xref: /linux/net/ipv4/fib_semantics.c (revision 497e6b37b0099dc415578488287fd84fb74433eb)
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 
34 #include <net/arp.h>
35 #include <net/inet_dscp.h>
36 #include <net/ip.h>
37 #include <net/protocol.h>
38 #include <net/route.h>
39 #include <net/tcp.h>
40 #include <net/sock.h>
41 #include <net/ip_fib.h>
42 #include <net/ip6_fib.h>
43 #include <net/nexthop.h>
44 #include <net/netlink.h>
45 #include <net/rtnh.h>
46 #include <net/lwtunnel.h>
47 #include <net/fib_notifier.h>
48 #include <net/addrconf.h>
49 
50 #include "fib_lookup.h"
51 
52 static DEFINE_SPINLOCK(fib_info_lock);
53 static struct hlist_head *fib_info_hash;
54 static struct hlist_head *fib_info_laddrhash;
55 static unsigned int fib_info_hash_size;
56 static unsigned int fib_info_hash_bits;
57 static unsigned int fib_info_cnt;
58 
59 #define DEVINDEX_HASHBITS 8
60 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
61 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
62 
63 /* for_nexthops and change_nexthops only used when nexthop object
64  * is not set in a fib_info. The logic within can reference fib_nh.
65  */
66 #ifdef CONFIG_IP_ROUTE_MULTIPATH
67 
68 #define for_nexthops(fi) {						\
69 	int nhsel; const struct fib_nh *nh;				\
70 	for (nhsel = 0, nh = (fi)->fib_nh;				\
71 	     nhsel < fib_info_num_path((fi));				\
72 	     nh++, nhsel++)
73 
74 #define change_nexthops(fi) {						\
75 	int nhsel; struct fib_nh *nexthop_nh;				\
76 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
77 	     nhsel < fib_info_num_path((fi));				\
78 	     nexthop_nh++, nhsel++)
79 
80 #else /* CONFIG_IP_ROUTE_MULTIPATH */
81 
82 /* Hope, that gcc will optimize it to get rid of dummy loop */
83 
84 #define for_nexthops(fi) {						\
85 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
86 	for (nhsel = 0; nhsel < 1; nhsel++)
87 
88 #define change_nexthops(fi) {						\
89 	int nhsel;							\
90 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
91 	for (nhsel = 0; nhsel < 1; nhsel++)
92 
93 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
94 
95 #define endfor_nexthops(fi) }
96 
97 
98 const struct fib_prop fib_props[RTN_MAX + 1] = {
99 	[RTN_UNSPEC] = {
100 		.error	= 0,
101 		.scope	= RT_SCOPE_NOWHERE,
102 	},
103 	[RTN_UNICAST] = {
104 		.error	= 0,
105 		.scope	= RT_SCOPE_UNIVERSE,
106 	},
107 	[RTN_LOCAL] = {
108 		.error	= 0,
109 		.scope	= RT_SCOPE_HOST,
110 	},
111 	[RTN_BROADCAST] = {
112 		.error	= 0,
113 		.scope	= RT_SCOPE_LINK,
114 	},
115 	[RTN_ANYCAST] = {
116 		.error	= 0,
117 		.scope	= RT_SCOPE_LINK,
118 	},
119 	[RTN_MULTICAST] = {
120 		.error	= 0,
121 		.scope	= RT_SCOPE_UNIVERSE,
122 	},
123 	[RTN_BLACKHOLE] = {
124 		.error	= -EINVAL,
125 		.scope	= RT_SCOPE_UNIVERSE,
126 	},
127 	[RTN_UNREACHABLE] = {
128 		.error	= -EHOSTUNREACH,
129 		.scope	= RT_SCOPE_UNIVERSE,
130 	},
131 	[RTN_PROHIBIT] = {
132 		.error	= -EACCES,
133 		.scope	= RT_SCOPE_UNIVERSE,
134 	},
135 	[RTN_THROW] = {
136 		.error	= -EAGAIN,
137 		.scope	= RT_SCOPE_UNIVERSE,
138 	},
139 	[RTN_NAT] = {
140 		.error	= -EINVAL,
141 		.scope	= RT_SCOPE_NOWHERE,
142 	},
143 	[RTN_XRESOLVE] = {
144 		.error	= -EINVAL,
145 		.scope	= RT_SCOPE_NOWHERE,
146 	},
147 };
148 
149 static void rt_fibinfo_free(struct rtable __rcu **rtp)
150 {
151 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
152 
153 	if (!rt)
154 		return;
155 
156 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
157 	 * because we waited an RCU grace period before calling
158 	 * free_fib_info_rcu()
159 	 */
160 
161 	dst_dev_put(&rt->dst);
162 	dst_release_immediate(&rt->dst);
163 }
164 
165 static void free_nh_exceptions(struct fib_nh_common *nhc)
166 {
167 	struct fnhe_hash_bucket *hash;
168 	int i;
169 
170 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
171 	if (!hash)
172 		return;
173 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
174 		struct fib_nh_exception *fnhe;
175 
176 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
177 		while (fnhe) {
178 			struct fib_nh_exception *next;
179 
180 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
181 
182 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
183 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
184 
185 			kfree(fnhe);
186 
187 			fnhe = next;
188 		}
189 	}
190 	kfree(hash);
191 }
192 
193 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
194 {
195 	int cpu;
196 
197 	if (!rtp)
198 		return;
199 
200 	for_each_possible_cpu(cpu) {
201 		struct rtable *rt;
202 
203 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
204 		if (rt) {
205 			dst_dev_put(&rt->dst);
206 			dst_release_immediate(&rt->dst);
207 		}
208 	}
209 	free_percpu(rtp);
210 }
211 
212 void fib_nh_common_release(struct fib_nh_common *nhc)
213 {
214 	netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
215 	lwtstate_put(nhc->nhc_lwtstate);
216 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
217 	rt_fibinfo_free(&nhc->nhc_rth_input);
218 	free_nh_exceptions(nhc);
219 }
220 EXPORT_SYMBOL_GPL(fib_nh_common_release);
221 
222 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
223 {
224 #ifdef CONFIG_IP_ROUTE_CLASSID
225 	if (fib_nh->nh_tclassid)
226 		atomic_dec(&net->ipv4.fib_num_tclassid_users);
227 #endif
228 	fib_nh_common_release(&fib_nh->nh_common);
229 }
230 
231 /* Release a nexthop info record */
232 static void free_fib_info_rcu(struct rcu_head *head)
233 {
234 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
235 
236 	if (fi->nh) {
237 		nexthop_put(fi->nh);
238 	} else {
239 		change_nexthops(fi) {
240 			fib_nh_release(fi->fib_net, nexthop_nh);
241 		} endfor_nexthops(fi);
242 	}
243 
244 	ip_fib_metrics_put(fi->fib_metrics);
245 
246 	kfree(fi);
247 }
248 
249 void free_fib_info(struct fib_info *fi)
250 {
251 	if (fi->fib_dead == 0) {
252 		pr_warn("Freeing alive fib_info %p\n", fi);
253 		return;
254 	}
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 && refcount_dec_and_test(&fi->fib_treeref)) {
264 		hlist_del(&fi->fib_hash);
265 
266 		/* Paired with READ_ONCE() in fib_create_info(). */
267 		WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
268 
269 		if (fi->fib_prefsrc)
270 			hlist_del(&fi->fib_lhash);
271 		if (fi->nh) {
272 			list_del(&fi->nh_list);
273 		} else {
274 			change_nexthops(fi) {
275 				if (!nexthop_nh->fib_nh_dev)
276 					continue;
277 				hlist_del(&nexthop_nh->nh_hash);
278 			} endfor_nexthops(fi)
279 		}
280 		fi->fib_dead = 1;
281 		fib_info_put(fi);
282 	}
283 	spin_unlock_bh(&fib_info_lock);
284 }
285 
286 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
287 {
288 	const struct fib_nh *onh;
289 
290 	if (fi->nh || ofi->nh)
291 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
292 
293 	if (ofi->fib_nhs == 0)
294 		return 0;
295 
296 	for_nexthops(fi) {
297 		onh = fib_info_nh(ofi, nhsel);
298 
299 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
300 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
301 		    nh->fib_nh_scope != onh->fib_nh_scope ||
302 #ifdef CONFIG_IP_ROUTE_MULTIPATH
303 		    nh->fib_nh_weight != onh->fib_nh_weight ||
304 #endif
305 #ifdef CONFIG_IP_ROUTE_CLASSID
306 		    nh->nh_tclassid != onh->nh_tclassid ||
307 #endif
308 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
309 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
310 			return -1;
311 
312 		if (nh->fib_nh_gw_family == AF_INET &&
313 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
314 			return -1;
315 
316 		if (nh->fib_nh_gw_family == AF_INET6 &&
