xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 76ce0265)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: semantics.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/uaccess.h>
13 #include <linux/bitops.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
17 #include <linux/mm.h>
18 #include <linux/string.h>
19 #include <linux/socket.h>
20 #include <linux/sockios.h>
21 #include <linux/errno.h>
22 #include <linux/in.h>
23 #include <linux/inet.h>
24 #include <linux/inetdevice.h>
25 #include <linux/netdevice.h>
26 #include <linux/if_arp.h>
27 #include <linux/proc_fs.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/netlink.h>
32 
33 #include <net/arp.h>
34 #include <net/ip.h>
35 #include <net/protocol.h>
36 #include <net/route.h>
37 #include <net/tcp.h>
38 #include <net/sock.h>
39 #include <net/ip_fib.h>
40 #include <net/ip6_fib.h>
41 #include <net/nexthop.h>
42 #include <net/netlink.h>
43 #include <net/rtnh.h>
44 #include <net/lwtunnel.h>
45 #include <net/fib_notifier.h>
46 #include <net/addrconf.h>
47 
48 #include "fib_lookup.h"
49 
50 static DEFINE_SPINLOCK(fib_info_lock);
51 static struct hlist_head *fib_info_hash;
52 static struct hlist_head *fib_info_laddrhash;
53 static unsigned int fib_info_hash_size;
54 static unsigned int fib_info_cnt;
55 
56 #define DEVINDEX_HASHBITS 8
57 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
58 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
59 
60 /* for_nexthops and change_nexthops only used when nexthop object
61  * is not set in a fib_info. The logic within can reference fib_nh.
62  */
63 #ifdef CONFIG_IP_ROUTE_MULTIPATH
64 
65 #define for_nexthops(fi) {						\
66 	int nhsel; const struct fib_nh *nh;				\
67 	for (nhsel = 0, nh = (fi)->fib_nh;				\
68 	     nhsel < fib_info_num_path((fi));				\
69 	     nh++, nhsel++)
70 
71 #define change_nexthops(fi) {						\
72 	int nhsel; struct fib_nh *nexthop_nh;				\
73 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
74 	     nhsel < fib_info_num_path((fi));				\
75 	     nexthop_nh++, nhsel++)
76 
77 #else /* CONFIG_IP_ROUTE_MULTIPATH */
78 
79 /* Hope, that gcc will optimize it to get rid of dummy loop */
80 
81 #define for_nexthops(fi) {						\
82 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
83 	for (nhsel = 0; nhsel < 1; nhsel++)
84 
85 #define change_nexthops(fi) {						\
86 	int nhsel;							\
87 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
88 	for (nhsel = 0; nhsel < 1; nhsel++)
89 
90 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
91 
92 #define endfor_nexthops(fi) }
93 
94 
95 const struct fib_prop fib_props[RTN_MAX + 1] = {
96 	[RTN_UNSPEC] = {
97 		.error	= 0,
98 		.scope	= RT_SCOPE_NOWHERE,
99 	},
100 	[RTN_UNICAST] = {
101 		.error	= 0,
102 		.scope	= RT_SCOPE_UNIVERSE,
103 	},
104 	[RTN_LOCAL] = {
105 		.error	= 0,
106 		.scope	= RT_SCOPE_HOST,
107 	},
108 	[RTN_BROADCAST] = {
109 		.error	= 0,
110 		.scope	= RT_SCOPE_LINK,
111 	},
112 	[RTN_ANYCAST] = {
113 		.error	= 0,
114 		.scope	= RT_SCOPE_LINK,
115 	},
116 	[RTN_MULTICAST] = {
117 		.error	= 0,
118 		.scope	= RT_SCOPE_UNIVERSE,
119 	},
120 	[RTN_BLACKHOLE] = {
121 		.error	= -EINVAL,
122 		.scope	= RT_SCOPE_UNIVERSE,
123 	},
124 	[RTN_UNREACHABLE] = {
125 		.error	= -EHOSTUNREACH,
126 		.scope	= RT_SCOPE_UNIVERSE,
127 	},
128 	[RTN_PROHIBIT] = {
129 		.error	= -EACCES,
130 		.scope	= RT_SCOPE_UNIVERSE,
131 	},
132 	[RTN_THROW] = {
133 		.error	= -EAGAIN,
134 		.scope	= RT_SCOPE_UNIVERSE,
135 	},
136 	[RTN_NAT] = {
137 		.error	= -EINVAL,
138 		.scope	= RT_SCOPE_NOWHERE,
139 	},
140 	[RTN_XRESOLVE] = {
141 		.error	= -EINVAL,
142 		.scope	= RT_SCOPE_NOWHERE,
143 	},
144 };
145 
146 static void rt_fibinfo_free(struct rtable __rcu **rtp)
147 {
148 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
149 
150 	if (!rt)
151 		return;
152 
153 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
154 	 * because we waited an RCU grace period before calling
155 	 * free_fib_info_rcu()
156 	 */
157 
158 	dst_dev_put(&rt->dst);
159 	dst_release_immediate(&rt->dst);
160 }
161 
162 static void free_nh_exceptions(struct fib_nh_common *nhc)
163 {
164 	struct fnhe_hash_bucket *hash;
165 	int i;
166 
167 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
168 	if (!hash)
169 		return;
170 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
171 		struct fib_nh_exception *fnhe;
172 
173 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
174 		while (fnhe) {
175 			struct fib_nh_exception *next;
176 
177 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
178 
179 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
180 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
181 
182 			kfree(fnhe);
183 
184 			fnhe = next;
185 		}
186 	}
187 	kfree(hash);
188 }
189 
190 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
191 {
192 	int cpu;
193 
194 	if (!rtp)
195 		return;
196 
197 	for_each_possible_cpu(cpu) {
198 		struct rtable *rt;
199 
200 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
201 		if (rt) {
202 			dst_dev_put(&rt->dst);
203 			dst_release_immediate(&rt->dst);
204 		}
205 	}
206 	free_percpu(rtp);
207 }
208 
209 void fib_nh_common_release(struct fib_nh_common *nhc)
210 {
211 	if (nhc->nhc_dev)
212 		dev_put(nhc->nhc_dev);
213 
214 	lwtstate_put(nhc->nhc_lwtstate);
215 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
216 	rt_fibinfo_free(&nhc->nhc_rth_input);
217 	free_nh_exceptions(nhc);
218 }
219 EXPORT_SYMBOL_GPL(fib_nh_common_release);
220 
221 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
222 {
223 #ifdef CONFIG_IP_ROUTE_CLASSID
224 	if (fib_nh->nh_tclassid)
225 		net->ipv4.fib_num_tclassid_users--;
226 #endif
227 	fib_nh_common_release(&fib_nh->nh_common);
228 }
229 
230 /* Release a nexthop info record */
231 static void free_fib_info_rcu(struct rcu_head *head)
232 {
233 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
234 
235 	if (fi->nh) {
236 		nexthop_put(fi->nh);
237 	} else {
238 		change_nexthops(fi) {
239 			fib_nh_release(fi->fib_net, nexthop_nh);
240 		} endfor_nexthops(fi);
241 	}
242 
243 	ip_fib_metrics_put(fi->fib_metrics);
244 
245 	kfree(fi);
246 }
247 
248 void free_fib_info(struct fib_info *fi)
249 {
250 	if (fi->fib_dead == 0) {
251 		pr_warn("Freeing alive fib_info %p\n", fi);
252 		return;
253 	}
254 	fib_info_cnt--;
255 
256 	call_rcu(&fi->rcu, free_fib_info_rcu);
257 }
258 EXPORT_SYMBOL_GPL(free_fib_info);
259 
260 void fib_release_info(struct fib_info *fi)
261 {
262 	spin_lock_bh(&fib_info_lock);
263 	if (fi && --fi->fib_treeref == 0) {
264 		hlist_del(&fi->fib_hash);
265 		if (fi->fib_prefsrc)
266 			hlist_del(&fi->fib_lhash);
267 		if (fi->nh) {
268 			list_del(&fi->nh_list);
269 		} else {
270 			change_nexthops(fi) {
271 				if (!nexthop_nh->fib_nh_dev)
272 					continue;
273 				hlist_del(&nexthop_nh->nh_hash);
274 			} endfor_nexthops(fi)
275 		}
276 		fi->fib_dead = 1;
277 		fib_info_put(fi);
278 	}
279 	spin_unlock_bh(&fib_info_lock);
280 }
281 
282 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
283 {
284 	const struct fib_nh *onh;
285 
286 	if (fi->nh || ofi->nh)
287 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
