist_for_each_entry(fi, head, fib_hash) {
407 		if (!net_eq(fi->fib_net, nfi->fib_net))
408 			continue;
409 		if (fi->fib_nhs != nfi->fib_nhs)
410 			continue;
411 		if (nfi->fib_protocol == fi->fib_protocol &&
412 		    nfi->fib_scope == fi->fib_scope &&
413 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
414 		    nfi->fib_priority == fi->fib_priority &&
415 		    nfi->fib_type == fi->fib_type &&
416 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
417 			   sizeof(u32) * RTAX_MAX) == 0 &&
418 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
419 		    nh_comp(fi, nfi) == 0)
420 			return fi;
421 	}
422 
423 	return NULL;
424 }
425 
426 /* Check, that the gateway is already configured.
427  * Used only by redirect accept routine.
428  */
429 int ip_fib_check_default(__be32 gw, struct net_device *dev)
430 {
431 	struct hlist_head *head;
432 	struct fib_nh *nh;
433 	unsigned int hash;
434 
435 	spin_lock(&fib_info_lock);
436 
437 	hash = fib_devindex_hashfn(dev->ifindex);
438 	head = &fib_info_devhash[hash];
439 	hlist_for_each_entry(nh, head, nh_hash) {
440 		if (nh->fib_nh_dev == dev &&
441 		    nh->fib_nh_gw4 == gw &&
442 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
443 			spin_unlock(&fib_info_lock);
444 			return 0;
445 		}
446 	}
447 
448 	spin_unlock(&fib_info_lock);
449 
450 	return -1;
451 }
452 
453 size_t fib_nlmsg_size(struct fib_info *fi)
454 {
455 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
456 			 + nla_total_size(4) /* RTA_TABLE */
457 			 + nla_total_size(4) /* RTA_DST */
458 			 + nla_total_size(4) /* RTA_PRIORITY */
459 			 + nla_total_size(4) /* RTA_PREFSRC */
460 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
461 	unsigned int nhs = fib_info_num_path(fi);
462 
463 	/* space for nested metrics */
464 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
465 
466 	if (fi->nh)
467 		payload += nla_total_size(4); /* RTA_NH_ID */
468 
469 	if (nhs) {
470 		size_t nh_encapsize = 0;
471 		/* Also handles the special case nhs == 1 */
472 
473 		/* each nexthop is packed in an attribute */
474 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
475 		unsigned int i;
476 
477 		/* may contain flow and gateway attribute */
478 		nhsize += 2 * nla_total_size(4);
479 
480 		/* grab encap info */
481 		for (i = 0; i < fib_info_num_path(fi); i++) {
482 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
483 
484 			if (nhc->nhc_lwtstate) {
485 				/* RTA_ENCAP_TYPE */
486 				nh_encapsize += lwtunnel_get_encap_size(
487 						nhc->nhc_lwtstate);
488 				/* RTA_ENCAP */
489 				nh_encapsize +=  nla_total_size(2);
490 			}
491 		}
492 
493 		/* all nexthops are packed in a nested attribute */
494 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
495 
496 	}
497 
498 	return payload;
499 }
500 
501 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
502 	       int dst_len, u32 tb_id, const struct nl_info *info,
503 	       unsigned int nlm_flags)
504 {
505 	struct fib_rt_info fri;
506 	struct sk_buff *skb;
507 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
508 	int err = -ENOBUFS;
509 
510 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
511 	if (!skb)
512 		goto errout;
513 
514 	fri.fi = fa->fa_info;
515 	fri.tb_id = tb_id;
516 	fri.dst = key;
517 	fri.dst_len = dst_len;
518 	fri.tos = fa->fa_tos;
519 	fri.type = fa->fa_type;
520 	fri.offload = fa->offload;
521 	fri.trap = fa->trap;
522 	fri.offload_failed = fa->offload_failed;
523 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
524 	if (err < 0) {
525 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
526 		WARN_ON(err == -EMSGSIZE);
527 		kfree_skb(skb);
528 		goto errout;
529 	}
530 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
531 		    info->nlh, GFP_KERNEL);
532 	return;
533 errout:
534 	if (err < 0)
535 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
536 }
537 
538 static int fib_detect_death(struct fib_info *fi, int order,
539 			    struct fib_info **last_resort, int *last_idx,
540 			    int dflt)
541 {
542 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
543 	struct neighbour *n;
544 	int state = NUD_NONE;
545 
546 	if (likely(nhc->nhc_gw_family == AF_INET))
547 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
548 	else if (nhc->nhc_gw_family == AF_INET6)
549 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
550 				 nhc->nhc_dev);
551 	else
552 		n = NULL;
553 
554 	if (n) {
555 		state = n->nud_state;
556 		neigh_release(n);
557 	} else {
558 		return 0;
559 	}
560 	if (state == NUD_REACHABLE)
561 		return 0;
562 	if ((state & NUD_VALID) && order != dflt)
563 		return 0;
564 	if ((state & NUD_VALID) ||
565 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
566 		*last_resort = fi;
567 		*last_idx = order;
568 	}
569 	return 1;
570 }
571 
572 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
573 		       struct nlattr *encap, u16 encap_type,
574 		       void *cfg, gfp_t gfp_flags,
575 		       struct netlink_ext_ack *extack)
576 {
577 	int err;
578 
579 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
580 						    gfp_flags);
581 	if (!nhc->nhc_pcpu_rth_output)
582 		return -ENOMEM;
583 
584 	if (encap) {
585 		struct lwtunnel_state *lwtstate;
586 
587 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
588 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
589 			err = -EINVAL;
590 			goto lwt_failure;
591 		}
592 		err = lwtunnel_build_state(net, encap_type, encap,
593 					   nhc->nhc_family, cfg, &lwtstate,
594 					   extack);
595 		if (err)
596 			goto lwt_failure;
597 
598 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
599 	}
600 
601 	return 0;
602 
603 lwt_failure:
604 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
605 	nhc->nhc_pcpu_rth_output = NULL;
606 	return err;
607 }
608 EXPORT_SYMBOL_GPL(fib_nh_common_init);
609 
610 int fib_nh_init(struct net *net, struct fib_nh *nh,
611 		struct fib_config *cfg, int nh_weight,
612 		struct netlink_ext_ack *extack)
613 {
614 	int err;
615 
