xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision f519cd13)
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 sk_buff *skb;
508 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
509 	int err = -ENOBUFS;
510 
511 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
512 	if (!skb)
513 		goto errout;
514 
515 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
516 			    fa->fa_type, key, dst_len,
517 			    fa->fa_tos, fa->fa_info, nlm_flags);
518 	if (err < 0) {
519 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
520 		WARN_ON(err == -EMSGSIZE);
521 		kfree_skb(skb);
522 		goto errout;
523 	}
524 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
525 		    info->nlh, GFP_KERNEL);
526 	return;
527 errout:
528 	if (err < 0)
529 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
530 }
531 
532 static int fib_detect_death(struct fib_info *fi, int order,
533 			    struct fib_info **last_resort, int *last_idx,
534 			    int dflt)
535 {
536 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
537 	struct neighbour *n;
538 	int state = NUD_NONE;
539 
540 	if (likely(nhc->nhc_gw_family == AF_INET))
541 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
542 	else if (nhc->nhc_gw_family == AF_INET6)
543 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
544 				 nhc->nhc_dev);
545 	else
546 		n = NULL;
547 
548 	if (n) {
549 		state = n->nud_state;
550 		neigh_release(n);
551 	} else {
552 		return 0;
553 	}
554 	if (state == NUD_REACHABLE)
555 		return 0;
556 	if ((state & NUD_VALID) && order != dflt)
557 		return 0;
558 	if ((state & NUD_VALID) ||
559 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
560 		*last_resort = fi;
561 		*last_idx = order;
562 	}
563 	return 1;
564 }
565 
566 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
567 		       u16 encap_type, void *cfg, gfp_t gfp_flags,
568 		       struct netlink_ext_ack *extack)
569 {
570 	int err;
571 
572 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
573 						    gfp_flags);
574 	if (!nhc->nhc_pcpu_rth_output)
575 		return -ENOMEM;
576 
577 	if (encap) {
578 		struct lwtunnel_state *lwtstate;
579 
580 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
581 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
582 			err = -EINVAL;
583 			goto lwt_failure;
584 		}
585 		err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family,
586 					   cfg, &lwtstate, extack);
587 		if (err)
588 			goto lwt_failure;
589 
590 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
591 	}
592 
593 	return 0;
594 
595 lwt_failure:
596 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
597 	nhc->nhc_pcpu_rth_output = NULL;
598 	return err;
599 }
600 EXPORT_SYMBOL_GPL(fib_nh_common_init);
601 
602 int fib_nh_init(struct net *net, struct fib_nh *nh,
603 		struct fib_config *cfg, int nh_weight,
604 		struct netlink_ext_ack *extack)
605 {
606 	int err;
607 
608 	nh->fib_nh_family = AF_INET;
609 
610 	err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
611 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
612 	if (err)
613 		return err;
614 
615 	nh->fib_nh_oif = cfg->fc_oif;
616 	nh->fib_nh_gw_family = cfg->fc_gw_family;
617 	if (cfg->fc_gw_family == AF_INET)
618 		nh->fib_nh_gw4 = cfg->fc_gw4;
619 	else if (cfg->fc_gw_family == AF_INET6)
620 		nh->fib_nh_gw6 = cfg->fc_gw6;
621 
622 	nh->fib_nh_flags = cfg->fc_flags;
623 
624 #ifdef CONFIG_IP_ROUTE_CLASSID
625 	nh->nh_tclassid = cfg->fc_flow;
626 	if (nh->nh_tclassid)
627 		net->ipv4.fib_num_tclassid_users++;
628 #endif
629 #ifdef CONFIG_IP_ROUTE_MULTIPATH
630 	nh->fib_nh_weight = nh_weight;
631 #endif
632 	return 0;
633 }
634 
635 #ifdef CONFIG_IP_ROUTE_MULTIPATH
636 
637 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
638 			      struct netlink_ext_ack *extack)
639 {
640 	int nhs = 0;
641 
642 	while (rtnh_ok(rtnh, remaining)) {
643 		nhs++;
644 		rtnh = rtnh_next(rtnh, &remaining);
645 	}
646 
647 	/* leftover implies invalid nexthop configuration, discard it */
648 	if (remaining > 0) {
649 		NL_SET_ERR_MSG(extack,
650 			       "Invalid nexthop configuration - extra data after nexthops");
651 		nhs = 0;
652 	}
653 
654 	return nhs;
655 }
656 
657 /* only called when fib_nh is integrated into fib_info */
658 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
659 		       int remaining, struct fib_config *cfg,
660 		       struct netlink_ext_ack *extack)
661 {
662 	struct net *net = fi->fib_net;
663 	struct fib_config fib_cfg;
664 	struct fib_nh *nh;
665 	int ret;
666 
667 	change_nexthops(fi) {
668 		int attrlen;
669 
670 		memset(&fib_cfg, 0, sizeof(fib_cfg));
671 
672 		if (!rtnh_ok(rtnh, remaining)) {
673 			NL_SET_ERR_MSG(extack,
674 				       "Invalid nexthop configuration - extra data after nexthop");
675 			return -EINVAL;
676 		}
677 
678 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
679 			NL_SET_ERR_MSG(extack,
680 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
681 			return -EINVAL;
682 		}
683 
684 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
685 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
686 
687 		attrlen = rtnh_attrlen(rtnh);
688 		if (attrlen > 0) {
689 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
690 
691 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
692 			nlav = nla_find(attrs, attrlen, RTA_VIA);
693 			if (nla && nlav) {
694 				NL_SET_ERR_MSG(extack,
695 					       "Nexthop configuration can not contain both GATEWAY and VIA");
696 				return -EINVAL;
697 			}
698 			if (nla) {
699 				fib_cfg.fc_gw4 = nla_get_in_addr(nla);
700 				if (fib_cfg.fc_gw4)
701 					fib_cfg.fc_gw_family = AF_INET;
702 			} else if (nlav) {
703 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
704 				if (ret)
705 					goto errout;
706 			}
707 
708 			nla = nla_find(attrs, attrlen, RTA_FLOW);
709 			if (nla)
710 				fib_cfg.fc_flow = nla_get_u32(nla);
711 
712 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
713 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
714 			if (nla)
715 				fib_cfg.fc_encap_type = nla_get_u16(nla);
716 		}
717 
718 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
719 				  rtnh->rtnh_hops + 1, extack);
720 		if (ret)
721 			goto errout;
722 
723 		rtnh = rtnh_next(rtnh, &remaining);
724 	} endfor_nexthops(fi);
725 
726 	ret = -EINVAL;
727 	nh = fib_info_nh(fi, 0);
728 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
729 		NL_SET_ERR_MSG(extack,
730 			       "Nexthop device index does not match RTA_OIF");
731 		goto errout;
732 	}
733 	if (cfg->fc_gw_family) {
734 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
735 		    (cfg->fc_gw_family == AF_INET &&
736 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
737 		    (cfg->fc_gw_family == AF_INET6 &&
738 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
739 			NL_SET_ERR_MSG(extack,
740 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
741 			goto errout;
742 		}
743 	}
744 #ifdef CONFIG_IP_ROUTE_CLASSID
