xref: /linux/net/ipv4/fib_semantics.c (revision 3ed8d344e061f382069c27705543c1882aca468a)
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(const struct net *net,
351 					   unsigned int val)
352 {
353 	return hash_32(val ^ net_hash_mix(net), fib_info_hash_bits);
354 }
355 
356 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
357 {
358 	unsigned int val;
359 
360 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
361 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
362 				fi->fib_priority);
363 
364 	if (fi->nh) {
365 		val ^= fib_devindex_hashfn(fi->nh->id);
366 	} else {
367 		for_nexthops(fi) {
368 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
369 		} endfor_nexthops(fi)
370 	}
371 
372 	return fib_info_hashfn_result(fi->fib_net, val);
373 }
374 
375 /* no metrics, only nexthop id */
376 static struct fib_info *fib_find_info_nh(struct net *net,
377 					 const struct fib_config *cfg)
378 {
379 	struct hlist_head *head;
380 	struct fib_info *fi;
381 	unsigned int hash;
382 
383 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
384 				 cfg->fc_protocol, cfg->fc_scope,
385 				 (__force u32)cfg->fc_prefsrc,
386 				 cfg->fc_priority);
387 	hash = fib_info_hashfn_result(net, hash);
388 	head = &fib_info_hash[hash];
389 
390 	hlist_for_each_entry(fi, head, fib_hash) {
391 		if (!net_eq(fi->fib_net, net))
392 			continue;
393 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
394 			continue;
395 		if (cfg->fc_protocol == fi->fib_protocol &&
396 		    cfg->fc_scope == fi->fib_scope &&
397 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
398 		    cfg->fc_priority == fi->fib_priority &&
399 		    cfg->fc_type == fi->fib_type &&
400 		    cfg->fc_table == fi->fib_tb_id &&
401 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
402 			return fi;
403 	}
404 
405 	return NULL;
406 }
407 
408 static struct fib_info *fib_find_info(struct fib_info *nfi)
409 {
410 	struct hlist_head *head;
411 	struct fib_info *fi;
412 	unsigned int hash;
413 
414 	hash = fib_info_hashfn(nfi);
415 	head = &fib_info_hash[hash];
416 
417 	hlist_for_each_entry(fi, head, fib_hash) {
418 		if (!net_eq(fi->fib_net, nfi->fib_net))
419 			continue;
420 		if (fi->fib_nhs != nfi->fib_nhs)
421 			continue;
422 		if (nfi->fib_protocol == fi->fib_protocol &&
423 		    nfi->fib_scope == fi->fib_scope &&
424 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
425 		    nfi->fib_priority == fi->fib_priority &&
426 		    nfi->fib_type == fi->fib_type &&
427 		    nfi->fib_tb_id == fi->fib_tb_id &&
428 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
429 			   sizeof(u32) * RTAX_MAX) == 0 &&
430 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
431 		    nh_comp(fi, nfi) == 0)
432 			return fi;
433 	}
434 
435 	return NULL;
436 }
437 
438 /* Check, that the gateway is already configured.
439  * Used only by redirect accept routine.
440  */
441 int ip_fib_check_default(__be32 gw, struct net_device *dev)
442 {
443 	struct hlist_head *head;
444 	struct fib_nh *nh;
445 
446 	spin_lock(&fib_info_lock);
447 
448 	head = fib_info_devhash_bucket(dev);
449 
450 	hlist_for_each_entry(nh, head, nh_hash) {
451 		if (nh->fib_nh_dev == dev &&
452 		    nh->fib_nh_gw4 == gw &&
453 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
454 			spin_unlock(&fib_info_lock);
455 			return 0;
456 		}
457 	}
458 
459 	spin_unlock(&fib_info_lock);
460 
461 	return -1;
462 }
463 
464 size_t fib_nlmsg_size(struct fib_info *fi)
465 {
466 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
467 			 + nla_total_size(4) /* RTA_TABLE */
468 			 + nla_total_size(4) /* RTA_DST */
469 			 + nla_total_size(4) /* RTA_PRIORITY */
470 			 + nla_total_size(4) /* RTA_PREFSRC */
471 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
472 	unsigned int nhs = fib_info_num_path(fi);
473 
474 	/* space for nested metrics */
475 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
476 
477 	if (fi->nh)
478 		payload += nla_total_size(4); /* RTA_NH_ID */
479 
480 	if (nhs) {
481 		size_t nh_encapsize = 0;
482 		/* Also handles the special case nhs == 1 */
483 
484 		/* each nexthop is packed in an attribute */
485 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
486 		unsigned int i;
487 
488 		/* may contain flow and gateway attribute */
489 		nhsize += 2 * nla_total_size(4);
490 
491 		/* grab encap info */
492 		for (i = 0; i < fib_info_num_path(fi); i++) {
493 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
494 
495 			if (nhc->nhc_lwtstate) {
496 				/* RTA_ENCAP_TYPE */
497 				nh_encapsize += lwtunnel_get_encap_size(
498 						nhc->nhc_lwtstate);
499 				/* RTA_ENCAP */
500 				nh_encapsize +=  nla_total_size(2);
501 			}
502 		}
503 
504 		/* all nexthops are packed in a nested attribute */
505 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
506 
507 	}
508 
509 	return payload;
510 }
511 
512 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
513 	       int dst_len, u32 tb_id, const struct nl_info *info,
514 	       unsigned int nlm_flags)
515 {
516 	struct fib_rt_info fri;
517 	struct sk_buff *skb;
518 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
519 	int err = -ENOBUFS;
520 