616 	nh->fib_nh_family = AF_INET;
617 
618 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
619 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
620 	if (err)
621 		return err;
622 
623 	nh->fib_nh_oif = cfg->fc_oif;
624 	nh->fib_nh_gw_family = cfg->fc_gw_family;
625 	if (cfg->fc_gw_family == AF_INET)
626 		nh->fib_nh_gw4 = cfg->fc_gw4;
627 	else if (cfg->fc_gw_family == AF_INET6)
628 		nh->fib_nh_gw6 = cfg->fc_gw6;
629 
630 	nh->fib_nh_flags = cfg->fc_flags;
631 
632 #ifdef CONFIG_IP_ROUTE_CLASSID
633 	nh->nh_tclassid = cfg->fc_flow;
634 	if (nh->nh_tclassid)
635 		net->ipv4.fib_num_tclassid_users++;
636 #endif
637 #ifdef CONFIG_IP_ROUTE_MULTIPATH
638 	nh->fib_nh_weight = nh_weight;
639 #endif
640 	return 0;
641 }
642 
643 #ifdef CONFIG_IP_ROUTE_MULTIPATH
644 
645 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
646 			      struct netlink_ext_ack *extack)
647 {
648 	int nhs = 0;
649 
650 	while (rtnh_ok(rtnh, remaining)) {
651 		nhs++;
652 		rtnh = rtnh_next(rtnh, &remaining);
653 	}
654 
655 	/* leftover implies invalid nexthop configuration, discard it */
656 	if (remaining > 0) {
657 		NL_SET_ERR_MSG(extack,
658 			       "Invalid nexthop configuration - extra data after nexthops");
659 		nhs = 0;
660 	}
661 
662 	return nhs;
663 }
664 
665 /* only called when fib_nh is integrated into fib_info */
666 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
667 		       int remaining, struct fib_config *cfg,
668 		       struct netlink_ext_ack *extack)
669 {
670 	struct net *net = fi->fib_net;
671 	struct fib_config fib_cfg;
672 	struct fib_nh *nh;
673 	int ret;
674 
675 	change_nexthops(fi) {
676 		int attrlen;
677 
678 		memset(&fib_cfg, 0, sizeof(fib_cfg));
679 
680 		if (!rtnh_ok(rtnh, remaining)) {
681 			NL_SET_ERR_MSG(extack,
682 				       "Invalid nexthop configuration - extra data after nexthop");
683 			return -EINVAL;
684 		}
685 
686 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
687 			NL_SET_ERR_MSG(extack,
688 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
689 			return -EINVAL;
690 		}
691 
692 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
693 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
694 
695 		attrlen = rtnh_attrlen(rtnh);
696 		if (attrlen > 0) {
697 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
698 
699 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
700 			nlav = nla_find(attrs, attrlen, RTA_VIA);
701 			if (nla && nlav) {
702 				NL_SET_ERR_MSG(extack,
703 					       "Nexthop configuration can not contain both GATEWAY and VIA");
704 				return -EINVAL;
705 			}
706 			if (nla) {
707 				fib_cfg.fc_gw4 = nla_get_in_addr(nla);
708 				if (fib_cfg.fc_gw4)
709 					fib_cfg.fc_gw_family = AF_INET;
710 			} else if (nlav) {
711 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
712 				if (ret)
713 					goto errout;
714 			}
715 
716 			nla = nla_find(attrs, attrlen, RTA_FLOW);
717 			if (nla)
718 				fib_cfg.fc_flow = nla_get_u32(nla);
719 
720 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
721 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
722 			if (nla)
723 				fib_cfg.fc_encap_type = nla_get_u16(nla);
724 		}
725 
726 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
727 				  rtnh->rtnh_hops + 1, extack);
728 		if (ret)
729 			goto errout;
730 
731 		rtnh = rtnh_next(rtnh, &remaining);
732 	} endfor_nexthops(fi);
733 
734 	ret = -EINVAL;
735 	nh = fib_info_nh(fi, 0);
736 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
737 		NL_SET_ERR_MSG(extack,
738 			       "Nexthop device index does not match RTA_OIF");
739 		goto errout;
740 	}
741 	if (cfg->fc_gw_family) {
742 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
743 		    (cfg->fc_gw_family == AF_INET &&
744 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
745 		    (cfg->fc_gw_family == AF_INET6 &&
746 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
747 			NL_SET_ERR_MSG(extack,
748 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
749 			goto errout;
750 		}
751 	}
752 #ifdef CONFIG_IP_ROUTE_CLASSID
753 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
754 		NL_SET_ERR_MSG(extack,
755 			       "Nexthop class id does not match RTA_FLOW");
756 		goto errout;
757 	}
758 #endif
759 	ret = 0;
760 errout:
761 	return ret;
762 }
763 
764 /* only called when fib_nh is integrated into fib_info */
765 static void fib_rebalance(struct fib_info *fi)
766 {
767 	int total;
768 	int w;
769 
770 	if (fib_info_num_path(fi) < 2)
771 		return;
772 
773 	total = 0;
774 	for_nexthops(fi) {
775 		if (nh->fib_nh_flags & RTNH_F_DEAD)
776 			continue;
777 
778 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
779 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
780 			continue;
781 
782 		total += nh->fib_nh_weight;
783 	} endfor_nexthops(fi);
784 
785 	w = 0;
786 	change_nexthops(fi) {
787 		int upper_bound;
788 
789 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
790 			upper_bound = -1;
791 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
792 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
793 			upper_bound = -1;
794 		} else {
795 			w += nexthop_nh->fib_nh_weight;
796 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
797 							    total) - 1;
798 		}
799 
800 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
801 	} endfor_nexthops(fi);
802 }
803 #else /* CONFIG_IP_ROUTE_MULTIPATH */
804 
805 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
806 		       int remaining, struct fib_config *cfg,
807 		       struct netlink_ext_ack *extack)
808 {
809 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
810 
811 	return -EINVAL;
812 }
813 
814 #define fib_rebalance(fi) do { } while (0)
815 
816 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
817 
818 static int fib_encap_match(struct net *net, u16 encap_type,
819 			   struct nlattr *encap,
820 			   const struct fib_nh *nh,
821 			   const struct fib_config *cfg,
822 			   struct netlink_ext_ack *extack)
823 {
824 	struct lwtunnel_state *lwtstate;