745 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
746 		NL_SET_ERR_MSG(extack,
747 			       "Nexthop class id does not match RTA_FLOW");
748 		goto errout;
749 	}
750 #endif
751 	ret = 0;
752 errout:
753 	return ret;
754 }
755 
756 /* only called when fib_nh is integrated into fib_info */
757 static void fib_rebalance(struct fib_info *fi)
758 {
759 	int total;
760 	int w;
761 
762 	if (fib_info_num_path(fi) < 2)
763 		return;
764 
765 	total = 0;
766 	for_nexthops(fi) {
767 		if (nh->fib_nh_flags & RTNH_F_DEAD)
768 			continue;
769 
770 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
771 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
772 			continue;
773 
774 		total += nh->fib_nh_weight;
775 	} endfor_nexthops(fi);
776 
777 	w = 0;
778 	change_nexthops(fi) {
779 		int upper_bound;
780 
781 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
782 			upper_bound = -1;
783 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
784 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
785 			upper_bound = -1;
786 		} else {
787 			w += nexthop_nh->fib_nh_weight;
788 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
789 							    total) - 1;
790 		}
791 
792 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
793 	} endfor_nexthops(fi);
794 }
795 #else /* CONFIG_IP_ROUTE_MULTIPATH */
796 
797 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
798 		       int remaining, struct fib_config *cfg,
799 		       struct netlink_ext_ack *extack)
800 {
801 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
802 
803 	return -EINVAL;
804 }
805 
806 #define fib_rebalance(fi) do { } while (0)
807 
808 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
809 
810 static int fib_encap_match(u16 encap_type,
811 			   struct nlattr *encap,
812 			   const struct fib_nh *nh,
813 			   const struct fib_config *cfg,
814 			   struct netlink_ext_ack *extack)
815 {
816 	struct lwtunnel_state *lwtstate;
817 	int ret, result = 0;
818 
819 	if (encap_type == LWTUNNEL_ENCAP_NONE)
820 		return 0;
821 
822 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
823 				   cfg, &lwtstate, extack);
824 	if (!ret) {
825 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
826 		lwtstate_free(lwtstate);
827 	}
828 
829 	return result;
830 }
831 
832 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
833 		 struct netlink_ext_ack *extack)
834 {
835 #ifdef CONFIG_IP_ROUTE_MULTIPATH
836 	struct rtnexthop *rtnh;
837 	int remaining;
838 #endif
839 
840 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
841 		return 1;
842 
843 	if (cfg->fc_nh_id) {
844 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
845 			return 0;
846 		return 1;
847 	}
848 
849 	if (cfg->fc_oif || cfg->fc_gw_family) {
850 		struct fib_nh *nh = fib_info_nh(fi, 0);
851 
852 		if (cfg->fc_encap) {
853 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
854 					    nh, cfg, extack))
855 				return 1;
856 		}
857 #ifdef CONFIG_IP_ROUTE_CLASSID
858 		if (cfg->fc_flow &&
859 		    cfg->fc_flow != nh->nh_tclassid)
860 			return 1;
861 #endif
862 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
863 		    (cfg->fc_gw_family &&
864 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
865 			return 1;
866 
867 		if (cfg->fc_gw_family == AF_INET &&
868 		    cfg->fc_gw4 != nh->fib_nh_gw4)
869 			return 1;
870 
871 		if (cfg->fc_gw_family == AF_INET6 &&
872 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
873 			return 1;
874 
875 		return 0;
876 	}
877 
878 #ifdef CONFIG_IP_ROUTE_MULTIPATH
879 	if (!cfg->fc_mp)
880 		return 0;
881 
882 	rtnh = cfg->fc_mp;
883 	remaining = cfg->fc_mp_len;
884 
885 	for_nexthops(fi) {
886 		int attrlen;
887 
888 		if (!rtnh_ok(rtnh, remaining))
889 			return -EINVAL;
890 
891 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
892 			return 1;
893 
894 		attrlen = rtnh_attrlen(rtnh);
895 		if (attrlen > 0) {
896 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
897 
898 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
899 			nlav = nla_find(attrs, attrlen, RTA_VIA);
900 			if (nla && nlav) {
901 				NL_SET_ERR_MSG(extack,
902 					       "Nexthop configuration can not contain both GATEWAY and VIA");
903 				return -EINVAL;
904 			}
905 
906 			if (nla) {
907 				if (nh->fib_nh_gw_family != AF_INET ||
908 				    nla_get_in_addr(nla) != nh->fib_nh_gw4)
909 					return 1;
910 			} else if (nlav) {
911 				struct fib_config cfg2;
912 				int err;
913 
914 				err = fib_gw_from_via(&cfg2, nlav, extack);
915 				if (err)
916 					return err;
917 
918 				switch (nh->fib_nh_gw_family) {
919 				case AF_INET:
920 					if (cfg2.fc_gw_family != AF_INET ||
921 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
922 						return 1;
923 					break;
924 				case AF_INET6:
925 					if (cfg2.fc_gw_family != AF_INET6 ||
926 					    ipv6_addr_cmp(&cfg2.fc_gw6,
927 							  &nh->fib_nh_gw6))
928 						return 1;
929 					break;
930 				}
931 			}
932 
933 #ifdef CONFIG_IP_ROUTE_CLASSID
934 			nla = nla_find(attrs, attrlen, RTA_FLOW);
935 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
936 				return 1;
937 #endif
938 		}
939 
940 		rtnh = rtnh_next(rtnh, &remaining);
941 	} endfor_nexthops(fi);
942 #endif
943 	return 0;
944 }
945 
946 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
947 {
948 	struct nlattr *nla;
949 	int remaining;
950 
951 	if (!cfg->fc_mx)
952 		return true;
953 
954 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
955 		int type = nla_type(nla);
956 		u32 fi_val, val;
957 
958 		if (!type)
959 			continue;
960 		if (type > RTAX_MAX)
961 			return false;
962 
963 		if (type == RTAX_CC_ALGO) {
964 			char tmp[TCP_CA_NAME_MAX];
965 			bool ecn_ca = false;
966 
967 			nla_strlcpy(tmp, nla, sizeof(tmp));
968 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
969 		} else {
970 			if (nla_len(nla) != sizeof(u32))
971 				return false;
972 			val = nla_get_u32(nla);
973 		}
974 
975 		fi_val = fi->fib_metrics->metrics[type - 1];
976 		if (type == RTAX_FEATURES)
977 			fi_val &= ~DST_FEATURE_ECN_CA;
978 
979 		if (fi_val != val)
980 			return false;
981 	}
982 
983 	return true;
984 }
985 
986 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
987 			      u32 table, struct netlink_ext_ack *extack)
988 {
989 	struct fib6_config cfg = {
990 		.fc_table = table,
991 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
992 		.fc_ifindex = nh->fib_nh_oif,
993 		.fc_gateway = nh->fib_nh_gw6,
994 	};
995 	struct fib6_nh fib6_nh = {};
996 	int err;
997 
998 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
999 	if (!err) {
1000 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1001 		dev_hold(nh->fib_nh_dev);
1002 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1003 		nh->fib_nh_scope = RT_SCOPE_LINK;
1004 
1005 		ipv6_stub->fib6_nh_release(&fib6_nh);
1006 	}
1007 
1008 	return err;
1009 }
1010 
1011 /*
1012  * Picture
1013  * -------
1014  *
1015  * Semantics of nexthop is very messy by historical reasons.