521 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
522 	if (!skb)
523 		goto errout;
524 
525 	fri.fi = fa->fa_info;
526 	fri.tb_id = tb_id;
527 	fri.dst = key;
528 	fri.dst_len = dst_len;
529 	fri.dscp = fa->fa_dscp;
530 	fri.type = fa->fa_type;
531 	fri.offload = READ_ONCE(fa->offload);
532 	fri.trap = READ_ONCE(fa->trap);
533 	fri.offload_failed = READ_ONCE(fa->offload_failed);
534 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
535 	if (err < 0) {
536 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
537 		WARN_ON(err == -EMSGSIZE);
538 		kfree_skb(skb);
539 		goto errout;
540 	}
541 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
542 		    info->nlh, GFP_KERNEL);
543 	return;
544 errout:
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 = READ_ONCE(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 		type = array_index_nospec(type, RTAX_MAX + 1);
1026 		if (type == RTAX_CC_ALGO) {
1027 			char tmp[TCP_CA_NAME_MAX];
1028 			bool ecn_ca = false;
1029 
1030 			nla_strscpy(tmp, nla, sizeof(tmp));
1031 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
1032 		} else {
1033 			if (nla_len(nla) != sizeof(u32))
1034 				return false;
1035 			val = nla_get_u32(nla);
1036 		}
1037 
1038 		fi_val = fi->fib_metrics->metrics[type - 1];
1039 		if (type == RTAX_FEATURES)
1040 			fi_val &= ~DST_FEATURE_ECN_CA;
1041 
1042 		if (fi_val != val)
1043 			return false;
1044 	}
1045 
1046 	return true;
1047 }
1048 
1049 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1050 			      u32 table, struct netlink_ext_ack *extack)
1051 {
1052 	struct fib6_config cfg = {
1053 		.fc_table = table,
1054 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1055 		.fc_ifindex = nh->fib_nh_oif,
1056 		.fc_gateway = nh->fib_nh_gw6,
1057 	};
1058 	struct fib6_nh fib6_nh = {};
1059 	int err;
1060 
1061 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1062 	if (!err) {
1063 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1064 		netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1065 			    GFP_KERNEL);
1066 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1067 		nh->fib_nh_scope = RT_SCOPE_LINK;
1068 
1069 		ipv6_stub->fib6_nh_release(&fib6_nh);
1070 	}
1071 
1072 	return err;
1073 }
1074 
1075 /*
1076  * Picture
1077  * -------
1078  *
1079  * Semantics of nexthop is very messy by historical reasons.
1080  * We have to take into account, that:
1081  * a) gateway can be actually local interface address,
1082  *    so that gatewayed route is direct.
1083  * b) gateway must be on-link address, possibly
1084  *    described not by an ifaddr, but also by a direct route.
1085  * c) If both gateway and interface are specified, they should not
1086  *    contradict.
1087  * d) If we use tunnel routes, gateway could be not on-link.
1088  *
1089  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1090  * results in pretty ugly and hairy code with obscure logic.
1091  *
1092  * I chose to generalized it instead, so that the size
1093  * of code does not increase practically, but it becomes
1094  * much more general.
1095  * Every prefix is assigned a "scope" value: "host" is local address,
1096  * "link" is direct route,
1097  * [ ... "site" ... "interior" ... ]
1098  * and "universe" is true gateway route with global meaning.
1099  *
1100  * Every prefix refers to a set of "nexthop"s (gw, oif),
1101  * where gw must have narrower scope. This recursion stops
1102  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1103  * which means that gw is forced to be on link.
1104  *
1105  * Code is still hairy, but now it is apparently logically
1106  * consistent and very flexible. F.e. as by-product it allows
1107  * to co-exists in peace independent exterior and interior
1108  * routing processes.
1109  *
1110  * Normally it looks as following.
1111  *
1112  * {universe prefix}  -> (gw, oif) [scope link]
1113  *		  |
1114  *		  |-> {link prefix} -> (gw, oif) [scope local]
1115  *					|
1116  *					|-> {local prefix} (terminal node)
1117  */
1118 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1119 			      u8 scope, struct netlink_ext_ack *extack)
1120 {
1121 	struct net_device *dev;
1122 	struct fib_result res;
1123 	int err = 0;
1124 
1125 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1126 		unsigned int addr_type;
1127 
1128 		if (scope >= RT_SCOPE_LINK) {
1129 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1130 			return -EINVAL;
1131 		}
1132 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1133 		if (!dev) {
1134 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1135 			return -ENODEV;
1136 		}
1137 		if (!(dev->flags & IFF_UP)) {
1138 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1139 			return -ENETDOWN;
1140 		}
1141 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1142 		if (addr_type != RTN_UNICAST) {
1143 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1144 			return -EINVAL;
1145 		}
1146 		if (!netif_carrier_ok(dev))
1147 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1148 		nh->fib_nh_dev = dev;
1149 		netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1150 		nh->fib_nh_scope = RT_SCOPE_LINK;
1151 		return 0;
1152 	}