288 
289 	if (ofi->fib_nhs == 0)
290 		return 0;
291 
292 	for_nexthops(fi) {
293 		onh = fib_info_nh(ofi, nhsel);
294 
295 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
296 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
297 		    nh->fib_nh_scope != onh->fib_nh_scope ||
298 #ifdef CONFIG_IP_ROUTE_MULTIPATH
299 		    nh->fib_nh_weight != onh->fib_nh_weight ||
300 #endif
301 #ifdef CONFIG_IP_ROUTE_CLASSID
302 		    nh->nh_tclassid != onh->nh_tclassid ||
303 #endif
304 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
305 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
306 			return -1;
307 
308 		if (nh->fib_nh_gw_family == AF_INET &&
309 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
310 			return -1;
311 
312 		if (nh->fib_nh_gw_family == AF_INET6 &&
313 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
314 			return -1;
315 	} endfor_nexthops(fi);
316 	return 0;
317 }
318 
319 static inline unsigned int fib_devindex_hashfn(unsigned int val)
320 {
321 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
322 
323 	return (val ^
324 		(val >> DEVINDEX_HASHBITS) ^
325 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
326 }
327 
328 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
329 				      u32 prefsrc, u32 priority)
330 {
331 	unsigned int val = init_val;
332 
333 	val ^= (protocol << 8) | scope;
334 	val ^= prefsrc;
335 	val ^= priority;
336 
337 	return val;
338 }
339 
340 static unsigned int fib_info_hashfn_result(unsigned int val)
341 {
342 	unsigned int mask = (fib_info_hash_size - 1);
343 
344 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
345 }
346 
347 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
348 {
349 	unsigned int val;
350 
351 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
352 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
353 				fi->fib_priority);
354 
355 	if (fi->nh) {
356 		val ^= fib_devindex_hashfn(fi->nh->id);
357 	} else {
358 		for_nexthops(fi) {
359 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
360 		} endfor_nexthops(fi)
361 	}
362 
363 	return fib_info_hashfn_result(val);
364 }
365 
366 /* no metrics, only nexthop id */
367 static struct fib_info *fib_find_info_nh(struct net *net,
368 					 const struct fib_config *cfg)
369 {
370 	struct hlist_head *head;
371 	struct fib_info *fi;
372 	unsigned int hash;
373 
374 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
375 				 cfg->fc_protocol, cfg->fc_scope,
376 				 (__force u32)cfg->fc_prefsrc,
377 				 cfg->fc_priority);
378 	hash = fib_info_hashfn_result(hash);
379 	head = &fib_info_hash[hash];
380 
381 	hlist_for_each_entry(fi, head, fib_hash) {
382 		if (!net_eq(fi->fib_net, net))
383 			continue;
384 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
385 			continue;
386 		if (cfg->fc_protocol == fi->fib_protocol &&
387 		    cfg->fc_scope == fi->fib_scope &&
388 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
389 		    cfg->fc_priority == fi->fib_priority &&
390 		    cfg->fc_type == fi->fib_type &&
391 		    cfg->fc_table == fi->fib_tb_id &&
392 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
393 			return fi;
394 	}
395 
396 	return NULL;
397 }
398 
399 static struct fib_info *fib_find_info(struct fib_info *nfi)
400 {
401 	struct hlist_head *head;
402 	struct fib_info *fi;
403 	unsigned int hash;
404 
405 	hash = fib_info_hashfn(nfi);
406 	head = &fib_info_hash[hash];
407 
408 	hlist_for_each_entry(fi, head, fib_hash) {
409 		if (!net_eq(fi->fib_net, nfi->fib_net))
410 			continue;
411 		if (fi->fib_nhs != nfi->fib_nhs)
412 			continue;
413 		if (nfi->fib_protocol == fi->fib_protocol &&
414 		    nfi->fib_scope == fi->fib_scope &&
415 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
416 		    nfi->fib_priority == fi->fib_priority &&
417 		    nfi->fib_type == fi->fib_type &&
418 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
419 			   sizeof(u32) * RTAX_MAX) == 0 &&
420 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
421 		    nh_comp(fi, nfi) == 0)
422 			return fi;
423 	}
424 
425 	return NULL;
426 }
427 
428 /* Check, that the gateway is already configured.
429  * Used only by redirect accept routine.
430  */
431 int ip_fib_check_default(__be32 gw, struct net_device *dev)
432 {
433 	struct hlist_head *head;
434 	struct fib_nh *nh;
435 	unsigned int hash;
436 
437 	spin_lock(&fib_info_lock);
438 
439 	hash = fib_devindex_hashfn(dev->ifindex);
440 	head = &fib_info_devhash[hash];
441 	hlist_for_each_entry(nh, head, nh_hash) {
442 		if (nh->fib_nh_dev == dev &&
443 		    nh->fib_nh_gw4 == gw &&
444 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
445 			spin_unlock(&fib_info_lock);
446 			return 0;
447 		}
448 	}
449 
450 	spin_unlock(&fib_info_lock);
451 
452 	return -1;
453 }
454 
455 static inline size_t fib_nlmsg_size(struct fib_info *fi)
456 {
457 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
458 			 + nla_total_size(4) /* RTA_TABLE */
459 			 + nla_total_size(4) /* RTA_DST */
460 			 + nla_total_size(4) /* RTA_PRIORITY */
461 			 + nla_total_size(4) /* RTA_PREFSRC */
462 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
463 	unsigned int nhs = fib_info_num_path(fi);
464 
465 	/* space for nested metrics */
466 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
467 
468 	if (fi->nh)
469 		payload += nla_total_size(4); /* RTA_NH_ID */
470 
471 	if (nhs) {
472 		size_t nh_encapsize = 0;
473 		/* Also handles the special case nhs == 1 */
474 
475 		/* each nexthop is packed in an attribute */
476 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
477 		unsigned int i;
478 
479 		/* may contain flow and gateway attribute */
480 		nhsize += 2 * nla_total_size(4);
481 
482 		/* grab encap info */
483 		for (i = 0; i < fib_info_num_path(fi); i++) {
484 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
485 
486 			if (nhc->nhc_lwtstate) {
487 				/* RTA_ENCAP_TYPE */
488 				nh_encapsize += lwtunnel_get_encap_size(
489 						nhc->nhc_lwtstate);
490 				/* RTA_ENCAP */
491 				nh_encapsize +=  nla_total_size(2);
492 			}
493 		}
494 
495 		/* all nexthops are packed in a nested attribute */
496 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
497 
498 	}
499 
500 	return payload;
501 }
502 
503 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
504 	       int dst_len, u32 tb_id, const struct nl_info *info,
505 	       unsigned int nlm_flags)
506 {
507 	struct fib_rt_info fri;
508 	struct sk_buff *skb;
509 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
510 	int err = -ENOBUFS;
511 
512 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
513 	if (!skb)
514 		goto errout;
515 
516 	fri.fi = fa->fa_info;
517 	fri.tb_id = tb_id;
518 	fri.dst = key;
519 	fri.dst_len = dst_len;
520 	fri.tos = fa->fa_tos;
521 	fri.type = fa->fa_type;
522 	fri.offload = fa->offload;
523 	fri.trap = fa->trap;
524 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
525 	if (err < 0) {
526 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
527 		WARN_ON(err == -EMSGSIZE);
528 		kfree_skb(skb);
529 		goto errout;
530 	}
531 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
