xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision e6486939)
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 	dev_put_track(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 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
427 			   sizeof(u32) * RTAX_MAX) == 0 &&
428 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
429 		    nh_comp(fi, nfi) == 0)
430 			return fi;
431 	}
432 
433 	return NULL;
434 }
435 
436 /* Check, that the gateway is already configured.
437  * Used only by redirect accept routine.
438  */
439 int ip_fib_check_default(__be32 gw, struct net_device *dev)
440 {
441 	struct hlist_head *head;
442 	struct fib_nh *nh;
443 
444 	spin_lock(&fib_info_lock);
445 
446 	head = fib_info_devhash_bucket(dev);
447 
448 	hlist_for_each_entry(nh, head, nh_hash) {
449 		if (nh->fib_nh_dev == dev &&
450 		    nh->fib_nh_gw4 == gw &&
451 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
452 			spin_unlock(&fib_info_lock);
453 			return 0;
454 		}
455 	}
456 
457 	spin_unlock(&fib_info_lock);
458 
459 	return -1;
460 }
461 
462 size_t fib_nlmsg_size(struct fib_info *fi)
463 {
464 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
465 			 + nla_total_size(4) /* RTA_TABLE */
466 			 + nla_total_size(4) /* RTA_DST */
467 			 + nla_total_size(4) /* RTA_PRIORITY */
468 			 + nla_total_size(4) /* RTA_PREFSRC */
469 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
470 	unsigned int nhs = fib_info_num_path(fi);
471 
472 	/* space for nested metrics */
473 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
474 
475 	if (fi->nh)
476 		payload += nla_total_size(4); /* RTA_NH_ID */
477 
478 	if (nhs) {
479 		size_t nh_encapsize = 0;
480 		/* Also handles the special case nhs == 1 */
481 
482 		/* each nexthop is packed in an attribute */
483 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
484 		unsigned int i;
485 
486 		/* may contain flow and gateway attribute */
487 		nhsize += 2 * nla_total_size(4);
488 
489 		/* grab encap info */
490 		for (i = 0; i < fib_info_num_path(fi); i++) {
491 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
492 
493 			if (nhc->nhc_lwtstate) {
494 				/* RTA_ENCAP_TYPE */
495 				nh_encapsize += lwtunnel_get_encap_size(
496 						nhc->nhc_lwtstate);
497 				/* RTA_ENCAP */
498 				nh_encapsize +=  nla_total_size(2);
499 			}
500 		}
501 
502 		/* all nexthops are packed in a nested attribute */
503 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
504 
505 	}
506 
507 	return payload;
508 }
509 
510 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
511 	       int dst_len, u32 tb_id, const struct nl_info *info,
512 	       unsigned int nlm_flags)
513 {
514 	struct fib_rt_info fri;
515 	struct sk_buff *skb;
516 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
517 	int err = -ENOBUFS;
518 
519 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
520 	if (!skb)
521 		goto errout;
522 
523 	fri.fi = fa->fa_info;
524 	fri.tb_id = tb_id;
525 	fri.dst = key;
526 	fri.dst_len = dst_len;
527 	fri.tos = inet_dscp_to_dsfield(fa->fa_dscp);
528 	fri.type = fa->fa_type;
529 	fri.offload = READ_ONCE(fa->offload);
530 	fri.trap = READ_ONCE(fa->trap);
531 	fri.offload_failed = READ_ONCE(fa->offload_failed);
532 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
533 	if (err < 0) {
534 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
535 		WARN_ON(err == -EMSGSIZE);
536 		kfree_skb(skb);
537 		goto errout;
538 	}
539 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
540 		    info->nlh, GFP_KERNEL);
541 	return;
542 errout:
543 	if (err < 0)
544 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
545 }
546 
547 static int fib_detect_death(struct fib_info *fi, int order,
548 			    struct fib_info **last_resort, int *last_idx,
549 			    int dflt)
550 {
551 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
552 	struct neighbour *n;
553 	int state = NUD_NONE;
554 
555 	if (likely(nhc->nhc_gw_family == AF_INET))
556 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
557 	else if (nhc->nhc_gw_family == AF_INET6)
558 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
559 				 nhc->nhc_dev);
560 	else
561 		n = NULL;
562 
563 	if (n) {
564 		state = n->nud_state;
565 		neigh_release(n);
566 	} else {
567 		return 0;
568 	}
569 	if (state == NUD_REACHABLE)
570 		return 0;
571 	if ((state & NUD_VALID) && order != dflt)
572 		return 0;
573 	if ((state & NUD_VALID) ||
574 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
575 		*last_resort = fi;
576 		*last_idx = order;
577 	}
578 	return 1;
579 }
580 
581 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
582 		       struct nlattr *encap, u16 encap_type,
583 		       void *cfg, gfp_t gfp_flags,
584 		       struct netlink_ext_ack *extack)
585 {
586 	int err;
587 
588 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
589 						    gfp_flags);
590 	if (!nhc->nhc_pcpu_rth_output)
591 		return -ENOMEM;
592 
593 	if (encap) {
594 		struct lwtunnel_state *lwtstate;
595 
596 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
597 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
598 			err = -EINVAL;
599 			goto lwt_failure;
600 		}
601 		err = lwtunnel_build_state(net, encap_type, encap,
602 					   nhc->nhc_family, cfg, &lwtstate,
603 					   extack);
604 		if (err)
605 			goto lwt_failure;
606 
607 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
608 	}
609 
610 	return 0;
611 
612 lwt_failure:
613 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
614 	nhc->nhc_pcpu_rth_output = NULL;
615 	return err;
616 }
617 EXPORT_SYMBOL_GPL(fib_nh_common_init);
618 
619 int fib_nh_init(struct net *net, struct fib_nh *nh,
620 		struct fib_config *cfg, int nh_weight,
621 		struct netlink_ext_ack *extack)
622 {
623 	int err;
624 
625 	nh->fib_nh_family = AF_INET;
626 
627 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
628 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
629 	if (err)