1016  * We have to take into account, that:
1017  * a) gateway can be actually local interface address,
1018  *    so that gatewayed route is direct.
1019  * b) gateway must be on-link address, possibly
1020  *    described not by an ifaddr, but also by a direct route.
1021  * c) If both gateway and interface are specified, they should not
1022  *    contradict.
1023  * d) If we use tunnel routes, gateway could be not on-link.
1024  *
1025  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1026  * results in pretty ugly and hairy code with obscure logic.
1027  *
1028  * I chose to generalized it instead, so that the size
1029  * of code does not increase practically, but it becomes
1030  * much more general.
1031  * Every prefix is assigned a "scope" value: "host" is local address,
1032  * "link" is direct route,
1033  * [ ... "site" ... "interior" ... ]
1034  * and "universe" is true gateway route with global meaning.
1035  *
1036  * Every prefix refers to a set of "nexthop"s (gw, oif),
1037  * where gw must have narrower scope. This recursion stops
1038  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1039  * which means that gw is forced to be on link.
1040  *
1041  * Code is still hairy, but now it is apparently logically
1042  * consistent and very flexible. F.e. as by-product it allows
1043  * to co-exists in peace independent exterior and interior
1044  * routing processes.
1045  *
1046  * Normally it looks as following.
1047  *
1048  * {universe prefix}  -> (gw, oif) [scope link]
1049  *		  |
1050  *		  |-> {link prefix} -> (gw, oif) [scope local]
1051  *					|
1052  *					|-> {local prefix} (terminal node)
1053  */
1054 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1055 			      u8 scope, struct netlink_ext_ack *extack)
1056 {
1057 	struct net_device *dev;
1058 	struct fib_result res;
1059 	int err = 0;
1060 
1061 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1062 		unsigned int addr_type;
1063 
1064 		if (scope >= RT_SCOPE_LINK) {
1065 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1066 			return -EINVAL;
1067 		}
1068 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1069 		if (!dev) {
1070 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1071 			return -ENODEV;
1072 		}
1073 		if (!(dev->flags & IFF_UP)) {
1074 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1075 			return -ENETDOWN;
1076 		}
1077 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1078 		if (addr_type != RTN_UNICAST) {
1079 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1080 			return -EINVAL;
1081 		}
1082 		if (!netif_carrier_ok(dev))
1083 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1084 		nh->fib_nh_dev = dev;
1085 		dev_hold(dev);
1086 		nh->fib_nh_scope = RT_SCOPE_LINK;
1087 		return 0;
1088 	}
1089 	rcu_read_lock();
1090 	{
1091 		struct fib_table *tbl = NULL;
1092 		struct flowi4 fl4 = {
1093 			.daddr = nh->fib_nh_gw4,
1094 			.flowi4_scope = scope + 1,
1095 			.flowi4_oif = nh->fib_nh_oif,
1096 			.flowi4_iif = LOOPBACK_IFINDEX,
1097 		};
1098 
1099 		/* It is not necessary, but requires a bit of thinking */
1100 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1101 			fl4.flowi4_scope = RT_SCOPE_LINK;
1102 
1103 		if (table)
1104 			tbl = fib_get_table(net, table);
1105 
1106 		if (tbl)
1107 			err = fib_table_lookup(tbl, &fl4, &res,
1108 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1109 					       FIB_LOOKUP_NOREF);
1110 
1111 		/* on error or if no table given do full lookup. This
1112 		 * is needed for example when nexthops are in the local
1113 		 * table rather than the given table
1114 		 */
1115 		if (!tbl || err) {
1116 			err = fib_lookup(net, &fl4, &res,
1117 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1118 		}
1119 
1120 		if (err) {
1121 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1122 			goto out;
1123 		}
1124 	}
1125 
1126 	err = -EINVAL;
1127 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1128 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1129 		goto out;
1130 	}
1131 	nh->fib_nh_scope = res.scope;
1132 	nh->fib_nh_oif = FIB_RES_OIF(res);
1133 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1134 	if (!dev) {
1135 		NL_SET_ERR_MSG(extack,
1136 			       "No egress device for nexthop gateway");
1137 		goto out;
1138 	}
1139 	dev_hold(dev);
1140 	if (!netif_carrier_ok(dev))
1141 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1142 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1143 out:
1144 	rcu_read_unlock();
1145 	return err;
1146 }
1147 
1148 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1149 			      struct netlink_ext_ack *extack)
1150 {
1151 	struct in_device *in_dev;
1152 	int err;
1153 
1154 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1155 		NL_SET_ERR_MSG(extack,
1156 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1157 		return -EINVAL;
1158 	}
1159 
1160 	rcu_read_lock();
1161 
1162 	err = -ENODEV;
1163 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1164 	if (!in_dev)
1165 		goto out;
1166 	err = -ENETDOWN;
1167 	if (!(in_dev->dev->flags & IFF_UP)) {
1168 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1169 		goto out;
1170 	}
1171 
1172 	nh->fib_nh_dev = in_dev->dev;
1173 	dev_hold(nh->fib_nh_dev);
1174 	nh->fib_nh_scope = RT_SCOPE_HOST;
1175 	if (!netif_carrier_ok(nh->fib_nh_dev))
1176 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1177 	err = 0;
1178 out:
1179 	rcu_read_unlock();
1180 	return err;
1181 }
1182 
1183 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1184 		 struct netlink_ext_ack *extack)
1185 {
1186 	int err;
1187 
1188 	if (nh->fib_nh_gw_family == AF_INET)
1189 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1190 	else if (nh->fib_nh_gw_family == AF_INET6)
1191 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1192 	else