1153 	rcu_read_lock();
1154 	{
1155 		struct fib_table *tbl = NULL;
1156 		struct flowi4 fl4 = {
1157 			.daddr = nh->fib_nh_gw4,
1158 			.flowi4_scope = scope + 1,
1159 			.flowi4_oif = nh->fib_nh_oif,
1160 			.flowi4_iif = LOOPBACK_IFINDEX,
1161 		};
1162 
1163 		/* It is not necessary, but requires a bit of thinking */
1164 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1165 			fl4.flowi4_scope = RT_SCOPE_LINK;
1166 
1167 		if (table && table != RT_TABLE_MAIN)
1168 			tbl = fib_get_table(net, table);
1169 
1170 		if (tbl)
1171 			err = fib_table_lookup(tbl, &fl4, &res,
1172 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1173 					       FIB_LOOKUP_NOREF);
1174 
1175 		/* on error or if no table given do full lookup. This
1176 		 * is needed for example when nexthops are in the local
1177 		 * table rather than the given table
1178 		 */
1179 		if (!tbl || err) {
1180 			err = fib_lookup(net, &fl4, &res,
1181 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1182 		}
1183 
1184 		if (err) {
1185 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1186 			goto out;
1187 		}
1188 	}
1189 
1190 	err = -EINVAL;
1191 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1192 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1193 		goto out;
1194 	}
1195 	nh->fib_nh_scope = res.scope;
1196 	nh->fib_nh_oif = FIB_RES_OIF(res);
1197 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1198 	if (!dev) {
1199 		NL_SET_ERR_MSG(extack,
1200 			       "No egress device for nexthop gateway");
1201 		goto out;
1202 	}
1203 	netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1204 	if (!netif_carrier_ok(dev))
1205 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1206 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1207 out:
1208 	rcu_read_unlock();
1209 	return err;
1210 }
1211 
1212 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1213 			      struct netlink_ext_ack *extack)
1214 {
1215 	struct in_device *in_dev;
1216 	int err;
1217 
1218 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1219 		NL_SET_ERR_MSG(extack,
1220 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1221 		return -EINVAL;
1222 	}
1223 
1224 	rcu_read_lock();
1225 
1226 	err = -ENODEV;
1227 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1228 	if (!in_dev)
1229 		goto out;
1230 	err = -ENETDOWN;
1231 	if (!(in_dev->dev->flags & IFF_UP)) {
1232 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1233 		goto out;
1234 	}
1235 
1236 	nh->fib_nh_dev = in_dev->dev;
1237 	netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1238 	nh->fib_nh_scope = RT_SCOPE_HOST;
1239 	if (!netif_carrier_ok(nh->fib_nh_dev))
1240 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1241 	err = 0;
1242 out:
1243 	rcu_read_unlock();
1244 	return err;
1245 }
1246 
1247 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1248 		 struct netlink_ext_ack *extack)
1249 {
1250 	int err;
1251 
1252 	if (nh->fib_nh_gw_family == AF_INET)
1253 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1254 	else if (nh->fib_nh_gw_family == AF_INET6)
1255 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1256 	else
1257 		err = fib_check_nh_nongw(net, nh, extack);
1258 
1259 	return err;
1260 }
1261 
1262 static struct hlist_head *
1263 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1264 {
1265 	u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1266 			   fib_info_hash_bits);
1267 
1268 	return &fib_info_laddrhash[slot];
1269 }
1270 
1271 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1272 			       struct hlist_head *new_laddrhash,
1273 			       unsigned int new_size)
1274 {
1275 	struct hlist_head *old_info_hash, *old_laddrhash;
1276 	unsigned int old_size = fib_info_hash_size;
1277 	unsigned int i;
1278 
1279 	spin_lock_bh(&fib_info_lock);
1280 	old_info_hash = fib_info_hash;
1281 	old_laddrhash = fib_info_laddrhash;
1282 	fib_info_hash_size = new_size;
1283 	fib_info_hash_bits = ilog2(new_size);
1284 
1285 	for (i = 0; i < old_size; i++) {
1286 		struct hlist_head *head = &fib_info_hash[i];
1287 		struct hlist_node *n;
1288 		struct fib_info *fi;
1289 
1290 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1291 			struct hlist_head *dest;
1292 			unsigned int new_hash;
1293 
1294 			new_hash = fib_info_hashfn(fi);
1295 			dest = &new_info_hash[new_hash];
1296 			hlist_add_head(&fi->fib_hash, dest);
1297 		}
1298 	}
1299 	fib_info_hash = new_info_hash;
1300 
1301 	fib_info_laddrhash = new_laddrhash;
1302 	for (i = 0; i < old_size; i++) {
1303 		struct hlist_head *lhead = &old_laddrhash[i];
1304 		struct hlist_node *n;
1305 		struct fib_info *fi;
1306 
1307 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1308 			struct hlist_head *ldest;
1309 
1310 			ldest = fib_info_laddrhash_bucket(fi->fib_net,
1311 							  fi->fib_prefsrc);
1312 			hlist_add_head(&fi->fib_lhash, ldest);
1313 		}
1314 	}
1315 
1316 	spin_unlock_bh(&fib_info_lock);
1317 
1318 	kvfree(old_info_hash);
1319 	kvfree(old_laddrhash);
1320 }
1321 
1322 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1323 				 unsigned char scope)
1324 {
1325 	struct fib_nh *nh;
1326 	__be32 saddr;
1327 
1328 	if (nhc->nhc_family != AF_INET)