825 	int ret, result = 0;
826 
827 	if (encap_type == LWTUNNEL_ENCAP_NONE)
828 		return 0;
829 
830 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
831 				   cfg, &lwtstate, extack);
832 	if (!ret) {
833 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
834 		lwtstate_free(lwtstate);
835 	}
836 
837 	return result;
838 }
839 
840 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
841 		 struct netlink_ext_ack *extack)
842 {
843 #ifdef CONFIG_IP_ROUTE_MULTIPATH
844 	struct rtnexthop *rtnh;
845 	int remaining;
846 #endif
847 
848 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
849 		return 1;
850 
851 	if (cfg->fc_nh_id) {
852 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
853 			return 0;
854 		return 1;
855 	}
856 
857 	if (cfg->fc_oif || cfg->fc_gw_family) {
858 		struct fib_nh *nh = fib_info_nh(fi, 0);
859 
860 		if (cfg->fc_encap) {
861 			if (fib_encap_match(net, cfg->fc_encap_type,
862 					    cfg->fc_encap, nh, cfg, extack))
863 				return 1;
864 		}
865 #ifdef CONFIG_IP_ROUTE_CLASSID
866 		if (cfg->fc_flow &&
867 		    cfg->fc_flow != nh->nh_tclassid)
868 			return 1;
869 #endif
870 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
871 		    (cfg->fc_gw_family &&
872 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
873 			return 1;
874 
875 		if (cfg->fc_gw_family == AF_INET &&
876 		    cfg->fc_gw4 != nh->fib_nh_gw4)
877 			return 1;
878 
879 		if (cfg->fc_gw_family == AF_INET6 &&
880 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
881 			return 1;
882 
883 		return 0;
884 	}
885 
886 #ifdef CONFIG_IP_ROUTE_MULTIPATH
887 	if (!cfg->fc_mp)
888 		return 0;
889 
890 	rtnh = cfg->fc_mp;
891 	remaining = cfg->fc_mp_len;
892 
893 	for_nexthops(fi) {
894 		int attrlen;
895 
896 		if (!rtnh_ok(rtnh, remaining))
897 			return -EINVAL;
898 
899 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
900 			return 1;
901 
902 		attrlen = rtnh_attrlen(rtnh);
903 		if (attrlen > 0) {
904 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
905 
906 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
907 			nlav = nla_find(attrs, attrlen, RTA_VIA);
908 			if (nla && nlav) {
909 				NL_SET_ERR_MSG(extack,
910 					       "Nexthop configuration can not contain both GATEWAY and VIA");
911 				return -EINVAL;
912 			}
913 
914 			if (nla) {
915 				if (nh->fib_nh_gw_family != AF_INET ||
916 				    nla_get_in_addr(nla) != nh->fib_nh_gw4)
917 					return 1;
918 			} else if (nlav) {
919 				struct fib_config cfg2;
920 				int err;
921 
922 				err = fib_gw_from_via(&cfg2, nlav, extack);
923 				if (err)
924 					return err;
925 
926 				switch (nh->fib_nh_gw_family) {
927 				case AF_INET:
928 					if (cfg2.fc_gw_family != AF_INET ||
929 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
930 						return 1;
931 					break;
932 				case AF_INET6:
933 					if (cfg2.fc_gw_family != AF_INET6 ||
934 					    ipv6_addr_cmp(&cfg2.fc_gw6,
935 							  &nh->fib_nh_gw6))
936 						return 1;
937 					break;
938 				}
939 			}
940 
941 #ifdef CONFIG_IP_ROUTE_CLASSID
942 			nla = nla_find(attrs, attrlen, RTA_FLOW);
943 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
944 				return 1;
945 #endif
946 		}
947 
948 		rtnh = rtnh_next(rtnh, &remaining);
949 	} endfor_nexthops(fi);
950 #endif
951 	return 0;
952 }
953 
954 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
955 {
956 	struct nlattr *nla;
957 	int remaining;
958 
959 	if (!cfg->fc_mx)
960 		return true;
961 
962 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
963 		int type = nla_type(nla);
964 		u32 fi_val, val;
965 
966 		if (!type)
967 			continue;
968 		if (type > RTAX_MAX)
969 			return false;
970 
971 		if (type == RTAX_CC_ALGO) {
972 			char tmp[TCP_CA_NAME_MAX];
973 			bool ecn_ca = false;
974 
975 			nla_strscpy(tmp, nla, sizeof(tmp));
976 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
977 		} else {
978 			if (nla_len(nla) != sizeof(u32))
979 				return false;
980 			val = nla_get_u32(nla);
981 		}
982 
983 		fi_val = fi->fib_metrics->metrics[type - 1];
984 		if (type == RTAX_FEATURES)
985 			fi_val &= ~DST_FEATURE_ECN_CA;
986 
987 		if (fi_val != val)
988 			return false;
989 	}
990 
991 	return true;
992 }
993 
994 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
995 			      u32 table, struct netlink_ext_ack *extack)
996 {
997 	struct fib6_config cfg = {
998 		.fc_table = table,
999 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1000 		.fc_ifindex = nh->fib_nh_oif,
1001 		.fc_gateway = nh->fib_nh_gw6,
1002 	};
1003 	struct fib6_nh fib6_nh = {};
1004 	int err;
1005 
1006 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1007 	if (!err) {
1008 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1009 		dev_hold(nh->fib_nh_dev);
1010 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1011 		nh->fib_nh_scope = RT_SCOPE_LINK;
1012 
1013 		ipv6_stub->fib6_nh_release(&fib6_nh);
1014 	}
1015 
1016 	return err;
1017 }
1018 
1019 /*
1020  * Picture
1021  * -------
1022  *
1023  * Semantics of nexthop is very messy by historical reasons.
1024  * We have to take into account, that:
1025  * a) gateway can be actually local interface address,
1026  *    so that gatewayed route is direct.
1027  * b) gateway must be on-link address, possibly
1028  *    described not by an ifaddr, but also by a direct route.
1029  * c) If both gateway and interface are specified, they should not
1030  *    contradict.
1031  * d) If we use tunnel routes, gateway could be not on-link.
1032  *
1033  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1034  * results in pretty ugly and hairy code with obscure logic.
1035  *
1036  * I chose to generalized it instead, so that the size
1037  * of code does not increase practically, but it becomes
1038  * much more general.
1039  * Every prefix is assigned a "scope" value: "host" is local address,
1040  * "link" is direct route,
1041  * [ ... "site" ... "interior" ... ]
1042  * and "universe" is true gateway route with global meaning.