1193 		err = fib_check_nh_nongw(net, nh, extack);
1194 
1195 	return err;
1196 }
1197 
1198 static inline unsigned int fib_laddr_hashfn(__be32 val)
1199 {
1200 	unsigned int mask = (fib_info_hash_size - 1);
1201 
1202 	return ((__force u32)val ^
1203 		((__force u32)val >> 7) ^
1204 		((__force u32)val >> 14)) & mask;
1205 }
1206 
1207 static struct hlist_head *fib_info_hash_alloc(int bytes)
1208 {
1209 	if (bytes <= PAGE_SIZE)
1210 		return kzalloc(bytes, GFP_KERNEL);
1211 	else
1212 		return (struct hlist_head *)
1213 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1214 					 get_order(bytes));
1215 }
1216 
1217 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1218 {
1219 	if (!hash)
1220 		return;
1221 
1222 	if (bytes <= PAGE_SIZE)
1223 		kfree(hash);
1224 	else
1225 		free_pages((unsigned long) hash, get_order(bytes));
1226 }
1227 
1228 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1229 			       struct hlist_head *new_laddrhash,
1230 			       unsigned int new_size)
1231 {
1232 	struct hlist_head *old_info_hash, *old_laddrhash;
1233 	unsigned int old_size = fib_info_hash_size;
1234 	unsigned int i, bytes;
1235 
1236 	spin_lock_bh(&fib_info_lock);
1237 	old_info_hash = fib_info_hash;
1238 	old_laddrhash = fib_info_laddrhash;
1239 	fib_info_hash_size = new_size;
1240 
1241 	for (i = 0; i < old_size; i++) {
1242 		struct hlist_head *head = &fib_info_hash[i];
1243 		struct hlist_node *n;
1244 		struct fib_info *fi;
1245 
1246 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1247 			struct hlist_head *dest;
1248 			unsigned int new_hash;
1249 
1250 			new_hash = fib_info_hashfn(fi);
1251 			dest = &new_info_hash[new_hash];
1252 			hlist_add_head(&fi->fib_hash, dest);
1253 		}
1254 	}
1255 	fib_info_hash = new_info_hash;
1256 
1257 	for (i = 0; i < old_size; i++) {
1258 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1259 		struct hlist_node *n;
1260 		struct fib_info *fi;
1261 
1262 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1263 			struct hlist_head *ldest;
1264 			unsigned int new_hash;
1265 
1266 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1267 			ldest = &new_laddrhash[new_hash];
1268 			hlist_add_head(&fi->fib_lhash, ldest);
1269 		}
1270 	}
1271 	fib_info_laddrhash = new_laddrhash;
1272 
1273 	spin_unlock_bh(&fib_info_lock);
1274 
1275 	bytes = old_size * sizeof(struct hlist_head *);
1276 	fib_info_hash_free(old_info_hash, bytes);
1277 	fib_info_hash_free(old_laddrhash, bytes);
1278 }
1279 
1280 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1281 				 unsigned char scope)
1282 {
1283 	struct fib_nh *nh;
1284 
1285 	if (nhc->nhc_family != AF_INET)
1286 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1287 
1288 	nh = container_of(nhc, struct fib_nh, nh_common);
1289 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1290 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1291 
1292 	return nh->nh_saddr;
1293 }
1294 
1295 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1296 {
1297 	struct fib_nh_common *nhc = res->nhc;
1298 
1299 	if (res->fi->fib_prefsrc)
1300 		return res->fi->fib_prefsrc;
1301 
1302 	if (nhc->nhc_family == AF_INET) {
1303 		struct fib_nh *nh;
1304 
1305 		nh = container_of(nhc, struct fib_nh, nh_common);
1306 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1307 			return nh->nh_saddr;
1308 	}
1309 
1310 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1311 }
1312 
1313 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1314 {
1315 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1316 	    fib_prefsrc != cfg->fc_dst) {
1317 		u32 tb_id = cfg->fc_table;
1318 		int rc;
1319 
1320 		if (tb_id == RT_TABLE_MAIN)
1321 			tb_id = RT_TABLE_LOCAL;
1322 
1323 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1324 					  fib_prefsrc, tb_id);
1325 
1326 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1327 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1328 						  fib_prefsrc, RT_TABLE_LOCAL);
1329 		}
1330 
1331 		if (rc != RTN_LOCAL)
1332 			return false;
1333 	}
1334 	return true;
1335 }
1336 
1337 struct fib_info *fib_create_info(struct fib_config *cfg,
1338 				 struct netlink_ext_ack *extack)
1339 {
1340 	int err;
1341 	struct fib_info *fi = NULL;
1342 	struct nexthop *nh = NULL;
1343 	struct fib_info *ofi;
1344 	int nhs = 1;
1345 	struct net *net = cfg->fc_nlinfo.nl_net;
1346 
1347 	if (cfg->fc_type > RTN_MAX)
1348 		goto err_inval;
1349 
1350 	/* Fast check to catch the most weird cases */
1351 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1352 		NL_SET_ERR_MSG(extack, "Invalid scope");
1353 		goto err_inval;
1354 	}
1355 
1356 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1357 		NL_SET_ERR_MSG(extack,
1358 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1359 		goto err_inval;
1360 	}
1361 
1362 	if (cfg->fc_nh_id) {
1363 		if (!cfg->fc_mx) {
1364 			fi = fib_find_info_nh(net, cfg);
1365 			if (fi) {
1366 				fi->fib_treeref++;
1367 				return fi;
1368 			}
1369 		}
1370 
1371 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1372 		if (!nh) {
1373 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1374 			goto err_inval;
1375 		}
1376 		nhs = 0;
1377 	}
1378 
1379 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1380 	if (cfg->fc_mp) {
1381 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1382 		if (nhs == 0)
1383 			goto err_inval;
1384 	}
1385 #endif
1386 
1387 	err = -ENOBUFS;
1388 	if (fib_info_cnt >= fib_info_hash_size) {
1389 		unsigned int new_size = fib_info_hash_size << 1;
1390 		struct hlist_head *new_info_hash;
1391 		struct hlist_head *new_laddrhash;
1392 		unsigned int bytes;
1393 
1394 		if (!new_size)
1395 			new_size = 16;
1396 		bytes = new_size * sizeof(struct hlist_head *);
1397 		new_info_hash = fib_info_hash_alloc(bytes);