1329 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1330 
1331 	nh = container_of(nhc, struct fib_nh, nh_common);
1332 	saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1333 
1334 	WRITE_ONCE(nh->nh_saddr, saddr);
1335 	WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid));
1336 
1337 	return saddr;
1338 }
1339 
1340 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1341 {
1342 	struct fib_nh_common *nhc = res->nhc;
1343 
1344 	if (res->fi->fib_prefsrc)
1345 		return res->fi->fib_prefsrc;
1346 
1347 	if (nhc->nhc_family == AF_INET) {
1348 		struct fib_nh *nh;
1349 
1350 		nh = container_of(nhc, struct fib_nh, nh_common);
1351 		if (READ_ONCE(nh->nh_saddr_genid) ==
1352 		    atomic_read(&net->ipv4.dev_addr_genid))
1353 			return READ_ONCE(nh->nh_saddr);
1354 	}
1355 
1356 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1357 }
1358 
1359 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1360 {
1361 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1362 	    fib_prefsrc != cfg->fc_dst) {
1363 		u32 tb_id = cfg->fc_table;
1364 		int rc;
1365 
1366 		if (tb_id == RT_TABLE_MAIN)
1367 			tb_id = RT_TABLE_LOCAL;
1368 
1369 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1370 					  fib_prefsrc, tb_id);
1371 
1372 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1373 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1374 						  fib_prefsrc, RT_TABLE_LOCAL);
1375 		}
1376 
1377 		if (rc != RTN_LOCAL)
1378 			return false;
1379 	}
1380 	return true;
1381 }
1382 
1383 struct fib_info *fib_create_info(struct fib_config *cfg,
1384 				 struct netlink_ext_ack *extack)
1385 {
1386 	int err;
1387 	struct fib_info *fi = NULL;
1388 	struct nexthop *nh = NULL;
1389 	struct fib_info *ofi;
1390 	int nhs = 1;
1391 	struct net *net = cfg->fc_nlinfo.nl_net;
1392 
1393 	if (cfg->fc_type > RTN_MAX)
1394 		goto err_inval;
1395 
1396 	/* Fast check to catch the most weird cases */
1397 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1398 		NL_SET_ERR_MSG(extack, "Invalid scope");
1399 		goto err_inval;
1400 	}
1401 
1402 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1403 		NL_SET_ERR_MSG(extack,
1404 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1405 		goto err_inval;
1406 	}
1407 
1408 	if (cfg->fc_nh_id) {
1409 		if (!cfg->fc_mx) {
1410 			fi = fib_find_info_nh(net, cfg);
1411 			if (fi) {
1412 				refcount_inc(&fi->fib_treeref);
1413 				return fi;
1414 			}
1415 		}
1416 
1417 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1418 		if (!nh) {
1419 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1420 			goto err_inval;
1421 		}
1422 		nhs = 0;
1423 	}
1424 
1425 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1426 	if (cfg->fc_mp) {
1427 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1428 		if (nhs == 0)
1429 			goto err_inval;
1430 	}
1431 #endif
1432 
1433 	err = -ENOBUFS;
1434 
1435 	/* Paired with WRITE_ONCE() in fib_release_info() */
1436 	if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1437 		unsigned int new_size = fib_info_hash_size << 1;
1438 		struct hlist_head *new_info_hash;
1439 		struct hlist_head *new_laddrhash;
1440 		size_t bytes;
1441 
1442 		if (!new_size)
1443 			new_size = 16;
1444 		bytes = (size_t)new_size * sizeof(struct hlist_head *);
1445 		new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1446 		new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1447 		if (!new_info_hash || !new_laddrhash) {
1448 			kvfree(new_info_hash);
1449 			kvfree(new_laddrhash);
1450 		} else {
1451 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1452 		}
1453 		if (!fib_info_hash_size)
1454 			goto failure;
1455 	}
1456 
1457 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1458 	if (!fi)
1459 		goto failure;
1460 	fi->fib_metrics = ip_fib_metrics_init(cfg->fc_mx, cfg->fc_mx_len, extack);
1461 	if (IS_ERR(fi->fib_metrics)) {
1462 		err = PTR_ERR(fi->fib_metrics);
1463 		kfree(fi);
1464 		return ERR_PTR(err);
1465 	}
1466 
1467 	fi->fib_net = net;
1468 	fi->fib_protocol = cfg->fc_protocol;
1469 	fi->fib_scope = cfg->fc_scope;
1470 	fi->fib_flags = cfg->fc_flags;
1471 	fi->fib_priority = cfg->fc_priority;
1472 	fi->fib_prefsrc = cfg->fc_prefsrc;
1473 	fi->fib_type = cfg->fc_type;
1474 	fi->fib_tb_id = cfg->fc_table;
1475 
1476 	fi->fib_nhs = nhs;
1477 	if (nh) {
1478 		if (!nexthop_get(nh)) {
1479 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1480 			err = -EINVAL;
1481 		} else {
1482 			err = 0;
1483 			fi->nh = nh;
1484 		}
1485 	} else {
1486 		change_nexthops(fi) {
1487 			nexthop_nh->nh_parent = fi;
1488 		} endfor_nexthops(fi)
1489 
1490 		if (cfg->fc_mp)
1491 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1492 					  extack);
1493 		else
1494 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1495 	}
1496 
1497 	if (err != 0)
1498 		goto failure;
1499 
1500 	if (fib_props[cfg->fc_type].error) {
1501 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1502 			NL_SET_ERR_MSG(extack,
1503 				       "Gateway, device and multipath can not be specified for this route type");
1504 			goto err_inval;
1505 		}
1506 		goto link_it;
1507 	} else {
1508 		switch (cfg->fc_type) {
1509 		case RTN_UNICAST:
1510 		case RTN_LOCAL:
1511 		case RTN_BROADCAST:
1512 		case RTN_ANYCAST:
1513 		case RTN_MULTICAST:
1514 			break;
1515 		default:
1516 			NL_SET_ERR_MSG(extack, "Invalid route type");
1517 			goto err_inval;
1518 		}
1519 	}
1520 
1521 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1522 		NL_SET_ERR_MSG(extack, "Invalid scope");
1523 		goto err_inval;
1524 	}
1525 
1526 	if (fi->nh) {
1527 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1528 		if (err)
1529 			goto failure;
1530 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1531 		struct fib_nh *nh = fi->fib_nh;
1532 
1533 		/* Local address is added. */
1534 		if (nhs != 1) {
1535 			NL_SET_ERR_MSG(extack,
1536 				       "Route with host scope can not have multiple nexthops");
1537 			goto err_inval;
1538 		}
1539 		if (nh->fib_nh_gw_family) {
1540 			NL_SET_ERR_MSG(extack,
1541 				       "Route with host scope can not have a gateway");
1542 			goto err_inval;
1543 		}
1544 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1545 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1546 		err = -ENODEV;
1547 		if (!nh->fib_nh_dev)
1548 			goto failure;
1549 		netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1550 				     GFP_KERNEL);
1551 	} else {
1552 		int linkdown = 0;
1553 
1554 		change_nexthops(fi) {
1555 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1556 					   cfg->fc_table, cfg->fc_scope,
1557 					   extack);
1558 			if (err != 0)
1559 				goto failure;
1560 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1561 				linkdown++;
1562 		} endfor_nexthops(fi)
1563 		if (linkdown == fi->fib_nhs)
1564 			fi->fib_flags |= RTNH_F_LINKDOWN;
1565 	}
1566 
1567 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1568 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1569 		goto err_inval;
1570 	}
1571 
1572 	if (!fi->nh) {
1573 		change_nexthops(fi) {
1574 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1575 						  fi->fib_scope);
1576 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1577 				fi->fib_nh_is_v6 = true;
1578 		} endfor_nexthops(fi)
1579 
1580 		fib_rebalance(fi);
1581 	}
1582 
1583 link_it:
1584 	ofi = fib_find_info(fi);
1585 	if (ofi) {
1586 		/* fib_table_lookup() should not see @fi yet. */
1587 		fi->fib_dead = 1;
1588 		free_fib_info(fi);
1589 		refcount_inc(&ofi->fib_treeref);
1590 		return ofi;
1591 	}
1592 
1593 	refcount_set(&fi->fib_treeref, 1);
1594 	refcount_set(&fi->fib_clntref, 1);
1595 	spin_lock_bh(&fib_info_lock);
1596 	fib_info_cnt++;
1597 	hlist_add_head(&fi->fib_hash,
1598 		       &fib_info_hash[fib_info_hashfn(fi)]);
1599 	if (fi->fib_prefsrc) {
1600 		struct hlist_head *head;
1601 
1602 		head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1603 		hlist_add_head(&fi->fib_lhash, head);
1604 	}
1605 	if (fi->nh) {
1606 		list_add(&fi->nh_list, &nh->fi_list);
1607 	} else {
1608 		change_nexthops(fi) {
1609 			struct hlist_head *head;
1610 
1611 			if (!nexthop_nh->fib_nh_dev)
1612 				continue;
1613 			head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1614 			hlist_add_head(&nexthop_nh->nh_hash, head);
1615 		} endfor_nexthops(fi)
1616 	}
1617 	spin_unlock_bh(&fib_info_lock);
1618 	return fi;
1619 
1620 err_inval:
1621 	err = -EINVAL;
1622 
1623 failure:
1624 	if (fi) {
1625 		/* fib_table_lookup() should not see @fi yet. */
1626 		fi->fib_dead = 1;
1627 		free_fib_info(fi);
1628 	}
1629 
1630 	return ERR_PTR(err);
1631 }
1632 
1633 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1634 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1635 {
1636 	if (nhc->nhc_flags & RTNH_F_DEAD)
1637 		*flags |= RTNH_F_DEAD;
1638 
1639 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1640 		*flags |= RTNH_F_LINKDOWN;
1641 
1642 		rcu_read_lock();
1643 		switch (nhc->nhc_family) {
1644 		case AF_INET:
1645 			if (ip_ignore_linkdown(nhc->nhc_dev))
1646 				*flags |= RTNH_F_DEAD;
1647 			break;
1648 		case AF_INET6:
1649 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1650 				*flags |= RTNH_F_DEAD;
1651 			break;
1652 		}
1653 		rcu_read_unlock();
1654 	}
1655 
1656 	switch (nhc->nhc_gw_family) {
1657 	case AF_INET:
1658 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1659 			goto nla_put_failure;
1660 		break;
1661 	case AF_INET6:
1662 		/* if gateway family does not match nexthop family
1663 		 * gateway is encoded as RTA_VIA
1664 		 */
1665 		if (rt_family != nhc->nhc_gw_family) {
1666 			int alen = sizeof(struct in6_addr);
1667 			struct nlattr *nla;
1668 			struct rtvia *via;
1669 
1670 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1671 			if (!nla)
1672 				goto nla_put_failure;
1673 
1674 			via = nla_data(nla);
1675 			via->rtvia_family = AF_INET6;
1676 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1677 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1678 					    &nhc->nhc_gw.ipv6) < 0) {
1679 			goto nla_put_failure;
1680 		}
1681 		break;
1682 	}
1683 
1684 	*flags |= (nhc->nhc_flags &
1685 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1686 
1687 	if (!skip_oif && nhc->nhc_dev &&
1688 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1689 		goto nla_put_failure;
1690 
1691 	if (nhc->nhc_lwtstate &&