630 		return err;
631 
632 	nh->fib_nh_oif = cfg->fc_oif;
633 	nh->fib_nh_gw_family = cfg->fc_gw_family;
634 	if (cfg->fc_gw_family == AF_INET)
635 		nh->fib_nh_gw4 = cfg->fc_gw4;
636 	else if (cfg->fc_gw_family == AF_INET6)
637 		nh->fib_nh_gw6 = cfg->fc_gw6;
638 
639 	nh->fib_nh_flags = cfg->fc_flags;
640 
641 #ifdef CONFIG_IP_ROUTE_CLASSID
642 	nh->nh_tclassid = cfg->fc_flow;
643 	if (nh->nh_tclassid)
644 		atomic_inc(&net->ipv4.fib_num_tclassid_users);
645 #endif
646 #ifdef CONFIG_IP_ROUTE_MULTIPATH
647 	nh->fib_nh_weight = nh_weight;
648 #endif
649 	return 0;
650 }
651 
652 #ifdef CONFIG_IP_ROUTE_MULTIPATH
653 
654 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
655 			      struct netlink_ext_ack *extack)
656 {
657 	int nhs = 0;
658 
659 	while (rtnh_ok(rtnh, remaining)) {
660 		nhs++;
661 		rtnh = rtnh_next(rtnh, &remaining);
662 	}
663 
664 	/* leftover implies invalid nexthop configuration, discard it */
665 	if (remaining > 0) {
666 		NL_SET_ERR_MSG(extack,
667 			       "Invalid nexthop configuration - extra data after nexthops");
668 		nhs = 0;
669 	}
670 
671 	return nhs;
672 }
673 
674 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
675 			    struct netlink_ext_ack *extack)
676 {
677 	if (nla_len(nla) < sizeof(*gw)) {
678 		NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
679 		return -EINVAL;
680 	}
681 
682 	*gw = nla_get_in_addr(nla);
683 
684 	return 0;
685 }
686 
687 /* only called when fib_nh is integrated into fib_info */
688 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
689 		       int remaining, struct fib_config *cfg,
690 		       struct netlink_ext_ack *extack)
691 {
692 	struct net *net = fi->fib_net;
693 	struct fib_config fib_cfg;
694 	struct fib_nh *nh;
695 	int ret;
696 
697 	change_nexthops(fi) {
698 		int attrlen;
699 
700 		memset(&fib_cfg, 0, sizeof(fib_cfg));
701 
702 		if (!rtnh_ok(rtnh, remaining)) {
703 			NL_SET_ERR_MSG(extack,
704 				       "Invalid nexthop configuration - extra data after nexthop");
705 			return -EINVAL;
706 		}
707 
708 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
709 			NL_SET_ERR_MSG(extack,
710 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
711 			return -EINVAL;
712 		}
713 
714 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
715 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
716 
717 		attrlen = rtnh_attrlen(rtnh);
718 		if (attrlen > 0) {
719 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
720 
721 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
722 			nlav = nla_find(attrs, attrlen, RTA_VIA);
723 			if (nla && nlav) {
724 				NL_SET_ERR_MSG(extack,
725 					       "Nexthop configuration can not contain both GATEWAY and VIA");
726 				return -EINVAL;
727 			}
728 			if (nla) {
729 				ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
730 						       extack);
731 				if (ret)
732 					goto errout;
733 
734 				if (fib_cfg.fc_gw4)
735 					fib_cfg.fc_gw_family = AF_INET;
736 			} else if (nlav) {
737 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
738 				if (ret)
739 					goto errout;
740 			}
741 
742 			nla = nla_find(attrs, attrlen, RTA_FLOW);
743 			if (nla) {
744 				if (nla_len(nla) < sizeof(u32)) {
745 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
746 					return -EINVAL;
747 				}
748 				fib_cfg.fc_flow = nla_get_u32(nla);
749 			}
750 
751 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
752 			/* RTA_ENCAP_TYPE length checked in
753 			 * lwtunnel_valid_encap_type_attr
754 			 */
755 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
756 			if (nla)
757 				fib_cfg.fc_encap_type = nla_get_u16(nla);
758 		}
759 
760 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
761 				  rtnh->rtnh_hops + 1, extack);
762 		if (ret)
763 			goto errout;
764 
765 		rtnh = rtnh_next(rtnh, &remaining);
766 	} endfor_nexthops(fi);
767 
768 	ret = -EINVAL;
769 	nh = fib_info_nh(fi, 0);
770 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
771 		NL_SET_ERR_MSG(extack,
772 			       "Nexthop device index does not match RTA_OIF");
773 		goto errout;
774 	}
775 	if (cfg->fc_gw_family) {
776 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
777 		    (cfg->fc_gw_family == AF_INET &&
778 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
779 		    (cfg->fc_gw_family == AF_INET6 &&
780 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
781 			NL_SET_ERR_MSG(extack,
782 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
783 			goto errout;
784 		}
785 	}
786 #ifdef CONFIG_IP_ROUTE_CLASSID
787 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
788 		NL_SET_ERR_MSG(extack,
789 			       "Nexthop class id does not match RTA_FLOW");
790 		goto errout;
791 	}
792 #endif
793 	ret = 0;
794 errout:
795 	return ret;
796 }
797 
798 /* only called when fib_nh is integrated into fib_info */
799 static void fib_rebalance(struct fib_info *fi)
800 {
801 	int total;
802 	int w;
803 
804 	if (fib_info_num_path(fi) < 2)
805 		return;
806 
807 	total = 0;
808 	for_nexthops(fi) {
809 		if (nh->fib_nh_flags & RTNH_F_DEAD)
810 			continue;
811 
812 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
813 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
814 			continue;
815 
816 		total += nh->fib_nh_weight;
817 	} endfor_nexthops(fi);
818 
819 	w = 0;
820 	change_nexthops(fi) {
821 		int upper_bound;
822 
823 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
824 			upper_bound = -1;
825 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
826 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
827 			upper_bound = -1;
828 		} else {
829 			w += nexthop_nh->fib_nh_weight;
830 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
831 							    total) - 1;
832 		}
833 
834 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
835 	} endfor_nexthops(fi);
836 }
837 #else /* CONFIG_IP_ROUTE_MULTIPATH */
838 
839 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