317 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
318 			return -1;
319 	} endfor_nexthops(fi);
320 	return 0;
321 }
322 
323 static inline unsigned int fib_devindex_hashfn(unsigned int val)
324 {
325 	return hash_32(val, DEVINDEX_HASHBITS);
326 }
327 
328 static struct hlist_head *
329 fib_info_devhash_bucket(const struct net_device *dev)
330 {
331 	u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
332 
333 	return &fib_info_devhash[fib_devindex_hashfn(val)];
334 }
335 
336 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
337 				      u32 prefsrc, u32 priority)
338 {
339 	unsigned int val = init_val;
340 
341 	val ^= (protocol << 8) | scope;
342 	val ^= prefsrc;
343 	val ^= priority;
344 
345 	return val;
346 }
347 
348 static unsigned int fib_info_hashfn_result(unsigned int val)
349 {
350 	unsigned int mask = (fib_info_hash_size - 1);
351 
352 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
353 }
354 
355 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
356 {
357 	unsigned int val;
358 
359 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
360 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
361 				fi->fib_priority);
362 
363 	if (fi->nh) {
364 		val ^= fib_devindex_hashfn(fi->nh->id);
365 	} else {
366 		for_nexthops(fi) {
367 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
368 		} endfor_nexthops(fi)
369 	}
370 
371 	return fib_info_hashfn_result(val);
372 }
373 
374 /* no metrics, only nexthop id */
375 static struct fib_info *fib_find_info_nh(struct net *net,
376 					 const struct fib_config *cfg)
377 {
378 	struct hlist_head *head;
379 	struct fib_info *fi;
380 	unsigned int hash;
381 
382 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
383 				 cfg->fc_protocol, cfg->fc_scope,
384 				 (__force u32)cfg->fc_prefsrc,
385 				 cfg->fc_priority);
386 	hash = fib_info_hashfn_result(hash);
387 	head = &fib_info_hash[hash];
388 
389 	hlist_for_each_entry(fi, head, fib_hash) {
390 		if (!net_eq(fi->fib_net, net))
391 			continue;
392 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
393 			continue;
394 		if (cfg->fc_protocol == fi->fib_protocol &&
395 		    cfg->fc_scope == fi->fib_scope &&
396 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
397 		    cfg->fc_priority == fi->fib_priority &&
398 		    cfg->fc_type == fi->fib_type &&
399 		    cfg->fc_table == fi->fib_tb_id &&
400 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
401 			return fi;
402 	}
403 
404 	return NULL;
405 }
406 
407 static struct fib_info *fib_find_info(struct fib_info *nfi)
408 {
409 	struct hlist_head *head;
410 	struct fib_info *fi;
411 	unsigned int hash;
412 
413 	hash = fib_info_hashfn(nfi);
414 	head = &fib_info_hash[hash];
415 
416 	hlist_for_each_entry(fi, head, fib_hash) {
417 		if (!net_eq(fi->fib_net, nfi->fib_net))
418 			continue;
419 		if (fi->fib_nhs != nfi->fib_nhs)
420 			continue;
421 		if (nfi->fib_protocol == fi->fib_protocol &&
422 		    nfi->fib_scope == fi->fib_scope &&
423 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
424 		    nfi->fib_priority == fi->fib_priority &&
425 		    nfi->fib_type == fi->fib_type &&
426 		    nfi->fib_tb_id == fi->fib_tb_id &&
427 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
428 			   sizeof(u32) * RTAX_MAX) == 0 &&
429 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
430 		    nh_comp(fi, nfi) == 0)
431 			return fi;
432 	}
433 
434 	return NULL;
435 }
436 
437 /* Check, that the gateway is already configured.
438  * Used only by redirect accept routine.
439  */
440 int ip_fib_check_default(__be32 gw, struct net_device *dev)
441 {
442 	struct hlist_head *head;
443 	struct fib_nh *nh;
444 
445 	spin_lock(&fib_info_lock);
446 
447 	head = fib_info_devhash_bucket(dev);
448 
449 	hlist_for_each_entry(nh, head, nh_hash) {
450 		if (nh->fib_nh_dev == dev &&
451 		    nh->fib_nh_gw4 == gw &&
452 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
453 			spin_unlock(&fib_info_lock);
454 			return 0;
455 		}
456 	}
457 
458 	spin_unlock(&fib_info_lock);
459 
460 	return -1;
461 }
462 
463 size_t fib_nlmsg_size(struct fib_info *fi)
464 {
465 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
466 			 + nla_total_size(4) /* RTA_TABLE */
467 			 + nla_total_size(4) /* RTA_DST */
468 			 + nla_total_size(4) /* RTA_PRIORITY */
469 			 + nla_total_size(4) /* RTA_PREFSRC */
470 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
471 	unsigned int nhs = fib_info_num_path(fi);
472 
473 	/* space for nested metrics */
474 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
475 
476 	if (fi->nh)
477 		payload += nla_total_size(4); /* RTA_NH_ID */
478 
479 	if (nhs) {
480 		size_t nh_encapsize = 0;
481 		/* Also handles the special case nhs == 1 */
482 
483 		/* each nexthop is packed in an attribute */
484 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
485 		unsigned int i;
486 
487 		/* may contain flow and gateway attribute */
488 		nhsize += 2 * nla_total_size(4);
489 
490 		/* grab encap info */
491 		for (i = 0; i < fib_info_num_path(fi); i++) {
492 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
493 
494 			if (nhc->nhc_lwtstate) {
495 				/* RTA_ENCAP_TYPE */
496 				nh_encapsize += lwtunnel_get_encap_size(
497 						nhc->nhc_lwtstate);
498 				/* RTA_ENCAP */
499 				nh_encapsize +=  nla_total_size(2);
500 			}
501 		}
502 
503 		/* all nexthops are packed in a nested attribute */
504 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
505 
506 	}
507 
508 	return payload;
509 }
510 
511 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
512 	       int dst_len, u32 tb_id, const struct nl_info *info,
513 	       unsigned int nlm_flags)
514 {
515 	struct fib_rt_info fri;
516 	struct sk_buff *skb;
517 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
518 	int err = -ENOBUFS;
519 
520 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
521 	if (!skb)
522 		goto errout;
523 
524 	fri.fi = fa->fa_info;
525 	fri.tb_id = tb_id;
526 	fri.dst = key;
527 	fri.dst_len = dst_len;
528 	fri.dscp = fa->fa_dscp;
529 	fri.type = fa->fa_type;
530 	fri.offload = READ_ONCE(fa->offload);
531 	fri.trap = READ_ONCE(fa->trap);
532 	fri.offload_failed = READ_ONCE(fa->offload_failed);
533 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
534 	if (err < 0) {
535 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
536 		WARN_ON(err == -EMSGSIZE);
537 		kfree_skb(skb);
538 		goto errout;
539 	}
540 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
541 		    info->nlh, GFP_KERNEL);
542 	return;
543 errout:
544 	if (err < 0)
545 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
546 }
547 
548 static int fib_detect_death(struct fib_info *fi, int order,
549 			    struct fib_info **last_resort, int *last_idx,
550 			    int dflt)
551 {
552 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
553 	struct neighbour *n;
554 	int state = NUD_NONE;
555 
556 	if (likely(nhc->nhc_gw_family == AF_INET))
557 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
558 	else if (nhc->nhc_gw_family == AF_INET6)
559 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
560 				 nhc->nhc_dev);
561 	else
562 		n = NULL;
563 
564 	if (n) {
565 		state = n->nud_state;
566 		neigh_release(n);
567 	} else {
568 		return 0;
569 	}
570 	if (state == NUD_REACHABLE)
571 		return 0;
572 	if ((state & NUD_VALID) && order != dflt)
573 		return 0;
574 	if ((state & NUD_VALID) ||
575 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