1692 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1693 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1694 		goto nla_put_failure;
1695 
1696 	return 0;
1697 
1698 nla_put_failure:
1699 	return -EMSGSIZE;
1700 }
1701 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1702 
1703 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1704 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1705 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1706 {
1707 	const struct net_device *dev = nhc->nhc_dev;
1708 	struct rtnexthop *rtnh;
1709 	unsigned char flags = 0;
1710 
1711 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1712 	if (!rtnh)
1713 		goto nla_put_failure;
1714 
1715 	rtnh->rtnh_hops = nh_weight - 1;
1716 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1717 
1718 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1719 		goto nla_put_failure;
1720 
1721 	rtnh->rtnh_flags = flags;
1722 
1723 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1724 		goto nla_put_failure;
1725 
1726 	/* length of rtnetlink header + attributes */
1727 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1728 
1729 	return 0;
1730 
1731 nla_put_failure:
1732 	return -EMSGSIZE;
1733 }
1734 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1735 #endif
1736 
1737 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1738 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1739 {
1740 	struct nlattr *mp;
1741 
1742 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1743 	if (!mp)
1744 		goto nla_put_failure;
1745 
1746 	if (unlikely(fi->nh)) {
1747 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1748 			goto nla_put_failure;
1749 		goto mp_end;
1750 	}
1751 
1752 	for_nexthops(fi) {
1753 		u32 nh_tclassid = 0;
1754 #ifdef CONFIG_IP_ROUTE_CLASSID
1755 		nh_tclassid = nh->nh_tclassid;
1756 #endif
1757 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1758 				    AF_INET, nh_tclassid) < 0)
1759 			goto nla_put_failure;
1760 	} endfor_nexthops(fi);
1761 
1762 mp_end:
1763 	nla_nest_end(skb, mp);
1764 
1765 	return 0;
1766 
1767 nla_put_failure:
1768 	return -EMSGSIZE;
1769 }
1770 #else
1771 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1772 {
1773 	return 0;
1774 }
1775 #endif
1776 
1777 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1778 		  const struct fib_rt_info *fri, unsigned int flags)
1779 {
1780 	unsigned int nhs = fib_info_num_path(fri->fi);
1781 	struct fib_info *fi = fri->fi;
1782 	u32 tb_id = fri->tb_id;
1783 	struct nlmsghdr *nlh;
1784 	struct rtmsg *rtm;
1785 
1786 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1787 	if (!nlh)
1788 		return -EMSGSIZE;
1789 
1790 	rtm = nlmsg_data(nlh);
1791 	rtm->rtm_family = AF_INET;
1792 	rtm->rtm_dst_len = fri->dst_len;
1793 	rtm->rtm_src_len = 0;
1794 	rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1795 	if (tb_id < 256)
1796 		rtm->rtm_table = tb_id;
1797 	else
1798 		rtm->rtm_table = RT_TABLE_COMPAT;
1799 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1800 		goto nla_put_failure;
1801 	rtm->rtm_type = fri->type;
1802 	rtm->rtm_flags = fi->fib_flags;
1803 	rtm->rtm_scope = fi->fib_scope;
1804 	rtm->rtm_protocol = fi->fib_protocol;
1805 
1806 	if (rtm->rtm_dst_len &&
1807 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1808 		goto nla_put_failure;
1809 	if (fi->fib_priority &&
1810 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1811 		goto nla_put_failure;
1812 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1813 		goto nla_put_failure;
1814 
1815 	if (fi->fib_prefsrc &&
1816 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1817 		goto nla_put_failure;
1818 
1819 	if (fi->nh) {
1820 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1821 			goto nla_put_failure;
1822 		if (nexthop_is_blackhole(fi->nh))
1823 			rtm->rtm_type = RTN_BLACKHOLE;
1824 		if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1825 			goto offload;
1826 	}
1827 
1828 	if (nhs == 1) {
1829 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1830 		unsigned char flags = 0;
1831 
1832 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1833 			goto nla_put_failure;
1834 
1835 		rtm->rtm_flags = flags;
1836 #ifdef CONFIG_IP_ROUTE_CLASSID
1837 		if (nhc->nhc_family == AF_INET) {
1838 			struct fib_nh *nh;
1839 
1840 			nh = container_of(nhc, struct fib_nh, nh_common);
1841 			if (nh->nh_tclassid &&
1842 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1843 				goto nla_put_failure;
1844 		}
1845 #endif
1846 	} else {
1847 		if (fib_add_multipath(skb, fi) < 0)
1848 			goto nla_put_failure;
1849 	}
1850 
1851 offload:
1852 	if (fri->offload)
1853 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1854 	if (fri->trap)
1855 		rtm->rtm_flags |= RTM_F_TRAP;
1856 	if (fri->offload_failed)
1857 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1858 
1859 	nlmsg_end(skb, nlh);
1860 	return 0;
1861 
1862 nla_put_failure:
1863 	nlmsg_cancel(skb, nlh);
1864 	return -EMSGSIZE;
1865 }
1866 
1867 /*
1868  * Update FIB if:
1869  * - local address disappeared -> we must delete all the entries
1870  *   referring to it.
1871  * - device went down -> we must shutdown all nexthops going via it.