1043  *
1044  * Every prefix refers to a set of "nexthop"s (gw, oif),
1045  * where gw must have narrower scope. This recursion stops
1046  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1047  * which means that gw is forced to be on link.
1048  *
1049  * Code is still hairy, but now it is apparently logically
1050  * consistent and very flexible. F.e. as by-product it allows
1051  * to co-exists in peace independent exterior and interior
1052  * routing processes.
1053  *
1054  * Normally it looks as following.
1055  *
1056  * {universe prefix}  -> (gw, oif) [scope link]
1057  *		  |
1058  *		  |-> {link prefix} -> (gw, oif) [scope local]
1059  *					|
1060  *					|-> {local prefix} (terminal node)
1061  */
1062 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1063 			      u8 scope, struct netlink_ext_ack *extack)
1064 {
1065 	struct net_device *dev;
1066 	struct fib_result res;
1067 	int err = 0;
1068 
1069 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1070 		unsigned int addr_type;
1071 
1072 		if (scope >= RT_SCOPE_LINK) {
1073 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1074 			return -EINVAL;
1075 		}
1076 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1077 		if (!dev) {
1078 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1079 			return -ENODEV;
1080 		}
1081 		if (!(dev->flags & IFF_UP)) {
1082 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1083 			return -ENETDOWN;
1084 		}
1085 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1086 		if (addr_type != RTN_UNICAST) {
1087 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1088 			return -EINVAL;
1089 		}
1090 		if (!netif_carrier_ok(dev))
1091 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1092 		nh->fib_nh_dev = dev;
1093 		dev_hold(dev);
1094 		nh->fib_nh_scope = RT_SCOPE_LINK;
1095 		return 0;
1096 	}
1097 	rcu_read_lock();
1098 	{
1099 		struct fib_table *tbl = NULL;
1100 		struct flowi4 fl4 = {
1101 			.daddr = nh->fib_nh_gw4,
1102 			.flowi4_scope = scope + 1,
1103 			.flowi4_oif = nh->fib_nh_oif,
1104 			.flowi4_iif = LOOPBACK_IFINDEX,
1105 		};
1106 
1107 		/* It is not necessary, but requires a bit of thinking */
1108 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1109 			fl4.flowi4_scope = RT_SCOPE_LINK;
1110 
1111 		if (table && table != RT_TABLE_MAIN)
1112 			tbl = fib_get_table(net, table);
1113 
1114 		if (tbl)
1115 			err = fib_table_lookup(tbl, &fl4, &res,
1116 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1117 					       FIB_LOOKUP_NOREF);
1118 
1119 		/* on error or if no table given do full lookup. This
1120 		 * is needed for example when nexthops are in the local
1121 		 * table rather than the given table
1122 		 */
1123 		if (!tbl || err) {
1124 			err = fib_lookup(net, &fl4, &res,
1125 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1126 		}
1127 
1128 		if (err) {
1129 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1130 			goto out;
1131 		}
1132 	}
1133 
1134 	err = -EINVAL;
1135 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1136 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1137 		goto out;
1138 	}
1139 	nh->fib_nh_scope = res.scope;
1140 	nh->fib_nh_oif = FIB_RES_OIF(res);
1141 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1142 	if (!dev) {
1143 		NL_SET_ERR_MSG(extack,
1144 			       "No egress device for nexthop gateway");
1145 		goto out;
1146 	}
1147 	dev_hold(dev);
1148 	if (!netif_carrier_ok(dev))
1149 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1150 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1151 out:
1152 	rcu_read_unlock();
1153 	return err;
1154 }
1155 
1156 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1157 			      struct netlink_ext_ack *extack)
1158 {
1159 	struct in_device *in_dev;
1160 	int err;
1161 
1162 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1163 		NL_SET_ERR_MSG(extack,
1164 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1165 		return -EINVAL;
1166 	}
1167 
1168 	rcu_read_lock();
1169 
1170 	err = -ENODEV;
1171 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1172 	if (!in_dev)
1173 		goto out;
1174 	err = -ENETDOWN;
1175 	if (!(in_dev->dev->flags & IFF_UP)) {
1176 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1177 		goto out;
1178 	}
1179 
1180 	nh->fib_nh_dev = in_dev->dev;
1181 	dev_hold(nh->fib_nh_dev);
1182 	nh->fib_nh_scope = RT_SCOPE_HOST;
1183 	if (!netif_carrier_ok(nh->fib_nh_dev))
1184 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1185 	err = 0;
1186 out:
1187 	rcu_read_unlock();
1188 	return err;
1189 }
1190 
1191 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1192 		 struct netlink_ext_ack *extack)
1193 {
1194 	int err;
1195 
1196 	if (nh->fib_nh_gw_family == AF_INET)
1197 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1198 	else if (nh->fib_nh_gw_family == AF_INET6)
1199 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1200 	else
1201 		err = fib_check_nh_nongw(net, nh, extack);
1202 
1203 	return err;
1204 }
1205 
1206 static inline unsigned int fib_laddr_hashfn(__be32 val)
1207 {
1208 	unsigned int mask = (fib_info_hash_size - 1);
1209 
1210 	return ((__force u32)val ^
1211 		((__force u32)val >> 7) ^
1212 		((__force u32)val >> 14)) & mask;
1213 }
1214 
1215 static struct hlist_head *fib_info_hash_alloc(int bytes)
1216 {
1217 	if (bytes <= PAGE_SIZE)
1218 		return kzalloc(bytes, GFP_KERNEL);
1219 	else
1220 		return (struct hlist_head *)
1221 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1222 					 get_order(bytes));
1223 }
1224 
1225 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1226 {
1227 	if (!hash)
1228 		return;
1229 
1230 	if (bytes <= PAGE_SIZE)
1231 		kfree(hash);
1232 	else
1233 		free_pages((unsigned long) hash, get_order(bytes));
1234 }
1235 
1236 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1237 			       struct hlist_head *new_laddrhash,
1238 			       unsigned int new_size)
1239 {