1398 		new_laddrhash = fib_info_hash_alloc(bytes);
1399 		if (!new_info_hash || !new_laddrhash) {
1400 			fib_info_hash_free(new_info_hash, bytes);
1401 			fib_info_hash_free(new_laddrhash, bytes);
1402 		} else
1403 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1404 
1405 		if (!fib_info_hash_size)
1406 			goto failure;
1407 	}
1408 
1409 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1410 	if (!fi)
1411 		goto failure;
1412 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1413 					      cfg->fc_mx_len, extack);
1414 	if (IS_ERR(fi->fib_metrics)) {
1415 		err = PTR_ERR(fi->fib_metrics);
1416 		kfree(fi);
1417 		return ERR_PTR(err);
1418 	}
1419 
1420 	fib_info_cnt++;
1421 	fi->fib_net = net;
1422 	fi->fib_protocol = cfg->fc_protocol;
1423 	fi->fib_scope = cfg->fc_scope;
1424 	fi->fib_flags = cfg->fc_flags;
1425 	fi->fib_priority = cfg->fc_priority;
1426 	fi->fib_prefsrc = cfg->fc_prefsrc;
1427 	fi->fib_type = cfg->fc_type;
1428 	fi->fib_tb_id = cfg->fc_table;
1429 
1430 	fi->fib_nhs = nhs;
1431 	if (nh) {
1432 		if (!nexthop_get(nh)) {
1433 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1434 			err = -EINVAL;
1435 		} else {
1436 			err = 0;
1437 			fi->nh = nh;
1438 		}
1439 	} else {
1440 		change_nexthops(fi) {
1441 			nexthop_nh->nh_parent = fi;
1442 		} endfor_nexthops(fi)
1443 
1444 		if (cfg->fc_mp)
1445 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1446 					  extack);
1447 		else
1448 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1449 	}
1450 
1451 	if (err != 0)
1452 		goto failure;
1453 
1454 	if (fib_props[cfg->fc_type].error) {
1455 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1456 			NL_SET_ERR_MSG(extack,
1457 				       "Gateway, device and multipath can not be specified for this route type");
1458 			goto err_inval;
1459 		}
1460 		goto link_it;
1461 	} else {
1462 		switch (cfg->fc_type) {
1463 		case RTN_UNICAST:
1464 		case RTN_LOCAL:
1465 		case RTN_BROADCAST:
1466 		case RTN_ANYCAST:
1467 		case RTN_MULTICAST:
1468 			break;
1469 		default:
1470 			NL_SET_ERR_MSG(extack, "Invalid route type");
1471 			goto err_inval;
1472 		}
1473 	}
1474 
1475 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1476 		NL_SET_ERR_MSG(extack, "Invalid scope");
1477 		goto err_inval;
1478 	}
1479 
1480 	if (fi->nh) {
1481 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1482 		if (err)
1483 			goto failure;
1484 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1485 		struct fib_nh *nh = fi->fib_nh;
1486 
1487 		/* Local address is added. */
1488 		if (nhs != 1) {
1489 			NL_SET_ERR_MSG(extack,
1490 				       "Route with host scope can not have multiple nexthops");
1491 			goto err_inval;
1492 		}
1493 		if (nh->fib_nh_gw_family) {
1494 			NL_SET_ERR_MSG(extack,
1495 				       "Route with host scope can not have a gateway");
1496 			goto err_inval;
1497 		}
1498 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1499 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1500 		err = -ENODEV;
1501 		if (!nh->fib_nh_dev)
1502 			goto failure;
1503 	} else {
1504 		int linkdown = 0;
1505 
1506 		change_nexthops(fi) {
1507 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1508 					   cfg->fc_table, cfg->fc_scope,
1509 					   extack);
1510 			if (err != 0)
1511 				goto failure;
1512 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1513 				linkdown++;
1514 		} endfor_nexthops(fi)
1515 		if (linkdown == fi->fib_nhs)
1516 			fi->fib_flags |= RTNH_F_LINKDOWN;
1517 	}
1518 
1519 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1520 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1521 		goto err_inval;
1522 	}
1523 
1524 	if (!fi->nh) {
1525 		change_nexthops(fi) {
1526 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1527 						  fi->fib_scope);
1528 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1529 				fi->fib_nh_is_v6 = true;
1530 		} endfor_nexthops(fi)
1531 
1532 		fib_rebalance(fi);
1533 	}
1534 
1535 link_it:
1536 	ofi = fib_find_info(fi);
1537 	if (ofi) {
1538 		fi->fib_dead = 1;
1539 		free_fib_info(fi);
1540 		ofi->fib_treeref++;
1541 		return ofi;
1542 	}
1543 
1544 	fi->fib_treeref++;
1545 	refcount_set(&fi->fib_clntref, 1);
1546 	spin_lock_bh(&fib_info_lock);
1547 	hlist_add_head(&fi->fib_hash,
1548 		       &fib_info_hash[fib_info_hashfn(fi)]);
1549 	if (fi->fib_prefsrc) {
1550 		struct hlist_head *head;
1551 
1552 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1553 		hlist_add_head(&fi->fib_lhash, head);
1554 	}
1555 	if (fi->nh) {
1556 		list_add(&fi->nh_list, &nh->fi_list);
1557 	} else {
1558 		change_nexthops(fi) {
1559 			struct hlist_head *head;
1560 			unsigned int hash;
1561 
1562 			if (!nexthop_nh->fib_nh_dev)
1563 				continue;
1564 			hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1565 			head = &fib_info_devhash[hash];
1566 			hlist_add_head(&nexthop_nh->nh_hash, head);
1567 		} endfor_nexthops(fi)
1568 	}
1569 	spin_unlock_bh(&fib_info_lock);
1570 	return fi;
1571 
1572 err_inval:
1573 	err = -EINVAL;
1574 
1575 failure:
1576 	if (fi) {
1577 		fi->fib_dead = 1;
1578 		free_fib_info(fi);
1579 	}
1580 
1581 	return ERR_PTR(err);
1582 }
1583 
1584 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1585 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1586 {
1587 	if (nhc->nhc_flags & RTNH_F_DEAD)
1588 		*flags |= RTNH_F_DEAD;
1589 
1590 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1591 		*flags |= RTNH_F_LINKDOWN;
1592 
1593 		rcu_read_lock();
1594 		switch (nhc->nhc_family) {
1595 		case AF_INET:
1596 			if (ip_ignore_linkdown(nhc->nhc_dev))
1597 				*flags |= RTNH_F_DEAD;
1598 			break;
1599 		case AF_INET6:
1600 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1601 				*flags |= RTNH_F_DEAD;
1602 			break;
1603 		}
1604 		rcu_read_unlock();