840 		       int remaining, struct fib_config *cfg,
841 		       struct netlink_ext_ack *extack)
842 {
843 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
844 
845 	return -EINVAL;
846 }
847 
848 #define fib_rebalance(fi) do { } while (0)
849 
850 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
851 
852 static int fib_encap_match(struct net *net, u16 encap_type,
853 			   struct nlattr *encap,
854 			   const struct fib_nh *nh,
855 			   const struct fib_config *cfg,
856 			   struct netlink_ext_ack *extack)
857 {
858 	struct lwtunnel_state *lwtstate;
859 	int ret, result = 0;
860 
861 	if (encap_type == LWTUNNEL_ENCAP_NONE)
862 		return 0;
863 
864 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
865 				   cfg, &lwtstate, extack);
866 	if (!ret) {
867 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
868 		lwtstate_free(lwtstate);
869 	}
870 
871 	return result;
872 }
873 
874 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
875 		 struct netlink_ext_ack *extack)
876 {
877 #ifdef CONFIG_IP_ROUTE_MULTIPATH
878 	struct rtnexthop *rtnh;
879 	int remaining;
880 #endif
881 
882 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
883 		return 1;
884 
885 	if (cfg->fc_nh_id) {
886 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
887 			return 0;
888 		return 1;
889 	}
890 
891 	if (cfg->fc_oif || cfg->fc_gw_family) {
892 		struct fib_nh *nh = fib_info_nh(fi, 0);
893 
894 		if (cfg->fc_encap) {
895 			if (fib_encap_match(net, cfg->fc_encap_type,
896 					    cfg->fc_encap, nh, cfg, extack))
897 				return 1;
898 		}
899 #ifdef CONFIG_IP_ROUTE_CLASSID
900 		if (cfg->fc_flow &&
901 		    cfg->fc_flow != nh->nh_tclassid)
902 			return 1;
903 #endif
904 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
905 		    (cfg->fc_gw_family &&
906 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
907 			return 1;
908 
909 		if (cfg->fc_gw_family == AF_INET &&
910 		    cfg->fc_gw4 != nh->fib_nh_gw4)
911 			return 1;
912 
913 		if (cfg->fc_gw_family == AF_INET6 &&
914 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
915 			return 1;
916 
917 		return 0;
918 	}
919 
920 #ifdef CONFIG_IP_ROUTE_MULTIPATH
921 	if (!cfg->fc_mp)
922 		return 0;
923 
924 	rtnh = cfg->fc_mp;
925 	remaining = cfg->fc_mp_len;
926 
927 	for_nexthops(fi) {
928 		int attrlen;
929 
930 		if (!rtnh_ok(rtnh, remaining))
931 			return -EINVAL;
932 
933 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
934 			return 1;
935 
936 		attrlen = rtnh_attrlen(rtnh);
937 		if (attrlen > 0) {
938 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
939 			int err;
940 
941 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
942 			nlav = nla_find(attrs, attrlen, RTA_VIA);
943 			if (nla && nlav) {
944 				NL_SET_ERR_MSG(extack,
945 					       "Nexthop configuration can not contain both GATEWAY and VIA");
946 				return -EINVAL;
947 			}
948 
949 			if (nla) {
950 				__be32 gw;
951 
952 				err = fib_gw_from_attr(&gw, nla, extack);
953 				if (err)
954 					return err;
955 
956 				if (nh->fib_nh_gw_family != AF_INET ||
957 				    gw != nh->fib_nh_gw4)
958 					return 1;
959 			} else if (nlav) {
960 				struct fib_config cfg2;
961 
962 				err = fib_gw_from_via(&cfg2, nlav, extack);
963 				if (err)
964 					return err;
965 
966 				switch (nh->fib_nh_gw_family) {
967 				case AF_INET:
968 					if (cfg2.fc_gw_family != AF_INET ||
969 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
970 						return 1;
971 					break;
972 				case AF_INET6:
973 					if (cfg2.fc_gw_family != AF_INET6 ||
974 					    ipv6_addr_cmp(&cfg2.fc_gw6,
975 							  &nh->fib_nh_gw6))
976 						return 1;
977 					break;
978 				}
979 			}
980 
981 #ifdef CONFIG_IP_ROUTE_CLASSID
982 			nla = nla_find(attrs, attrlen, RTA_FLOW);
983 			if (nla) {
984 				if (nla_len(nla) < sizeof(u32)) {
985 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
986 					return -EINVAL;
987 				}
988 				if (nla_get_u32(nla) != nh->nh_tclassid)
989 					return 1;
990 			}
991 #endif
992 		}
993 
994 		rtnh = rtnh_next(rtnh, &remaining);
995 	} endfor_nexthops(fi);
996 #endif
997 	return 0;
998 }
999 
1000 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1001 {
1002 	struct nlattr *nla;
1003 	int remaining;
1004 
1005 	if (!cfg->fc_mx)
1006 		return true;
1007 
1008 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1009 		int type = nla_type(nla);
1010 		u32 fi_val, val;
1011 
1012 		if (!type)
1013 			continue;
1014 		if (type > RTAX_MAX)
1015 			return false;
1016 
1017 		if (type == RTAX_CC_ALGO) {
1018 			char tmp[TCP_CA_NAME_MAX];
1019 			bool ecn_ca = false;
1020 
1021 			nla_strscpy(tmp, nla, sizeof(tmp));
1022 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1023 		} else {
1024 			if (nla_len(nla) != sizeof(u32))
1025 				return false;
1026 			val = nla_get_u32(nla);
1027 		}
1028 
1029 		fi_val = fi->fib_metrics->metrics[type - 1];
1030 		if (type == RTAX_FEATURES)
1031 			fi_val &= ~DST_FEATURE_ECN_CA;
1032 
1033 		if (fi_val != val)
1034 			return false;
1035 	}
1036 
1037 	return true;
1038 }
1039 
1040 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1041 			      u32 table, struct netlink_ext_ack *extack)
1042 {
1043 	struct fib6_config cfg = {
1044 		.fc_table = table,
1045 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1046 		.fc_ifindex = nh->fib_nh_oif,
1047 		.fc_gateway = nh->fib_nh_gw6,
1048 	};
1049 	struct fib6_nh fib6_nh = {};
1050 	int err;
1051 
1052 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1053 	if (!err) {
1054 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1055 		dev_hold_track(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_KERNEL);
1056 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1057 		nh->fib_nh_scope = RT_SCOPE_LINK;
1058 
1059 		ipv6_stub->fib6_nh_release(&fib6_nh);
1060 	}
1061 
1062 	return err;
1063 }
1064 
1065 /*
1066  * Picture
1067  * -------
1068  *
1069  * Semantics of nexthop is very messy by historical reasons.