576 		*last_resort = fi;
577 		*last_idx = order;
578 	}
579 	return 1;
580 }
581 
582 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
583 		       struct nlattr *encap, u16 encap_type,
584 		       void *cfg, gfp_t gfp_flags,
585 		       struct netlink_ext_ack *extack)
586 {
587 	int err;
588 
589 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
590 						    gfp_flags);
591 	if (!nhc->nhc_pcpu_rth_output)
592 		return -ENOMEM;
593 
594 	if (encap) {
595 		struct lwtunnel_state *lwtstate;
596 
597 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
598 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
599 			err = -EINVAL;
600 			goto lwt_failure;
601 		}
602 		err = lwtunnel_build_state(net, encap_type, encap,
603 					   nhc->nhc_family, cfg, &lwtstate,
604 					   extack);
605 		if (err)
606 			goto lwt_failure;
607 
608 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
609 	}
610 
611 	return 0;
612 
613 lwt_failure:
614 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
615 	nhc->nhc_pcpu_rth_output = NULL;
616 	return err;
617 }
618 EXPORT_SYMBOL_GPL(fib_nh_common_init);
619 
620 int fib_nh_init(struct net *net, struct fib_nh *nh,
621 		struct fib_config *cfg, int nh_weight,
622 		struct netlink_ext_ack *extack)
623 {
624 	int err;
625 
626 	nh->fib_nh_family = AF_INET;
627 
628 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
629 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
630 	if (err)
631 		return err;
632 
633 	nh->fib_nh_oif = cfg->fc_oif;
634 	nh->fib_nh_gw_family = cfg->fc_gw_family;
635 	if (cfg->fc_gw_family == AF_INET)
636 		nh->fib_nh_gw4 = cfg->fc_gw4;
637 	else if (cfg->fc_gw_family == AF_INET6)
638 		nh->fib_nh_gw6 = cfg->fc_gw6;
639 
640 	nh->fib_nh_flags = cfg->fc_flags;
641 
642 #ifdef CONFIG_IP_ROUTE_CLASSID
643 	nh->nh_tclassid = cfg->fc_flow;
644 	if (nh->nh_tclassid)
645 		atomic_inc(&net->ipv4.fib_num_tclassid_users);
646 #endif
647 #ifdef CONFIG_IP_ROUTE_MULTIPATH
648 	nh->fib_nh_weight = nh_weight;
649 #endif
650 	return 0;
651 }
652 
653 #ifdef CONFIG_IP_ROUTE_MULTIPATH
654 
655 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
656 			      struct netlink_ext_ack *extack)
657 {
658 	int nhs = 0;
659 
660 	while (rtnh_ok(rtnh, remaining)) {
661 		nhs++;
662 		rtnh = rtnh_next(rtnh, &remaining);
663 	}
664 
665 	/* leftover implies invalid nexthop configuration, discard it */
666 	if (remaining > 0) {
667 		NL_SET_ERR_MSG(extack,
668 			       "Invalid nexthop configuration - extra data after nexthops");
669 		nhs = 0;
670 	}
671 
672 	return nhs;
673 }
674 
675 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
676 			    struct netlink_ext_ack *extack)
677 {
678 	if (nla_len(nla) < sizeof(*gw)) {
679 		NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
680 		return -EINVAL;
681 	}
682 
683 	*gw = nla_get_in_addr(nla);
684 
685 	return 0;
686 }
687 
688 /* only called when fib_nh is integrated into fib_info */
689 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
690 		       int remaining, struct fib_config *cfg,
691 		       struct netlink_ext_ack *extack)
692 {
693 	struct net *net = fi->fib_net;
694 	struct fib_config fib_cfg;
695 	struct fib_nh *nh;
696 	int ret;
697 
698 	change_nexthops(fi) {
699 		int attrlen;
700 
701 		memset(&fib_cfg, 0, sizeof(fib_cfg));
702 
703 		if (!rtnh_ok(rtnh, remaining)) {
704 			NL_SET_ERR_MSG(extack,
705 				       "Invalid nexthop configuration - extra data after nexthop");
706 			return -EINVAL;
707 		}
708 
709 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
710 			NL_SET_ERR_MSG(extack,
711 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
712 			return -EINVAL;
713 		}
714 
715 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
716 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
717 
718 		attrlen = rtnh_attrlen(rtnh);
719 		if (attrlen > 0) {
720 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
721 
722 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
723 			nlav = nla_find(attrs, attrlen, RTA_VIA);
724 			if (nla && nlav) {
725 				NL_SET_ERR_MSG(extack,
726 					       "Nexthop configuration can not contain both GATEWAY and VIA");
727 				return -EINVAL;
728 			}
729 			if (nla) {
730 				ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
731 						       extack);
732 				if (ret)
733 					goto errout;
734 
735 				if (fib_cfg.fc_gw4)
736 					fib_cfg.fc_gw_family = AF_INET;
737 			} else if (nlav) {
738 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
739 				if (ret)
740 					goto errout;
741 			}
742 
743 			nla = nla_find(attrs, attrlen, RTA_FLOW);
744 			if (nla) {
745 				if (nla_len(nla) < sizeof(u32)) {
746 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
747 					return -EINVAL;
748 				}
749 				fib_cfg.fc_flow = nla_get_u32(nla);
750 			}
751 
752 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
753 			/* RTA_ENCAP_TYPE length checked in
754 			 * lwtunnel_valid_encap_type_attr
755 			 */
756 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
757 			if (nla)
758 				fib_cfg.fc_encap_type = nla_get_u16(nla);
759 		}
760 
761 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
762 				  rtnh->rtnh_hops + 1, extack);
763 		if (ret)
764 			goto errout;
765 
766 		rtnh = rtnh_next(rtnh, &remaining);
767 	} endfor_nexthops(fi);
768 
769 	ret = -EINVAL;
770 	nh = fib_info_nh(fi, 0);
771 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
772 		NL_SET_ERR_MSG(extack,
773 			       "Nexthop device index does not match RTA_OIF");
774 		goto errout;
775 	}
776 	if (cfg->fc_gw_family) {
777 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
778 		    (cfg->fc_gw_family == AF_INET &&
779 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
780 		    (cfg->fc_gw_family == AF_INET6 &&
781 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
782 			NL_SET_ERR_MSG(extack,
783 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
784 			goto errout;
785 		}
786 	}
787 #ifdef CONFIG_IP_ROUTE_CLASSID
788 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
789 		NL_SET_ERR_MSG(extack,
790 			       "Nexthop class id does not match RTA_FLOW");
791 		goto errout;
792 	}
793 #endif
794 	ret = 0;
795 errout:
796 	return ret;
797 }
798 
799 /* only called when fib_nh is integrated into fib_info */
800 static void fib_rebalance(struct fib_info *fi)
801 {
802 	int total;
803 	int w;
804 
805 	if (fib_info_num_path(fi) < 2)
806 		return;
807 
808 	total = 0;
809 	for_nexthops(fi) {
810 		if (nh->fib_nh_flags & RTNH_F_DEAD)
811 			continue;
812 
813 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
814 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
815 			continue;
816 
817 		total += nh->fib_nh_weight;
818 	} endfor_nexthops(fi);
819 
820 	w = 0;
821 	change_nexthops(fi) {
822 		int upper_bound;
823 
824 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
825 			upper_bound = -1;
826 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
827 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
828 			upper_bound = -1;
829 		} else {
830 			w += nexthop_nh->fib_nh_weight;
831 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
832 							    total) - 1;
833 		}
834 
835 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
836 	} endfor_nexthops(fi);
837 }
838 #else /* CONFIG_IP_ROUTE_MULTIPATH */
839 
840 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
841 		       int remaining, struct fib_config *cfg,
842 		       struct netlink_ext_ack *extack)
