xref: /linux/net/ipv4/fib_semantics.c (revision 55d0969c451159cff86949b38c39171cab962069)
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 #include <linux/hash.h>
33 #include <linux/nospec.h>
34 
35 #include <net/arp.h>
36 #include <net/inet_dscp.h>
37 #include <net/ip.h>
38 #include <net/protocol.h>
39 #include <net/route.h>
40 #include <net/tcp.h>
41 #include <net/sock.h>
42 #include <net/ip_fib.h>
43 #include <net/ip6_fib.h>
44 #include <net/nexthop.h>
45 #include <net/netlink.h>
46 #include <net/rtnh.h>
47 #include <net/lwtunnel.h>
48 #include <net/fib_notifier.h>
49 #include <net/addrconf.h>
50 
51 #include "fib_lookup.h"
52 
53 static DEFINE_SPINLOCK(fib_info_lock);
54 static struct hlist_head *fib_info_hash;
55 static struct hlist_head *fib_info_laddrhash;
56 static unsigned int fib_info_hash_size;
57 static unsigned int fib_info_hash_bits;
58 static unsigned int fib_info_cnt;
59 
60 #define DEVINDEX_HASHBITS 8
61 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
62 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
63 
64 /* for_nexthops and change_nexthops only used when nexthop object
65  * is not set in a fib_info. The logic within can reference fib_nh.
66  */
67 #ifdef CONFIG_IP_ROUTE_MULTIPATH
68 
69 #define for_nexthops(fi) {						\
70 	int nhsel; const struct fib_nh *nh;				\
71 	for (nhsel = 0, nh = (fi)->fib_nh;				\
72 	     nhsel < fib_info_num_path((fi));				\
73 	     nh++, nhsel++)
74 
75 #define change_nexthops(fi) {						\
76 	int nhsel; struct fib_nh *nexthop_nh;				\
77 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
78 	     nhsel < fib_info_num_path((fi));				\
79 	     nexthop_nh++, nhsel++)
80 
81 #else /* CONFIG_IP_ROUTE_MULTIPATH */
82 
83 /* Hope, that gcc will optimize it to get rid of dummy loop */
84 
85 #define for_nexthops(fi) {						\
86 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
87 	for (nhsel = 0; nhsel < 1; nhsel++)
88 
89 #define change_nexthops(fi) {						\
90 	int nhsel;							\
91 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
92 	for (nhsel = 0; nhsel < 1; nhsel++)
93 
94 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
95 
96 #define endfor_nexthops(fi) }
97 
98 
99 const struct fib_prop fib_props[RTN_MAX + 1] = {
100 	[RTN_UNSPEC] = {
101 		.error	= 0,
102 		.scope	= RT_SCOPE_NOWHERE,
103 	},
104 	[RTN_UNICAST] = {
105 		.error	= 0,
106 		.scope	= RT_SCOPE_UNIVERSE,
107 	},
108 	[RTN_LOCAL] = {
109 		.error	= 0,
110 		.scope	= RT_SCOPE_HOST,
111 	},
112 	[RTN_BROADCAST] = {
113 		.error	= 0,
114 		.scope	= RT_SCOPE_LINK,
115 	},
116 	[RTN_ANYCAST] = {
117 		.error	= 0,
118 		.scope	= RT_SCOPE_LINK,
119 	},
120 	[RTN_MULTICAST] = {
121 		.error	= 0,
122 		.scope	= RT_SCOPE_UNIVERSE,
123 	},
124 	[RTN_BLACKHOLE] = {
125 		.error	= -EINVAL,
126 		.scope	= RT_SCOPE_UNIVERSE,
127 	},
128 	[RTN_UNREACHABLE] = {
129 		.error	= -EHOSTUNREACH,
130 		.scope	= RT_SCOPE_UNIVERSE,
131 	},
132 	[RTN_PROHIBIT] = {
133 		.error	= -EACCES,
134 		.scope	= RT_SCOPE_UNIVERSE,
135 	},
136 	[RTN_THROW] = {
137 		.error	= -EAGAIN,
138 		.scope	= RT_SCOPE_UNIVERSE,
139 	},
140 	[RTN_NAT] = {
141 		.error	= -EINVAL,
142 		.scope	= RT_SCOPE_NOWHERE,
143 	},
144 	[RTN_XRESOLVE] = {
145 		.error	= -EINVAL,
146 		.scope	= RT_SCOPE_NOWHERE,
147 	},
148 };
149 
150 static void rt_fibinfo_free(struct rtable __rcu **rtp)
151 {
152 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
153 
154 	if (!rt)
155 		return;
156 
157 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
158 	 * because we waited an RCU grace period before calling
159 	 * free_fib_info_rcu()
160 	 */
161 
162 	dst_dev_put(&rt->dst);
163 	dst_release_immediate(&rt->dst);
164 }
165 
166 static void free_nh_exceptions(struct fib_nh_common *nhc)
167 {
168 	struct fnhe_hash_bucket *hash;
169 	int i;
170 
171 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
172 	if (!hash)
173 		return;
174 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
175 		struct fib_nh_exception *fnhe;
176 
177 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
178 		while (fnhe) {
179 			struct fib_nh_exception *next;
180 
181 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
182 
183 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
184 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
185 
186 			kfree(fnhe);
187 
188 			fnhe = next;
189 		}
190 	}
191 	kfree(hash);
192 }
193 
194 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
195 {
196 	int cpu;
197 
198 	if (!rtp)
199 		return;
200 
201 	for_each_possible_cpu(cpu) {
202 		struct rtable *rt;
203 
204 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
205 		if (rt) {
206 			dst_dev_put(&rt->dst);
207 			dst_release_immediate(&rt->dst);
208 		}
209 	}
210 	free_percpu(rtp);
211 }
212 
213 void fib_nh_common_release(struct fib_nh_common *nhc)
214 {
215 	netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
216 	lwtstate_put(nhc->nhc_lwtstate);
217 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
218 	rt_fibinfo_free(&nhc->nhc_rth_input);
219 	free_nh_exceptions(nhc);
220 }
221 EXPORT_SYMBOL_GPL(fib_nh_common_release);
222 
223 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
224 {
225 #ifdef CONFIG_IP_ROUTE_CLASSID
226 	if (fib_nh->nh_tclassid)
227 		atomic_dec(&net->ipv4.fib_num_tclassid_users);
228 #endif
229 	fib_nh_common_release(&fib_nh->nh_common);
230 }
231 
232 /* Release a nexthop info record */
233 static void free_fib_info_rcu(struct rcu_head *head)
234 {
235 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
236 
237 	if (fi->nh) {
238 		nexthop_put(fi->nh);
239 	} else {
240 		change_nexthops(fi) {
241 			fib_nh_release(fi->fib_net, nexthop_nh);
242 		} endfor_nexthops(fi);
243 	}
244 
245 	ip_fib_metrics_put(fi->fib_metrics);
246 
247 	kfree(fi);
248 }
249 
250 void free_fib_info(struct fib_info *fi)
251 {
252 	if (fi->fib_dead == 0) {
253 		pr_warn("Freeing alive fib_info %p\n", fi);
254 		return;
255 	}
256 
257 	call_rcu_hurry(&fi->rcu, free_fib_info_rcu);
258 }
259 EXPORT_SYMBOL_GPL(free_fib_info);
260 
261 void fib_release_info(struct fib_info *fi)
262 {
263 	spin_lock_bh(&fib_info_lock);
264 	if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
265 		hlist_del(&fi->fib_hash);
266 
267 		/* Paired with READ_ONCE() in fib_create_info(). */
268 		WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
269 
270 		if (fi->fib_prefsrc)
271 			hlist_del(&fi->fib_lhash);
272 		if (fi->nh) {
273 			list_del(&fi->nh_list);
274 		} else {
275 			change_nexthops(fi) {
276 				if (!nexthop_nh->fib_nh_dev)
277 					continue;
278 				hlist_del(&nexthop_nh->nh_hash);
279 			} endfor_nexthops(fi)
280 		}
281 		/* Paired with READ_ONCE() from fib_table_lookup() */
282 		WRITE_ONCE(fi->fib_dead, 1);
283 		fib_info_put(fi);
284 	}
285 	spin_unlock_bh(&fib_info_lock);
286 }
287 
288 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
289 {
290 	const struct fib_nh *onh;
291 
292 	if (fi->nh || ofi->nh)
293 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
294 
295 	if (ofi->fib_nhs == 0)
296 		return 0;
297 
298 	for_nexthops(fi) {
299 		onh = fib_info_nh(ofi, nhsel);
300 
301 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
302 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
303 		    nh->fib_nh_scope != onh->fib_nh_scope ||
304 #ifdef CONFIG_IP_ROUTE_MULTIPATH
305 		    nh->fib_nh_weight != onh->fib_nh_weight ||
306 #endif
307 #ifdef CONFIG_IP_ROUTE_CLASSID
308 		    nh->nh_tclassid != onh->nh_tclassid ||
309 #endif
310 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
311 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