1070  * We have to take into account, that:
1071  * a) gateway can be actually local interface address,
1072  *    so that gatewayed route is direct.
1073  * b) gateway must be on-link address, possibly
1074  *    described not by an ifaddr, but also by a direct route.
1075  * c) If both gateway and interface are specified, they should not
1076  *    contradict.
1077  * d) If we use tunnel routes, gateway could be not on-link.
1078  *
1079  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1080  * results in pretty ugly and hairy code with obscure logic.
1081  *
1082  * I chose to generalized it instead, so that the size
1083  * of code does not increase practically, but it becomes
1084  * much more general.
1085  * Every prefix is assigned a "scope" value: "host" is local address,
1086  * "link" is direct route,
1087  * [ ... "site" ... "interior" ... ]
1088  * and "universe" is true gateway route with global meaning.
1089  *
1090  * Every prefix refers to a set of "nexthop"s (gw, oif),
1091  * where gw must have narrower scope. This recursion stops
1092  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1093  * which means that gw is forced to be on link.
1094  *
1095  * Code is still hairy, but now it is apparently logically
1096  * consistent and very flexible. F.e. as by-product it allows
1097  * to co-exists in peace independent exterior and interior
1098  * routing processes.
1099  *
1100  * Normally it looks as following.
1101  *
1102  * {universe prefix}  -> (gw, oif) [scope link]
1103  *		  |
1104  *		  |-> {link prefix} -> (gw, oif) [scope local]
1105  *					|
1106  *					|-> {local prefix} (terminal node)
1107  */
1108 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1109 			      u8 scope, struct netlink_ext_ack *extack)
1110 {
1111 	struct net_device *dev;
1112 	struct fib_result res;
1113 	int err = 0;
1114 
1115 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1116 		unsigned int addr_type;
1117 
1118 		if (scope >= RT_SCOPE_LINK) {
1119 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1120 			return -EINVAL;
1121 		}
1122 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1123 		if (!dev) {
1124 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1125 			return -ENODEV;
1126 		}
1127 		if (!(dev->flags & IFF_UP)) {
1128 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1129 			return -ENETDOWN;
1130 		}
1131 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1132 		if (addr_type != RTN_UNICAST) {
1133 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1134 			return -EINVAL;
1135 		}
1136 		if (!netif_carrier_ok(dev))
1137 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1138 		nh->fib_nh_dev = dev;
1139 		dev_hold_track(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1140 		nh->fib_nh_scope = RT_SCOPE_LINK;
1141 		return 0;
1142 	}
1143 	rcu_read_lock();
1144 	{
1145 		struct fib_table *tbl = NULL;
1146 		struct flowi4 fl4 = {
1147 			.daddr = nh->fib_nh_gw4,
1148 			.flowi4_scope = scope + 1,
1149 			.flowi4_oif = nh->fib_nh_oif,
1150 			.flowi4_iif = LOOPBACK_IFINDEX,
1151 		};
1152 
1153 		/* It is not necessary, but requires a bit of thinking */
1154 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1155 			fl4.flowi4_scope = RT_SCOPE_LINK;
1156 
1157 		if (table && table != RT_TABLE_MAIN)
1158 			tbl = fib_get_table(net, table);
1159 
1160 		if (tbl)
1161 			err = fib_table_lookup(tbl, &fl4, &res,
1162 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1163 					       FIB_LOOKUP_NOREF);
1164 
1165 		/* on error or if no table given do full lookup. This
1166 		 * is needed for example when nexthops are in the local
1167 		 * table rather than the given table
1168 		 */
1169 		if (!tbl || err) {
1170 			err = fib_lookup(net, &fl4, &res,
1171 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1172 		}
1173 
1174 		if (err) {
1175 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1176 			goto out;
1177 		}
1178 	}
1179 
1180 	err = -EINVAL;
1181 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1182 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1183 		goto out;
1184 	}
1185 	nh->fib_nh_scope = res.scope;
1186 	nh->fib_nh_oif = FIB_RES_OIF(res);
1187 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1188 	if (!dev) {
1189 		NL_SET_ERR_MSG(extack,
1190 			       "No egress device for nexthop gateway");
1191 		goto out;
1192 	}
1193 	dev_hold_track(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1194 	if (!netif_carrier_ok(dev))
1195 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1196 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1197 out:
1198 	rcu_read_unlock();
1199 	return err;
1200 }
1201 
1202 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1203 			      struct netlink_ext_ack *extack)
1204 {
1205 	struct in_device *in_dev;
1206 	int err;
1207 
1208 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1209 		NL_SET_ERR_MSG(extack,
1210 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1211 		return -EINVAL;
1212 	}
1213 
1214 	rcu_read_lock();
1215 
1216 	err = -ENODEV;
1217 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1218 	if (!in_dev)
1219 		goto out;
1220 	err = -ENETDOWN;
1221 	if (!(in_dev->dev->flags & IFF_UP)) {
1222 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1223 		goto out;