843 {
844 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
845 
846 	return -EINVAL;
847 }
848 
849 #define fib_rebalance(fi) do { } while (0)
850 
851 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
852 
853 static int fib_encap_match(struct net *net, u16 encap_type,
854 			   struct nlattr *encap,
855 			   const struct fib_nh *nh,
856 			   const struct fib_config *cfg,
857 			   struct netlink_ext_ack *extack)
858 {
859 	struct lwtunnel_state *lwtstate;
860 	int ret, result = 0;
861 
862 	if (encap_type == LWTUNNEL_ENCAP_NONE)
863 		return 0;
864 
865 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
866 				   cfg, &lwtstate, extack);
867 	if (!ret) {
868 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
869 		lwtstate_free(lwtstate);
870 	}
871 
872 	return result;
873 }
874 
875 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
876 		 struct netlink_ext_ack *extack)
877 {
878 #ifdef CONFIG_IP_ROUTE_MULTIPATH
879 	struct rtnexthop *rtnh;
880 	int remaining;
881 #endif
882 
883 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
884 		return 1;
885 
886 	if (cfg->fc_nh_id) {
887 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
888 			return 0;
889 		return 1;
890 	}
891 
892 	if (fi->nh) {
893 		if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
894 			return 1;
895 		return 0;
896 	}
897 
898 	if (cfg->fc_oif || cfg->fc_gw_family) {
899 		struct fib_nh *nh;
900 
901 		nh = fib_info_nh(fi, 0);
902 		if (cfg->fc_encap) {
903 			if (fib_encap_match(net, cfg->fc_encap_type,
904 					    cfg->fc_encap, nh, cfg, extack))
905 				return 1;
906 		}
907 #ifdef CONFIG_IP_ROUTE_CLASSID
908 		if (cfg->fc_flow &&
909 		    cfg->fc_flow != nh->nh_tclassid)
910 			return 1;
911 #endif
912 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
913 		    (cfg->fc_gw_family &&
914 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
915 			return 1;
916 
917 		if (cfg->fc_gw_family == AF_INET &&
918 		    cfg->fc_gw4 != nh->fib_nh_gw4)
919 			return 1;
920 
921 		if (cfg->fc_gw_family == AF_INET6 &&
922 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
923 			return 1;
924 
925 		return 0;
926 	}
927 
928 #ifdef CONFIG_IP_ROUTE_MULTIPATH
929 	if (!cfg->fc_mp)
930 		return 0;
931 
932 	rtnh = cfg->fc_mp;
933 	remaining = cfg->fc_mp_len;
934 
935 	for_nexthops(fi) {
936 		int attrlen;
937 
938 		if (!rtnh_ok(rtnh, remaining))
939 			return -EINVAL;
940 
941 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
942 			return 1;
943 
944 		attrlen = rtnh_attrlen(rtnh);
945 		if (attrlen > 0) {
946 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
947 			int err;
948 
949 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
950 			nlav = nla_find(attrs, attrlen, RTA_VIA);
951 			if (nla && nlav) {
952 				NL_SET_ERR_MSG(extack,
953 					       "Nexthop configuration can not contain both GATEWAY and VIA");
954 				return -EINVAL;
955 			}
956 
957 			if (nla) {
958 				__be32 gw;
959 
960 				err = fib_gw_from_attr(&gw, nla, extack);
961 				if (err)
962 					return err;
963 
964 				if (nh->fib_nh_gw_family != AF_INET ||
965 				    gw != nh->fib_nh_gw4)
966 					return 1;
967 			} else if (nlav) {
968 				struct fib_config cfg2;
969 
970 				err = fib_gw_from_via(&cfg2, nlav, extack);
971 				if (err)
972 					return err;
973 
974 				switch (nh->fib_nh_gw_family) {
975 				case AF_INET:
976 					if (cfg2.fc_gw_family != AF_INET ||
977 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
978 						return 1;
979 					break;
980 				case AF_INET6:
981 					if (cfg2.fc_gw_family != AF_INET6 ||
982 					    ipv6_addr_cmp(&cfg2.fc_gw6,
983 							  &nh->fib_nh_gw6))
984 						return 1;
985 					break;
986 				}
987 			}
988 
989 #ifdef CONFIG_IP_ROUTE_CLASSID
990 			nla = nla_find(attrs, attrlen, RTA_FLOW);
991 			if (nla) {
992 				if (nla_len(nla) < sizeof(u32)) {
993 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
994 					return -EINVAL;
995 				}
996 				if (nla_get_u32(nla) != nh->nh_tclassid)
997 					return 1;
998 			}
999 #endif
1000 		}
1001 
1002 		rtnh = rtnh_next(rtnh, &remaining);
1003 	} endfor_nexthops(fi);
1004 #endif
1005 	return 0;
1006 }
1007 
1008 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1009 {
1010 	struct nlattr *nla;
1011 	int remaining;
1012 
1013 	if (!cfg->fc_mx)
1014 		return true;
1015 
1016 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1017 		int type = nla_type(nla);
1018 		u32 fi_val, val;
1019 
1020 		if (!type)
1021 			continue;
1022 		if (type > RTAX_MAX)
1023 			return false;
1024 
1025 		if (type == RTAX_CC_ALGO) {
1026 			char tmp[TCP_CA_NAME_MAX];
1027 			bool ecn_ca = false;
1028 
1029 			nla_strscpy(tmp, nla, sizeof(tmp));
1030 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1031 		} else {
1032 			if (nla_len(nla) != sizeof(u32))
1033 				return false;
1034 			val = nla_get_u32(nla);
1035 		}
1036 
1037 		fi_val = fi->fib_metrics->metrics[type - 1];
1038 		if (type == RTAX_FEATURES)
1039 			fi_val &= ~DST_FEATURE_ECN_CA;
1040 
1041 		if (fi_val != val)
1042 			return false;
1043 	}
1044 
1045 	return true;
1046 }
1047 
1048 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1049 			      u32 table, struct netlink_ext_ack *extack)
1050 {
1051 	struct fib6_config cfg = {
1052 		.fc_table = table,
1053 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1054 		.fc_ifindex = nh->fib_nh_oif,
1055 		.fc_gateway = nh->fib_nh_gw6,
1056 	};
1057 	struct fib6_nh fib6_nh = {};
1058 	int err;
1059 
1060 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1061 	if (!err) {
1062 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1063 		netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1064 			    GFP_KERNEL);
1065 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1066 		nh->fib_nh_scope = RT_SCOPE_LINK;
1067 
1068 		ipv6_stub->fib6_nh_release(&fib6_nh);
1069 	}
1070 
1071 	return err;
1072 }
1073 
1074 /*
1075  * Picture
1076  * -------
1077  *
1078  * Semantics of nexthop is very messy by historical reasons.
1079  * We have to take into account, that:
1080  * a) gateway can be actually local interface address,
1081  *    so that gatewayed route is direct.
1082  * b) gateway must be on-link address, possibly
1083  *    described not by an ifaddr, but also by a direct route.
1084  * c) If both gateway and interface are specified, they should not
1085  *    contradict.
1086  * d) If we use tunnel routes, gateway could be not on-link.
1087  *
1088  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1089  * results in pretty ugly and hairy code with obscure logic.
1090  *
1091  * I chose to generalized it instead, so that the size
1092  * of code does not increase practically, but it becomes
1093  * much more general.
1094  * Every prefix is assigned a "scope" value: "host" is local address,
1095  * "link" is direct route,
1096  * [ ... "site" ... "interior" ... ]
1097  * and "universe" is true gateway route with global meaning.
1098  *
1099  * Every prefix refers to a set of "nexthop"s (gw, oif),
1100  * where gw must have narrower scope. This recursion stops
1101  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1102  * which means that gw is forced to be on link.
1103  *
1104  * Code is still hairy, but now it is apparently logically
1105  * consistent and very flexible. F.e. as by-product it allows
1106  * to co-exists in peace independent exterior and interior
1107  * routing processes.
1108  *
1109  * Normally it looks as following.