532 		    info->nlh, GFP_KERNEL);
533 	return;
534 errout:
535 	if (err < 0)
536 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
537 }
538 
539 static int fib_detect_death(struct fib_info *fi, int order,
540 			    struct fib_info **last_resort, int *last_idx,
541 			    int dflt)
542 {
543 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
544 	struct neighbour *n;
545 	int state = NUD_NONE;
546 
547 	if (likely(nhc->nhc_gw_family == AF_INET))
548 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
549 	else if (nhc->nhc_gw_family == AF_INET6)
550 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
551 				 nhc->nhc_dev);
552 	else
553 		n = NULL;
554 
555 	if (n) {
556 		state = n->nud_state;
557 		neigh_release(n);
558 	} else {
559 		return 0;
560 	}
561 	if (state == NUD_REACHABLE)
562 		return 0;
563 	if ((state & NUD_VALID) && order != dflt)
564 		return 0;
565 	if ((state & NUD_VALID) ||
566 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
567 		*last_resort = fi;
568 		*last_idx = order;
569 	}
570 	return 1;
571 }
572 
573 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
574 		       u16 encap_type, 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(encap_type, encap, nhc->nhc_family,
593 					   cfg, &lwtstate, extack);
594 		if (err)
595 			goto lwt_failure;
596 
597 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
598 	}
599 
600 	return 0;
601 
602 lwt_failure:
603 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
604 	nhc->nhc_pcpu_rth_output = NULL;
605 	return err;
606 }
607 EXPORT_SYMBOL_GPL(fib_nh_common_init);
608 
609 int fib_nh_init(struct net *net, struct fib_nh *nh,
610 		struct fib_config *cfg, int nh_weight,
611 		struct netlink_ext_ack *extack)
612 {
613 	int err;
614 
615 	nh->fib_nh_family = AF_INET;
616 
617 	err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
618 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
619 	if (err)
620 		return err;
621 
622 	nh->fib_nh_oif = cfg->fc_oif;
623 	nh->fib_nh_gw_family = cfg->fc_gw_family;
624 	if (cfg->fc_gw_family == AF_INET)
625 		nh->fib_nh_gw4 = cfg->fc_gw4;
626 	else if (cfg->fc_gw_family == AF_INET6)
627 		nh->fib_nh_gw6 = cfg->fc_gw6;
628 
629 	nh->fib_nh_flags = cfg->fc_flags;
630 
631 #ifdef CONFIG_IP_ROUTE_CLASSID
632 	nh->nh_tclassid = cfg->fc_flow;
633 	if (nh->nh_tclassid)
634 		net->ipv4.fib_num_tclassid_users++;
635 #endif
636 #ifdef CONFIG_IP_ROUTE_MULTIPATH
637 	nh->fib_nh_weight = nh_weight;
638 #endif
639 	return 0;
640 }
641 
642 #ifdef CONFIG_IP_ROUTE_MULTIPATH
643 
644 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
645 			      struct netlink_ext_ack *extack)
646 {
647 	int nhs = 0;
648 
649 	while (rtnh_ok(rtnh, remaining)) {
650 		nhs++;
651 		rtnh = rtnh_next(rtnh, &remaining);
652 	}
653 
654 	/* leftover implies invalid nexthop configuration, discard it */
655 	if (remaining > 0) {
656 		NL_SET_ERR_MSG(extack,
657 			       "Invalid nexthop configuration - extra data after nexthops");
658 		nhs = 0;
659 	}
660 
661 	return nhs;
662 }
663 
664 /* only called when fib_nh is integrated into fib_info */
665 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
666 		       int remaining, struct fib_config *cfg,
667 		       struct netlink_ext_ack *extack)
668 {
669 	struct net *net = fi->fib_net;
670 	struct fib_config fib_cfg;
671 	struct fib_nh *nh;
672 	int ret;
673 
674 	change_nexthops(fi) {
675 		int attrlen;
676 
677 		memset(&fib_cfg, 0, sizeof(fib_cfg));
678 
679 		if (!rtnh_ok(rtnh, remaining)) {
680 			NL_SET_ERR_MSG(extack,
681 				       "Invalid nexthop configuration - extra data after nexthop");
682 			return -EINVAL;
683 		}
684 
685 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
686 			NL_SET_ERR_MSG(extack,
687 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
688 			return -EINVAL;
689 		}
690 
691 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
692 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
693 
694 		attrlen = rtnh_attrlen(rtnh);
695 		if (attrlen > 0) {
696 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
697 
698 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
699 			nlav = nla_find(attrs, attrlen, RTA_VIA);
700 			if (nla && nlav) {
701 				NL_SET_ERR_MSG(extack,
702 					       "Nexthop configuration can not contain both GATEWAY and VIA");
703 				return -EINVAL;
704 			}
705 			if (nla) {
706 				fib_cfg.fc_gw4 = nla_get_in_addr(nla);
707 				if (fib_cfg.fc_gw4)
708 					fib_cfg.fc_gw_family = AF_INET;
709 			} else if (nlav) {
710 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
711 				if (ret)
712 					goto errout;
713 			}
714 
715 			nla = nla_find(attrs, attrlen, RTA_FLOW);
716 			if (nla)
717 				fib_cfg.fc_flow = nla_get_u32(nla);
718 
719 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
720 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
721 			if (nla)
722 				fib_cfg.fc_encap_type = nla_get_u16(nla);
723 		}
724 
725 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
726 				  rtnh->rtnh_hops + 1, extack);
727 		if (ret)
728 			goto errout;
729 
730 		rtnh = rtnh_next(rtnh, &remaining);
731 	} endfor_nexthops(fi);
732 
733 	ret = -EINVAL;
734 	nh = fib_info_nh(fi, 0);
735 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
736 		NL_SET_ERR_MSG(extack,
737 			       "Nexthop device index does not match RTA_OIF");
738 		goto errout;
739 	}
740 	if (cfg->fc_gw_family) {
741 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
742 		    (cfg->fc_gw_family == AF_INET &&
743 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
744 		    (cfg->fc_gw_family == AF_INET6 &&
745 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
746 			NL_SET_ERR_MSG(extack,
747 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
748 			goto errout;
749 		}
750 	}
751 #ifdef CONFIG_IP_ROUTE_CLASSID
752 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
753 		NL_SET_ERR_MSG(extack,
754 			       "Nexthop class id does not match RTA_FLOW");
755 		goto errout;
756 	}
757 #endif
758 	ret = 0;
759 errout:
760 	return ret;
761 }
762 
763 /* only called when fib_nh is integrated into fib_info */
764 static void fib_rebalance(struct fib_info *fi)
765 {
766 	int total;
767 	int w;
768 
769 	if (fib_info_num_path(fi) < 2)
770 		return;
771 
772 	total = 0;
773 	for_nexthops(fi) {
774 		if (nh->fib_nh_flags & RTNH_F_DEAD)
775 			continue;
776 
777 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
778 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
779 			continue;
780 
781 		total += nh->fib_nh_weight;
782 	} endfor_nexthops(fi);
783 
784 	w = 0;
785 	change_nexthops(fi) {
786 		int upper_bound;
787 
788 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
789 			upper_bound = -1;
790 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
791 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
792 			upper_bound = -1;
793 		} else {
794 			w += nexthop_nh->fib_nh_weight;
795 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
796 							    total) - 1;
797 		}
798 
799 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
800 	} endfor_nexthops(fi);
801 }
802 #else /* CONFIG_IP_ROUTE_MULTIPATH */