1224 	}
1225 
1226 	nh->fib_nh_dev = in_dev->dev;
1227 	dev_hold_track(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1228 	nh->fib_nh_scope = RT_SCOPE_HOST;
1229 	if (!netif_carrier_ok(nh->fib_nh_dev))
1230 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1231 	err = 0;
1232 out:
1233 	rcu_read_unlock();
1234 	return err;
1235 }
1236 
1237 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1238 		 struct netlink_ext_ack *extack)
1239 {
1240 	int err;
1241 
1242 	if (nh->fib_nh_gw_family == AF_INET)
1243 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1244 	else if (nh->fib_nh_gw_family == AF_INET6)
1245 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1246 	else
1247 		err = fib_check_nh_nongw(net, nh, extack);
1248 
1249 	return err;
1250 }
1251 
1252 static struct hlist_head *
1253 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1254 {
1255 	u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1256 			   fib_info_hash_bits);
1257 
1258 	return &fib_info_laddrhash[slot];
1259 }
1260 
1261 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1262 			       struct hlist_head *new_laddrhash,
1263 			       unsigned int new_size)
1264 {
1265 	struct hlist_head *old_info_hash, *old_laddrhash;
1266 	unsigned int old_size = fib_info_hash_size;
1267 	unsigned int i;
1268 
1269 	spin_lock_bh(&fib_info_lock);
1270 	old_info_hash = fib_info_hash;
1271 	old_laddrhash = fib_info_laddrhash;
1272 	fib_info_hash_size = new_size;
1273 	fib_info_hash_bits = ilog2(new_size);
1274 
1275 	for (i = 0; i < old_size; i++) {
1276 		struct hlist_head *head = &fib_info_hash[i];
1277 		struct hlist_node *n;
1278 		struct fib_info *fi;
1279 
1280 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1281 			struct hlist_head *dest;
1282 			unsigned int new_hash;
1283 
1284 			new_hash = fib_info_hashfn(fi);
1285 			dest = &new_info_hash[new_hash];
1286 			hlist_add_head(&fi->fib_hash, dest);
1287 		}
1288 	}
1289 	fib_info_hash = new_info_hash;
1290 
1291 	fib_info_laddrhash = new_laddrhash;
1292 	for (i = 0; i < old_size; i++) {
1293 		struct hlist_head *lhead = &old_laddrhash[i];
1294 		struct hlist_node *n;
1295 		struct fib_info *fi;
1296 
1297 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1298 			struct hlist_head *ldest;
1299 
1300 			ldest = fib_info_laddrhash_bucket(fi->fib_net,
1301 							  fi->fib_prefsrc);
1302 			hlist_add_head(&fi->fib_lhash, ldest);
1303 		}
1304 	}
1305 
1306 	spin_unlock_bh(&fib_info_lock);
1307 
1308 	kvfree(old_info_hash);
1309 	kvfree(old_laddrhash);
1310 }
1311 
1312 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1313 				 unsigned char scope)
1314 {
1315 	struct fib_nh *nh;
1316 
1317 	if (nhc->nhc_family != AF_INET)
1318 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1319 
1320 	nh = container_of(nhc, struct fib_nh, nh_common);
1321 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1322 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1323 
1324 	return nh->nh_saddr;
1325 }
1326 
1327 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1328 {
1329 	struct fib_nh_common *nhc = res->nhc;
1330 
1331 	if (res->fi->fib_prefsrc)
1332 		return res->fi->fib_prefsrc;
1333 
1334 	if (nhc->nhc_family == AF_INET) {
1335 		struct fib_nh *nh;
1336 
1337 		nh = container_of(nhc, struct fib_nh, nh_common);
1338 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1339 			return nh->nh_saddr;
1340 	}
1341 
1342 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1343 }
1344 
1345 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1346 {
1347 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1348 	    fib_prefsrc != cfg->fc_dst) {
1349 		u32 tb_id = cfg->fc_table;
1350 		int rc;
1351 
1352 		if (tb_id == RT_TABLE_MAIN)
1353 			tb_id = RT_TABLE_LOCAL;
1354 
1355 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1356 					  fib_prefsrc, tb_id);
1357 
1358 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1359 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1360 						  fib_prefsrc, RT_TABLE_LOCAL);
1361 		}
1362 
1363 		if (rc != RTN_LOCAL)
1364 			return false;
1365 	}
1366 	return true;
1367 }
1368 
1369 struct fib_info *fib_create_info(struct fib_config *cfg,
1370 				 struct netlink_ext_ack *extack)
1371 {
1372 	int err;
1373 	struct fib_info *fi = NULL;
1374 	struct nexthop *nh = NULL;
1375 	struct fib_info *ofi;
1376 	int nhs = 1;
1377 	struct net *net = cfg->fc_nlinfo.nl_net;
1378 
1379 	if (cfg->fc_type > RTN_MAX)
1380 		goto err_inval;
1381 
1382 	/* Fast check to catch the most weird cases */
1383 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1384 		NL_SET_ERR_MSG(extack, "Invalid scope");
1385 		goto err_inval;
1386 	}
1387 
1388 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1389 		NL_SET_ERR_MSG(extack,
1390 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1391 		goto err_inval;
1392 	}
1393 
1394 	if (cfg->fc_nh_id) {
1395 		if (!cfg->fc_mx) {
1396 			fi = fib_find_info_nh(net, cfg);
1397 			if (fi) {
1398 				refcount_inc(&fi->fib_treeref);
1399 				return fi;
1400 			}
1401 		}
1402 
1403 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1404 		if (!nh) {
1405 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1406 			goto err_inval;
1407 		}
1408 		nhs = 0;
1409 	}
1410 