312 			return -1;
313 
314 		if (nh->fib_nh_gw_family == AF_INET &&
315 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
316 			return -1;
317 
318 		if (nh->fib_nh_gw_family == AF_INET6 &&
319 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
320 			return -1;
321 	} endfor_nexthops(fi);
322 	return 0;
323 }
324 
325 static inline unsigned int fib_devindex_hashfn(unsigned int val)
326 {
327 	return hash_32(val, DEVINDEX_HASHBITS);
328 }
329 
330 static struct hlist_head *
331 fib_info_devhash_bucket(const struct net_device *dev)
332 {
333 	u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
334 
335 	return &fib_info_devhash[fib_devindex_hashfn(val)];
336 }
337 
338 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
339 				      u32 prefsrc, u32 priority)
340 {
341 	unsigned int val = init_val;
342 
343 	val ^= (protocol << 8) | scope;
344 	val ^= prefsrc;
345 	val ^= priority;
346 
347 	return val;
348 }
349 
350 static unsigned int fib_info_hashfn_result(unsigned int val)
351 {
352 	unsigned int mask = (fib_info_hash_size - 1);
353 
354 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
355 }
356 
357 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
358 {
359 	unsigned int val;
360 
361 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
362 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
363 				fi->fib_priority);
364 
365 	if (fi->nh) {
366 		val ^= fib_devindex_hashfn(fi->nh->id);
367 	} else {
368 		for_nexthops(fi) {
369 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
370 		} endfor_nexthops(fi)
371 	}
372 
373 	return fib_info_hashfn_result(val);
374 }
375 
376 /* no metrics, only nexthop id */
377 static struct fib_info *fib_find_info_nh(struct net *net,
378 					 const struct fib_config *cfg)
379 {
380 	struct hlist_head *head;
381 	struct fib_info *fi;
382 	unsigned int hash;
383 
384 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
385 				 cfg->fc_protocol, cfg->fc_scope,
386 				 (__force u32)cfg->fc_prefsrc,
387 				 cfg->fc_priority);
388 	hash = fib_info_hashfn_result(hash);
389 	head = &fib_info_hash[hash];
390 
391 	hlist_for_each_entry(fi, head, fib_hash) {
392 		if (!net_eq(fi->fib_net, net))
393 			continue;
394 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
395 			continue;
396 		if (cfg->fc_protocol == fi->fib_protocol &&
397 		    cfg->fc_scope == fi->fib_scope &&
398 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
399 		    cfg->fc_priority == fi->fib_priority &&
400 		    cfg->fc_type == fi->fib_type &&
401 		    cfg->fc_table == fi->fib_tb_id &&
402 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
403 			return fi;
404 	}
405 
406 	return NULL;
407 }
408 
409 static struct fib_info *fib_find_info(struct fib_info *nfi)
410 {
411 	struct hlist_head *head;
412 	struct fib_info *fi;
413 	unsigned int hash;
414 
415 	hash = fib_info_hashfn(nfi);
416 	head = &fib_info_hash[hash];
417 
418 	hlist_for_each_entry(fi, head, fib_hash) {
419 		if (!net_eq(fi->fib_net, nfi->fib_net))
420 			continue;
421 		if (fi->fib_nhs != nfi->fib_nhs)
422 			continue;
423 		if (nfi->fib_protocol == fi->fib_protocol &&
424 		    nfi->fib_scope == fi->fib_scope &&
425 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
426 		    nfi->fib_priority == fi->fib_priority &&
427 		    nfi->fib_type == fi->fib_type &&
428 		    nfi->fib_tb_id == fi->fib_tb_id &&
429 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
430 			   sizeof(u32) * RTAX_MAX) == 0 &&
431 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
432 		    nh_comp(fi, nfi) == 0)
433 			return fi;
434 	}
435 
436 	return NULL;
437 }
438 
439 /* Check, that the gateway is already configured.
440  * Used only by redirect accept routine.
441  */
442 int ip_fib_check_default(__be32 gw, struct net_device *dev)
443 {
444 	struct hlist_head *head;
445 	struct fib_nh *nh;
446 
447 	spin_lock(&fib_info_lock);
448 
449 	head = fib_info_devhash_bucket(dev);
450 
451 	hlist_for_each_entry(nh, head, nh_hash) {
452 		if (nh->fib_nh_dev == dev &&
453 		    nh->fib_nh_gw4 == gw &&
454 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
455 			spin_unlock(&fib_info_lock);
456 			return 0;
457 		}
458 	}
459 
460 	spin_unlock(&fib_info_lock);
461 
462 	return -1;
463 }
464 
465 size_t fib_nlmsg_size(struct fib_info *fi)
466 {
467 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
468 			 + nla_total_size(4) /* RTA_TABLE */
469 			 + nla_total_size(4) /* RTA_DST */
470 			 + nla_total_size(4) /* RTA_PRIORITY */
471 			 + nla_total_size(4) /* RTA_PREFSRC */
472 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
473 	unsigned int nhs = fib_info_num_path(fi);
474 
475 	/* space for nested metrics */
476 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
477 
478 	if (fi->nh)
479 		payload += nla_total_size(4); /* RTA_NH_ID */
480 
481 	if (nhs) {
482 		size_t nh_encapsize = 0;
483 		/* Also handles the special case nhs == 1 */
484 
485 		/* each nexthop is packed in an attribute */
486 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
487 		unsigned int i;
488 
489 		/* may contain flow and gateway attribute */
490 		nhsize += 2 * nla_total_size(4);
491 
492 		/* grab encap info */
493 		for (i = 0; i < fib_info_num_path(fi); i++) {
494 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
495 
496 			if (nhc->nhc_lwtstate) {
497 				/* RTA_ENCAP_TYPE */
498 				nh_encapsize += lwtunnel_get_encap_size(
499 						nhc->nhc_lwtstate);
500 				/* RTA_ENCAP */
501 				nh_encapsize +=  nla_total_size(2);
502 			}
503 		}
504 
505 		/* all nexthops are packed in a nested attribute */
506 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
507 
508 	}
509 
510 	return payload;
511 }
512 
513 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
514 	       int dst_len, u32 tb_id, const struct nl_info *info,
515 	       unsigned int nlm_flags)
516 {
517 	struct fib_rt_info fri;
518 	struct sk_buff *skb;
519 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
520 	int err = -ENOBUFS;
521 
522 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
523 	if (!skb)
524 		goto errout;
525 
526 	fri.fi = fa->fa_info;
527 	fri.tb_id = tb_id;
528 	fri.dst = key;
529 	fri.dst_len = dst_len;
530 	fri.dscp = fa->fa_dscp;
531 	fri.type = fa->fa_type;
532 	fri.offload = READ_ONCE(fa->offload);
533 	fri.trap = READ_ONCE(fa->trap);
534 	fri.offload_failed = READ_ONCE(fa->offload_failed);
535 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
536 	if (err < 0) {
537 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
538 		WARN_ON(err == -EMSGSIZE);
539 		kfree_skb(skb);
540 		goto errout;
541 	}
542 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
543 		    info->nlh, GFP_KERNEL);
544 	return;
545 errout:
546 	rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
547 }
548 
549 static int fib_detect_death(struct fib_info *fi, int order,
550 			    struct fib_info **last_resort, int *last_idx,
551 			    int dflt)
552 {
553 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
554 	struct neighbour *n;
555 	int state = NUD_NONE;
556 
557 	if (likely(nhc->nhc_gw_family == AF_INET))
558 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
559 	else if (nhc->nhc_gw_family == AF_INET6)
560 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
561 				 nhc->nhc_dev);
562 	else
563 		n = NULL;
564 
565 	if (n) {
566 		state = READ_ONCE(n->nud_state);
567 		neigh_release(n);
568 	} else {
569 		return 0;
570 	}
571 	if (state == NUD_REACHABLE)
572 		return 0;
573 	if ((state & NUD_VALID) && order != dflt)