1605 	}
1606 
1607 	switch (nhc->nhc_gw_family) {
1608 	case AF_INET:
1609 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1610 			goto nla_put_failure;
1611 		break;
1612 	case AF_INET6:
1613 		/* if gateway family does not match nexthop family
1614 		 * gateway is encoded as RTA_VIA
1615 		 */
1616 		if (rt_family != nhc->nhc_gw_family) {
1617 			int alen = sizeof(struct in6_addr);
1618 			struct nlattr *nla;
1619 			struct rtvia *via;
1620 
1621 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1622 			if (!nla)
1623 				goto nla_put_failure;
1624 
1625 			via = nla_data(nla);
1626 			via->rtvia_family = AF_INET6;
1627 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1628 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1629 					    &nhc->nhc_gw.ipv6) < 0) {
1630 			goto nla_put_failure;
1631 		}
1632 		break;
1633 	}
1634 
1635 	*flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1636 	if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1637 		*flags |= RTNH_F_OFFLOAD;
1638 
1639 	if (!skip_oif && nhc->nhc_dev &&
1640 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1641 		goto nla_put_failure;
1642 
1643 	if (nhc->nhc_lwtstate &&
1644 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1645 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1646 		goto nla_put_failure;
1647 
1648 	return 0;
1649 
1650 nla_put_failure:
1651 	return -EMSGSIZE;
1652 }
1653 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1654 
1655 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1656 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1657 		    int nh_weight, u8 rt_family)
1658 {
1659 	const struct net_device *dev = nhc->nhc_dev;
1660 	struct rtnexthop *rtnh;
1661 	unsigned char flags = 0;
1662 
1663 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1664 	if (!rtnh)
1665 		goto nla_put_failure;
1666 
1667 	rtnh->rtnh_hops = nh_weight - 1;
1668 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1669 
1670 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1671 		goto nla_put_failure;
1672 
1673 	rtnh->rtnh_flags = flags;
1674 
1675 	/* length of rtnetlink header + attributes */
1676 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1677 
1678 	return 0;
1679 
1680 nla_put_failure:
1681 	return -EMSGSIZE;
1682 }
1683 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1684 #endif
1685 
1686 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1687 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1688 {
1689 	struct nlattr *mp;
1690 
1691 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1692 	if (!mp)
1693 		goto nla_put_failure;
1694 
1695 	if (unlikely(fi->nh)) {
1696 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1697 			goto nla_put_failure;
1698 		goto mp_end;
1699 	}
1700 
1701 	for_nexthops(fi) {
1702 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1703 				    AF_INET) < 0)
1704 			goto nla_put_failure;
1705 #ifdef CONFIG_IP_ROUTE_CLASSID
1706 		if (nh->nh_tclassid &&
1707 		    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1708 			goto nla_put_failure;
1709 #endif
1710 	} endfor_nexthops(fi);
1711 
1712 mp_end:
1713 	nla_nest_end(skb, mp);
1714 
1715 	return 0;
1716 
1717 nla_put_failure:
1718 	return -EMSGSIZE;
1719 }
1720 #else
1721 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1722 {
1723 	return 0;
1724 }
1725 #endif
1726 
1727 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1728 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1729 		  struct fib_info *fi, unsigned int flags)
1730 {
1731 	unsigned int nhs = fib_info_num_path(fi);
1732 	struct nlmsghdr *nlh;
1733 	struct rtmsg *rtm;
1734 
1735 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1736 	if (!nlh)
1737 		return -EMSGSIZE;
1738 
1739 	rtm = nlmsg_data(nlh);
1740 	rtm->rtm_family = AF_INET;
1741 	rtm->rtm_dst_len = dst_len;
1742 	rtm->rtm_src_len = 0;
1743 	rtm->rtm_tos = tos;
1744 	if (tb_id < 256)
1745 		rtm->rtm_table = tb_id;
1746 	else
1747 		rtm->rtm_table = RT_TABLE_COMPAT;
1748 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1749 		goto nla_put_failure;
1750 	rtm->rtm_type = type;
1751 	rtm->rtm_flags = fi->fib_flags;
1752 	rtm->rtm_scope = fi->fib_scope;
1753 	rtm->rtm_protocol = fi->fib_protocol;
1754 
1755 	if (rtm->rtm_dst_len &&
1756 	    nla_put_in_addr(skb, RTA_DST, dst))
1757 		goto nla_put_failure;
1758 	if (fi->fib_priority &&
1759 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1760 		goto nla_put_failure;
1761 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1762 		goto nla_put_failure;
1763 
1764 	if (fi->fib_prefsrc &&
1765 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1766 		goto nla_put_failure;
1767 
1768 	if (fi->nh) {
1769 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1770 			goto nla_put_failure;
1771 		if (nexthop_is_blackhole(fi->nh))
1772 			rtm->rtm_type = RTN_BLACKHOLE;
1773 	}
1774 
1775 	if (nhs == 1) {
1776 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1777 		unsigned char flags = 0;
1778 
1779 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1780 			goto nla_put_failure;
1781 
1782 		rtm->rtm_flags = flags;
1783 #ifdef CONFIG_IP_ROUTE_CLASSID
1784 		if (nhc->nhc_family == AF_INET) {
1785 			struct fib_nh *nh;
1786 
1787 			nh = container_of(nhc, struct fib_nh, nh_common);
1788 			if (nh->nh_tclassid &&
1789 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1790 				goto nla_put_failure;
1791 		}
1792 #endif
1793 	} else {
1794 		if (fib_add_multipath(skb, fi) < 0)
1795 			goto nla_put_failure;
1796 	}
1797 
1798 	nlmsg_end(skb, nlh);
1799 	return 0;
1800 
1801 nla_put_failure:
1802 	nlmsg_cancel(skb, nlh);
1803 	return -EMSGSIZE;
1804 }
1805 
1806 /*
1807  * Update FIB if:
1808  * - local address disappeared -> we must delete all the entries
1809  *   referring to it.