1110  *
1111  * {universe prefix}  -> (gw, oif) [scope link]
1112  *		  |
1113  *		  |-> {link prefix} -> (gw, oif) [scope local]
1114  *					|
1115  *					|-> {local prefix} (terminal node)
1116  */
1117 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1118 			      u8 scope, struct netlink_ext_ack *extack)
1119 {
1120 	struct net_device *dev;
1121 	struct fib_result res;
1122 	int err = 0;
1123 
1124 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1125 		unsigned int addr_type;
1126 
1127 		if (scope >= RT_SCOPE_LINK) {
1128 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1129 			return -EINVAL;
1130 		}
1131 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1132 		if (!dev) {
1133 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1134 			return -ENODEV;
1135 		}
1136 		if (!(dev->flags & IFF_UP)) {
1137 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1138 			return -ENETDOWN;
1139 		}
1140 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1141 		if (addr_type != RTN_UNICAST) {
1142 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1143 			return -EINVAL;
1144 		}
1145 		if (!netif_carrier_ok(dev))
1146 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1147 		nh->fib_nh_dev = dev;
1148 		netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1149 		nh->fib_nh_scope = RT_SCOPE_LINK;
1150 		return 0;
1151 	}
1152 	rcu_read_lock();
1153 	{
1154 		struct fib_table *tbl = NULL;
1155 		struct flowi4 fl4 = {
1156 			.daddr = nh->fib_nh_gw4,
1157 			.flowi4_scope = scope + 1,
1158 			.flowi4_oif = nh->fib_nh_oif,
1159 			.flowi4_iif = LOOPBACK_IFINDEX,
1160 		};
1161 
1162 		/* It is not necessary, but requires a bit of thinking */
1163 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1164 			fl4.flowi4_scope = RT_SCOPE_LINK;
1165 
1166 		if (table && table != RT_TABLE_MAIN)
1167 			tbl = fib_get_table(net, table);
1168 
1169 		if (tbl)
1170 			err = fib_table_lookup(tbl, &fl4, &res,
1171 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1172 					       FIB_LOOKUP_NOREF);
1173 
1174 		/* on error or if no table given do full lookup. This
1175 		 * is needed for example when nexthops are in the local
1176 		 * table rather than the given table
1177 		 */
1178 		if (!tbl || err) {
1179 			err = fib_lookup(net, &fl4, &res,
1180 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1181 		}
1182 
1183 		if (err) {
1184 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1185 			goto out;
1186 		}
1187 	}
1188 
1189 	err = -EINVAL;
1190 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1191 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1192 		goto out;
1193 	}
1194 	nh->fib_nh_scope = res.scope;
1195 	nh->fib_nh_oif = FIB_RES_OIF(res);
1196 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1197 	if (!dev) {
1198 		NL_SET_ERR_MSG(extack,
1199 			       "No egress device for nexthop gateway");
1200 		goto out;
1201 	}
1202 	netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1203 	if (!netif_carrier_ok(dev))
1204 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1205 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1206 out:
1207 	rcu_read_unlock();
1208 	return err;
1209 }
1210 
1211 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1212 			      struct netlink_ext_ack *extack)
1213 {
1214 	struct in_device *in_dev;
1215 	int err;
1216 
1217 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1218 		NL_SET_ERR_MSG(extack,
1219 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1220 		return -EINVAL;
1221 	}
1222 
1223 	rcu_read_lock();
1224 
1225 	err = -ENODEV;
1226 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1227 	if (!in_dev)
1228 		goto out;
1229 	err = -ENETDOWN;
1230 	if (!(in_dev->dev->flags & IFF_UP)) {
1231 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1232 		goto out;
1233 	}
1234 
1235 	nh->fib_nh_dev = in_dev->dev;
1236 	netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1237 	nh->fib_nh_scope = RT_SCOPE_HOST;
1238 	if (!netif_carrier_ok(nh->fib_nh_dev))
1239 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1240 	err = 0;
1241 out:
1242 	rcu_read_unlock();
1243 	return err;
1244 }
1245 
1246 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1247 		 struct netlink_ext_ack *extack)
1248 {
1249 	int err;
1250 
1251 	if (nh->fib_nh_gw_family == AF_INET)
1252 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1253 	else if (nh->fib_nh_gw_family == AF_INET6)
1254 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1255 	else
1256 		err = fib_check_nh_nongw(net, nh, extack);
1257 
1258 	return err;
1259 }
1260 
1261 static struct hlist_head *
1262 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1263 {
1264 	u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1265 			   fib_info_hash_bits);
1266 
1267 	return &fib_info_laddrhash[slot];
1268 }
1269 
1270 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1271 			       struct hlist_head *new_laddrhash,
1272 			       unsigned int new_size)
1273 {
1274 	struct hlist_head *old_info_hash, *old_laddrhash;
1275 	unsigned int old_size = fib_info_hash_size;
1276 	unsigned int i;
1277 
1278 	spin_lock_bh(&fib_info_lock);
1279 	old_info_hash = fib_info_hash;
1280 	old_laddrhash = fib_info_laddrhash;
1281 	fib_info_hash_size = new_size;
1282 	fib_info_hash_bits = ilog2(new_size);
1283 
1284 	for (i = 0; i < old_size; i++) {
1285 		struct hlist_head *head = &fib_info_hash[i];
1286 		struct hlist_node *n;
1287 		struct fib_info *fi;
1288 
1289 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1290 			struct hlist_head *dest;
1291 			unsigned int new_hash;
1292 
1293 			new_hash = fib_info_hashfn(fi);
1294 			dest = &new_info_hash[new_hash];
1295 			hlist_add_head(&fi->fib_hash, dest);
1296 		}
1297 	}
1298 	fib_info_hash = new_info_hash;
1299 
1300 	fib_info_laddrhash = new_laddrhash;
1301 	for (i = 0; i < old_size; i++) {
1302 		struct hlist_head *lhead = &old_laddrhash[i];
1303 		struct hlist_node *n;
1304 		struct fib_info *fi;
1305 
1306 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1307 			struct hlist_head *ldest;
1308 
1309 			ldest = fib_info_laddrhash_bucket(fi->fib_net,
1310 							  fi->fib_prefsrc);
1311 			hlist_add_head(&fi->fib_lhash, ldest);
1312 		}
1313 	}
1314 
1315 	spin_unlock_bh(&fib_info_lock);
1316 
1317 	kvfree(old_info_hash);
1318 	kvfree(old_laddrhash);
1319 }
1320 
1321 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1322 				 unsigned char scope)
1323 {
1324 	struct fib_nh *nh;
1325 
1326 	if (nhc->nhc_family != AF_INET)
1327 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1328 
1329 	nh = container_of(nhc, struct fib_nh, nh_common);
1330 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1331 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1332 
1333 	return nh->nh_saddr;
1334 }
1335 
1336 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1337 {
1338 	struct fib_nh_common *nhc = res->nhc;
1339 
1340 	if (res->fi->fib_prefsrc)
1341 		return res->fi->fib_prefsrc;
1342 
1343 	if (nhc->nhc_family == AF_INET) {
1344 		struct fib_nh *nh;
1345 
1346 		nh = container_of(nhc, struct fib_nh, nh_common);
1347 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1348 			return nh->nh_saddr;
1349 	}
1350 
1351 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1352 }
1353 
1354 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1355 {
1356 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1357 	    fib_prefsrc != cfg->fc_dst) {
1358 		u32 tb_id = cfg->fc_table;
1359 		int rc;
1360 
1361 		if (tb_id == RT_TABLE_MAIN)
1362 			tb_id = RT_TABLE_LOCAL;
1363 
1364 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1365 					  fib_prefsrc, tb_id);
1366 
1367 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1368 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1369 						  fib_prefsrc, RT_TABLE_LOCAL);
1370 		}
1371 
1372 		if (rc != RTN_LOCAL)
1373 			return false;
1374 	}
1375 	return true;
1376 }
1377 
1378 struct fib_info *fib_create_info(struct fib_config *cfg,
1379 				 struct netlink_ext_ack *extack)
1380 {
1381 	int err;
1382 	struct fib_info *fi = NULL;
1383 	struct nexthop *nh = NULL;
1384 	struct fib_info *ofi;
1385 	int nhs = 1;
1386 	struct net *net = cfg->fc_nlinfo.nl_net;
1387 
1388 	if (cfg->fc_type > RTN_MAX)
1389 		goto err_inval;