1411 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1412 	if (cfg->fc_mp) {
1413 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1414 		if (nhs == 0)
1415 			goto err_inval;
1416 	}
1417 #endif
1418 
1419 	err = -ENOBUFS;
1420 
1421 	/* Paired with WRITE_ONCE() in fib_release_info() */
1422 	if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1423 		unsigned int new_size = fib_info_hash_size << 1;
1424 		struct hlist_head *new_info_hash;
1425 		struct hlist_head *new_laddrhash;
1426 		size_t bytes;
1427 
1428 		if (!new_size)
1429 			new_size = 16;
1430 		bytes = (size_t)new_size * sizeof(struct hlist_head *);
1431 		new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1432 		new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1433 		if (!new_info_hash || !new_laddrhash) {
1434 			kvfree(new_info_hash);
1435 			kvfree(new_laddrhash);
1436 		} else {
1437 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1438 		}
1439 		if (!fib_info_hash_size)
1440 			goto failure;
1441 	}
1442 
1443 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1444 	if (!fi)
1445 		goto failure;
1446 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1447 					      cfg->fc_mx_len, extack);
1448 	if (IS_ERR(fi->fib_metrics)) {
1449 		err = PTR_ERR(fi->fib_metrics);
1450 		kfree(fi);
1451 		return ERR_PTR(err);
1452 	}
1453 
1454 	fi->fib_net = net;
1455 	fi->fib_protocol = cfg->fc_protocol;
1456 	fi->fib_scope = cfg->fc_scope;
1457 	fi->fib_flags = cfg->fc_flags;
1458 	fi->fib_priority = cfg->fc_priority;
1459 	fi->fib_prefsrc = cfg->fc_prefsrc;
1460 	fi->fib_type = cfg->fc_type;
1461 	fi->fib_tb_id = cfg->fc_table;
1462 
1463 	fi->fib_nhs = nhs;
1464 	if (nh) {
1465 		if (!nexthop_get(nh)) {
1466 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1467 			err = -EINVAL;
1468 		} else {
1469 			err = 0;
1470 			fi->nh = nh;
1471 		}
1472 	} else {
1473 		change_nexthops(fi) {
1474 			nexthop_nh->nh_parent = fi;
1475 		} endfor_nexthops(fi)
1476 
1477 		if (cfg->fc_mp)
1478 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1479 					  extack);
1480 		else
1481 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1482 	}
1483 
1484 	if (err != 0)
1485 		goto failure;
1486 
1487 	if (fib_props[cfg->fc_type].error) {
1488 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1489 			NL_SET_ERR_MSG(extack,
1490 				       "Gateway, device and multipath can not be specified for this route type");
1491 			goto err_inval;
1492 		}
1493 		goto link_it;
1494 	} else {
1495 		switch (cfg->fc_type) {
1496 		case RTN_UNICAST:
1497 		case RTN_LOCAL:
1498 		case RTN_BROADCAST:
1499 		case RTN_ANYCAST:
1500 		case RTN_MULTICAST:
1501 			break;
1502 		default:
1503 			NL_SET_ERR_MSG(extack, "Invalid route type");
1504 			goto err_inval;
1505 		}
1506 	}
1507 
1508 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1509 		NL_SET_ERR_MSG(extack, "Invalid scope");
1510 		goto err_inval;
1511 	}
1512 
1513 	if (fi->nh) {
1514 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1515 		if (err)
1516 			goto failure;
1517 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1518 		struct fib_nh *nh = fi->fib_nh;
1519 
1520 		/* Local address is added. */
1521 		if (nhs != 1) {
1522 			NL_SET_ERR_MSG(extack,
1523 				       "Route with host scope can not have multiple nexthops");
1524 			goto err_inval;
1525 		}
1526 		if (nh->fib_nh_gw_family) {
1527 			NL_SET_ERR_MSG(extack,
1528 				       "Route with host scope can not have a gateway");
1529 			goto err_inval;
1530 		}
1531 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1532 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1533 		err = -ENODEV;
1534 		if (!nh->fib_nh_dev)
1535 			goto failure;
1536 		netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1537 				     GFP_KERNEL);
1538 	} else {
1539 		int linkdown = 0;
1540 
1541 		change_nexthops(fi) {
1542 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1543 					   cfg->fc_table, cfg->fc_scope,
1544 					   extack);
1545 			if (err != 0)
1546 				goto failure;
1547 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1548 				linkdown++;
1549 		} endfor_nexthops(fi)
1550 		if (linkdown == fi->fib_nhs)
1551 			fi->fib_flags |= RTNH_F_LINKDOWN;
1552 	}
1553 
1554 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1555 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1556 		goto err_inval;
1557 	}
1558 
1559 	if (!fi->nh) {
1560 		change_nexthops(fi) {
1561 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1562 						  fi->fib_scope);
1563 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1564 				fi->fib_nh_is_v6 = true;
1565 		} endfor_nexthops(fi)
1566 
1567 		fib_rebalance(fi);
1568 	}
1569 
1570 link_it:
1571 	ofi = fib_find_info(fi);
1572 	if (ofi) {
1573 		fi->fib_dead = 1;
1574 		free_fib_info(fi);
1575 		refcount_inc(&ofi->fib_treeref);
1576 		return ofi;
1577 	}
1578 
1579 	refcount_set(&fi->fib_treeref, 1);
1580 	refcount_set(&fi->fib_clntref, 1);
1581 	spin_lock_bh(&fib_info_lock);
1582 	fib_info_cnt++;
1583 	hlist_add_head(&fi->fib_hash,
1584 		       &fib_info_hash[fib_info_hashfn(fi)]);
1585 	if (fi->fib_prefsrc) {
1586 		struct hlist_head *head;
1587 
1588 		head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1589 		hlist_add_head(&fi->fib_lhash, head);
1590 	}
1591 	if (fi->nh) {