1810  * - device went down -> we must shutdown all nexthops going via it.
1811  */
1812 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1813 {
1814 	int ret = 0;
1815 	unsigned int hash = fib_laddr_hashfn(local);
1816 	struct hlist_head *head = &fib_info_laddrhash[hash];
1817 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1818 	struct net *net = dev_net(dev);
1819 	struct fib_info *fi;
1820 
1821 	if (!fib_info_laddrhash || local == 0)
1822 		return 0;
1823 
1824 	hlist_for_each_entry(fi, head, fib_lhash) {
1825 		if (!net_eq(fi->fib_net, net) ||
1826 		    fi->fib_tb_id != tb_id)
1827 			continue;
1828 		if (fi->fib_prefsrc == local) {
1829 			fi->fib_flags |= RTNH_F_DEAD;
1830 			ret++;
1831 		}
1832 	}
1833 	return ret;
1834 }
1835 
1836 static int call_fib_nh_notifiers(struct fib_nh *nh,
1837 				 enum fib_event_type event_type)
1838 {
1839 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1840 	struct fib_nh_notifier_info info = {
1841 		.fib_nh = nh,
1842 	};
1843 
1844 	switch (event_type) {
1845 	case FIB_EVENT_NH_ADD:
1846 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1847 			break;
1848 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1849 			break;
1850 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1851 					   &info.info);
1852 	case FIB_EVENT_NH_DEL:
1853 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1854 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1855 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1856 						   event_type, &info.info);
1857 	default:
1858 		break;
1859 	}
1860 
1861 	return NOTIFY_DONE;
1862 }
1863 
1864 /* Update the PMTU of exceptions when:
1865  * - the new MTU of the first hop becomes smaller than the PMTU
1866  * - the old MTU was the same as the PMTU, and it limited discovery of
1867  *   larger MTUs on the path. With that limit raised, we can now
1868  *   discover larger MTUs
1869  * A special case is locked exceptions, for which the PMTU is smaller
1870  * than the minimal accepted PMTU:
1871  * - if the new MTU is greater than the PMTU, don't make any change
1872  * - otherwise, unlock and set PMTU
1873  */
1874 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1875 {
1876 	struct fnhe_hash_bucket *bucket;
1877 	int i;
1878 
1879 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1880 	if (!bucket)
1881 		return;
1882 
1883 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1884 		struct fib_nh_exception *fnhe;
1885 
1886 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1887 		     fnhe;
1888 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1889 			if (fnhe->fnhe_mtu_locked) {
1890 				if (new <= fnhe->fnhe_pmtu) {
1891 					fnhe->fnhe_pmtu = new;
1892 					fnhe->fnhe_mtu_locked = false;
1893 				}
1894 			} else if (new < fnhe->fnhe_pmtu ||
1895 				   orig == fnhe->fnhe_pmtu) {
1896 				fnhe->fnhe_pmtu = new;
1897 			}
1898 		}
1899 	}
1900 }
1901 
1902 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1903 {
1904 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1905 	struct hlist_head *head = &fib_info_devhash[hash];
1906 	struct fib_nh *nh;
1907 
1908 	hlist_for_each_entry(nh, head, nh_hash) {
1909 		if (nh->fib_nh_dev == dev)
1910 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1911 	}
1912 }
1913 
1914 /* Event              force Flags           Description
1915  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1916  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1917  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1918  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1919  *
1920  * only used when fib_nh is built into fib_info
1921  */
1922 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1923 {
1924 	int ret = 0;
1925 	int scope = RT_SCOPE_NOWHERE;
1926 	struct fib_info *prev_fi = NULL;
1927 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1928 	struct hlist_head *head = &fib_info_devhash[hash];
1929 	struct fib_nh *nh;
1930 
1931 	if (force)
1932 		scope = -1;
1933 
1934 	hlist_for_each_entry(nh, head, nh_hash) {
1935 		struct fib_info *fi = nh->nh_parent;
1936 		int dead;
1937 
1938 		BUG_ON(!fi->fib_nhs);
1939 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1940 			continue;
1941 		prev_fi = fi;
1942 		dead = 0;
1943 		change_nexthops(fi) {
1944 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1945 				dead++;
1946 			else if (nexthop_nh->fib_nh_dev == dev &&
1947 				 nexthop_nh->fib_nh_scope != scope) {
1948 				switch (event) {
1949 				case NETDEV_DOWN:
1950 				case NETDEV_UNREGISTER:
1951 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1952 					/* fall through */
1953 				case NETDEV_CHANGE:
1954 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1955 					break;
1956 				}
1957 				call_fib_nh_notifiers(nexthop_nh,
1958 						      FIB_EVENT_NH_DEL);
1959 				dead++;
1960 			}
1961 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1962 			if (event == NETDEV_UNREGISTER &&
1963 			    nexthop_nh->fib_nh_dev == dev) {
1964 				dead = fi->fib_nhs;
1965 				break;
1966 			}
1967 #endif
1968 		} endfor_nexthops(fi)
1969 		if (dead == fi->fib_nhs) {
1970 			switch (event) {
1971 			case NETDEV_DOWN:
1972 			case NETDEV_UNREGISTER:
1973 				fi->fib_flags |= RTNH_F_DEAD;
1974 				/* fall through */
1975 			case NETDEV_CHANGE:
1976 				fi->fib_flags |= RTNH_F_LINKDOWN;
1977 				break;
1978 			}
1979 			ret++;
1980 		}
1981 
1982 		fib_rebalance(fi);
1983 	}
1984 
1985 	return ret;
1986 }
1987 
1988 /* Must be invoked inside of an RCU protected region.  */
1989 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1990 {
1991 	struct fib_info *fi = NULL, *last_resort = NULL;
1992 	struct hlist_head *fa_head = res->fa_head;
1993 	struct fib_table *tb = res->table;
1994 	u8 slen = 32 - res->prefixlen;