803 
804 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
805 		       int remaining, struct fib_config *cfg,
806 		       struct netlink_ext_ack *extack)
807 {
808 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
809 
810 	return -EINVAL;
811 }
812 
813 #define fib_rebalance(fi) do { } while (0)
814 
815 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
816 
817 static int fib_encap_match(u16 encap_type,
818 			   struct nlattr *encap,
819 			   const struct fib_nh *nh,
820 			   const struct fib_config *cfg,
821 			   struct netlink_ext_ack *extack)
822 {
823 	struct lwtunnel_state *lwtstate;
824 	int ret, result = 0;
825 
826 	if (encap_type == LWTUNNEL_ENCAP_NONE)
827 		return 0;
828 
829 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
830 				   cfg, &lwtstate, extack);
831 	if (!ret) {
832 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
833 		lwtstate_free(lwtstate);
834 	}
835 
836 	return result;
837 }
838 
839 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
840 		 struct netlink_ext_ack *extack)
841 {
842 #ifdef CONFIG_IP_ROUTE_MULTIPATH
843 	struct rtnexthop *rtnh;
844 	int remaining;
845 #endif
846 
847 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
848 		return 1;
849 
850 	if (cfg->fc_nh_id) {
851 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
852 			return 0;
853 		return 1;
854 	}
855 
856 	if (cfg->fc_oif || cfg->fc_gw_family) {
857 		struct fib_nh *nh = fib_info_nh(fi, 0);
858 
859 		if (cfg->fc_encap) {
860 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
861 					    nh, cfg, extack))
862 				return 1;
863 		}
864 #ifdef CONFIG_IP_ROUTE_CLASSID
865 		if (cfg->fc_flow &&
866 		    cfg->fc_flow != nh->nh_tclassid)
867 			return 1;
868 #endif
869 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
870 		    (cfg->fc_gw_family &&
871 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
872 			return 1;
873 
874 		if (cfg->fc_gw_family == AF_INET &&
875 		    cfg->fc_gw4 != nh->fib_nh_gw4)
876 			return 1;
877 
878 		if (cfg->fc_gw_family == AF_INET6 &&
879 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
880 			return 1;
881 
882 		return 0;
883 	}
884 
885 #ifdef CONFIG_IP_ROUTE_MULTIPATH
886 	if (!cfg->fc_mp)
887 		return 0;
888 
889 	rtnh = cfg->fc_mp;
890 	remaining = cfg->fc_mp_len;
891 
892 	for_nexthops(fi) {
893 		int attrlen;
894 
895 		if (!rtnh_ok(rtnh, remaining))
896 			return -EINVAL;
897 
898 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
899 			return 1;
900 
901 		attrlen = rtnh_attrlen(rtnh);
902 		if (attrlen > 0) {
903 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
904 
905 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
906 			nlav = nla_find(attrs, attrlen, RTA_VIA);
907 			if (nla && nlav) {
908 				NL_SET_ERR_MSG(extack,
909 					       "Nexthop configuration can not contain both GATEWAY and VIA");
910 				return -EINVAL;
911 			}
912 
913 			if (nla) {
914 				if (nh->fib_nh_gw_family != AF_INET ||
915 				    nla_get_in_addr(nla) != nh->fib_nh_gw4)
916 					return 1;
917 			} else if (nlav) {
918 				struct fib_config cfg2;
919 				int err;
920 
921 				err = fib_gw_from_via(&cfg2, nlav, extack);
922 				if (err)
923 					return err;
924 
925 				switch (nh->fib_nh_gw_family) {
926 				case AF_INET:
927 					if (cfg2.fc_gw_family != AF_INET ||
928 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
929 						return 1;
930 					break;
931 				case AF_INET6:
932 					if (cfg2.fc_gw_family != AF_INET6 ||
933 					    ipv6_addr_cmp(&cfg2.fc_gw6,
934 							  &nh->fib_nh_gw6))
935 						return 1;
936 					break;
937 				}
938 			}
939 
940 #ifdef CONFIG_IP_ROUTE_CLASSID
941 			nla = nla_find(attrs, attrlen, RTA_FLOW);
942 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
943 				return 1;
944 #endif
945 		}
946 
947 		rtnh = rtnh_next(rtnh, &remaining);
948 	} endfor_nexthops(fi);
949 #endif
950 	return 0;
951 }
952 
953 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
954 {
955 	struct nlattr *nla;
956 	int remaining;
957 
958 	if (!cfg->fc_mx)
959 		return true;
960 
961 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
962 		int type = nla_type(nla);
963 		u32 fi_val, val;
964 
965 		if (!type)
966 			continue;
967 		if (type > RTAX_MAX)
968 			return false;
969 
970 		if (type == RTAX_CC_ALGO) {
971 			char tmp[TCP_CA_NAME_MAX];
972 			bool ecn_ca = false;
973 
974 			nla_strlcpy(tmp, nla, sizeof(tmp));
975 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
976 		} else {
977 			if (nla_len(nla) != sizeof(u32))
978 				return false;
979 			val = nla_get_u32(nla);
980 		}
981 
982 		fi_val = fi->fib_metrics->metrics[type - 1];
983 		if (type == RTAX_FEATURES)
984 			fi_val &= ~DST_FEATURE_ECN_CA;
985 
986 		if (fi_val != val)
987 			return false;
988 	}
989 
990 	return true;
991 }
992 
993 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
994 			      u32 table, struct netlink_ext_ack *extack)
995 {
996 	struct fib6_config cfg = {
997 		.fc_table = table,
998 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
999 		.fc_ifindex = nh->fib_nh_oif,
1000 		.fc_gateway = nh->fib_nh_gw6,
1001 	};
1002 	struct fib6_nh fib6_nh = {};
1003 	int err;
1004 
1005 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1006 	if (!err) {
1007 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1008 		dev_hold(nh->fib_nh_dev);
1009 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1010 		nh->fib_nh_scope = RT_SCOPE_LINK;
1011 
1012 		ipv6_stub->fib6_nh_release(&fib6_nh);
1013 	}
1014 
1015 	return err;
1016 }
1017 
1018 /*
1019  * Picture
1020  * -------
1021  *
1022  * Semantics of nexthop is very messy by historical reasons.
1023  * We have to take into account, that:
1024  * a) gateway can be actually local interface address,
1025  *    so that gatewayed route is direct.
1026  * b) gateway must be on-link address, possibly
1027  *    described not by an ifaddr, but also by a direct route.
1028  * c) If both gateway and interface are specified, they should not
1029  *    contradict.
1030  * d) If we use tunnel routes, gateway could be not on-link.
1031  *
1032  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1033  * results in pretty ugly and hairy code with obscure logic.
1034  *
1035  * I chose to generalized it instead, so that the size
1036  * of code does not increase practically, but it becomes
1037  * much more general.
1038  * Every prefix is assigned a "scope" value: "host" is local address,
1039  * "link" is direct route,
1040  * [ ... "site" ... "interior" ... ]
1041  * and "universe" is true gateway route with global meaning.
1042  *
1043  * Every prefix refers to a set of "nexthop"s (gw, oif),
1044  * where gw must have narrower scope. This recursion stops
1045  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1046  * which means that gw is forced to be on link.
1047  *
1048  * Code is still hairy, but now it is apparently logically
1049  * consistent and very flexible. F.e. as by-product it allows
1050  * to co-exists in peace independent exterior and interior
1051  * routing processes.
1052  *
1053  * Normally it looks as following.