574 		return 0;
575 	if ((state & NUD_VALID) ||
576 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
577 		*last_resort = fi;
578 		*last_idx = order;
579 	}
580 	return 1;
581 }
582 
583 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
584 		       struct nlattr *encap, u16 encap_type,
585 		       void *cfg, gfp_t gfp_flags,
586 		       struct netlink_ext_ack *extack)
587 {
588 	int err;
589 
590 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
591 						    gfp_flags);
592 	if (!nhc->nhc_pcpu_rth_output)
593 		return -ENOMEM;
594 
595 	if (encap) {
596 		struct lwtunnel_state *lwtstate;
597 
598 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
599 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
600 			err = -EINVAL;
601 			goto lwt_failure;
602 		}
603 		err = lwtunnel_build_state(net, encap_type, encap,
604 					   nhc->nhc_family, cfg, &lwtstate,
605 					   extack);
606 		if (err)
607 			goto lwt_failure;
608 
609 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
610 	}
611 
612 	return 0;
613 
614 lwt_failure:
615 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
616 	nhc->nhc_pcpu_rth_output = NULL;
617 	return err;
618 }
619 EXPORT_SYMBOL_GPL(fib_nh_common_init);
620 
621 int fib_nh_init(struct net *net, struct fib_nh *nh,
622 		struct fib_config *cfg, int nh_weight,
623 		struct netlink_ext_ack *extack)
624 {
625 	int err;
626 
627 	nh->fib_nh_family = AF_INET;
628 
629 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
630 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
631 	if (err)
632 		return err;
633 
634 	nh->fib_nh_oif = cfg->fc_oif;
635 	nh->fib_nh_gw_family = cfg->fc_gw_family;
636 	if (cfg->fc_gw_family == AF_INET)
637 		nh->fib_nh_gw4 = cfg->fc_gw4;
638 	else if (cfg->fc_gw_family == AF_INET6)
639 		nh->fib_nh_gw6 = cfg->fc_gw6;
640 
641 	nh->fib_nh_flags = cfg->fc_flags;
642 
643 #ifdef CONFIG_IP_ROUTE_CLASSID
644 	nh->nh_tclassid = cfg->fc_flow;
645 	if (nh->nh_tclassid)
646 		atomic_inc(&net->ipv4.fib_num_tclassid_users);
647 #endif
648 #ifdef CONFIG_IP_ROUTE_MULTIPATH
649 	nh->fib_nh_weight = nh_weight;
650 #endif
651 	return 0;
652 }
653 
654 #ifdef CONFIG_IP_ROUTE_MULTIPATH
655 
656 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
657 			      struct netlink_ext_ack *extack)
658 {
659 	int nhs = 0;
660 
661 	while (rtnh_ok(rtnh, remaining)) {
662 		nhs++;
663 		rtnh = rtnh_next(rtnh, &remaining);
664 	}
665 
666 	/* leftover implies invalid nexthop configuration, discard it */
667 	if (remaining > 0) {
668 		NL_SET_ERR_MSG(extack,
669 			       "Invalid nexthop configuration - extra data after nexthops");
670 		nhs = 0;
671 	}
672 
673 	return nhs;
674 }
675 
676 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
677 			    struct netlink_ext_ack *extack)
678 {
679 	if (nla_len(nla) < sizeof(*gw)) {
680 		NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
681 		return -EINVAL;
682 	}
683 
684 	*gw = nla_get_in_addr(nla);
685 
686 	return 0;
687 }
688 
689 /* only called when fib_nh is integrated into fib_info */
690 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
691 		       int remaining, struct fib_config *cfg,
692 		       struct netlink_ext_ack *extack)
693 {
694 	struct net *net = fi->fib_net;
695 	struct fib_config fib_cfg;
696 	struct fib_nh *nh;
697 	int ret;
698 
699 	change_nexthops(fi) {
700 		int attrlen;
701 
702 		memset(&fib_cfg, 0, sizeof(fib_cfg));
703 
704 		if (!rtnh_ok(rtnh, remaining)) {
705 			NL_SET_ERR_MSG(extack,
706 				       "Invalid nexthop configuration - extra data after nexthop");
707 			return -EINVAL;
708 		}
709 
710 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
711 			NL_SET_ERR_MSG(extack,
712 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
713 			return -EINVAL;
714 		}
715 
716 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
717 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
718 
719 		attrlen = rtnh_attrlen(rtnh);
720 		if (attrlen > 0) {
721 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
722 
723 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
724 			nlav = nla_find(attrs, attrlen, RTA_VIA);
725 			if (nla && nlav) {
726 				NL_SET_ERR_MSG(extack,
727 					       "Nexthop configuration can not contain both GATEWAY and VIA");
728 				return -EINVAL;
729 			}
730 			if (nla) {
731 				ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
732 						       extack);
733 				if (ret)
734 					goto errout;
735 
736 				if (fib_cfg.fc_gw4)
737 					fib_cfg.fc_gw_family = AF_INET;
738 			} else if (nlav) {
739 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
740 				if (ret)
741 					goto errout;
742 			}
743 
744 			nla = nla_find(attrs, attrlen, RTA_FLOW);
745 			if (nla) {
746 				if (nla_len(nla) < sizeof(u32)) {
747 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
748 					return -EINVAL;
749 				}
750 				fib_cfg.fc_flow = nla_get_u32(nla);
751 			}
752 
753 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
754 			/* RTA_ENCAP_TYPE length checked in
755 			 * lwtunnel_valid_encap_type_attr
756 			 */
757 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
758 			if (nla)
759 				fib_cfg.fc_encap_type = nla_get_u16(nla);
760 		}
761 
762 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
763 				  rtnh->rtnh_hops + 1, extack);
764 		if (ret)
765 			goto errout;
766 
767 		rtnh = rtnh_next(rtnh, &remaining);
768 	} endfor_nexthops(fi);
769 
770 	ret = -EINVAL;
771 	nh = fib_info_nh(fi, 0);
772 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
773 		NL_SET_ERR_MSG(extack,
774 			       "Nexthop device index does not match RTA_OIF");
775 		goto errout;
776 	}
777 	if (cfg->fc_gw_family) {
778 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
779 		    (cfg->fc_gw_family == AF_INET &&
780 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
781 		    (cfg->fc_gw_family == AF_INET6 &&
782 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
783 			NL_SET_ERR_MSG(extack,
784 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
785 			goto errout;
786 		}
787 	}
788 #ifdef CONFIG_IP_ROUTE_CLASSID
789 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
790 		NL_SET_ERR_MSG(extack,
791 			       "Nexthop class id does not match RTA_FLOW");
792 		goto errout;
793 	}
794 #endif
795 	ret = 0;
796 errout:
797 	return ret;
798 }
799 
800 /* only called when fib_nh is integrated into fib_info */
801 static void fib_rebalance(struct fib_info *fi)
802 {
803 	int total;
804 	int w;
805 
806 	if (fib_info_num_path(fi) < 2)
807 		return;
808 
809 	total = 0;
810 	for_nexthops(fi) {
811 		if (nh->fib_nh_flags & RTNH_F_DEAD)
812 			continue;
813 
814 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
815 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
816 			continue;
817 
818 		total += nh->fib_nh_weight;
819 	} endfor_nexthops(fi);
820 
821 	w = 0;
822 	change_nexthops(fi) {
823 		int upper_bound;
824 
825 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
826 			upper_bound = -1;
827 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
828 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
829 			upper_bound = -1;
830 		} else {
831 			w += nexthop_nh->fib_nh_weight;
832 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
833 							    total) - 1;
834 		}
835 
836 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
837 	} endfor_nexthops(fi);
838 }
839 #else /* CONFIG_IP_ROUTE_MULTIPATH */
840 
841 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
842 		       int remaining, struct fib_config *cfg,