1995 	int order = -1, last_idx = -1;
1996 	struct fib_alias *fa, *fa1 = NULL;
1997 	u32 last_prio = res->fi->fib_priority;
1998 	u8 last_tos = 0;
1999 
2000 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2001 		struct fib_info *next_fi = fa->fa_info;
2002 		struct fib_nh *nh;
2003 
2004 		if (fa->fa_slen != slen)
2005 			continue;
2006 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2007 			continue;
2008 		if (fa->tb_id != tb->tb_id)
2009 			continue;
2010 		if (next_fi->fib_priority > last_prio &&
2011 		    fa->fa_tos == last_tos) {
2012 			if (last_tos)
2013 				continue;
2014 			break;
2015 		}
2016 		if (next_fi->fib_flags & RTNH_F_DEAD)
2017 			continue;
2018 		last_tos = fa->fa_tos;
2019 		last_prio = next_fi->fib_priority;
2020 
2021 		if (next_fi->fib_scope != res->scope ||
2022 		    fa->fa_type != RTN_UNICAST)
2023 			continue;
2024 
2025 		nh = fib_info_nh(next_fi, 0);
2026 		if (!nh->fib_nh_gw4 || nh->fib_nh_scope != RT_SCOPE_LINK)
2027 			continue;
2028 
2029 		fib_alias_accessed(fa);
2030 
2031 		if (!fi) {
2032 			if (next_fi != res->fi)
2033 				break;
2034 			fa1 = fa;
2035 		} else if (!fib_detect_death(fi, order, &last_resort,
2036 					     &last_idx, fa1->fa_default)) {
2037 			fib_result_assign(res, fi);
2038 			fa1->fa_default = order;
2039 			goto out;
2040 		}
2041 		fi = next_fi;
2042 		order++;
2043 	}
2044 
2045 	if (order <= 0 || !fi) {
2046 		if (fa1)
2047 			fa1->fa_default = -1;
2048 		goto out;
2049 	}
2050 
2051 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2052 			      fa1->fa_default)) {
2053 		fib_result_assign(res, fi);
2054 		fa1->fa_default = order;
2055 		goto out;
2056 	}
2057 
2058 	if (last_idx >= 0)
2059 		fib_result_assign(res, last_resort);
2060 	fa1->fa_default = last_idx;
2061 out:
2062 	return;
2063 }
2064 
2065 /*
2066  * Dead device goes up. We wake up dead nexthops.
2067  * It takes sense only on multipath routes.
2068  *
2069  * only used when fib_nh is built into fib_info
2070  */
2071 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2072 {
2073 	struct fib_info *prev_fi;
2074 	unsigned int hash;
2075 	struct hlist_head *head;
2076 	struct fib_nh *nh;
2077 	int ret;
2078 
2079 	if (!(dev->flags & IFF_UP))
2080 		return 0;
2081 
2082 	if (nh_flags & RTNH_F_DEAD) {
2083 		unsigned int flags = dev_get_flags(dev);
2084 
2085 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2086 			nh_flags |= RTNH_F_LINKDOWN;
2087 	}
2088 
2089 	prev_fi = NULL;
2090 	hash = fib_devindex_hashfn(dev->ifindex);
2091 	head = &fib_info_devhash[hash];
2092 	ret = 0;
2093 
2094 	hlist_for_each_entry(nh, head, nh_hash) {
2095 		struct fib_info *fi = nh->nh_parent;
2096 		int alive;
2097 
2098 		BUG_ON(!fi->fib_nhs);
2099 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2100 			continue;
2101 
2102 		prev_fi = fi;
2103 		alive = 0;
2104 		change_nexthops(fi) {
2105 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2106 				alive++;
2107 				continue;
2108 			}
2109 			if (!nexthop_nh->fib_nh_dev ||
2110 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2111 				continue;
2112 			if (nexthop_nh->fib_nh_dev != dev ||
2113 			    !__in_dev_get_rtnl(dev))
2114 				continue;
2115 			alive++;
2116 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2117 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2118 		} endfor_nexthops(fi)
2119 
2120 		if (alive > 0) {
2121 			fi->fib_flags &= ~nh_flags;
2122 			ret++;
2123 		}
2124 
2125 		fib_rebalance(fi);
2126 	}
2127 
2128 	return ret;
2129 }
2130 
2131 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2132 static bool fib_good_nh(const struct fib_nh *nh)
2133 {
2134 	int state = NUD_REACHABLE;
2135 
2136 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2137 		struct neighbour *n;
2138 
2139 		rcu_read_lock_bh();
2140 
2141 		if (likely(nh->fib_nh_gw_family == AF_INET))
2142 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2143 						   (__force u32)nh->fib_nh_gw4);
2144 		else if (nh->fib_nh_gw_family == AF_INET6)
2145 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2146 							   &nh->fib_nh_gw6);
2147 		else
2148 			n = NULL;
2149 		if (n)
2150 			state = n->nud_state;
2151 
2152 		rcu_read_unlock_bh();
2153 	}
2154 
2155 	return !!(state & NUD_VALID);
2156 }
2157 
2158 void fib_select_multipath(struct fib_result *res, int hash)
2159 {
2160 	struct fib_info *fi = res->fi;
2161 	struct net *net = fi->fib_net;
2162 	bool first = false;
2163 
2164 	if (unlikely(res->fi->nh)) {
2165 		nexthop_path_fib_result(res, hash);
2166 		return;
2167 	}
2168 
2169 	change_nexthops(fi) {
2170 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2171 			if (!fib_good_nh(nexthop_nh))
2172 				continue;
2173 			if (!first) {
2174 				res->nh_sel = nhsel;
2175 				res->nhc = &nexthop_nh->nh_common;
2176 				first = true;
2177 			}
2178 		}
2179 
2180 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2181 			continue;
2182 
2183 		res->nh_sel = nhsel;
2184 		res->nhc = &nexthop_nh->nh_common;
2185 		return;
2186 	} endfor_nexthops(fi);
2187 }
2188 #endif
2189 
2190 void fib_select_path(struct net *net, struct fib_result *res,
2191 		     struct flowi4 *fl4, const struct sk_buff *skb)
2192 {
2193 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2194 		goto check_saddr;
2195 
2196 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2197 	if (fib_info_num_path(res->fi) > 1) {
2198 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2199 
2200 		fib_select_multipath(res, h);
2201 	}
2202 	else
2203 #endif
2204 	if (!res->prefixlen &&
2205 	    res->table->tb_num_default > 1 &&
2206 	    res->type == RTN_UNICAST)
2207 		fib_select_default(fl4, res);
2208 
2209 check_saddr:
2210 	if (!fl4->saddr)
2211 		fl4->saddr = fib_result_prefsrc(net, res);
2212 }
2213