1054  *
1055  * {universe prefix}  -> (gw, oif) [scope link]
1056  *		  |
1057  *		  |-> {link prefix} -> (gw, oif) [scope local]
1058  *					|
1059  *					|-> {local prefix} (terminal node)
1060  */
1061 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1062 			      u8 scope, struct netlink_ext_ack *extack)
1063 {
1064 	struct net_device *dev;
1065 	struct fib_result res;
1066 	int err = 0;
1067 
1068 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1069 		unsigned int addr_type;
1070 
1071 		if (scope >= RT_SCOPE_LINK) {
1072 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1073 			return -EINVAL;
1074 		}
1075 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1076 		if (!dev) {
1077 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1078 			return -ENODEV;
1079 		}
1080 		if (!(dev->flags & IFF_UP)) {
1081 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1082 			return -ENETDOWN;
1083 		}
1084 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1085 		if (addr_type != RTN_UNICAST) {
1086 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1087 			return -EINVAL;
1088 		}
1089 		if (!netif_carrier_ok(dev))
1090 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1091 		nh->fib_nh_dev = dev;
1092 		dev_hold(dev);
1093 		nh->fib_nh_scope = RT_SCOPE_LINK;
1094 		return 0;
1095 	}
1096 	rcu_read_lock();
1097 	{
1098 		struct fib_table *tbl = NULL;
1099 		struct flowi4 fl4 = {
1100 			.daddr = nh->fib_nh_gw4,
1101 			.flowi4_scope = scope + 1,
1102 			.flowi4_oif = nh->fib_nh_oif,
1103 			.flowi4_iif = LOOPBACK_IFINDEX,
1104 		};
1105 
1106 		/* It is not necessary, but requires a bit of thinking */
1107 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1108 			fl4.flowi4_scope = RT_SCOPE_LINK;
1109 
1110 		if (table)
1111 			tbl = fib_get_table(net, table);
1112 
1113 		if (tbl)
1114 			err = fib_table_lookup(tbl, &fl4, &res,
1115 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1116 					       FIB_LOOKUP_NOREF);
1117 
1118 		/* on error or if no table given do full lookup. This
1119 		 * is needed for example when nexthops are in the local
1120 		 * table rather than the given table
1121 		 */
1122 		if (!tbl || err) {
1123 			err = fib_lookup(net, &fl4, &res,
1124 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1125 		}
1126 
1127 		if (err) {
1128 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1129 			goto out;
1130 		}
1131 	}
1132 
1133 	err = -EINVAL;
1134 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1135 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1136 		goto out;
1137 	}
1138 	nh->fib_nh_scope = res.scope;
1139 	nh->fib_nh_oif = FIB_RES_OIF(res);
1140 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1141 	if (!dev) {
1142 		NL_SET_ERR_MSG(extack,
1143 			       "No egress device for nexthop gateway");
1144 		goto out;
1145 	}
1146 	dev_hold(dev);
1147 	if (!netif_carrier_ok(dev))
1148 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1149 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1150 out:
1151 	rcu_read_unlock();
1152 	return err;
1153 }
1154 
1155 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1156 			      struct netlink_ext_ack *extack)
1157 {
1158 	struct in_device *in_dev;
1159 	int err;
1160 
1161 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1162 		NL_SET_ERR_MSG(extack,
1163 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1164 		return -EINVAL;
1165 	}
1166 
1167 	rcu_read_lock();
1168 
1169 	err = -ENODEV;
1170 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1171 	if (!in_dev)
1172 		goto out;
1173 	err = -ENETDOWN;
1174 	if (!(in_dev->dev->flags & IFF_UP)) {
1175 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1176 		goto out;
1177 	}
1178 
1179 	nh->fib_nh_dev = in_dev->dev;
1180 	dev_hold(nh->fib_nh_dev);
1181 	nh->fib_nh_scope = RT_SCOPE_HOST;
1182 	if (!netif_carrier_ok(nh->fib_nh_dev))
1183 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1184 	err = 0;
1185 out:
1186 	rcu_read_unlock();
1187 	return err;
1188 }
1189 
1190 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1191 		 struct netlink_ext_ack *extack)
1192 {
1193 	int err;
1194 
1195 	if (nh->fib_nh_gw_family == AF_INET)
1196 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1197 	else if (nh->fib_nh_gw_family == AF_INET6)
1198 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1199 	else
1200 		err = fib_check_nh_nongw(net, nh, extack);
1201 
1202 	return err;
1203 }
1204 
1205 static inline unsigned int fib_laddr_hashfn(__be32 val)
1206 {
1207 	unsigned int mask = (fib_info_hash_size - 1);
1208 
1209 	return ((__force u32)val ^
1210 		((__force u32)val >> 7) ^
1211 		((__force u32)val >> 14)) & mask;
1212 }
1213 
1214 static struct hlist_head *fib_info_hash_alloc(int bytes)
1215 {
1216 	if (bytes <= PAGE_SIZE)
1217 		return kzalloc(bytes, GFP_KERNEL);
1218 	else
1219 		return (struct hlist_head *)
1220 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
1221 					 get_order(bytes));
1222 }
1223 
1224 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1225 {
1226 	if (!hash)
1227 		return;
1228 
1229 	if (bytes <= PAGE_SIZE)
1230 		kfree(hash);
1231 	else
1232 		free_pages((unsigned long) hash, get_order(bytes));
1233 }
1234 
1235 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1236 			       struct hlist_head *new_laddrhash,
1237 			       unsigned int new_size)
1238 {
1239 	struct hlist_head *old_info_hash, *old_laddrhash;
1240 	unsigned int old_size = fib_info_hash_size;
1241 	unsigned int i, bytes;
1242 
1243 	spin_lock_bh(&fib_info_lock);
1244 	old_info_hash = fib_info_hash;
1245 	old_laddrhash = fib_info_laddrhash;
1246 	fib_info_hash_size = new_size;
1247 
1248 	for (i = 0; i < old_size; i++) {
1249 		struct hlist_head *head = &fib_info_hash[i];
1250 		struct hlist_node *n;
1251 		struct fib_info *fi;
1252 
1253 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1254 			struct hlist_head *dest;
1255 			unsigned int new_hash;
1256 
1257 			new_hash = fib_info_hashfn(fi);
1258 			dest = &new_info_hash[new_hash];
1259 			hlist_add_head(&fi->fib_hash, dest);
1260 		}
1261 	}
1262 	fib_info_hash = new_info_hash;
1263 
1264 	for (i = 0; i < old_size; i++) {
1265 		struct hlist_head *lhead = &fib_info_laddrhash[i];
1266 		struct hlist_node *n;
1267 		struct fib_info *fi;
1268 
1269 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1270 			struct hlist_head *ldest;
1271 			unsigned int new_hash;
1272 
1273 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1274 			ldest = &new_laddrhash[new_hash];
1275 			hlist_add_head(&fi->fib_lhash, ldest);
1276 		}
1277 	}
1278 	fib_info_laddrhash = new_laddrhash;
1279 
1280 	spin_unlock_bh(&fib_info_lock);
1281 
1282 	bytes = old_size * sizeof(struct hlist_head *);
1283 	fib_info_hash_free(old_info_hash, bytes);
1284 	fib_info_hash_free(old_laddrhash, bytes);
1285 }
1286 
1287 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1288 				 unsigned char scope)
1289 {
1290 	struct fib_nh *nh;
1291 
1292 	if (nhc->nhc_family != AF_INET)
1293 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1294 
1295 	nh = container_of(nhc, struct fib_nh, nh_common);
1296 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1297 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1298 
1299 	return nh->nh_saddr;
1300 }
1301 
1302 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1303 {
1304 	struct fib_nh_common *nhc = res->nhc;
1305 
1306 	if (res->fi->fib_prefsrc)
1307 		return res->fi->fib_prefsrc;
1308 
1309 	if (nhc->nhc_family == AF_INET) {
1310 		struct fib_nh *nh;
1311 
1312 		nh = container_of(nhc, struct fib_nh, nh_common);
1313 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1314 			return nh->nh_saddr;
1315 	}
1316 
1317 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1318 }
1319 
1320 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1321 {
1322 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1323 	    fib_prefsrc != cfg->fc_dst) {
1324 		u32 tb_id = cfg->fc_table;
1325 		int rc;
1326 
1327 		if (tb_id == RT_TABLE_MAIN)
1328 			tb_id = RT_TABLE_LOCAL;
1329 
1330 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1331 					  fib_prefsrc, tb_id);