1390 
1391 	/* Fast check to catch the most weird cases */
1392 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1393 		NL_SET_ERR_MSG(extack, "Invalid scope");
1394 		goto err_inval;
1395 	}
1396 
1397 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1398 		NL_SET_ERR_MSG(extack,
1399 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1400 		goto err_inval;
1401 	}
1402 
1403 	if (cfg->fc_nh_id) {
1404 		if (!cfg->fc_mx) {
1405 			fi = fib_find_info_nh(net, cfg);
1406 			if (fi) {
1407 				refcount_inc(&fi->fib_treeref);
1408 				return fi;
1409 			}
1410 		}
1411 
1412 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1413 		if (!nh) {
1414 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1415 			goto err_inval;
1416 		}
1417 		nhs = 0;
1418 	}
1419 
1420 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1421 	if (cfg->fc_mp) {
1422 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1423 		if (nhs == 0)
1424 			goto err_inval;
1425 	}
1426 #endif
1427 
1428 	err = -ENOBUFS;
1429 
1430 	/* Paired with WRITE_ONCE() in fib_release_info() */
1431 	if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1432 		unsigned int new_size = fib_info_hash_size << 1;
1433 		struct hlist_head *new_info_hash;
1434 		struct hlist_head *new_laddrhash;
1435 		size_t bytes;
1436 
1437 		if (!new_size)
1438 			new_size = 16;
1439 		bytes = (size_t)new_size * sizeof(struct hlist_head *);
1440 		new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1441 		new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1442 		if (!new_info_hash || !new_laddrhash) {
1443 			kvfree(new_info_hash);
1444 			kvfree(new_laddrhash);
1445 		} else {
1446 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1447 		}
1448 		if (!fib_info_hash_size)
1449 			goto failure;
1450 	}
1451 
1452 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1453 	if (!fi)
1454 		goto failure;
1455 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1456 					      cfg->fc_mx_len, extack);
1457 	if (IS_ERR(fi->fib_metrics)) {
1458 		err = PTR_ERR(fi->fib_metrics);
1459 		kfree(fi);
1460 		return ERR_PTR(err);
1461 	}
1462 
1463 	fi->fib_net = net;
1464 	fi->fib_protocol = cfg->fc_protocol;
1465 	fi->fib_scope = cfg->fc_scope;
1466 	fi->fib_flags = cfg->fc_flags;
1467 	fi->fib_priority = cfg->fc_priority;
1468 	fi->fib_prefsrc = cfg->fc_prefsrc;
1469 	fi->fib_type = cfg->fc_type;
1470 	fi->fib_tb_id = cfg->fc_table;
1471 
1472 	fi->fib_nhs = nhs;
1473 	if (nh) {
1474 		if (!nexthop_get(nh)) {
1475 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1476 			err = -EINVAL;
1477 		} else {
1478 			err = 0;
1479 			fi->nh = nh;
1480 		}
1481 	} else {
1482 		change_nexthops(fi) {
1483 			nexthop_nh->nh_parent = fi;
1484 		} endfor_nexthops(fi)
1485 
1486 		if (cfg->fc_mp)
1487 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1488 					  extack);
1489 		else
1490 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1491 	}
1492 
1493 	if (err != 0)
1494 		goto failure;
1495 
1496 	if (fib_props[cfg->fc_type].error) {
1497 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1498 			NL_SET_ERR_MSG(extack,
1499 				       "Gateway, device and multipath can not be specified for this route type");
1500 			goto err_inval;
1501 		}
1502 		goto link_it;
1503 	} else {
1504 		switch (cfg->fc_type) {
1505 		case RTN_UNICAST:
1506 		case RTN_LOCAL:
1507 		case RTN_BROADCAST:
1508 		case RTN_ANYCAST:
1509 		case RTN_MULTICAST:
1510 			break;
1511 		default:
1512 			NL_SET_ERR_MSG(extack, "Invalid route type");
1513 			goto err_inval;
1514 		}
1515 	}
1516 
1517 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1518 		NL_SET_ERR_MSG(extack, "Invalid scope");
1519 		goto err_inval;
1520 	}
1521 
1522 	if (fi->nh) {
1523 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1524 		if (err)
1525 			goto failure;
1526 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1527 		struct fib_nh *nh = fi->fib_nh;
1528 
1529 		/* Local address is added. */
1530 		if (nhs != 1) {
1531 			NL_SET_ERR_MSG(extack,
1532 				       "Route with host scope can not have multiple nexthops");
1533 			goto err_inval;
1534 		}
1535 		if (nh->fib_nh_gw_family) {
1536 			NL_SET_ERR_MSG(extack,
1537 				       "Route with host scope can not have a gateway");
1538 			goto err_inval;
1539 		}
1540 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1541 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1542 		err = -ENODEV;
1543 		if (!nh->fib_nh_dev)
1544 			goto failure;
1545 		netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1546 				     GFP_KERNEL);
1547 	} else {
1548 		int linkdown = 0;
1549 
1550 		change_nexthops(fi) {
1551 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1552 					   cfg->fc_table, cfg->fc_scope,
1553 					   extack);
1554 			if (err != 0)
1555 				goto failure;
1556 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1557 				linkdown++;
1558 		} endfor_nexthops(fi)
1559 		if (linkdown == fi->fib_nhs)
1560 			fi->fib_flags |= RTNH_F_LINKDOWN;
1561 	}
1562 
1563 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1564 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1565 		goto err_inval;
1566 	}
1567 
1568 	if (!fi->nh) {
1569 		change_nexthops(fi) {
1570 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1571 						  fi->fib_scope);
1572 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1573 				fi->fib_nh_is_v6 = true;
1574 		} endfor_nexthops(fi)
1575 
1576 		fib_rebalance(fi);
1577 	}
1578 
1579 link_it:
1580 	ofi = fib_find_info(fi);
1581 	if (ofi) {
1582 		fi->fib_dead = 1;
1583 		free_fib_info(fi);
1584 		refcount_inc(&ofi->fib_treeref);
1585 		return ofi;
1586 	}
1587 
1588 	refcount_set(&fi->fib_treeref, 1);
1589 	refcount_set(&fi->fib_clntref, 1);
1590 	spin_lock_bh(&fib_info_lock);
1591 	fib_info_cnt++;
1592 	hlist_add_head(&fi->fib_hash,
1593 		       &fib_info_hash[fib_info_hashfn(fi)]);
1594 	if (fi->fib_prefsrc) {
1595 		struct hlist_head *head;
1596 
1597 		head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1598 		hlist_add_head(&fi->fib_lhash, head);
1599 	}
1600 	if (fi->nh) {
1601 		list_add(&fi->nh_list, &nh->fi_list);
1602 	} else {
1603 		change_nexthops(fi) {
1604 			struct hlist_head *head;
1605 
1606 			if (!nexthop_nh->fib_nh_dev)
1607 				continue;
1608 			head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1609 			hlist_add_head(&nexthop_nh->nh_hash, head);
1610 		} endfor_nexthops(fi)
1611 	}
1612 	spin_unlock_bh(&fib_info_lock);
1613 	return fi;
1614 
1615 err_inval:
1616 	err = -EINVAL;
1617 
1618 failure:
1619 	if (fi) {
1620 		fi->fib_dead = 1;
1621 		free_fib_info(fi);
1622 	}
1623 
1624 	return ERR_PTR(err);
1625 }
1626 
1627 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1628 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1629 {
1630 	if (nhc->nhc_flags & RTNH_F_DEAD)
1631 		*flags |= RTNH_F_DEAD;
1632 
1633 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1634 		*flags |= RTNH_F_LINKDOWN;
1635 
1636 		rcu_read_lock();
1637 		switch (nhc->nhc_family) {
1638 		case AF_INET:
1639 			if (ip_ignore_linkdown(nhc->nhc_dev))
1640 				*flags |= RTNH_F_DEAD;
1641 			break;
1642 		case AF_INET6:
1643 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1644 				*flags |= RTNH_F_DEAD;
1645 			break;
1646 		}
1647 		rcu_read_unlock();
1648 	}
1649 
1650 	switch (nhc->nhc_gw_family) {
1651 	case AF_INET:
1652 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1653 			goto nla_put_failure;
1654 		break;
1655 	case AF_INET6:
1656 		/* if gateway family does not match nexthop family
1657 		 * gateway is encoded as RTA_VIA
1658 		 */
1659 		if (rt_family != nhc->nhc_gw_family) {
1660 			int alen = sizeof(struct in6_addr);
1661 			struct nlattr *nla;
1662 			struct rtvia *via;
1663 
1664 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1665 			if (!nla)
1666 				goto nla_put_failure;
1667 
1668 			via = nla_data(nla);
1669 			via->rtvia_family = AF_INET6;
1670 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1671 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1672 					    &nhc->nhc_gw.ipv6) < 0) {
1673 			goto nla_put_failure;
1674 		}
1675 		break;
1676 	}
1677 
1678 	*flags |= (nhc->nhc_flags &
1679 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1680 
1681 	if (!skip_oif && nhc->nhc_dev &&