843 		       struct netlink_ext_ack *extack)
844 {
845 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
846 
847 	return -EINVAL;
848 }
849 
850 #define fib_rebalance(fi) do { } while (0)
851 
852 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
853 
854 static int fib_encap_match(struct net *net, u16 encap_type,
855 			   struct nlattr *encap,
856 			   const struct fib_nh *nh,
857 			   const struct fib_config *cfg,
858 			   struct netlink_ext_ack *extack)
859 {
860 	struct lwtunnel_state *lwtstate;
861 	int ret, result = 0;
862 
863 	if (encap_type == LWTUNNEL_ENCAP_NONE)
864 		return 0;
865 
866 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
867 				   cfg, &lwtstate, extack);
868 	if (!ret) {
869 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
870 		lwtstate_free(lwtstate);
871 	}
872 
873 	return result;
874 }
875 
876 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
877 		 struct netlink_ext_ack *extack)
878 {
879 #ifdef CONFIG_IP_ROUTE_MULTIPATH
880 	struct rtnexthop *rtnh;
881 	int remaining;
882 #endif
883 
884 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
885 		return 1;
886 
887 	if (cfg->fc_nh_id) {
888 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
889 			return 0;
890 		return 1;
891 	}
892 
893 	if (fi->nh) {
894 		if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
895 			return 1;
896 		return 0;
897 	}
898 
899 	if (cfg->fc_oif || cfg->fc_gw_family) {
900 		struct fib_nh *nh;
901 
902 		nh = fib_info_nh(fi, 0);
903 		if (cfg->fc_encap) {
904 			if (fib_encap_match(net, cfg->fc_encap_type,
905 					    cfg->fc_encap, nh, cfg, extack))
906 				return 1;
907 		}
908 #ifdef CONFIG_IP_ROUTE_CLASSID
909 		if (cfg->fc_flow &&
910 		    cfg->fc_flow != nh->nh_tclassid)
911 			return 1;
912 #endif
913 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
914 		    (cfg->fc_gw_family &&
915 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
916 			return 1;
917 
918 		if (cfg->fc_gw_family == AF_INET &&
919 		    cfg->fc_gw4 != nh->fib_nh_gw4)
920 			return 1;
921 
922 		if (cfg->fc_gw_family == AF_INET6 &&
923 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
924 			return 1;
925 
926 		return 0;
927 	}
928 
929 #ifdef CONFIG_IP_ROUTE_MULTIPATH
930 	if (!cfg->fc_mp)
931 		return 0;
932 
933 	rtnh = cfg->fc_mp;
934 	remaining = cfg->fc_mp_len;
935 
936 	for_nexthops(fi) {
937 		int attrlen;
938 
939 		if (!rtnh_ok(rtnh, remaining))
940 			return -EINVAL;
941 
942 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
943 			return 1;
944 
945 		attrlen = rtnh_attrlen(rtnh);
946 		if (attrlen > 0) {
947 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
948 			int err;
949 
950 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
951 			nlav = nla_find(attrs, attrlen, RTA_VIA);
952 			if (nla && nlav) {
953 				NL_SET_ERR_MSG(extack,
954 					       "Nexthop configuration can not contain both GATEWAY and VIA");
955 				return -EINVAL;
956 			}
957 
958 			if (nla) {
959 				__be32 gw;
960 
961 				err = fib_gw_from_attr(&gw, nla, extack);
962 				if (err)
963 					return err;
964 
965 				if (nh->fib_nh_gw_family != AF_INET ||
966 				    gw != nh->fib_nh_gw4)
967 					return 1;
968 			} else if (nlav) {
969 				struct fib_config cfg2;
970 
971 				err = fib_gw_from_via(&cfg2, nlav, extack);
972 				if (err)
973 					return err;
974 
975 				switch (nh->fib_nh_gw_family) {
976 				case AF_INET:
977 					if (cfg2.fc_gw_family != AF_INET ||
978 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
979 						return 1;
980 					break;
981 				case AF_INET6:
982 					if (cfg2.fc_gw_family != AF_INET6 ||
983 					    ipv6_addr_cmp(&cfg2.fc_gw6,
984 							  &nh->fib_nh_gw6))
985 						return 1;
986 					break;
987 				}
988 			}
989 
990 #ifdef CONFIG_IP_ROUTE_CLASSID
991 			nla = nla_find(attrs, attrlen, RTA_FLOW);
992 			if (nla) {
993 				if (nla_len(nla) < sizeof(u32)) {
994 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
995 					return -EINVAL;
996 				}
997 				if (nla_get_u32(nla) != nh->nh_tclassid)
998 					return 1;
999 			}
1000 #endif
1001 		}
1002 
1003 		rtnh = rtnh_next(rtnh, &remaining);
1004 	} endfor_nexthops(fi);
1005 #endif
1006 	return 0;
1007 }
1008 
1009 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1010 {
1011 	struct nlattr *nla;
1012 	int remaining;
1013 
1014 	if (!cfg->fc_mx)
1015 		return true;
1016 
1017 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1018 		int type = nla_type(nla);
1019 		u32 fi_val, val;
1020 
1021 		if (!type)
1022 			continue;
1023 		if (type > RTAX_MAX)
1024 			return false;
1025 
1026 		type = array_index_nospec(type, RTAX_MAX + 1);
1027 		if (type == RTAX_CC_ALGO) {
1028 			char tmp[TCP_CA_NAME_MAX];
1029 			bool ecn_ca = false;
1030 
1031 			nla_strscpy(tmp, nla, sizeof(tmp));
1032 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
1033 		} else {
1034 			if (nla_len(nla) != sizeof(u32))
1035 				return false;
1036 			val = nla_get_u32(nla);
1037 		}
1038 
1039 		fi_val = fi->fib_metrics->metrics[type - 1];
1040 		if (type == RTAX_FEATURES)
1041 			fi_val &= ~DST_FEATURE_ECN_CA;
1042 
1043 		if (fi_val != val)
1044 			return false;
1045 	}
1046 
1047 	return true;
1048 }
1049 
1050 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1051 			      u32 table, struct netlink_ext_ack *extack)
1052 {
1053 	struct fib6_config cfg = {
1054 		.fc_table = table,
1055 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1056 		.fc_ifindex = nh->fib_nh_oif,
1057 		.fc_gateway = nh->fib_nh_gw6,
1058 	};
1059 	struct fib6_nh fib6_nh = {};
1060 	int err;
1061 
1062 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1063 	if (!err) {
1064 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1065 		netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1066 			    GFP_KERNEL);
1067 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1068 		nh->fib_nh_scope = RT_SCOPE_LINK;
1069 
1070 		ipv6_stub->fib6_nh_release(&fib6_nh);
1071 	}
1072 
1073 	return err;
1074 }
1075 
1076 /*
1077  * Picture
1078  * -------
1079  *
1080  * Semantics of nexthop is very messy by historical reasons.
1081  * We have to take into account, that:
1082  * a) gateway can be actually local interface address,
1083  *    so that gatewayed route is direct.
1084  * b) gateway must be on-link address, possibly
1085  *    described not by an ifaddr, but also by a direct route.
1086  * c) If both gateway and interface are specified, they should not
1087  *    contradict.
1088  * d) If we use tunnel routes, gateway could be not on-link.
1089  *
1090  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1091  * results in pretty ugly and hairy code with obscure logic.
1092  *
1093  * I chose to generalized it instead, so that the size
1094  * of code does not increase practically, but it becomes
1095  * much more general.
1096  * Every prefix is assigned a "scope" value: "host" is local address,
1097  * "link" is direct route,
1098  * [ ... "site" ... "interior" ... ]
1099  * and "universe" is true gateway route with global meaning.
1100  *
1101  * Every prefix refers to a set of "nexthop"s (gw, oif),
1102  * where gw must have narrower scope. This recursion stops
1103  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1104  * which means that gw is forced to be on link.
1105  *
1106  * Code is still hairy, but now it is apparently logically
1107  * consistent and very flexible. F.e. as by-product it allows
1108  * to co-exists in peace independent exterior and interior
1109  * routing processes.
1110  *
1111  * Normally it looks as following.