1332 
1333 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1334 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1335 						  fib_prefsrc, RT_TABLE_LOCAL);
1336 		}
1337 
1338 		if (rc != RTN_LOCAL)
1339 			return false;
1340 	}
1341 	return true;
1342 }
1343 
1344 struct fib_info *fib_create_info(struct fib_config *cfg,
1345 				 struct netlink_ext_ack *extack)
1346 {
1347 	int err;
1348 	struct fib_info *fi = NULL;
1349 	struct nexthop *nh = NULL;
1350 	struct fib_info *ofi;
1351 	int nhs = 1;
1352 	struct net *net = cfg->fc_nlinfo.nl_net;
1353 
1354 	if (cfg->fc_type > RTN_MAX)
1355 		goto err_inval;
1356 
1357 	/* Fast check to catch the most weird cases */
1358 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1359 		NL_SET_ERR_MSG(extack, "Invalid scope");
1360 		goto err_inval;
1361 	}
1362 
1363 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1364 		NL_SET_ERR_MSG(extack,
1365 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1366 		goto err_inval;
1367 	}
1368 
1369 	if (cfg->fc_nh_id) {
1370 		if (!cfg->fc_mx) {
1371 			fi = fib_find_info_nh(net, cfg);
1372 			if (fi) {
1373 				fi->fib_treeref++;
1374 				return fi;
1375 			}
1376 		}
1377 
1378 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1379 		if (!nh) {
1380 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1381 			goto err_inval;
1382 		}
1383 		nhs = 0;
1384 	}
1385 
1386 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1387 	if (cfg->fc_mp) {
1388 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1389 		if (nhs == 0)
1390 			goto err_inval;
1391 	}
1392 #endif
1393 
1394 	err = -ENOBUFS;
1395 	if (fib_info_cnt >= fib_info_hash_size) {
1396 		unsigned int new_size = fib_info_hash_size << 1;
1397 		struct hlist_head *new_info_hash;
1398 		struct hlist_head *new_laddrhash;
1399 		unsigned int bytes;
1400 
1401 		if (!new_size)
1402 			new_size = 16;
1403 		bytes = new_size * sizeof(struct hlist_head *);
1404 		new_info_hash = fib_info_hash_alloc(bytes);
1405 		new_laddrhash = fib_info_hash_alloc(bytes);
1406 		if (!new_info_hash || !new_laddrhash) {
1407 			fib_info_hash_free(new_info_hash, bytes);
1408 			fib_info_hash_free(new_laddrhash, bytes);
1409 		} else
1410 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1411 
1412 		if (!fib_info_hash_size)
1413 			goto failure;
1414 	}
1415 
1416 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1417 	if (!fi)
1418 		goto failure;
1419 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1420 					      cfg->fc_mx_len, extack);
1421 	if (IS_ERR(fi->fib_metrics)) {
1422 		err = PTR_ERR(fi->fib_metrics);
1423 		kfree(fi);
1424 		return ERR_PTR(err);
1425 	}
1426 
1427 	fib_info_cnt++;
1428 	fi->fib_net = net;
1429 	fi->fib_protocol = cfg->fc_protocol;
1430 	fi->fib_scope = cfg->fc_scope;
1431 	fi->fib_flags = cfg->fc_flags;
1432 	fi->fib_priority = cfg->fc_priority;
1433 	fi->fib_prefsrc = cfg->fc_prefsrc;
1434 	fi->fib_type = cfg->fc_type;
1435 	fi->fib_tb_id = cfg->fc_table;
1436 
1437 	fi->fib_nhs = nhs;
1438 	if (nh) {
1439 		if (!nexthop_get(nh)) {
1440 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1441 			err = -EINVAL;
1442 		} else {
1443 			err = 0;
1444 			fi->nh = nh;
1445 		}
1446 	} else {
1447 		change_nexthops(fi) {
1448 			nexthop_nh->nh_parent = fi;
1449 		} endfor_nexthops(fi)
1450 
1451 		if (cfg->fc_mp)
1452 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1453 					  extack);
1454 		else
1455 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1456 	}
1457 
1458 	if (err != 0)
1459 		goto failure;
1460 
1461 	if (fib_props[cfg->fc_type].error) {
1462 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1463 			NL_SET_ERR_MSG(extack,
1464 				       "Gateway, device and multipath can not be specified for this route type");
1465 			goto err_inval;
1466 		}
1467 		goto link_it;
1468 	} else {
1469 		switch (cfg->fc_type) {
1470 		case RTN_UNICAST:
1471 		case RTN_LOCAL:
1472 		case RTN_BROADCAST:
1473 		case RTN_ANYCAST:
1474 		case RTN_MULTICAST:
1475 			break;
1476 		default:
1477 			NL_SET_ERR_MSG(extack, "Invalid route type");
1478 			goto err_inval;
1479 		}
1480 	}
1481 
1482 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1483 		NL_SET_ERR_MSG(extack, "Invalid scope");
1484 		goto err_inval;
1485 	}
1486 
1487 	if (fi->nh) {
1488 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1489 		if (err)
1490 			goto failure;
1491 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1492 		struct fib_nh *nh = fi->fib_nh;
1493 
1494 		/* Local address is added. */
1495 		if (nhs != 1) {
1496 			NL_SET_ERR_MSG(extack,
1497 				       "Route with host scope can not have multiple nexthops");
1498 			goto err_inval;
1499 		}
1500 		if (nh->fib_nh_gw_family) {
1501 			NL_SET_ERR_MSG(extack,
1502 				       "Route with host scope can not have a gateway");
1503 			goto err_inval;
1504 		}
1505 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1506 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1507 		err = -ENODEV;
1508 		if (!nh->fib_nh_dev)
1509 			goto failure;
1510 	} else {
1511 		int linkdown = 0;
1512 
1513 		change_nexthops(fi) {
1514 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1515 					   cfg->fc_table, cfg->fc_scope,
1516 					   extack);
1517 			if (err != 0)
1518 				goto failure;
1519 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1520 				linkdown++;
1521 		} endfor_nexthops(fi)
1522 		if (linkdown == fi->fib_nhs)
1523 			fi->fib_flags |= RTNH_F_LINKDOWN;
1524 	}
1525 
1526 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1527 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1528 		goto err_inval;
1529 	}
1530 
1531 	if (!fi->nh) {
1532 		change_nexthops(fi) {
1533 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1534 						  fi->fib_scope);
1535 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1536 				fi->fib_nh_is_v6 = true;
1537 		} endfor_nexthops(fi)
1538 
1539 		fib_rebalance(fi);
1540 	}
1541 
1542 link_it:
1543 	ofi = fib_find_info(fi);
1544 	if (ofi) {
1545 		fi->fib_dead = 1;
1546 		free_fib_info(fi);
1547 		ofi->fib_treeref++;
1548 		return ofi;
1549 	}
1550 
1551 	fi->fib_treeref++;
1552 	refcount_set(&fi->fib_clntref, 1);
1553 	spin_lock_bh(&fib_info_lock);
1554 	hlist_add_head(&fi->fib_hash,
1555 		       &fib_info_hash[fib_info_hashfn(fi)]);
1556 	if (fi->fib_prefsrc) {
1557 		struct hlist_head *head;
1558 
1559 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1560 		hlist_add_head(&fi->fib_lhash, head);
1561 	}
1562 	if (fi->nh) {
1563 		list_add(&fi->nh_list, &nh->fi_list);
1564 	} else {
1565 		change_nexthops(fi) {
1566 			struct hlist_head *head;
1567 			unsigned int hash;
1568 
1569 			if (!nexthop_nh->fib_nh_dev)
1570 				continue;
1571 			hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1572 			head = &fib_info_devhash[hash];
1573 			hlist_add_head(&nexthop_nh->nh_hash, head);
1574 		} endfor_nexthops(fi)
1575 	}
1576 	spin_unlock_bh(&fib_info_lock);
1577 	return fi;
1578 
1579 err_inval:
1580 	err = -EINVAL;
1581 
1582 failure:
1583 	if (fi) {
1584 		fi->fib_dead = 1;
1585 		free_fib_info(fi);
1586 	}
1587 
1588 	return ERR_PTR(err);
1589 }
1590 
1591 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1592 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1593 {
1594 	if (nhc->nhc_flags & RTNH_F_DEAD)
1595 		*flags |= RTNH_F_DEAD;
1596 
1597 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1598 		*flags |= RTNH_F_LINKDOWN;
1599 
1600 		rcu_read_lock();
1601 		switch (nhc->nhc_family) {
1602 		case AF_INET:
1603 			if (ip_ignore_linkdown(nhc->nhc_dev))
1604 				*flags |= RTNH_F_DEAD;
1605 			break;
1606 		case AF_INET6:
1607 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1608 				*flags |= RTNH_F_DEAD;
1609 			break;
1610 		}
1611 		rcu_read_unlock();
1612 	}
1613 
1614 	switch (nhc->nhc_gw_family) {
1615 	case AF_INET:
1616 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1617 			goto nla_put_failure;
1618 		break;
1619 	case AF_INET6:
1620 		/* if gateway family does not match nexthop family
1621 		 * gateway is encoded as RTA_VIA
1622 		 */