1682 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1683 		goto nla_put_failure;
1684 
1685 	if (nhc->nhc_lwtstate &&
1686 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1687 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1688 		goto nla_put_failure;
1689 
1690 	return 0;
1691 
1692 nla_put_failure:
1693 	return -EMSGSIZE;
1694 }
1695 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1696 
1697 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1698 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1699 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1700 {
1701 	const struct net_device *dev = nhc->nhc_dev;
1702 	struct rtnexthop *rtnh;
1703 	unsigned char flags = 0;
1704 
1705 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1706 	if (!rtnh)
1707 		goto nla_put_failure;
1708 
1709 	rtnh->rtnh_hops = nh_weight - 1;
1710 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1711 
1712 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1713 		goto nla_put_failure;
1714 
1715 	rtnh->rtnh_flags = flags;
1716 
1717 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1718 		goto nla_put_failure;
1719 
1720 	/* length of rtnetlink header + attributes */
1721 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1722 
1723 	return 0;
1724 
1725 nla_put_failure:
1726 	return -EMSGSIZE;
1727 }
1728 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1729 #endif
1730 
1731 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1732 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1733 {
1734 	struct nlattr *mp;
1735 
1736 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1737 	if (!mp)
1738 		goto nla_put_failure;
1739 
1740 	if (unlikely(fi->nh)) {
1741 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1742 			goto nla_put_failure;
1743 		goto mp_end;
1744 	}
1745 
1746 	for_nexthops(fi) {
1747 		u32 nh_tclassid = 0;
1748 #ifdef CONFIG_IP_ROUTE_CLASSID
1749 		nh_tclassid = nh->nh_tclassid;
1750 #endif
1751 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1752 				    AF_INET, nh_tclassid) < 0)
1753 			goto nla_put_failure;
1754 	} endfor_nexthops(fi);
1755 
1756 mp_end:
1757 	nla_nest_end(skb, mp);
1758 
1759 	return 0;
1760 
1761 nla_put_failure:
1762 	return -EMSGSIZE;
1763 }
1764 #else
1765 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1766 {
1767 	return 0;
1768 }
1769 #endif
1770 
1771 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1772 		  const struct fib_rt_info *fri, unsigned int flags)
1773 {
1774 	unsigned int nhs = fib_info_num_path(fri->fi);
1775 	struct fib_info *fi = fri->fi;
1776 	u32 tb_id = fri->tb_id;
1777 	struct nlmsghdr *nlh;
1778 	struct rtmsg *rtm;
1779 
1780 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1781 	if (!nlh)
1782 		return -EMSGSIZE;
1783 
1784 	rtm = nlmsg_data(nlh);
1785 	rtm->rtm_family = AF_INET;
1786 	rtm->rtm_dst_len = fri->dst_len;
1787 	rtm->rtm_src_len = 0;
1788 	rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1789 	if (tb_id < 256)
1790 		rtm->rtm_table = tb_id;
1791 	else
1792 		rtm->rtm_table = RT_TABLE_COMPAT;
1793 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1794 		goto nla_put_failure;
1795 	rtm->rtm_type = fri->type;
1796 	rtm->rtm_flags = fi->fib_flags;
1797 	rtm->rtm_scope = fi->fib_scope;
1798 	rtm->rtm_protocol = fi->fib_protocol;
1799 
1800 	if (rtm->rtm_dst_len &&
1801 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1802 		goto nla_put_failure;
1803 	if (fi->fib_priority &&
1804 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1805 		goto nla_put_failure;
1806 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1807 		goto nla_put_failure;
1808 
1809 	if (fi->fib_prefsrc &&
1810 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1811 		goto nla_put_failure;
1812 
1813 	if (fi->nh) {
1814 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1815 			goto nla_put_failure;
1816 		if (nexthop_is_blackhole(fi->nh))
1817 			rtm->rtm_type = RTN_BLACKHOLE;
1818 		if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1819 			goto offload;
1820 	}
1821 
1822 	if (nhs == 1) {
1823 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1824 		unsigned char flags = 0;
1825 
1826 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1827 			goto nla_put_failure;
1828 
1829 		rtm->rtm_flags = flags;
1830 #ifdef CONFIG_IP_ROUTE_CLASSID
1831 		if (nhc->nhc_family == AF_INET) {
1832 			struct fib_nh *nh;
1833 
1834 			nh = container_of(nhc, struct fib_nh, nh_common);
1835 			if (nh->nh_tclassid &&
1836 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1837 				goto nla_put_failure;
1838 		}
1839 #endif
1840 	} else {
1841 		if (fib_add_multipath(skb, fi) < 0)
1842 			goto nla_put_failure;
1843 	}
1844 
1845 offload:
1846 	if (fri->offload)
1847 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1848 	if (fri->trap)
1849 		rtm->rtm_flags |= RTM_F_TRAP;
1850 	if (fri->offload_failed)
1851 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1852 
1853 	nlmsg_end(skb, nlh);
1854 	return 0;
1855 
1856 nla_put_failure:
1857 	nlmsg_cancel(skb, nlh);
1858 	return -EMSGSIZE;
1859 }
1860 
1861 /*
1862  * Update FIB if:
1863  * - local address disappeared -> we must delete all the entries
1864  *   referring to it.
1865  * - device went down -> we must shutdown all nexthops going via it.
1866  */
1867 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1868 {
1869 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1870 	struct net *net = dev_net(dev);
1871 	struct hlist_head *head;
1872 	struct fib_info *fi;
1873 	int ret = 0;
1874 
1875 	if (!fib_info_laddrhash || local == 0)
1876 		return 0;
1877 
1878 	head = fib_info_laddrhash_bucket(net, local);
1879 	hlist_for_each_entry(fi, head, fib_lhash) {
1880 		if (!net_eq(fi->fib_net, net) ||
1881 		    fi->fib_tb_id != tb_id)
1882 			continue;
1883 		if (fi->fib_prefsrc == local) {
1884 			fi->fib_flags |= RTNH_F_DEAD;
1885 			ret++;
1886 		}
1887 	}
1888 	return ret;
1889 }
1890 
1891 static int call_fib_nh_notifiers(struct fib_nh *nh,
1892 				 enum fib_event_type event_type)
1893 {
1894 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1895 	struct fib_nh_notifier_info info = {
1896 		.fib_nh = nh,
1897 	};
1898 
1899 	switch (event_type) {
1900 	case FIB_EVENT_NH_ADD:
1901 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1902 			break;
1903 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1904 			break;
1905 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1906 					   &info.info);
1907 	case FIB_EVENT_NH_DEL:
1908 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1909 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1910 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1911 						   event_type, &info.info);
1912 		break;
1913 	default:
1914 		break;
1915 	}
1916 
1917 	return NOTIFY_DONE;
1918 }
1919 
1920 /* Update the PMTU of exceptions when:
1921  * - the new MTU of the first hop becomes smaller than the PMTU
1922  * - the old MTU was the same as the PMTU, and it limited discovery of
1923  *   larger MTUs on the path. With that limit raised, we can now
1924  *   discover larger MTUs
1925  * A special case is locked exceptions, for which the PMTU is smaller
1926  * than the minimal accepted PMTU:
1927  * - if the new MTU is greater than the PMTU, don't make any change
1928  * - otherwise, unlock and set PMTU
1929  */
1930 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1931 {
1932 	struct fnhe_hash_bucket *bucket;
1933 	int i;
1934 
1935 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1936 	if (!bucket)
1937 		return;
1938 
1939 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1940 		struct fib_nh_exception *fnhe;
1941 
1942 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1943 		     fnhe;
1944 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1945 			if (fnhe->fnhe_mtu_locked) {
1946 				if (new <= fnhe->fnhe_pmtu) {
1947 					fnhe->fnhe_pmtu = new;
1948 					fnhe->fnhe_mtu_locked = false;
1949 				}
1950 			} else if (new < fnhe->fnhe_pmtu ||
1951 				   orig == fnhe->fnhe_pmtu) {
1952 				fnhe->fnhe_pmtu = new;
1953 			}
1954 		}
1955 	}
1956 }
1957 
1958 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1959 {
1960 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1961 	struct fib_nh *nh;
1962 