1872  */
1873 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1874 {
1875 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1876 	struct net *net = dev_net(dev);
1877 	struct hlist_head *head;
1878 	struct fib_info *fi;
1879 	int ret = 0;
1880 
1881 	if (!fib_info_laddrhash || local == 0)
1882 		return 0;
1883 
1884 	head = fib_info_laddrhash_bucket(net, local);
1885 	hlist_for_each_entry(fi, head, fib_lhash) {
1886 		if (!net_eq(fi->fib_net, net) ||
1887 		    fi->fib_tb_id != tb_id)
1888 			continue;
1889 		if (fi->fib_prefsrc == local) {
1890 			fi->fib_flags |= RTNH_F_DEAD;
1891 			fi->pfsrc_removed = true;
1892 			ret++;
1893 		}
1894 	}
1895 	return ret;
1896 }
1897 
1898 static int call_fib_nh_notifiers(struct fib_nh *nh,
1899 				 enum fib_event_type event_type)
1900 {
1901 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1902 	struct fib_nh_notifier_info info = {
1903 		.fib_nh = nh,
1904 	};
1905 
1906 	switch (event_type) {
1907 	case FIB_EVENT_NH_ADD:
1908 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1909 			break;
1910 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1911 			break;
1912 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1913 					   &info.info);
1914 	case FIB_EVENT_NH_DEL:
1915 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1916 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1917 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1918 						   event_type, &info.info);
1919 		break;
1920 	default:
1921 		break;
1922 	}
1923 
1924 	return NOTIFY_DONE;
1925 }
1926 
1927 /* Update the PMTU of exceptions when:
1928  * - the new MTU of the first hop becomes smaller than the PMTU
1929  * - the old MTU was the same as the PMTU, and it limited discovery of
1930  *   larger MTUs on the path. With that limit raised, we can now
1931  *   discover larger MTUs
1932  * A special case is locked exceptions, for which the PMTU is smaller
1933  * than the minimal accepted PMTU:
1934  * - if the new MTU is greater than the PMTU, don't make any change
1935  * - otherwise, unlock and set PMTU
1936  */
1937 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1938 {
1939 	struct fnhe_hash_bucket *bucket;
1940 	int i;
1941 
1942 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1943 	if (!bucket)
1944 		return;
1945 
1946 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1947 		struct fib_nh_exception *fnhe;
1948 
1949 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1950 		     fnhe;
1951 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1952 			if (fnhe->fnhe_mtu_locked) {
1953 				if (new <= fnhe->fnhe_pmtu) {
1954 					fnhe->fnhe_pmtu = new;
1955 					fnhe->fnhe_mtu_locked = false;
1956 				}
1957 			} else if (new < fnhe->fnhe_pmtu ||
1958 				   orig == fnhe->fnhe_pmtu) {
1959 				fnhe->fnhe_pmtu = new;
1960 			}
1961 		}
1962 	}
1963 }
1964 
1965 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1966 {
1967 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1968 	struct fib_nh *nh;
1969 
1970 	hlist_for_each_entry(nh, head, nh_hash) {
1971 		if (nh->fib_nh_dev == dev)
1972 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1973 	}
1974 }
1975 
1976 /* Event              force Flags           Description
1977  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1978  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1979  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1980  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1981  *
1982  * only used when fib_nh is built into fib_info
1983  */
1984 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1985 {
1986 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1987 	struct fib_info *prev_fi = NULL;
1988 	int scope = RT_SCOPE_NOWHERE;
1989 	struct fib_nh *nh;
1990 	int ret = 0;
1991 
1992 	if (force)
1993 		scope = -1;
1994 
1995 	hlist_for_each_entry(nh, head, nh_hash) {
1996 		struct fib_info *fi = nh->nh_parent;
1997 		int dead;
1998 
1999 		BUG_ON(!fi->fib_nhs);
2000 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2001 			continue;
2002 		prev_fi = fi;
2003 		dead = 0;
2004 		change_nexthops(fi) {
2005 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2006 				dead++;
2007 			else if (nexthop_nh->fib_nh_dev == dev &&
2008 				 nexthop_nh->fib_nh_scope != scope) {
2009 				switch (event) {
2010 				case NETDEV_DOWN:
2011 				case NETDEV_UNREGISTER:
2012 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2013 					fallthrough;
2014 				case NETDEV_CHANGE:
2015 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2016 					break;
2017 				}
2018 				call_fib_nh_notifiers(nexthop_nh,
2019 						      FIB_EVENT_NH_DEL);
2020 				dead++;
2021 			}
2022 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2023 			if (event == NETDEV_UNREGISTER &&
2024 			    nexthop_nh->fib_nh_dev == dev) {
2025 				dead = fi->fib_nhs;
2026 				break;
2027 			}
2028 #endif
2029 		} endfor_nexthops(fi)
2030 		if (dead == fi->fib_nhs) {
2031 			switch (event) {
2032 			case NETDEV_DOWN:
2033 			case NETDEV_UNREGISTER:
2034 				fi->fib_flags |= RTNH_F_DEAD;
2035 				fallthrough;
2036 			case NETDEV_CHANGE:
2037 				fi->fib_flags |= RTNH_F_LINKDOWN;
2038 				break;
2039 			}
2040 			ret++;
2041 		}
2042 
2043 		fib_rebalance(fi);
2044 	}
2045 
2046 	return ret;
2047 }
2048 
2049 /* Must be invoked inside of an RCU protected region.  */
2050 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2051 {
2052 	struct fib_info *fi = NULL, *last_resort = NULL;
2053 	struct hlist_head *fa_head = res->fa_head;
2054 	struct fib_table *tb = res->table;
2055 	u8 slen = 32 - res->prefixlen;
2056 	int order = -1, last_idx = -1;
2057 	struct fib_alias *fa, *fa1 = NULL;
2058 	u32 last_prio = res->fi->fib_priority;
2059 	dscp_t last_dscp = 0;
2060 
2061 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2062 		struct fib_info *next_fi = fa->fa_info;
2063 		struct fib_nh_common *nhc;
2064 
2065 		if (fa->fa_slen != slen)
2066 			continue;
2067 		if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp))
2068 			continue;
2069 		if (fa->tb_id != tb->tb_id)
2070 			continue;
2071 		if (next_fi->fib_priority > last_prio &&
2072 		    fa->fa_dscp == last_dscp) {
2073 			if (last_dscp)
2074 				continue;
2075 			break;
2076 		}
2077 		if (next_fi->fib_flags & RTNH_F_DEAD)
2078 			continue;
2079 		last_dscp = fa->fa_dscp;
2080 		last_prio = next_fi->fib_priority;
2081 
2082 		if (next_fi->fib_scope != res->scope ||
2083 		    fa->fa_type != RTN_UNICAST)
2084 			continue;
2085 
2086 		nhc = fib_info_nhc(next_fi, 0);
2087 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2088 			continue;
2089 
2090 		fib_alias_accessed(fa);
2091 
2092 		if (!fi) {
2093 			if (next_fi != res->fi)
2094 				break;
2095 			fa1 = fa;
2096 		} else if (!fib_detect_death(fi, order, &last_resort,
2097 					     &last_idx, fa1->fa_default)) {
2098 			fib_result_assign(res, fi);
2099 			fa1->fa_default = order;
2100 			goto out;
2101 		}
2102 		fi = next_fi;
2103 		order++;
2104 	}
2105 
2106 	if (order <= 0 || !fi) {
2107 		if (fa1)
2108 			fa1->fa_default = -1;
2109 		goto out;
2110 	}
2111 
2112 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2113 			      fa1->fa_default)) {
2114 		fib_result_assign(res, fi);
2115 		fa1->fa_default = order;
2116 		goto out;
2117 	}
2118 
2119 	if (last_idx >= 0)
2120 		fib_result_assign(res, last_resort);
2121 	fa1->fa_default = last_idx;
2122 out:
2123 	return;
2124 }
2125 
2126 /*
2127  * Dead device goes up. We wake up dead nexthops.