1623 		if (rt_family != nhc->nhc_gw_family) {
1624 			int alen = sizeof(struct in6_addr);
1625 			struct nlattr *nla;
1626 			struct rtvia *via;
1627 
1628 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1629 			if (!nla)
1630 				goto nla_put_failure;
1631 
1632 			via = nla_data(nla);
1633 			via->rtvia_family = AF_INET6;
1634 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1635 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1636 					    &nhc->nhc_gw.ipv6) < 0) {
1637 			goto nla_put_failure;
1638 		}
1639 		break;
1640 	}
1641 
1642 	*flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1643 	if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1644 		*flags |= RTNH_F_OFFLOAD;
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)
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 	/* length of rtnetlink header + attributes */
1683 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1684 
1685 	return 0;
1686 
1687 nla_put_failure:
1688 	return -EMSGSIZE;
1689 }
1690 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1691 #endif
1692 
1693 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1694 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1695 {
1696 	struct nlattr *mp;
1697 
1698 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1699 	if (!mp)
1700 		goto nla_put_failure;
1701 
1702 	if (unlikely(fi->nh)) {
1703 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1704 			goto nla_put_failure;
1705 		goto mp_end;
1706 	}
1707 
1708 	for_nexthops(fi) {
1709 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1710 				    AF_INET) < 0)
1711 			goto nla_put_failure;
1712 #ifdef CONFIG_IP_ROUTE_CLASSID
1713 		if (nh->nh_tclassid &&
1714 		    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1715 			goto nla_put_failure;
1716 #endif
1717 	} endfor_nexthops(fi);
1718 
1719 mp_end:
1720 	nla_nest_end(skb, mp);
1721 
1722 	return 0;
1723 
1724 nla_put_failure:
1725 	return -EMSGSIZE;
1726 }
1727 #else
1728 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1729 {
1730 	return 0;
1731 }
1732 #endif
1733 
1734 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1735 		  struct fib_rt_info *fri, unsigned int flags)
1736 {
1737 	unsigned int nhs = fib_info_num_path(fri->fi);
1738 	struct fib_info *fi = fri->fi;
1739 	u32 tb_id = fri->tb_id;
1740 	struct nlmsghdr *nlh;
1741 	struct rtmsg *rtm;
1742 
1743 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1744 	if (!nlh)
1745 		return -EMSGSIZE;
1746 
1747 	rtm = nlmsg_data(nlh);
1748 	rtm->rtm_family = AF_INET;
1749 	rtm->rtm_dst_len = fri->dst_len;
1750 	rtm->rtm_src_len = 0;
1751 	rtm->rtm_tos = fri->tos;
1752 	if (tb_id < 256)
1753 		rtm->rtm_table = tb_id;
1754 	else
1755 		rtm->rtm_table = RT_TABLE_COMPAT;
1756 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1757 		goto nla_put_failure;
1758 	rtm->rtm_type = fri->type;
1759 	rtm->rtm_flags = fi->fib_flags;
1760 	rtm->rtm_scope = fi->fib_scope;
1761 	rtm->rtm_protocol = fi->fib_protocol;
1762 
1763 	if (rtm->rtm_dst_len &&
1764 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1765 		goto nla_put_failure;
1766 	if (fi->fib_priority &&
1767 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1768 		goto nla_put_failure;
1769 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1770 		goto nla_put_failure;
1771 
1772 	if (fi->fib_prefsrc &&
1773 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1774 		goto nla_put_failure;
1775 
1776 	if (fi->nh) {
1777 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1778 			goto nla_put_failure;
1779 		if (nexthop_is_blackhole(fi->nh))
1780 			rtm->rtm_type = RTN_BLACKHOLE;
1781 	}
1782 
1783 	if (nhs == 1) {
1784 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1785 		unsigned char flags = 0;
1786 
1787 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1788 			goto nla_put_failure;
1789 
1790 		rtm->rtm_flags = flags;
1791 #ifdef CONFIG_IP_ROUTE_CLASSID
1792 		if (nhc->nhc_family == AF_INET) {
1793 			struct fib_nh *nh;
1794 
1795 			nh = container_of(nhc, struct fib_nh, nh_common);
1796 			if (nh->nh_tclassid &&
1797 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1798 				goto nla_put_failure;
1799 		}
1800 #endif
1801 	} else {
1802 		if (fib_add_multipath(skb, fi) < 0)
1803 			goto nla_put_failure;
1804 	}
1805 
1806 	if (fri->offload)
1807 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1808 	if (fri->trap)
1809 		rtm->rtm_flags |= RTM_F_TRAP;
1810 
1811 	nlmsg_end(skb, nlh);
1812 	return 0;
1813 
1814 nla_put_failure:
1815 	nlmsg_cancel(skb, nlh);
1816 	return -EMSGSIZE;
1817 }
1818 
1819 /*
1820  * Update FIB if:
1821  * - local address disappeared -> we must delete all the entries
1822  *   referring to it.
1823  * - device went down -> we must shutdown all nexthops going via it.
1824  */
1825 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1826 {
1827 	int ret = 0;
1828 	unsigned int hash = fib_laddr_hashfn(local);
1829 	struct hlist_head *head = &fib_info_laddrhash[hash];
1830 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1831 	struct net *net = dev_net(dev);
1832 	struct fib_info *fi;
1833 
1834 	if (!fib_info_laddrhash || local == 0)
1835 		return 0;
1836 
1837 	hlist_for_each_entry(fi, head, fib_lhash) {
1838 		if (!net_eq(fi->fib_net, net) ||
1839 		    fi->fib_tb_id != tb_id)
1840 			continue;
1841 		if (fi->fib_prefsrc == local) {
1842 			fi->fib_flags |= RTNH_F_DEAD;
1843 			ret++;
1844 		}
1845 	}
1846 	return ret;
1847 }
1848 
1849 static int call_fib_nh_notifiers(struct fib_nh *nh,
1850 				 enum fib_event_type event_type)
1851 {
1852 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1853 	struct fib_nh_notifier_info info = {
1854 		.fib_nh = nh,
1855 	};
1856 
1857 	switch (event_type) {
1858 	case FIB_EVENT_NH_ADD:
1859 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1860 			break;
1861 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1862 			break;
1863 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1864 					   &info.info);
1865 	case FIB_EVENT_NH_DEL:
1866 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1867 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1868 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1869 						   event_type, &info.info);
1870 	default:
1871 		break;
1872 	}
1873 
1874 	return NOTIFY_DONE;
1875 }
1876 
1877 /* Update the PMTU of exceptions when:
1878  * - the new MTU of the first hop becomes smaller than the PMTU
1879  * - the old MTU was the same as the PMTU, and it limited discovery of
1880  *   larger MTUs on the path. With that limit raised, we can now
1881  *   discover larger MTUs
1882  * A special case is locked exceptions, for which the PMTU is smaller
1883  * than the minimal accepted PMTU:
1884  * - if the new MTU is greater than the PMTU, don't make any change
1885  * - otherwise, unlock and set PMTU
1886  */
1887 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1888 {
1889 	struct fnhe_hash_bucket *bucket;
1890 	int i;
1891 
1892 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1893 	if (!bucket)
1894 		return;
1895 
1896 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1897 		struct fib_nh_exception *fnhe;
1898 
1899 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1900 		     fnhe;
1901 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1902 			if (fnhe->fnhe_mtu_locked) {
1903 				if (new <= fnhe->fnhe_pmtu) {
1904 					fnhe->fnhe_pmtu = new;
1905 					fnhe->fnhe_mtu_locked = false;
1906 				}
1907 			} else if (new < fnhe->fnhe_pmtu ||
1908 				   orig == fnhe->fnhe_pmtu) {
1909 				fnhe->fnhe_pmtu = new;
1910 			}
1911 		}
1912 	}
1913 }
1914 
1915 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1916 {
1917 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1918 	struct hlist_head *head = &fib_info_devhash[hash];
1919 	struct fib_nh *nh;
1920 
1921 	hlist_for_each_entry(nh, head, nh_hash) {
1922 		if (nh->fib_nh_dev == dev)
1923 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1924 	}
1925 }
1926 
1927 /* Event              force Flags           Description