1240 	struct hlist_head *old_info_hash, *old_laddrhash;
1241 	unsigned int old_size = fib_info_hash_size;
1242 	unsigned int i, bytes;
1243 
1244 	spin_lock_bh(&fib_info_lock);
1245 	old_info_hash = fib_info_hash;
1246 	old_laddrhash = fib_info_laddrhash;
1247 	fib_info_hash_size = new_size;
1248 
1249 	for (i = 0; i < old_size; i++) {
1250 		struct hlist_head *head = &fib_info_hash[i];
1251 		struct hlist_node *n;
1252 		struct fib_info *fi;
1253 
1254 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1255 			struct hlist_head *dest;
1256 			unsigned int new_hash;
1257 
1258 			new_hash = fib_info_hashfn(fi);
1259 			dest = &new_info_hash[new_hash];
1260 			hlist_add_head(&fi->fib_hash, dest);
1261 		}
1262 	}
1263 	fib_info_hash = new_info_hash;
1264 
1265 	for (i = 0; i < old_size; i++) {
1266 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1267 		struct hlist_node *n;
1268 		struct fib_info *fi;
1269 
1270 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1271 			struct hlist_head *ldest;
1272 			unsigned int new_hash;
1273 
1274 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1275 			ldest = &new_laddrhash[new_hash];
1276 			hlist_add_head(&fi->fib_lhash, ldest);
1277 		}
1278 	}
1279 	fib_info_laddrhash = new_laddrhash;
1280 
1281 	spin_unlock_bh(&fib_info_lock);
1282 
1283 	bytes = old_size * sizeof(struct hlist_head *);
1284 	fib_info_hash_free(old_info_hash, bytes);
1285 	fib_info_hash_free(old_laddrhash, bytes);
1286 }
1287 
1288 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1289 				 unsigned char scope)
1290 {
1291 	struct fib_nh *nh;
1292 
1293 	if (nhc->nhc_family != AF_INET)
1294 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1295 
1296 	nh = container_of(nhc, struct fib_nh, nh_common);
1297 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1298 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1299 
1300 	return nh->nh_saddr;
1301 }
1302 
1303 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1304 {
1305 	struct fib_nh_common *nhc = res->nhc;
1306 
1307 	if (res->fi->fib_prefsrc)
1308 		return res->fi->fib_prefsrc;
1309 
1310 	if (nhc->nhc_family == AF_INET) {
1311 		struct fib_nh *nh;
1312 
1313 		nh = container_of(nhc, struct fib_nh, nh_common);
1314 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1315 			return nh->nh_saddr;
1316 	}
1317 
1318 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1319 }
1320 
1321 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1322 {
1323 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1324 	    fib_prefsrc != cfg->fc_dst) {
1325 		u32 tb_id = cfg->fc_table;
1326 		int rc;
1327 
1328 		if (tb_id == RT_TABLE_MAIN)
1329 			tb_id = RT_TABLE_LOCAL;
1330 
1331 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1332 					  fib_prefsrc, tb_id);
1333 
1334 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1335 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1336 						  fib_prefsrc, RT_TABLE_LOCAL);
1337 		}
1338 
1339 		if (rc != RTN_LOCAL)
1340 			return false;
1341 	}
1342 	return true;
1343 }
1344 
1345 struct fib_info *fib_create_info(struct fib_config *cfg,
1346 				 struct netlink_ext_ack *extack)
1347 {
1348 	int err;
1349 	struct fib_info *fi = NULL;
1350 	struct nexthop *nh = NULL;
1351 	struct fib_info *ofi;
1352 	int nhs = 1;
1353 	struct net *net = cfg->fc_nlinfo.nl_net;
1354 
1355 	if (cfg->fc_type > RTN_MAX)
1356 		goto err_inval;
1357 
1358 	/* Fast check to catch the most weird cases */
1359 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1360 		NL_SET_ERR_MSG(extack, "Invalid scope");
1361 		goto err_inval;
1362 	}
1363 
1364 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1365 		NL_SET_ERR_MSG(extack,
1366 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1367 		goto err_inval;
1368 	}
1369 
1370 	if (cfg->fc_nh_id) {
1371 		if (!cfg->fc_mx) {
1372 			fi = fib_find_info_nh(net, cfg);
1373 			if (fi) {
1374 				refcount_inc(&fi->fib_treeref);
1375 				return fi;
1376 			}
1377 		}
1378 
1379 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1380 		if (!nh) {
1381 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1382 			goto err_inval;
1383 		}
1384 		nhs = 0;
1385 	}
1386 
1387 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1388 	if (cfg->fc_mp) {
1389 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1390 		if (nhs == 0)
1391 			goto err_inval;
1392 	}
1393 #endif
1394 
1395 	err = -ENOBUFS;
1396 	if (fib_info_cnt >= fib_info_hash_size) {
1397 		unsigned int new_size = fib_info_hash_size << 1;
1398 		struct hlist_head *new_info_hash;
1399 		struct hlist_head *new_laddrhash;
1400 		unsigned int bytes;
1401 
1402 		if (!new_size)
1403 			new_size = 16;
1404 		bytes = new_size * sizeof(struct hlist_head *);
1405 		new_info_hash = fib_info_hash_alloc(bytes);
1406 		new_laddrhash = fib_info_hash_alloc(bytes);
1407 		if (!new_info_hash || !new_laddrhash) {
1408 			fib_info_hash_free(new_info_hash, bytes);
1409 			fib_info_hash_free(new_laddrhash, bytes);
1410 		} else
1411 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1412 
1413 		if (!fib_info_hash_size)
1414 			goto failure;
1415 	}
1416 
1417 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1418 	if (!fi)
1419 		goto failure;
1420 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1421 					      cfg->fc_mx_len, extack);
1422 	if (IS_ERR(fi->fib_metrics)) {
1423 		err = PTR_ERR(fi->fib_metrics);
1424 		kfree(fi);
1425 		return ERR_PTR(err);
1426 	}
1427 
1428 	fib_info_cnt++;
1429 	fi->fib_net = net;
1430 	fi->fib_protocol = cfg->fc_protocol;
1431 	fi->fib_scope = cfg->fc_scope;
1432 	fi->fib_flags = cfg->fc_flags;
1433 	fi->fib_priority = cfg->fc_priority;
1434 	fi->fib_prefsrc = cfg->fc_prefsrc;
1435 	fi->fib_type = cfg->fc_type;
1436 	fi->fib_tb_id = cfg->fc_table;
1437 
1438 	fi->fib_nhs = nhs;
1439 	if (nh) {
1440 		if (!nexthop_get(nh)) {
1441 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1442 			err = -EINVAL;
1443 		} else {