1112  *
1113  * {universe prefix}  -> (gw, oif) [scope link]
1114  *		  |
1115  *		  |-> {link prefix} -> (gw, oif) [scope local]
1116  *					|
1117  *					|-> {local prefix} (terminal node)
1118  */
1119 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1120 			      u8 scope, struct netlink_ext_ack *extack)
1121 {
1122 	struct net_device *dev;
1123 	struct fib_result res;
1124 	int err = 0;
1125 
1126 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1127 		unsigned int addr_type;
1128 
1129 		if (scope >= RT_SCOPE_LINK) {
1130 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1131 			return -EINVAL;
1132 		}
1133 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1134 		if (!dev) {
1135 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1136 			return -ENODEV;
1137 		}
1138 		if (!(dev->flags & IFF_UP)) {
1139 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1140 			return -ENETDOWN;
1141 		}
1142 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1143 		if (addr_type != RTN_UNICAST) {
1144 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1145 			return -EINVAL;
1146 		}
1147 		if (!netif_carrier_ok(dev))
1148 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1149 		nh->fib_nh_dev = dev;
1150 		netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1151 		nh->fib_nh_scope = RT_SCOPE_LINK;
1152 		return 0;
1153 	}
1154 	rcu_read_lock();
1155 	{
1156 		struct fib_table *tbl = NULL;
1157 		struct flowi4 fl4 = {
1158 			.daddr = nh->fib_nh_gw4,
1159 			.flowi4_scope = scope + 1,
1160 			.flowi4_oif = nh->fib_nh_oif,
1161 			.flowi4_iif = LOOPBACK_IFINDEX,
1162 		};
1163 
1164 		/* It is not necessary, but requires a bit of thinking */
1165 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1166 			fl4.flowi4_scope = RT_SCOPE_LINK;
1167 
1168 		if (table && table != RT_TABLE_MAIN)
1169 			tbl = fib_get_table(net, table);
1170 
1171 		if (tbl)
1172 			err = fib_table_lookup(tbl, &fl4, &res,
1173 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1174 					       FIB_LOOKUP_NOREF);
1175 
1176 		/* on error or if no table given do full lookup. This
1177 		 * is needed for example when nexthops are in the local
1178 		 * table rather than the given table
1179 		 */
1180 		if (!tbl || err) {
1181 			err = fib_lookup(net, &fl4, &res,
1182 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1183 		}
1184 
1185 		if (err) {
1186 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1187 			goto out;
1188 		}
1189 	}
1190 
1191 	err = -EINVAL;
1192 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1193 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1194 		goto out;
1195 	}
1196 	nh->fib_nh_scope = res.scope;
1197 	nh->fib_nh_oif = FIB_RES_OIF(res);
1198 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1199 	if (!dev) {
1200 		NL_SET_ERR_MSG(extack,
1201 			       "No egress device for nexthop gateway");
1202 		goto out;
1203 	}
1204 	netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1205 	if (!netif_carrier_ok(dev))
1206 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1207 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1208 out:
1209 	rcu_read_unlock();
1210 	return err;
1211 }
1212 
1213 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1214 			      struct netlink_ext_ack *extack)
1215 {
1216 	struct in_device *in_dev;
1217 	int err;
1218 
1219 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1220 		NL_SET_ERR_MSG(extack,
1221 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1222 		return -EINVAL;
1223 	}
1224 
1225 	rcu_read_lock();
1226 
1227 	err = -ENODEV;
1228 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1229 	if (!in_dev)
1230 		goto out;
1231 	err = -ENETDOWN;
1232 	if (!(in_dev->dev->flags & IFF_UP)) {
1233 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1234 		goto out;
1235 	}
1236 
1237 	nh->fib_nh_dev = in_dev->dev;
1238 	netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1239 	nh->fib_nh_scope = RT_SCOPE_HOST;
1240 	if (!netif_carrier_ok(nh->fib_nh_dev))
1241 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1242 	err = 0;
1243 out:
1244 	rcu_read_unlock();
1245 	return err;
1246 }
1247 
1248 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1249 		 struct netlink_ext_ack *extack)
1250 {
1251 	int err;
1252 
1253 	if (nh->fib_nh_gw_family == AF_INET)
1254 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1255 	else if (nh->fib_nh_gw_family == AF_INET6)
1256 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1257 	else
1258 		err = fib_check_nh_nongw(net, nh, extack);
1259 
1260 	return err;
1261 }
1262 
1263 static struct hlist_head *
1264 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1265 {
1266 	u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1267 			   fib_info_hash_bits);
1268 
1269 	return &fib_info_laddrhash[slot];
1270 }
1271 
1272 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1273 			       struct hlist_head *new_laddrhash,
1274 			       unsigned int new_size)
1275 {
1276 	struct hlist_head *old_info_hash, *old_laddrhash;
1277 	unsigned int old_size = fib_info_hash_size;
1278 	unsigned int i;
1279 
1280 	spin_lock_bh(&fib_info_lock);
1281 	old_info_hash = fib_info_hash;
1282 	old_laddrhash = fib_info_laddrhash;
1283 	fib_info_hash_size = new_size;
1284 	fib_info_hash_bits = ilog2(new_size);
1285 
1286 	for (i = 0; i < old_size; i++) {
1287 		struct hlist_head *head = &fib_info_hash[i];
1288 		struct hlist_node *n;
1289 		struct fib_info *fi;
1290 
1291 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1292 			struct hlist_head *dest;
1293 			unsigned int new_hash;
1294 
1295 			new_hash = fib_info_hashfn(fi);
1296 			dest = &new_info_hash[new_hash];
1297 			hlist_add_head(&fi->fib_hash, dest);
1298 		}
1299 	}
1300 	fib_info_hash = new_info_hash;
1301 
1302 	fib_info_laddrhash = new_laddrhash;
1303 	for (i = 0; i < old_size; i++) {
1304 		struct hlist_head *lhead = &old_laddrhash[i];
1305 		struct hlist_node *n;
1306 		struct fib_info *fi;
1307 
1308 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1309 			struct hlist_head *ldest;
1310 
1311 			ldest = fib_info_laddrhash_bucket(fi->fib_net,
1312 							  fi->fib_prefsrc);
1313 			hlist_add_head(&fi->fib_lhash, ldest);
1314 		}
1315 	}
1316 
1317 	spin_unlock_bh(&fib_info_lock);
1318 
1319 	kvfree(old_info_hash);
1320 	kvfree(old_laddrhash);
1321 }
1322 
1323 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1324 				 unsigned char scope)
1325 {
1326 	struct fib_nh *nh;
1327 	__be32 saddr;
1328 
1329 	if (nhc->nhc_family != AF_INET)
1330 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1331 
1332 	nh = container_of(nhc, struct fib_nh, nh_common);
1333 	saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1334 
1335 	WRITE_ONCE(nh->nh_saddr, saddr);
1336 	WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid));
1337 
1338 	return saddr;
1339 }
1340 
1341 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1342 {
1343 	struct fib_nh_common *nhc = res->nhc;
1344 
1345 	if (res->fi->fib_prefsrc)
1346 		return res->fi->fib_prefsrc;
1347 
1348 	if (nhc->nhc_family == AF_INET) {
1349 		struct fib_nh *nh;
1350 
1351 		nh = container_of(nhc, struct fib_nh, nh_common);
1352 		if (READ_ONCE(nh->nh_saddr_genid) ==
1353 		    atomic_read(&net->ipv4.dev_addr_genid))
1354 			return READ_ONCE(nh->nh_saddr);
1355 	}
1356 
1357 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1358 }
1359 
1360 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1361 {
1362 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1363 	    fib_prefsrc != cfg->fc_dst) {
1364 		u32 tb_id = cfg->fc_table;
1365 		int rc;
1366 
1367 		if (tb_id == RT_TABLE_MAIN)
1368 			tb_id = RT_TABLE_LOCAL;
1369 
1370 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1371 					  fib_prefsrc, tb_id);
1372 
1373 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1374 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1375 						  fib_prefsrc, RT_TABLE_LOCAL);
1376 		}
1377 
1378 		if (rc != RTN_LOCAL)
1379 			return false;
1380 	}
1381 	return true;
1382 }
1383 
1384 struct fib_info *fib_create_info(struct fib_config *cfg,
1385 				 struct netlink_ext_ack *extack)
1386 {
1387 	int err;
1388 	struct fib_info *fi = NULL;
1389 	struct nexthop *nh = NULL;
1390 	struct fib_info *ofi;
1391 	int nhs = 1;
1392 	struct net *net = cfg->fc_nlinfo.nl_net;
1393 
1394 	if (cfg->fc_type > RTN_MAX)
1395 		goto err_inval;
1396 