1963 	hlist_for_each_entry(nh, head, nh_hash) {
1964 		if (nh->fib_nh_dev == dev)
1965 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1966 	}
1967 }
1968 
1969 /* Event              force Flags           Description
1970  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1971  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1972  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1973  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1974  *
1975  * only used when fib_nh is built into fib_info
1976  */
1977 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1978 {
1979 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1980 	struct fib_info *prev_fi = NULL;
1981 	int scope = RT_SCOPE_NOWHERE;
1982 	struct fib_nh *nh;
1983 	int ret = 0;
1984 
1985 	if (force)
1986 		scope = -1;
1987 
1988 	hlist_for_each_entry(nh, head, nh_hash) {
1989 		struct fib_info *fi = nh->nh_parent;
1990 		int dead;
1991 
1992 		BUG_ON(!fi->fib_nhs);
1993 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1994 			continue;
1995 		prev_fi = fi;
1996 		dead = 0;
1997 		change_nexthops(fi) {
1998 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1999 				dead++;
2000 			else if (nexthop_nh->fib_nh_dev == dev &&
2001 				 nexthop_nh->fib_nh_scope != scope) {
2002 				switch (event) {
2003 				case NETDEV_DOWN:
2004 				case NETDEV_UNREGISTER:
2005 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2006 					fallthrough;
2007 				case NETDEV_CHANGE:
2008 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2009 					break;
2010 				}
2011 				call_fib_nh_notifiers(nexthop_nh,
2012 						      FIB_EVENT_NH_DEL);
2013 				dead++;
2014 			}
2015 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2016 			if (event == NETDEV_UNREGISTER &&
2017 			    nexthop_nh->fib_nh_dev == dev) {
2018 				dead = fi->fib_nhs;
2019 				break;
2020 			}
2021 #endif
2022 		} endfor_nexthops(fi)
2023 		if (dead == fi->fib_nhs) {
2024 			switch (event) {
2025 			case NETDEV_DOWN:
2026 			case NETDEV_UNREGISTER:
2027 				fi->fib_flags |= RTNH_F_DEAD;
2028 				fallthrough;
2029 			case NETDEV_CHANGE:
2030 				fi->fib_flags |= RTNH_F_LINKDOWN;
2031 				break;
2032 			}
2033 			ret++;
2034 		}
2035 
2036 		fib_rebalance(fi);
2037 	}
2038 
2039 	return ret;
2040 }
2041 
2042 /* Must be invoked inside of an RCU protected region.  */
2043 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2044 {
2045 	struct fib_info *fi = NULL, *last_resort = NULL;
2046 	struct hlist_head *fa_head = res->fa_head;
2047 	struct fib_table *tb = res->table;
2048 	u8 slen = 32 - res->prefixlen;
2049 	int order = -1, last_idx = -1;
2050 	struct fib_alias *fa, *fa1 = NULL;
2051 	u32 last_prio = res->fi->fib_priority;
2052 	dscp_t last_dscp = 0;
2053 
2054 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2055 		struct fib_info *next_fi = fa->fa_info;
2056 		struct fib_nh_common *nhc;
2057 
2058 		if (fa->fa_slen != slen)
2059 			continue;
2060 		if (fa->fa_dscp &&
2061 		    fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2062 			continue;
2063 		if (fa->tb_id != tb->tb_id)
2064 			continue;
2065 		if (next_fi->fib_priority > last_prio &&
2066 		    fa->fa_dscp == last_dscp) {
2067 			if (last_dscp)
2068 				continue;
2069 			break;
2070 		}
2071 		if (next_fi->fib_flags & RTNH_F_DEAD)
2072 			continue;
2073 		last_dscp = fa->fa_dscp;
2074 		last_prio = next_fi->fib_priority;
2075 
2076 		if (next_fi->fib_scope != res->scope ||
2077 		    fa->fa_type != RTN_UNICAST)
2078 			continue;
2079 
2080 		nhc = fib_info_nhc(next_fi, 0);
2081 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2082 			continue;
2083 
2084 		fib_alias_accessed(fa);
2085 
2086 		if (!fi) {
2087 			if (next_fi != res->fi)
2088 				break;
2089 			fa1 = fa;
2090 		} else if (!fib_detect_death(fi, order, &last_resort,
2091 					     &last_idx, fa1->fa_default)) {
2092 			fib_result_assign(res, fi);
2093 			fa1->fa_default = order;
2094 			goto out;
2095 		}
2096 		fi = next_fi;
2097 		order++;
2098 	}
2099 
2100 	if (order <= 0 || !fi) {
2101 		if (fa1)
2102 			fa1->fa_default = -1;
2103 		goto out;
2104 	}
2105 
2106 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2107 			      fa1->fa_default)) {
2108 		fib_result_assign(res, fi);
2109 		fa1->fa_default = order;
2110 		goto out;
2111 	}
2112 
2113 	if (last_idx >= 0)
2114 		fib_result_assign(res, last_resort);
2115 	fa1->fa_default = last_idx;
2116 out:
2117 	return;
2118 }
2119 
2120 /*
2121  * Dead device goes up. We wake up dead nexthops.
2122  * It takes sense only on multipath routes.
2123  *
2124  * only used when fib_nh is built into fib_info
2125  */
2126 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2127 {
2128 	struct fib_info *prev_fi;
2129 	struct hlist_head *head;
2130 	struct fib_nh *nh;
2131 	int ret;
2132 
2133 	if (!(dev->flags & IFF_UP))
2134 		return 0;
2135 
2136 	if (nh_flags & RTNH_F_DEAD) {
2137 		unsigned int flags = dev_get_flags(dev);
2138 
2139 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2140 			nh_flags |= RTNH_F_LINKDOWN;
2141 	}
2142 
2143 	prev_fi = NULL;
2144 	head = fib_info_devhash_bucket(dev);
2145 	ret = 0;
2146 
2147 	hlist_for_each_entry(nh, head, nh_hash) {
2148 		struct fib_info *fi = nh->nh_parent;
2149 		int alive;
2150 
2151 		BUG_ON(!fi->fib_nhs);
2152 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2153 			continue;
2154 
2155 		prev_fi = fi;
2156 		alive = 0;
2157 		change_nexthops(fi) {
2158 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2159 				alive++;
2160 				continue;
2161 			}
2162 			if (!nexthop_nh->fib_nh_dev ||
2163 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2164 				continue;
2165 			if (nexthop_nh->fib_nh_dev != dev ||
2166 			    !__in_dev_get_rtnl(dev))
2167 				continue;
2168 			alive++;
2169 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2170 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2171 		} endfor_nexthops(fi)
2172 
2173 		if (alive > 0) {
2174 			fi->fib_flags &= ~nh_flags;
2175 			ret++;
2176 		}
2177 
2178 		fib_rebalance(fi);
2179 	}
2180 
2181 	return ret;
2182 }
2183 
2184 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2185 static bool fib_good_nh(const struct fib_nh *nh)
2186 {
2187 	int state = NUD_REACHABLE;
2188 
2189 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2190 		struct neighbour *n;
2191 
2192 		rcu_read_lock_bh();
2193 
2194 		if (likely(nh->fib_nh_gw_family == AF_INET))
2195 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2196 						   (__force u32)nh->fib_nh_gw4);
2197 		else if (nh->fib_nh_gw_family == AF_INET6)
2198 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2199 							   &nh->fib_nh_gw6);
2200 		else
2201 			n = NULL;
2202 		if (n)
2203 			state = n->nud_state;
2204 
2205 		rcu_read_unlock_bh();
2206 	}
2207 
2208 	return !!(state & NUD_VALID);
2209 }
2210 
2211 void fib_select_multipath(struct fib_result *res, int hash)
2212 {
2213 	struct fib_info *fi = res->fi;
2214 	struct net *net = fi->fib_net;
2215 	bool first = false;
2216 
2217 	if (unlikely(res->fi->nh)) {
2218 		nexthop_path_fib_result(res, hash);
2219 		return;
2220 	}
2221 
2222 	change_nexthops(fi) {
2223 		if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2224 			if (!fib_good_nh(nexthop_nh))
2225 				continue;
2226 			if (!first) {
2227 				res->nh_sel = nhsel;
2228 				res->nhc = &nexthop_nh->nh_common;
2229 				first = true;
2230 			}
2231 		}
2232 
2233 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2234 			continue;
2235 
2236 		res->nh_sel = nhsel;
2237 		res->nhc = &nexthop_nh->nh_common;
2238 		return;
2239 	} endfor_nexthops(fi);
2240 }
2241 #endif
2242 
2243 void fib_select_path(struct net *net, struct fib_result *res,
2244 		     struct flowi4 *fl4, const struct sk_buff *skb)
2245 {
2246 	if (fl4->flowi4_oif)
2247 		goto check_saddr;
2248 
2249 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2250 	if (fib_info_num_path(res->fi) > 1) {
2251 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2252 
2253 		fib_select_multipath(res, h);
2254 	}
2255 	else
2256 #endif
2257 	if (!res->prefixlen &&
2258 	    res->table->tb_num_default > 1 &&
2259 	    res->type == RTN_UNICAST)
2260 		fib_select_default(fl4, res);
2261 
2262 check_saddr:
2263 	if (!fl4->saddr)
2264 		fl4->saddr = fib_result_prefsrc(net, res);
2265 }
2266