2128  * It takes sense only on multipath routes.
2129  *
2130  * only used when fib_nh is built into fib_info
2131  */
2132 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2133 {
2134 	struct fib_info *prev_fi;
2135 	struct hlist_head *head;
2136 	struct fib_nh *nh;
2137 	int ret;
2138 
2139 	if (!(dev->flags & IFF_UP))
2140 		return 0;
2141 
2142 	if (nh_flags & RTNH_F_DEAD) {
2143 		unsigned int flags = dev_get_flags(dev);
2144 
2145 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2146 			nh_flags |= RTNH_F_LINKDOWN;
2147 	}
2148 
2149 	prev_fi = NULL;
2150 	head = fib_info_devhash_bucket(dev);
2151 	ret = 0;
2152 
2153 	hlist_for_each_entry(nh, head, nh_hash) {
2154 		struct fib_info *fi = nh->nh_parent;
2155 		int alive;
2156 
2157 		BUG_ON(!fi->fib_nhs);
2158 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2159 			continue;
2160 
2161 		prev_fi = fi;
2162 		alive = 0;
2163 		change_nexthops(fi) {
2164 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2165 				alive++;
2166 				continue;
2167 			}
2168 			if (!nexthop_nh->fib_nh_dev ||
2169 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2170 				continue;
2171 			if (nexthop_nh->fib_nh_dev != dev ||
2172 			    !__in_dev_get_rtnl(dev))
2173 				continue;
2174 			alive++;
2175 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2176 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2177 		} endfor_nexthops(fi)
2178 
2179 		if (alive > 0) {
2180 			fi->fib_flags &= ~nh_flags;
2181 			ret++;
2182 		}
2183 
2184 		fib_rebalance(fi);
2185 	}
2186 
2187 	return ret;
2188 }
2189 
2190 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2191 static bool fib_good_nh(const struct fib_nh *nh)
2192 {
2193 	int state = NUD_REACHABLE;
2194 
2195 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2196 		struct neighbour *n;
2197 
2198 		rcu_read_lock();
2199 
2200 		if (likely(nh->fib_nh_gw_family == AF_INET))
2201 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2202 						   (__force u32)nh->fib_nh_gw4);
2203 		else if (nh->fib_nh_gw_family == AF_INET6)
2204 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2205 							   &nh->fib_nh_gw6);
2206 		else
2207 			n = NULL;
2208 		if (n)
2209 			state = READ_ONCE(n->nud_state);
2210 
2211 		rcu_read_unlock();
2212 	}
2213 
2214 	return !!(state & NUD_VALID);
2215 }
2216 
2217 void fib_select_multipath(struct fib_result *res, int hash)
2218 {
2219 	struct fib_info *fi = res->fi;
2220 	struct net *net = fi->fib_net;
2221 	bool first = false;
2222 
2223 	if (unlikely(res->fi->nh)) {
2224 		nexthop_path_fib_result(res, hash);
2225 		return;
2226 	}
2227 
2228 	change_nexthops(fi) {
2229 		if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2230 			if (!fib_good_nh(nexthop_nh))
2231 				continue;
2232 			if (!first) {
2233 				res->nh_sel = nhsel;
2234 				res->nhc = &nexthop_nh->nh_common;
2235 				first = true;
2236 			}
2237 		}
2238 
2239 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2240 			continue;
2241 
2242 		res->nh_sel = nhsel;
2243 		res->nhc = &nexthop_nh->nh_common;
2244 		return;
2245 	} endfor_nexthops(fi);
2246 }
2247 #endif
2248 
2249 void fib_select_path(struct net *net, struct fib_result *res,
2250 		     struct flowi4 *fl4, const struct sk_buff *skb)
2251 {
2252 	if (fl4->flowi4_oif)
2253 		goto check_saddr;
2254 
2255 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2256 	if (fib_info_num_path(res->fi) > 1) {
2257 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2258 
2259 		fib_select_multipath(res, h);
2260 	}
2261 	else
2262 #endif
2263 	if (!res->prefixlen &&
2264 	    res->table->tb_num_default > 1 &&
2265 	    res->type == RTN_UNICAST)
2266 		fib_select_default(fl4, res);
2267 
2268 check_saddr:
2269 	if (!fl4->saddr) {
2270 		struct net_device *l3mdev;
2271 
2272 		l3mdev = dev_get_by_index_rcu(net, fl4->flowi4_l3mdev);
2273 
2274 		if (!l3mdev ||
2275 		    l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) == l3mdev)
2276 			fl4->saddr = fib_result_prefsrc(net, res);
2277 		else
2278 			fl4->saddr = inet_select_addr(l3mdev, 0, RT_SCOPE_LINK);
2279 	}
2280 }
2281