1397 	/* Fast check to catch the most weird cases */
1398 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1399 		NL_SET_ERR_MSG(extack, "Invalid scope");
1400 		goto err_inval;
1401 	}
1402 
1403 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1404 		NL_SET_ERR_MSG(extack,
1405 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1406 		goto err_inval;
1407 	}
1408 
1409 	if (cfg->fc_nh_id) {
1410 		if (!cfg->fc_mx) {
1411 			fi = fib_find_info_nh(net, cfg);
1412 			if (fi) {
1413 				refcount_inc(&fi->fib_treeref);
1414 				return fi;
1415 			}
1416 		}
1417 
1418 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1419 		if (!nh) {
1420 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1421 			goto err_inval;
1422 		}
1423 		nhs = 0;
1424 	}
1425 
1426 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1427 	if (cfg->fc_mp) {
1428 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1429 		if (nhs == 0)
1430 			goto err_inval;
1431 	}
1432 #endif
1433 
1434 	err = -ENOBUFS;
1435 
1436 	/* Paired with WRITE_ONCE() in fib_release_info() */
1437 	if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1438 		unsigned int new_size = fib_info_hash_size << 1;
1439 		struct hlist_head *new_info_hash;
1440 		struct hlist_head *new_laddrhash;
1441 		size_t bytes;
1442 
1443 		if (!new_size)
1444 			new_size = 16;
1445 		bytes = (size_t)new_size * sizeof(struct hlist_head *);
1446 		new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1447 		new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1448 		if (!new_info_hash || !new_laddrhash) {
1449 			kvfree(new_info_hash);
1450 			kvfree(new_laddrhash);
1451 		} else {
1452 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1453 		}
1454 		if (!fib_info_hash_size)
1455 			goto failure;
1456 	}
1457 
1458 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1459 	if (!fi)
1460 		goto failure;
1461 	fi->fib_metrics = ip_fib_metrics_init(cfg->fc_mx, cfg->fc_mx_len, extack);
1462 	if (IS_ERR(fi->fib_metrics)) {
1463 		err = PTR_ERR(fi->fib_metrics);
1464 		kfree(fi);
1465 		return ERR_PTR(err);
1466 	}
1467 
1468 	fi->fib_net = net;
1469 	fi->fib_protocol = cfg->fc_protocol;
1470 	fi->fib_scope = cfg->fc_scope;
1471 	fi->fib_flags = cfg->fc_flags;
1472 	fi->fib_priority = cfg->fc_priority;
1473 	fi->fib_prefsrc = cfg->fc_prefsrc;
1474 	fi->fib_type = cfg->fc_type;
1475 	fi->fib_tb_id = cfg->fc_table;
1476 
1477 	fi->fib_nhs = nhs;
1478 	if (nh) {
1479 		if (!nexthop_get(nh)) {
1480 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1481 			err = -EINVAL;
1482 		} else {
1483 			err = 0;
1484 			fi->nh = nh;
1485 		}
1486 	} else {
1487 		change_nexthops(fi) {
1488 			nexthop_nh->nh_parent = fi;
1489 		} endfor_nexthops(fi)
1490 
1491 		if (cfg->fc_mp)
1492 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1493 					  extack);
1494 		else
1495 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1496 	}
1497 
1498 	if (err != 0)
1499 		goto failure;
1500 
1501 	if (fib_props[cfg->fc_type].error) {
1502 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1503 			NL_SET_ERR_MSG(extack,
1504 				       "Gateway, device and multipath can not be specified for this route type");
1505 			goto err_inval;
1506 		}
1507 		goto link_it;
1508 	} else {
1509 		switch (cfg->fc_type) {
1510 		case RTN_UNICAST:
1511 		case RTN_LOCAL:
1512 		case RTN_BROADCAST:
1513 		case RTN_ANYCAST:
1514 		case RTN_MULTICAST:
1515 			break;
1516 		default:
1517 			NL_SET_ERR_MSG(extack, "Invalid route type");
1518 			goto err_inval;
1519 		}
1520 	}
1521 
1522 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1523 		NL_SET_ERR_MSG(extack, "Invalid scope");
1524 		goto err_inval;
1525 	}
1526 
1527 	if (fi->nh) {
1528 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1529 		if (err)
1530 			goto failure;
1531 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1532 		struct fib_nh *nh = fi->fib_nh;
1533 
1534 		/* Local address is added. */
1535 		if (nhs != 1) {
1536 			NL_SET_ERR_MSG(extack,
1537 				       "Route with host scope can not have multiple nexthops");
1538 			goto err_inval;
1539 		}
1540 		if (nh->fib_nh_gw_family) {
1541 			NL_SET_ERR_MSG(extack,
1542 				       "Route with host scope can not have a gateway");
1543 			goto err_inval;
1544 		}
1545 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1546 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1547 		err = -ENODEV;
1548 		if (!nh->fib_nh_dev)
1549 			goto failure;
1550 		netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1551 				     GFP_KERNEL);
1552 	} else {
1553 		int linkdown = 0;
1554 
1555 		change_nexthops(fi) {
1556 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1557 					   cfg->fc_table, cfg->fc_scope,
1558 					   extack);
1559 			if (err != 0)
1560 				goto failure;
1561 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1562 				linkdown++;
1563 		} endfor_nexthops(fi)
1564 		if (linkdown == fi->fib_nhs)
1565 			fi->fib_flags |= RTNH_F_LINKDOWN;
1566 	}
1567 
1568 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1569 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1570 		goto err_inval;
1571 	}
1572 
1573 	if (!fi->nh) {
1574 		change_nexthops(fi) {
1575 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1576 						  fi->fib_scope);
1577 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1578 				fi->fib_nh_is_v6 = true;
1579 		} endfor_nexthops(fi)
1580 
1581 		fib_rebalance(fi);
1582 	}
1583 
1584 link_it:
1585 	ofi = fib_find_info(fi);
1586 	if (ofi) {
1587 		/* fib_table_lookup() should not see @fi yet. */
1588 		fi->fib_dead = 1;
1589 		free_fib_info(fi);
1590 		refcount_inc(&ofi->fib_treeref);
1591 		return ofi;
1592 	}
1593 
1594 	refcount_set(&fi->fib_treeref, 1);
1595 	refcount_set(&fi->fib_clntref, 1);
1596 	spin_lock_bh(&fib_info_lock);
1597 	fib_info_cnt++;
1598 	hlist_add_head(&fi->fib_hash,
1599 		       &fib_info_hash[fib_info_hashfn(fi)]);
1600 	if (fi->fib_prefsrc) {
1601 		struct hlist_head *head;
1602 
1603 		head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1604 		hlist_add_head(&fi->fib_lhash, head);
1605 	}
1606 	if (fi->nh) {
1607 		list_add(&fi->nh_list, &nh->fi_list);
1608 	} else {
1609 		change_nexthops(fi) {
1610 			struct hlist_head *head;
1611 
1612 			if (!nexthop_nh->fib_nh_dev)
1613 				continue;
1614 			head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1615 			hlist_add_head(&nexthop_nh->nh_hash, head);
1616 		} endfor_nexthops(fi)
1617 	}
1618 	spin_unlock_bh(&fib_info_lock);
1619 	return fi;
1620 
1621 err_inval:
1622 	err = -EINVAL;
1623 
1624 failure:
1625 	if (fi) {
1626 		/* fib_table_lookup() should not see @fi yet. */
1627 		fi->fib_dead = 1;
1628 		free_fib_info(fi);
1629 	}
1630 
1631 	return ERR_PTR(err);
1632 }
1633 
1634 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1635 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1636 {
1637 	if (nhc->nhc_flags & RTNH_F_DEAD)
1638 		*flags |= RTNH_F_DEAD;
1639 
1640 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1641 		*flags |= RTNH_F_LINKDOWN;
1642 
1643 		rcu_read_lock();
1644 		switch (nhc->nhc_family) {
1645 		case AF_INET:
1646 			if (ip_ignore_linkdown(nhc->nhc_dev))
1647 				*flags |= RTNH_F_DEAD;
1648 			break;
1649 		case AF_INET6:
1650 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1651 				*flags |= RTNH_F_DEAD;
1652 			break;
1653 		}
1654 		rcu_read_unlock();
1655 	}
1656 
1657 	switch (nhc->nhc_gw_family) {
1658 	case AF_INET:
1659 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1660 			goto nla_put_failure;
1661 		break;
1662 	case AF_INET6:
1663 		/* if gateway family does not match nexthop family
1664 		 * gateway is encoded as RTA_VIA
1665 		 */
1666 		if (rt_family != nhc->nhc_gw_family) {
1667 			int alen = sizeof(struct in6_addr);
1668 			struct nlattr *nla;
1669 			struct rtvia *via;
1670 
1671 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1672 			if (!nla)
1673 				goto nla_put_failure;
1674 
1675 			via = nla_data(nla);
1676 			via->rtvia_family = AF_INET6;
1677 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1678 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1679 					    &nhc->nhc_gw.ipv6) < 0) {
1680 			goto nla_put_failure;
1681 		}
1682 		break;
1683 	}
1684 
1685 	*flags |= (nhc->nhc_flags &
1686 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1687 
1688 	if (!skip_oif && nhc->nhc_dev &&
1689 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1690 		goto nla_put_failure;