1444 			err = 0;
1445 			fi->nh = nh;
1446 		}
1447 	} else {
1448 		change_nexthops(fi) {
1449 			nexthop_nh->nh_parent = fi;
1450 		} endfor_nexthops(fi)
1451 
1452 		if (cfg->fc_mp)
1453 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1454 					  extack);
1455 		else
1456 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1457 	}
1458 
1459 	if (err != 0)
1460 		goto failure;
1461 
1462 	if (fib_props[cfg->fc_type].error) {
1463 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1464 			NL_SET_ERR_MSG(extack,
1465 				       "Gateway, device and multipath can not be specified for this route type");
1466 			goto err_inval;
1467 		}
1468 		goto link_it;
1469 	} else {
1470 		switch (cfg->fc_type) {
1471 		case RTN_UNICAST:
1472 		case RTN_LOCAL:
1473 		case RTN_BROADCAST:
1474 		case RTN_ANYCAST:
1475 		case RTN_MULTICAST:
1476 			break;
1477 		default:
1478 			NL_SET_ERR_MSG(extack, "Invalid route type");
1479 			goto err_inval;
1480 		}
1481 	}
1482 
1483 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1484 		NL_SET_ERR_MSG(extack, "Invalid scope");
1485 		goto err_inval;
1486 	}
1487 
1488 	if (fi->nh) {
1489 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1490 		if (err)
1491 			goto failure;
1492 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1493 		struct fib_nh *nh = fi->fib_nh;
1494 
1495 		/* Local address is added. */
1496 		if (nhs != 1) {
1497 			NL_SET_ERR_MSG(extack,
1498 				       "Route with host scope can not have multiple nexthops");
1499 			goto err_inval;
1500 		}
1501 		if (nh->fib_nh_gw_family) {
1502 			NL_SET_ERR_MSG(extack,
1503 				       "Route with host scope can not have a gateway");
1504 			goto err_inval;
1505 		}
1506 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1507 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1508 		err = -ENODEV;
1509 		if (!nh->fib_nh_dev)
1510 			goto failure;
1511 	} else {
1512 		int linkdown = 0;
1513 
1514 		change_nexthops(fi) {
1515 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1516 					   cfg->fc_table, cfg->fc_scope,
1517 					   extack);
1518 			if (err != 0)
1519 				goto failure;
1520 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1521 				linkdown++;
1522 		} endfor_nexthops(fi)
1523 		if (linkdown == fi->fib_nhs)
1524 			fi->fib_flags |= RTNH_F_LINKDOWN;
1525 	}
1526 
1527 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1528 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1529 		goto err_inval;
1530 	}
1531 
1532 	if (!fi->nh) {
1533 		change_nexthops(fi) {
1534 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1535 						  fi->fib_scope);
1536 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1537 				fi->fib_nh_is_v6 = true;
1538 		} endfor_nexthops(fi)
1539 
1540 		fib_rebalance(fi);
1541 	}
1542 
1543 link_it:
1544 	ofi = fib_find_info(fi);
1545 	if (ofi) {
1546 		fi->fib_dead = 1;
1547 		free_fib_info(fi);
1548 		refcount_inc(&ofi->fib_treeref);
1549 		return ofi;
1550 	}
1551 
1552 	refcount_set(&fi->fib_treeref, 1);
1553 	refcount_set(&fi->fib_clntref, 1);
1554 	spin_lock_bh(&fib_info_lock);
1555 	hlist_add_head(&fi->fib_hash,
1556 		       &fib_info_hash[fib_info_hashfn(fi)]);
1557 	if (fi->fib_prefsrc) {
1558 		struct hlist_head *head;
1559 
1560 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1561 		hlist_add_head(&fi->fib_lhash, head);
1562 	}
1563 	if (fi->nh) {
1564 		list_add(&fi->nh_list, &nh->fi_list);
1565 	} else {
1566 		change_nexthops(fi) {
1567 			struct hlist_head *head;
1568 			unsigned int hash;
1569 
1570 			if (!nexthop_nh->fib_nh_dev)
1571 				continue;
1572 			hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1573 			head = &fib_info_devhash[hash];
1574 			hlist_add_head(&nexthop_nh->nh_hash, head);
1575 		} endfor_nexthops(fi)
1576 	}
1577 	spin_unlock_bh(&fib_info_lock);
1578 	return fi;
1579 
1580 err_inval:
1581 	err = -EINVAL;
1582 
1583 failure:
1584 	if (fi) {
1585 		fi->fib_dead = 1;
1586 		free_fib_info(fi);
1587 	}
1588 
1589 	return ERR_PTR(err);
1590 }
1591 
1592 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1593 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1594 {
1595 	if (nhc->nhc_flags & RTNH_F_DEAD)
1596 		*flags |= RTNH_F_DEAD;
1597 
1598 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1599 		*flags |= RTNH_F_LINKDOWN;
1600 
1601 		rcu_read_lock();
1602 		switch (nhc->nhc_family) {
1603 		case AF_INET:
1604 			if (ip_ignore_linkdown(nhc->nhc_dev))
1605 				*flags |= RTNH_F_DEAD;
1606 			break;
1607 		case AF_INET6:
1608 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1609 				*flags |= RTNH_F_DEAD;
1610 			break;
1611 		}
1612 		rcu_read_unlock();
1613 	}
1614 
1615 	switch (nhc->nhc_gw_family) {
1616 	case AF_INET:
1617 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1618 			goto nla_put_failure;
1619 		break;
1620 	case AF_INET6:
1621 		/* if gateway family does not match nexthop family
1622 		 * gateway is encoded as RTA_VIA
1623 		 */
1624 		if (rt_family != nhc->nhc_gw_family) {
1625 			int alen = sizeof(struct in6_addr);
1626 			struct nlattr *nla;
1627 			struct rtvia *via;
1628 
1629 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1630 			if (!nla)
1631 				goto nla_put_failure;
1632 
1633 			via = nla_data(nla);
1634 			via->rtvia_family = AF_INET6;
1635 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1636 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1637 					    &nhc->nhc_gw.ipv6) < 0) {
1638 			goto nla_put_failure;
1639 		}
1640 		break;
1641 	}
1642 
1643 	*flags |= (nhc->nhc_flags &
1644 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1645 
1646 	if (!skip_oif && nhc->nhc_dev &&
1647 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1648 		goto nla_put_failure;
1649 
1650 	if (nhc->nhc_lwtstate &&
1651 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1652 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1653 		goto nla_put_failure;
1654 
1655 	return 0;