1928  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1929  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1930  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1931  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1932  *
1933  * only used when fib_nh is built into fib_info
1934  */
1935 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1936 {
1937 	int ret = 0;
1938 	int scope = RT_SCOPE_NOWHERE;
1939 	struct fib_info *prev_fi = NULL;
1940 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1941 	struct hlist_head *head = &fib_info_devhash[hash];
1942 	struct fib_nh *nh;
1943 
1944 	if (force)
1945 		scope = -1;
1946 
1947 	hlist_for_each_entry(nh, head, nh_hash) {
1948 		struct fib_info *fi = nh->nh_parent;
1949 		int dead;
1950 
1951 		BUG_ON(!fi->fib_nhs);
1952 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1953 			continue;
1954 		prev_fi = fi;
1955 		dead = 0;
1956 		change_nexthops(fi) {
1957 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1958 				dead++;
1959 			else if (nexthop_nh->fib_nh_dev == dev &&
1960 				 nexthop_nh->fib_nh_scope != scope) {
1961 				switch (event) {
1962 				case NETDEV_DOWN:
1963 				case NETDEV_UNREGISTER:
1964 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1965 					/* fall through */
1966 				case NETDEV_CHANGE:
1967 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1968 					break;
1969 				}
1970 				call_fib_nh_notifiers(nexthop_nh,
1971 						      FIB_EVENT_NH_DEL);
1972 				dead++;
1973 			}
1974 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1975 			if (event == NETDEV_UNREGISTER &&
1976 			    nexthop_nh->fib_nh_dev == dev) {
1977 				dead = fi->fib_nhs;
1978 				break;
1979 			}
1980 #endif
1981 		} endfor_nexthops(fi)
1982 		if (dead == fi->fib_nhs) {
1983 			switch (event) {
1984 			case NETDEV_DOWN:
1985 			case NETDEV_UNREGISTER:
1986 				fi->fib_flags |= RTNH_F_DEAD;
1987 				/* fall through */
1988 			case NETDEV_CHANGE:
1989 				fi->fib_flags |= RTNH_F_LINKDOWN;
1990 				break;
1991 			}
1992 			ret++;
1993 		}
1994 
1995 		fib_rebalance(fi);
1996 	}
1997 
1998 	return ret;
1999 }
2000 
2001 /* Must be invoked inside of an RCU protected region.  */
2002 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2003 {
2004 	struct fib_info *fi = NULL, *last_resort = NULL;
2005 	struct hlist_head *fa_head = res->fa_head;
2006 	struct fib_table *tb = res->table;
2007 	u8 slen = 32 - res->prefixlen;
2008 	int order = -1, last_idx = -1;
2009 	struct fib_alias *fa, *fa1 = NULL;
2010 	u32 last_prio = res->fi->fib_priority;
2011 	u8 last_tos = 0;
2012 
2013 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2014 		struct fib_info *next_fi = fa->fa_info;
2015 		struct fib_nh *nh;
2016 
2017 		if (fa->fa_slen != slen)
2018 			continue;
2019 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2020 			continue;
2021 		if (fa->tb_id != tb->tb_id)
2022 			continue;
2023 		if (next_fi->fib_priority > last_prio &&
2024 		    fa->fa_tos == last_tos) {
2025 			if (last_tos)
2026 				continue;
2027 			break;
2028 		}
2029 		if (next_fi->fib_flags & RTNH_F_DEAD)
2030 			continue;
2031 		last_tos = fa->fa_tos;
2032 		last_prio = next_fi->fib_priority;
2033 
2034 		if (next_fi->fib_scope != res->scope ||
2035 		    fa->fa_type != RTN_UNICAST)
2036 			continue;
2037 
2038 		nh = fib_info_nh(next_fi, 0);
2039 		if (!nh->fib_nh_gw4 || nh->fib_nh_scope != RT_SCOPE_LINK)
2040 			continue;
2041 
2042 		fib_alias_accessed(fa);
2043 
2044 		if (!fi) {
2045 			if (next_fi != res->fi)
2046 				break;
2047 			fa1 = fa;
2048 		} else if (!fib_detect_death(fi, order, &last_resort,
2049 					     &last_idx, fa1->fa_default)) {
2050 			fib_result_assign(res, fi);
2051 			fa1->fa_default = order;
2052 			goto out;
2053 		}
2054 		fi = next_fi;
2055 		order++;
2056 	}
2057 
2058 	if (order <= 0 || !fi) {
2059 		if (fa1)
2060 			fa1->fa_default = -1;
2061 		goto out;
2062 	}
2063 
2064 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2065 			      fa1->fa_default)) {
2066 		fib_result_assign(res, fi);
2067 		fa1->fa_default = order;
2068 		goto out;
2069 	}
2070 
2071 	if (last_idx >= 0)
2072 		fib_result_assign(res, last_resort);
2073 	fa1->fa_default = last_idx;
2074 out:
2075 	return;
2076 }
2077 
2078 /*
2079  * Dead device goes up. We wake up dead nexthops.
2080  * It takes sense only on multipath routes.
2081  *
2082  * only used when fib_nh is built into fib_info
2083  */
2084 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2085 {
2086 	struct fib_info *prev_fi;
2087 	unsigned int hash;
2088 	struct hlist_head *head;
2089 	struct fib_nh *nh;
2090 	int ret;
2091 
2092 	if (!(dev->flags & IFF_UP))
2093 		return 0;
2094 
2095 	if (nh_flags & RTNH_F_DEAD) {
2096 		unsigned int flags = dev_get_flags(dev);
2097 
2098 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2099 			nh_flags |= RTNH_F_LINKDOWN;
2100 	}
2101 
2102 	prev_fi = NULL;
2103 	hash = fib_devindex_hashfn(dev->ifindex);
2104 	head = &fib_info_devhash[hash];
2105 	ret = 0;
2106 
2107 	hlist_for_each_entry(nh, head, nh_hash) {
2108 		struct fib_info *fi = nh->nh_parent;
2109 		int alive;
2110 
2111 		BUG_ON(!fi->fib_nhs);
2112 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2113 			continue;
2114 
2115 		prev_fi = fi;
2116 		alive = 0;
2117 		change_nexthops(fi) {
2118 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2119 				alive++;
2120 				continue;
2121 			}
2122 			if (!nexthop_nh->fib_nh_dev ||
2123 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2124 				continue;
2125 			if (nexthop_nh->fib_nh_dev != dev ||
2126 			    !__in_dev_get_rtnl(dev))
2127 				continue;
2128 			alive++;
2129 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2130 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2131 		} endfor_nexthops(fi)
2132 
2133 		if (alive > 0) {
2134 			fi->fib_flags &= ~nh_flags;
2135 			ret++;
2136 		}
2137 
2138 		fib_rebalance(fi);
2139 	}
2140 
2141 	return ret;
2142 }
2143 
2144 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2145 static bool fib_good_nh(const struct fib_nh *nh)
2146 {
2147 	int state = NUD_REACHABLE;
2148 
2149 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2150 		struct neighbour *n;
2151 
2152 		rcu_read_lock_bh();
2153 
2154 		if (likely(nh->fib_nh_gw_family == AF_INET))
2155 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2156 						   (__force u32)nh->fib_nh_gw4);
2157 		else if (nh->fib_nh_gw_family == AF_INET6)
2158 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2159 							   &nh->fib_nh_gw6);
2160 		else
2161 			n = NULL;
2162 		if (n)
2163 			state = n->nud_state;
2164 
2165 		rcu_read_unlock_bh();
2166 	}
2167 
2168 	return !!(state & NUD_VALID);
2169 }
2170 
2171 void fib_select_multipath(struct fib_result *res, int hash)
2172 {
2173 	struct fib_info *fi = res->fi;
2174 	struct net *net = fi->fib_net;
2175 	bool first = false;
2176 
2177 	if (unlikely(res->fi->nh)) {
2178 		nexthop_path_fib_result(res, hash);
2179 		return;
2180 	}
2181 
2182 	change_nexthops(fi) {
2183 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2184 			if (!fib_good_nh(nexthop_nh))
2185 				continue;
2186 			if (!first) {
2187 				res->nh_sel = nhsel;
2188 				res->nhc = &nexthop_nh->nh_common;
2189 				first = true;
2190 			}
2191 		}
2192 
2193 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2194 			continue;
2195 
2196 		res->nh_sel = nhsel;
2197 		res->nhc = &nexthop_nh->nh_common;
2198 		return;
2199 	} endfor_nexthops(fi);
2200 }
2201 #endif
2202 
2203 void fib_select_path(struct net *net, struct fib_result *res,
2204 		     struct flowi4 *fl4, const struct sk_buff *skb)
2205 {
2206 	if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2207 		goto check_saddr;
2208 
2209 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2210 	if (fib_info_num_path(res->fi) > 1) {
2211 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2212 
2213 		fib_select_multipath(res, h);
2214 	}
2215 	else
2216 #endif
2217 	if (!res->prefixlen &&
2218 	    res->table->tb_num_default > 1 &&
2219 	    res->type == RTN_UNICAST)
2220 		fib_select_default(fl4, res);
2221 
2222 check_saddr:
2223 	if (!fl4->saddr)
2224 		fl4->saddr = fib_result_prefsrc(net, res);
2225 }
2226