1691 
1692 	if (nhc->nhc_lwtstate &&
1693 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1694 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1695 		goto nla_put_failure;
1696 
1697 	return 0;
1698 
1699 nla_put_failure:
1700 	return -EMSGSIZE;
1701 }
1702 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1703 
1704 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1705 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1706 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1707 {
1708 	const struct net_device *dev = nhc->nhc_dev;
1709 	struct rtnexthop *rtnh;
1710 	unsigned char flags = 0;
1711 
1712 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1713 	if (!rtnh)
1714 		goto nla_put_failure;
1715 
1716 	rtnh->rtnh_hops = nh_weight - 1;
1717 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1718 
1719 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1720 		goto nla_put_failure;
1721 
1722 	rtnh->rtnh_flags = flags;
1723 
1724 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1725 		goto nla_put_failure;
1726 
1727 	/* length of rtnetlink header + attributes */
1728 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1729 
1730 	return 0;
1731 
1732 nla_put_failure:
1733 	return -EMSGSIZE;
1734 }
1735 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1736 #endif
1737 
1738 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1739 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1740 {
1741 	struct nlattr *mp;
1742 
1743 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1744 	if (!mp)
1745 		goto nla_put_failure;
1746 
1747 	if (unlikely(fi->nh)) {
1748 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1749 			goto nla_put_failure;
1750 		goto mp_end;
1751 	}
1752 
1753 	for_nexthops(fi) {
1754 		u32 nh_tclassid = 0;
1755 #ifdef CONFIG_IP_ROUTE_CLASSID
1756 		nh_tclassid = nh->nh_tclassid;
1757 #endif
1758 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1759 				    AF_INET, nh_tclassid) < 0)
1760 			goto nla_put_failure;
1761 	} endfor_nexthops(fi);
1762 
1763 mp_end:
1764 	nla_nest_end(skb, mp);
1765 
1766 	return 0;
1767 
1768 nla_put_failure:
1769 	return -EMSGSIZE;
1770 }
1771 #else
1772 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1773 {
1774 	return 0;
1775 }
1776 #endif
1777 
1778 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1779 		  const struct fib_rt_info *fri, unsigned int flags)
1780 {
1781 	unsigned int nhs = fib_info_num_path(fri->fi);
1782 	struct fib_info *fi = fri->fi;
1783 	u32 tb_id = fri->tb_id;
1784 	struct nlmsghdr *nlh;
1785 	struct rtmsg *rtm;
1786 
1787 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1788 	if (!nlh)
1789 		return -EMSGSIZE;
1790 
1791 	rtm = nlmsg_data(nlh);
1792 	rtm->rtm_family = AF_INET;
1793 	rtm->rtm_dst_len = fri->dst_len;
1794 	rtm->rtm_src_len = 0;
1795 	rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1796 	if (tb_id < 256)
1797 		rtm->rtm_table = tb_id;
1798 	else
1799 		rtm->rtm_table = RT_TABLE_COMPAT;
1800 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1801 		goto nla_put_failure;
1802 	rtm->rtm_type = fri->type;
1803 	rtm->rtm_flags = fi->fib_flags;
1804 	rtm->rtm_scope = fi->fib_scope;
1805 	rtm->rtm_protocol = fi->fib_protocol;
1806 
1807 	if (rtm->rtm_dst_len &&
1808 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1809 		goto nla_put_failure;
1810 	if (fi->fib_priority &&
1811 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1812 		goto nla_put_failure;
1813 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1814 		goto nla_put_failure;
1815 
1816 	if (fi->fib_prefsrc &&
1817 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1818 		goto nla_put_failure;
1819 
1820 	if (fi->nh) {
1821 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1822 			goto nla_put_failure;
1823 		if (nexthop_is_blackhole(fi->nh))
1824 			rtm->rtm_type = RTN_BLACKHOLE;
1825 		if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1826 			goto offload;
1827 	}
1828 
1829 	if (nhs == 1) {
1830 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1831 		unsigned char flags = 0;
1832 
1833 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1834 			goto nla_put_failure;
1835 
1836 		rtm->rtm_flags = flags;
1837 #ifdef CONFIG_IP_ROUTE_CLASSID
1838 		if (nhc->nhc_family == AF_INET) {
1839 			struct fib_nh *nh;
1840 
1841 			nh = container_of(nhc, struct fib_nh, nh_common);
1842 			if (nh->nh_tclassid &&
1843 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1844 				goto nla_put_failure;
1845 		}
1846 #endif
1847 	} else {
1848 		if (fib_add_multipath(skb, fi) < 0)
1849 			goto nla_put_failure;
1850 	}
1851 
1852 offload:
1853 	if (fri->offload)
1854 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1855 	if (fri->trap)
1856 		rtm->rtm_flags |= RTM_F_TRAP;
1857 	if (fri->offload_failed)
1858 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1859 
1860 	nlmsg_end(skb, nlh);
1861 	return 0;
1862 
1863 nla_put_failure:
1864 	nlmsg_cancel(skb, nlh);
1865 	return -EMSGSIZE;
1866 }
1867 
1868 /*
1869  * Update FIB if:
1870  * - local address disappeared -> we must delete all the entries
1871  *   referring to it.
1872  * - device went down -> we must shutdown all nexthops going via it.
1873  */
1874 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1875 {
1876 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1877 	struct net *net = dev_net(dev);
1878 	struct hlist_head *head;
1879 	struct fib_info *fi;
1880 	int ret = 0;
1881 
1882 	if (!fib_info_laddrhash || local == 0)
1883 		return 0;
1884 
1885 	head = fib_info_laddrhash_bucket(net, local);
1886 	hlist_for_each_entry(fi, head, fib_lhash) {
1887 		if (!net_eq(fi->fib_net, net) ||
1888 		    fi->fib_tb_id != tb_id)
1889 			continue;
1890 		if (fi->fib_prefsrc == local) {
1891 			fi->fib_flags |= RTNH_F_DEAD;
1892 			fi->pfsrc_removed = true;
1893 			ret++;
1894 		}
1895 	}
1896 	return ret;
1897 }
1898 
1899 static int call_fib_nh_notifiers(struct fib_nh *nh,
1900 				 enum fib_event_type event_type)
1901 {
1902 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1903 	struct fib_nh_notifier_info info = {
1904 		.fib_nh = nh,
1905 	};
1906 
1907 	switch (event_type) {
1908 	case FIB_EVENT_NH_ADD:
1909 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1910 			break;
1911 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1912 			break;
1913 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1914 					   &info.info);
1915 	case FIB_EVENT_NH_DEL:
1916 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1917 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1918 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1919 						   event_type, &info.info);
1920 		break;
1921 	default:
1922 		break;
1923 	}
1924 
1925 	return NOTIFY_DONE;
1926 }
1927 
1928 /* Update the PMTU of exceptions when:
1929  * - the new MTU of the first hop becomes smaller than the PMTU
1930  * - the old MTU was the same as the PMTU, and it limited discovery of
1931  *   larger MTUs on the path. With that limit raised, we can now
1932  *   discover larger MTUs
1933  * A special case is locked exceptions, for which the PMTU is smaller
1934  * than the minimal accepted PMTU:
1935  * - if the new MTU is greater than the PMTU, don't make any change
1936  * - otherwise, unlock and set PMTU
1937  */
1938 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1939 {
1940 	struct fnhe_hash_bucket *bucket;
1941 	int i;
1942 
1943 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1944 	if (!bucket)
1945 		return;
1946 
1947 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1948 		struct fib_nh_exception *fnhe;
1949 
1950 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1951 		     fnhe;
1952 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1953 			if (fnhe->fnhe_mtu_locked) {
1954 				if (new <= fnhe->fnhe_pmtu) {
1955 					fnhe->fnhe_pmtu = new;
1956 					fnhe->fnhe_mtu_locked = false;
1957 				}
1958 			} else if (new < fnhe->fnhe_pmtu ||
1959 				   orig == fnhe->fnhe_pmtu) {
1960 				fnhe->fnhe_pmtu = new;
1961 			}
1962 		}
1963 	}
1964 }
1965 
1966 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1967 {
1968 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1969 	struct fib_nh *nh;
1970 
1971 	hlist_for_each_entry(nh, head, nh_hash) {
1972 		if (nh->fib_nh_dev == dev)
1973 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1974 	}
1975 }
1976 