1656 
1657 nla_put_failure:
1658 	return -EMSGSIZE;
1659 }
1660 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1661 
1662 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1663 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1664 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1665 {
1666 	const struct net_device *dev = nhc->nhc_dev;
1667 	struct rtnexthop *rtnh;
1668 	unsigned char flags = 0;
1669 
1670 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1671 	if (!rtnh)
1672 		goto nla_put_failure;
1673 
1674 	rtnh->rtnh_hops = nh_weight - 1;
1675 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1676 
1677 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1678 		goto nla_put_failure;
1679 
1680 	rtnh->rtnh_flags = flags;
1681 
1682 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1683 		goto nla_put_failure;
1684 
1685 	/* length of rtnetlink header + attributes */
1686 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1687 
1688 	return 0;
1689 
1690 nla_put_failure:
1691 	return -EMSGSIZE;
1692 }
1693 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1694 #endif
1695 
1696 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1697 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1698 {
1699 	struct nlattr *mp;
1700 
1701 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1702 	if (!mp)
1703 		goto nla_put_failure;
1704 
1705 	if (unlikely(fi->nh)) {
1706 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1707 			goto nla_put_failure;
1708 		goto mp_end;
1709 	}
1710 
1711 	for_nexthops(fi) {
1712 		u32 nh_tclassid = 0;
1713 #ifdef CONFIG_IP_ROUTE_CLASSID
1714 		nh_tclassid = nh->nh_tclassid;
1715 #endif
1716 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1717 				    AF_INET, nh_tclassid) < 0)
1718 			goto nla_put_failure;
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 		break;
1878 	default:
1879 		break;
1880 	}
1881 
1882 	return NOTIFY_DONE;
1883 }
1884 
1885 /* Update the PMTU of exceptions when:
1886  * - the new MTU of the first hop becomes smaller than the PMTU
1887  * - the old MTU was the same as the PMTU, and it limited discovery of
1888  *   larger MTUs on the path. With that limit raised, we can now
1889  *   discover larger MTUs
1890  * A special case is locked exceptions, for which the PMTU is smaller
1891  * than the minimal accepted PMTU:
1892  * - if the new MTU is greater than the PMTU, don't make any change
1893  * - otherwise, unlock and set PMTU
1894  */
1895 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1896 {
1897 	struct fnhe_hash_bucket *bucket;
1898 	int i;
1899 
1900 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1901 	if (!bucket)
1902 		return;
1903 
1904 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1905 		struct fib_nh_exception *fnhe;
1906 
1907 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1908 		     fnhe;
1909 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1910 			if (fnhe->fnhe_mtu_locked) {
1911 				if (new <= fnhe->fnhe_pmtu) {
1912 					fnhe->fnhe_pmtu = new;
1913 					fnhe->fnhe_mtu_locked = false;
1914 				}
1915 			} else if (new < fnhe->fnhe_pmtu ||
1916 				   orig == fnhe->fnhe_pmtu) {
1917 				fnhe->fnhe_pmtu = new;
1918 			}
1919 		}
1920 	}
1921 }
1922 
1923 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1924 {
1925 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1926 	struct hlist_head *head = &fib_info_devhash[hash];
1927 	struct fib_nh *nh;
1928 
1929 	hlist_for_each_entry(nh, head, nh_hash) {
1930 		if (nh->fib_nh_dev == dev)
1931 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1932 	}
1933 }
1934 
1935 /* Event              force Flags           Description
1936  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1937  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1938  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1939  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1940  *
1941  * only used when fib_nh is built into fib_info
1942  */
1943 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1944 {
1945 	int ret = 0;
1946 	int scope = RT_SCOPE_NOWHERE;
1947 	struct fib_info *prev_fi = NULL;
1948 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1949 	struct hlist_head *head = &fib_info_devhash[hash];
1950 	struct fib_nh *nh;
1951 
1952 	if (force)
1953 		scope = -1;
1954 
1955 	hlist_for_each_entry(nh, head, nh_hash) {
1956 		struct fib_info *fi = nh->nh_parent;
1957 		int dead;
1958 
1959 		BUG_ON(!fi->fib_nhs);
1960 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1961 			continue;
1962 		prev_fi = fi;
1963 		dead = 0;
1964 		change_nexthops(fi) {
1965 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1966 				dead++;
1967 			else if (nexthop_nh->fib_nh_dev == dev &&
1968 				 nexthop_nh->fib_nh_scope != scope) {
1969 				switch (event) {
1970 				case NETDEV_DOWN:
1971 				case NETDEV_UNREGISTER:
1972 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1973 					fallthrough;
1974 				case NETDEV_CHANGE:
1975 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1976 					break;
1977 				}
1978 				call_fib_nh_notifiers(nexthop_nh,
1979 						      FIB_EVENT_NH_DEL);
1980 				dead++;
1981 			}
1982 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1983 			if (event == NETDEV_UNREGISTER &&
1984 			    nexthop_nh->fib_nh_dev == dev) {
1985 				dead = fi->fib_nhs;
1986 				break;
1987 			}
1988 #endif
1989 		} endfor_nexthops(fi)
1990 		if (dead == fi->fib_nhs) {
1991 			switch (event) {
1992 			case NETDEV_DOWN:
1993 			case NETDEV_UNREGISTER:
1994 				fi->fib_flags |= RTNH_F_DEAD;
1995 				fallthrough;
1996 			case NETDEV_CHANGE:
1997 				fi->fib_flags |= RTNH_F_LINKDOWN;
1998 				break;
1999 			}
2000 			ret++;
2001 		}
2002 
2003 		fib_rebalance(fi);
2004 	}
2005 
2006 	return ret;
2007 }
2008 
2009 /* Must be invoked inside of an RCU protected region.  */
2010 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2011 {