1592 		list_add(&fi->nh_list, &nh->fi_list);
1593 	} else {
1594 		change_nexthops(fi) {
1595 			struct hlist_head *head;
1596 
1597 			if (!nexthop_nh->fib_nh_dev)
1598 				continue;
1599 			head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1600 			hlist_add_head(&nexthop_nh->nh_hash, head);
1601 		} endfor_nexthops(fi)
1602 	}
1603 	spin_unlock_bh(&fib_info_lock);
1604 	return fi;
1605 
1606 err_inval:
1607 	err = -EINVAL;
1608 
1609 failure:
1610 	if (fi) {
1611 		fi->fib_dead = 1;
1612 		free_fib_info(fi);
1613 	}
1614 
1615 	return ERR_PTR(err);
1616 }
1617 
1618 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1619 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1620 {
1621 	if (nhc->nhc_flags & RTNH_F_DEAD)
1622 		*flags |= RTNH_F_DEAD;
1623 
1624 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1625 		*flags |= RTNH_F_LINKDOWN;
1626 
1627 		rcu_read_lock();
1628 		switch (nhc->nhc_family) {
1629 		case AF_INET:
1630 			if (ip_ignore_linkdown(nhc->nhc_dev))
1631 				*flags |= RTNH_F_DEAD;
1632 			break;
1633 		case AF_INET6:
1634 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1635 				*flags |= RTNH_F_DEAD;
1636 			break;
1637 		}
1638 		rcu_read_unlock();
1639 	}
1640 
1641 	switch (nhc->nhc_gw_family) {
1642 	case AF_INET:
1643 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1644 			goto nla_put_failure;
1645 		break;
1646 	case AF_INET6:
1647 		/* if gateway family does not match nexthop family
1648 		 * gateway is encoded as RTA_VIA
1649 		 */
1650 		if (rt_family != nhc->nhc_gw_family) {
1651 			int alen = sizeof(struct in6_addr);
1652 			struct nlattr *nla;
1653 			struct rtvia *via;
1654 
1655 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1656 			if (!nla)
1657 				goto nla_put_failure;
1658 
1659 			via = nla_data(nla);
1660 			via->rtvia_family = AF_INET6;
1661 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1662 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1663 					    &nhc->nhc_gw.ipv6) < 0) {
1664 			goto nla_put_failure;
1665 		}
1666 		break;
1667 	}
1668 
1669 	*flags |= (nhc->nhc_flags &
1670 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1671 
1672 	if (!skip_oif && nhc->nhc_dev &&
1673 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1674 		goto nla_put_failure;
1675 
1676 	if (nhc->nhc_lwtstate &&
1677 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1678 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1679 		goto nla_put_failure;
1680 
1681 	return 0;
1682 
1683 nla_put_failure:
1684 	return -EMSGSIZE;
1685 }
1686 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1687 
1688 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1689 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1690 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1691 {
1692 	const struct net_device *dev = nhc->nhc_dev;
1693 	struct rtnexthop *rtnh;
1694 	unsigned char flags = 0;
1695 
1696 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1697 	if (!rtnh)
1698 		goto nla_put_failure;
1699 
1700 	rtnh->rtnh_hops = nh_weight - 1;
1701 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1702 
1703 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1704 		goto nla_put_failure;
1705 
1706 	rtnh->rtnh_flags = flags;
1707 
1708 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1709 		goto nla_put_failure;
1710 
1711 	/* length of rtnetlink header + attributes */
1712 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1713 
1714 	return 0;
1715 
1716 nla_put_failure:
1717 	return -EMSGSIZE;
1718 }
1719 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1720 #endif
1721 
1722 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1723 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1724 {
1725 	struct nlattr *mp;
1726 
1727 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1728 	if (!mp)
1729 		goto nla_put_failure;
1730 
1731 	if (unlikely(fi->nh)) {
1732 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1733 			goto nla_put_failure;
1734 		goto mp_end;
1735 	}
1736 
1737 	for_nexthops(fi) {
1738 		u32 nh_tclassid = 0;
1739 #ifdef CONFIG_IP_ROUTE_CLASSID
1740 		nh_tclassid = nh->nh_tclassid;
1741 #endif
1742 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1743 				    AF_INET, nh_tclassid) < 0)
1744 			goto nla_put_failure;
1745 	} endfor_nexthops(fi);
1746 
1747 mp_end:
1748 	nla_nest_end(skb, mp);
1749 
1750 	return 0;
1751 
1752 nla_put_failure:
1753 	return -EMSGSIZE;
1754 }
1755 #else
1756 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1757 {
1758 	return 0;
1759 }
1760 #endif
1761 
1762 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1763 		  const struct fib_rt_info *fri, unsigned int flags)
1764 {
1765 	unsigned int nhs = fib_info_num_path(fri->fi);
1766 	struct fib_info *fi = fri->fi;
1767 	u32 tb_id = fri->tb_id;
1768 	struct nlmsghdr *nlh;
1769 	struct rtmsg *rtm;
1770 
1771 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1772 	if (!nlh)
1773 		return -EMSGSIZE;
1774 
1775 	rtm = nlmsg_data(nlh);
1776 	rtm->rtm_family = AF_INET;
1777 	rtm->rtm_dst_len = fri->dst_len;
1778 	rtm->rtm_src_len = 0;
1779 	rtm->rtm_tos = fri->tos;
1780 	if (tb_id < 256)
1781 		rtm->rtm_table = tb_id;
1782 	else
1783 		rtm->rtm_table = RT_TABLE_COMPAT;
1784 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1785 		goto nla_put_failure;