1977 /* Event              force Flags           Description
1978  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1979  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1980  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1981  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1982  *
1983  * only used when fib_nh is built into fib_info
1984  */
1985 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1986 {
1987 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1988 	struct fib_info *prev_fi = NULL;
1989 	int scope = RT_SCOPE_NOWHERE;
1990 	struct fib_nh *nh;
1991 	int ret = 0;
1992 
1993 	if (force)
1994 		scope = -1;
1995 
1996 	hlist_for_each_entry(nh, head, nh_hash) {
1997 		struct fib_info *fi = nh->nh_parent;
1998 		int dead;
1999 
2000 		BUG_ON(!fi->fib_nhs);
2001 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2002 			continue;
2003 		prev_fi = fi;
2004 		dead = 0;
2005 		change_nexthops(fi) {
2006 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2007 				dead++;
2008 			else if (nexthop_nh->fib_nh_dev == dev &&
2009 				 nexthop_nh->fib_nh_scope != scope) {
2010 				switch (event) {
2011 				case NETDEV_DOWN:
2012 				case NETDEV_UNREGISTER:
2013 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2014 					fallthrough;
2015 				case NETDEV_CHANGE:
2016 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2017 					break;
2018 				}
2019 				call_fib_nh_notifiers(nexthop_nh,
2020 						      FIB_EVENT_NH_DEL);
2021 				dead++;
2022 			}
2023 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2024 			if (event == NETDEV_UNREGISTER &&
2025 			    nexthop_nh->fib_nh_dev == dev) {
2026 				dead = fi->fib_nhs;
2027 				break;
2028 			}
2029 #endif
2030 		} endfor_nexthops(fi)
2031 		if (dead == fi->fib_nhs) {
2032 			switch (event) {
2033 			case NETDEV_DOWN:
2034 			case NETDEV_UNREGISTER:
2035 				fi->fib_flags |= RTNH_F_DEAD;
2036 				fallthrough;
2037 			case NETDEV_CHANGE:
2038 				fi->fib_flags |= RTNH_F_LINKDOWN;
2039 				break;
2040 			}
2041 			ret++;
2042 		}
2043 
2044 		fib_rebalance(fi);
2045 	}
2046 
2047 	return ret;
2048 }
2049 
2050 /* Must be invoked inside of an RCU protected region.  */
2051 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2052 {
2053 	struct fib_info *fi = NULL, *last_resort = NULL;
2054 	struct hlist_head *fa_head = res->fa_head;
2055 	struct fib_table *tb = res->table;
2056 	u8 slen = 32 - res->prefixlen;
2057 	int order = -1, last_idx = -1;
2058 	struct fib_alias *fa, *fa1 = NULL;
2059 	u32 last_prio = res->fi->fib_priority;
2060 	dscp_t last_dscp = 0;
2061 
2062 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2063 		struct fib_info *next_fi = fa->fa_info;
2064 		struct fib_nh_common *nhc;
2065 
2066 		if (fa->fa_slen != slen)
2067 			continue;
2068 		if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp))
2069 			continue;
2070 		if (fa->tb_id != tb->tb_id)
2071 			continue;
2072 		if (next_fi->fib_priority > last_prio &&
2073 		    fa->fa_dscp == last_dscp) {
2074 			if (last_dscp)
2075 				continue;
2076 			break;
2077 		}
2078 		if (next_fi->fib_flags & RTNH_F_DEAD)
2079 			continue;
2080 		last_dscp = fa->fa_dscp;
2081 		last_prio = next_fi->fib_priority;
2082 
2083 		if (next_fi->fib_scope != res->scope ||
2084 		    fa->fa_type != RTN_UNICAST)
2085 			continue;
2086 
2087 		nhc = fib_info_nhc(next_fi, 0);
2088 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2089 			continue;
2090 
2091 		fib_alias_accessed(fa);
2092 
2093 		if (!fi) {
2094 			if (next_fi != res->fi)
2095 				break;
2096 			fa1 = fa;
2097 		} else if (!fib_detect_death(fi, order, &last_resort,
2098 					     &last_idx, fa1->fa_default)) {
2099 			fib_result_assign(res, fi);
2100 			fa1->fa_default = order;
2101 			goto out;
2102 		}
2103 		fi = next_fi;
2104 		order++;
2105 	}
2106 
2107 	if (order <= 0 || !fi) {
2108 		if (fa1)
2109 			fa1->fa_default = -1;
2110 		goto out;
2111 	}
2112 
2113 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2114 			      fa1->fa_default)) {
2115 		fib_result_assign(res, fi);
2116 		fa1->fa_default = order;
2117 		goto out;
2118 	}
2119 
2120 	if (last_idx >= 0)
2121 		fib_result_assign(res, last_resort);
2122 	fa1->fa_default = last_idx;
2123 out:
2124 	return;
2125 }
2126 
2127 /*
2128  * Dead device goes up. We wake up dead nexthops.
2129  * It takes sense only on multipath routes.
2130  *
2131  * only used when fib_nh is built into fib_info
2132  */
2133 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2134 {
2135 	struct fib_info *prev_fi;
2136 	struct hlist_head *head;
2137 	struct fib_nh *nh;
2138 	int ret;
2139 
2140 	if (!(dev->flags & IFF_UP))
2141 		return 0;
2142 
2143 	if (nh_flags & RTNH_F_DEAD) {
2144 		unsigned int flags = dev_get_flags(dev);
2145 
2146 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2147 			nh_flags |= RTNH_F_LINKDOWN;
2148 	}
2149 
2150 	prev_fi = NULL;
2151 	head = fib_info_devhash_bucket(dev);
2152 	ret = 0;
2153 
2154 	hlist_for_each_entry(nh, head, nh_hash) {
2155 		struct fib_info *fi = nh->nh_parent;
2156 		int alive;
2157 
2158 		BUG_ON(!fi->fib_nhs);
2159 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2160 			continue;
2161 
2162 		prev_fi = fi;
2163 		alive = 0;
2164 		change_nexthops(fi) {
2165 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2166 				alive++;
2167 				continue;
2168 			}
2169 			if (!nexthop_nh->fib_nh_dev ||
2170 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2171 				continue;
2172 			if (nexthop_nh->fib_nh_dev != dev ||
2173 			    !__in_dev_get_rtnl(dev))
2174 				continue;
2175 			alive++;
2176 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2177 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2178 		} endfor_nexthops(fi)
2179 
2180 		if (alive > 0) {
2181 			fi->fib_flags &= ~nh_flags;
2182 			ret++;
2183 		}
2184 
2185 		fib_rebalance(fi);
2186 	}
2187 
2188 	return ret;
2189 }
2190 
2191 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2192 static bool fib_good_nh(const struct fib_nh *nh)
2193 {
2194 	int state = NUD_REACHABLE;
2195 
2196 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2197 		struct neighbour *n;
2198 
2199 		rcu_read_lock();
2200 
2201 		if (likely(nh->fib_nh_gw_family == AF_INET))
2202 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2203 						   (__force u32)nh->fib_nh_gw4);
2204 		else if (nh->fib_nh_gw_family == AF_INET6)
2205 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2206 							   &nh->fib_nh_gw6);
2207 		else
2208 			n = NULL;
2209 		if (n)
2210 			state = READ_ONCE(n->nud_state);
2211 
2212 		rcu_read_unlock();
2213 	}
2214 
2215 	return !!(state & NUD_VALID);
2216 }
2217 
2218 void fib_select_multipath(struct fib_result *res, int hash)
2219 {
2220 	struct fib_info *fi = res->fi;
2221 	struct net *net = fi->fib_net;
2222 	bool first = false;
2223 
2224 	if (unlikely(res->fi->nh)) {
2225 		nexthop_path_fib_result(res, hash);
2226 		return;
2227 	}
2228 
2229 	change_nexthops(fi) {
2230 		if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2231 			if (!fib_good_nh(nexthop_nh))
2232 				continue;
2233 			if (!first) {
2234 				res->nh_sel = nhsel;
2235 				res->nhc = &nexthop_nh->nh_common;
2236 				first = true;
2237 			}
2238 		}
2239 
2240 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2241 			continue;
2242 
2243 		res->nh_sel = nhsel;
2244 		res->nhc = &nexthop_nh->nh_common;
2245 		return;
2246 	} endfor_nexthops(fi);
2247 }
2248 #endif
2249 
2250 void fib_select_path(struct net *net, struct fib_result *res,
2251 		     struct flowi4 *fl4, const struct sk_buff *skb)
2252 {
2253 	if (fl4->flowi4_oif)
2254 		goto check_saddr;
2255 
2256 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2257 	if (fib_info_num_path(res->fi) > 1) {
2258 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2259 
2260 		fib_select_multipath(res, h);
2261 	}
2262 	else
2263 #endif
2264 	if (!res->prefixlen &&
2265 	    res->table->tb_num_default > 1 &&
2266 	    res->type == RTN_UNICAST)
2267 		fib_select_default(fl4, res);
2268 
2269 check_saddr:
2270 	if (!fl4->saddr) {
2271 		struct net_device *l3mdev;
2272 
2273 		l3mdev = dev_get_by_index_rcu(net, fl4->flowi4_l3mdev);
2274 
2275 		if (!l3mdev ||
2276 		    l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) == l3mdev)
2277 			fl4->saddr = fib_result_prefsrc(net, res);
2278 		else
2279 			fl4->saddr = inet_select_addr(l3mdev, 0, RT_SCOPE_LINK);
2280 	}
2281 }
2282