2012 	struct fib_info *fi = NULL, *last_resort = NULL;
2013 	struct hlist_head *fa_head = res->fa_head;
2014 	struct fib_table *tb = res->table;
2015 	u8 slen = 32 - res->prefixlen;
2016 	int order = -1, last_idx = -1;
2017 	struct fib_alias *fa, *fa1 = NULL;
2018 	u32 last_prio = res->fi->fib_priority;
2019 	u8 last_tos = 0;
2020 
2021 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2022 		struct fib_info *next_fi = fa->fa_info;
2023 		struct fib_nh_common *nhc;
2024 
2025 		if (fa->fa_slen != slen)
2026 			continue;
2027 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2028 			continue;
2029 		if (fa->tb_id != tb->tb_id)
2030 			continue;
2031 		if (next_fi->fib_priority > last_prio &&
2032 		    fa->fa_tos == last_tos) {
2033 			if (last_tos)
2034 				continue;
2035 			break;
2036 		}
2037 		if (next_fi->fib_flags & RTNH_F_DEAD)
2038 			continue;
2039 		last_tos = fa->fa_tos;
2040 		last_prio = next_fi->fib_priority;
2041 
2042 		if (next_fi->fib_scope != res->scope ||
2043 		    fa->fa_type != RTN_UNICAST)
2044 			continue;
2045 
2046 		nhc = fib_info_nhc(next_fi, 0);
2047 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2048 			continue;
2049 
2050 		fib_alias_accessed(fa);
2051 
2052 		if (!fi) {
2053 			if (next_fi != res->fi)
2054 				break;
2055 			fa1 = fa;
2056 		} else if (!fib_detect_death(fi, order, &last_resort,
2057 					     &last_idx, fa1->fa_default)) {
2058 			fib_result_assign(res, fi);
2059 			fa1->fa_default = order;
2060 			goto out;
2061 		}
2062 		fi = next_fi;
2063 		order++;
2064 	}
2065 
2066 	if (order <= 0 || !fi) {
2067 		if (fa1)
2068 			fa1->fa_default = -1;
2069 		goto out;
2070 	}
2071 
2072 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2073 			      fa1->fa_default)) {
2074 		fib_result_assign(res, fi);
2075 		fa1->fa_default = order;
2076 		goto out;
2077 	}
2078 
2079 	if (last_idx >= 0)
2080 		fib_result_assign(res, last_resort);
2081 	fa1->fa_default = last_idx;
2082 out:
2083 	return;
2084 }
2085 
2086 /*
2087  * Dead device goes up. We wake up dead nexthops.
2088  * It takes sense only on multipath routes.
2089  *
2090  * only used when fib_nh is built into fib_info
2091  */
2092 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2093 {
2094 	struct fib_info *prev_fi;
2095 	unsigned int hash;
2096 	struct hlist_head *head;
2097 	struct fib_nh *nh;
2098 	int ret;
2099 
2100 	if (!(dev->flags & IFF_UP))
2101 		return 0;
2102 
2103 	if (nh_flags & RTNH_F_DEAD) {
2104 		unsigned int flags = dev_get_flags(dev);
2105 
2106 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2107 			nh_flags |= RTNH_F_LINKDOWN;
2108 	}
2109 
2110 	prev_fi = NULL;
2111 	hash = fib_devindex_hashfn(dev->ifindex);
2112 	head = &fib_info_devhash[hash];
2113 	ret = 0;
2114 
2115 	hlist_for_each_entry(nh, head, nh_hash) {
2116 		struct fib_info *fi = nh->nh_parent;
2117 		int alive;
2118 
2119 		BUG_ON(!fi->fib_nhs);
2120 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2121 			continue;
2122 
2123 		prev_fi = fi;
2124 		alive = 0;
2125 		change_nexthops(fi) {
2126 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2127 				alive++;
2128 				continue;
2129 			}
2130 			if (!nexthop_nh->fib_nh_dev ||
2131 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2132 				continue;
2133 			if (nexthop_nh->fib_nh_dev != dev ||
2134 			    !__in_dev_get_rtnl(dev))
2135 				continue;
2136 			alive++;
2137 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2138 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2139 		} endfor_nexthops(fi)
2140 
2141 		if (alive > 0) {
2142 			fi->fib_flags &= ~nh_flags;
2143 			ret++;
2144 		}
2145 
2146 		fib_rebalance(fi);
2147 	}
2148 
2149 	return ret;
2150 }
2151 
2152 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2153 static bool fib_good_nh(const struct fib_nh *nh)
2154 {
2155 	int state = NUD_REACHABLE;
2156 
2157 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2158 		struct neighbour *n;
2159 
2160 		rcu_read_lock_bh();
2161 
2162 		if (likely(nh->fib_nh_gw_family == AF_INET))
2163 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2164 						   (__force u32)nh->fib_nh_gw4);
2165 		else if (nh->fib_nh_gw_family == AF_INET6)
2166 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2167 							   &nh->fib_nh_gw6);
2168 		else
2169 			n = NULL;
2170 		if (n)
2171 			state = n->nud_state;
2172 
2173 		rcu_read_unlock_bh();
2174 	}
2175 
2176 	return !!(state & NUD_VALID);
2177 }
2178 
2179 void fib_select_multipath(struct fib_result *res, int hash)
2180 {
2181 	struct fib_info *fi = res->fi;
2182 	struct net *net = fi->fib_net;
2183 	bool first = false;
2184 
2185 	if (unlikely(res->fi->nh)) {
2186 		nexthop_path_fib_result(res, hash);
2187 		return;
2188 	}
2189 
2190 	change_nexthops(fi) {
2191 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2192 			if (!fib_good_nh(nexthop_nh))
2193 				continue;
2194 			if (!first) {
2195 				res->nh_sel = nhsel;
2196 				res->nhc = &nexthop_nh->nh_common;
2197 				first = true;
2198 			}
2199 		}
2200 
2201 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2202 			continue;
2203 
2204 		res->nh_sel = nhsel;
2205 		res->nhc = &nexthop_nh->nh_common;
2206 		return;
2207 	} endfor_nexthops(fi);
2208 }
2209 #endif
2210 
2211 void fib_select_path(struct net *net, struct fib_result *res,
2212 		     struct flowi4 *fl4, const struct sk_buff *skb)
2213 {
2214 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2215 		goto check_saddr;
2216 
2217 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2218 	if (fib_info_num_path(res->fi) > 1) {
2219 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2220 
2221 		fib_select_multipath(res, h);
2222 	}
2223 	else
2224 #endif
2225 	if (!res->prefixlen &&
2226 	    res->table->tb_num_default > 1 &&
2227 	    res->type == RTN_UNICAST)
2228 		fib_select_default(fl4, res);
2229 
2230 check_saddr:
2231 	if (!fl4->saddr)
2232 		fl4->saddr = fib_result_prefsrc(net, res);
2233 }
2234