1786 	rtm->rtm_type = fri->type;
1787 	rtm->rtm_flags = fi->fib_flags;
1788 	rtm->rtm_scope = fi->fib_scope;
1789 	rtm->rtm_protocol = fi->fib_protocol;
1790 
1791 	if (rtm->rtm_dst_len &&
1792 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1793 		goto nla_put_failure;
1794 	if (fi->fib_priority &&
1795 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1796 		goto nla_put_failure;
1797 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1798 		goto nla_put_failure;
1799 
1800 	if (fi->fib_prefsrc &&
1801 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1802 		goto nla_put_failure;
1803 
1804 	if (fi->nh) {
1805 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1806 			goto nla_put_failure;
1807 		if (nexthop_is_blackhole(fi->nh))
1808 			rtm->rtm_type = RTN_BLACKHOLE;
1809 		if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode)
1810 			goto offload;
1811 	}
1812 
1813 	if (nhs == 1) {
1814 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1815 		unsigned char flags = 0;
1816 
1817 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1818 			goto nla_put_failure;
1819 
1820 		rtm->rtm_flags = flags;
1821 #ifdef CONFIG_IP_ROUTE_CLASSID
1822 		if (nhc->nhc_family == AF_INET) {
1823 			struct fib_nh *nh;
1824 
1825 			nh = container_of(nhc, struct fib_nh, nh_common);
1826 			if (nh->nh_tclassid &&
1827 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1828 				goto nla_put_failure;
1829 		}
1830 #endif
1831 	} else {
1832 		if (fib_add_multipath(skb, fi) < 0)
1833 			goto nla_put_failure;
1834 	}
1835 
1836 offload:
1837 	if (fri->offload)
1838 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1839 	if (fri->trap)
1840 		rtm->rtm_flags |= RTM_F_TRAP;
1841 	if (fri->offload_failed)
1842 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1843 
1844 	nlmsg_end(skb, nlh);
1845 	return 0;
1846 
1847 nla_put_failure:
1848 	nlmsg_cancel(skb, nlh);
1849 	return -EMSGSIZE;
1850 }
1851 
1852 /*
1853  * Update FIB if:
1854  * - local address disappeared -> we must delete all the entries
1855  *   referring to it.
1856  * - device went down -> we must shutdown all nexthops going via it.
1857  */
1858 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1859 {
1860 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1861 	struct net *net = dev_net(dev);
1862 	struct hlist_head *head;
1863 	struct fib_info *fi;
1864 	int ret = 0;
1865 
1866 	if (!fib_info_laddrhash || local == 0)
1867 		return 0;
1868 
1869 	head = fib_info_laddrhash_bucket(net, local);
1870 	hlist_for_each_entry(fi, head, fib_lhash) {
1871 		if (!net_eq(fi->fib_net, net) ||
1872 		    fi->fib_tb_id != tb_id)
1873 			continue;
1874 		if (fi->fib_prefsrc == local) {
1875 			fi->fib_flags |= RTNH_F_DEAD;
1876 			ret++;
1877 		}
1878 	}
1879 	return ret;
1880 }
1881 
1882 static int call_fib_nh_notifiers(struct fib_nh *nh,
1883 				 enum fib_event_type event_type)
1884 {
1885 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1886 	struct fib_nh_notifier_info info = {
1887 		.fib_nh = nh,
1888 	};
1889 
1890 	switch (event_type) {
1891 	case FIB_EVENT_NH_ADD:
1892 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1893 			break;
1894 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1895 			break;
1896 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1897 					   &info.info);
1898 	case FIB_EVENT_NH_DEL:
1899 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1900 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1901 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1902 						   event_type, &info.info);
1903 		break;
1904 	default:
1905 		break;
1906 	}
1907 
1908 	return NOTIFY_DONE;
1909 }
1910 
1911 /* Update the PMTU of exceptions when:
1912  * - the new MTU of the first hop becomes smaller than the PMTU
1913  * - the old MTU was the same as the PMTU, and it limited discovery of
1914  *   larger MTUs on the path. With that limit raised, we can now
1915  *   discover larger MTUs
1916  * A special case is locked exceptions, for which the PMTU is smaller
1917  * than the minimal accepted PMTU:
1918  * - if the new MTU is greater than the PMTU, don't make any change
1919  * - otherwise, unlock and set PMTU
1920  */
1921 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1922 {
1923 	struct fnhe_hash_bucket *bucket;
1924 	int i;
1925 
1926 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1927 	if (!bucket)
1928 		return;
1929 
1930 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1931 		struct fib_nh_exception *fnhe;
1932 
1933 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1934 		     fnhe;
1935 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1936 			if (fnhe->fnhe_mtu_locked) {
1937 				if (new <= fnhe->fnhe_pmtu) {
1938 					fnhe->fnhe_pmtu = new;
1939 					fnhe->fnhe_mtu_locked = false;
1940 				}
1941 			} else if (new < fnhe->fnhe_pmtu ||
1942 				   orig == fnhe->fnhe_pmtu) {
1943 				fnhe->fnhe_pmtu = new;
1944 			}
1945 		}
1946 	}
1947 }
1948 
1949 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1950 {
1951 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1952 	struct fib_nh *nh;
1953 
1954 	hlist_for_each_entry(nh, head, nh_hash) {
1955 		if (nh->fib_nh_dev == dev)
1956 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1957 	}
1958 }
1959 
1960 /* Event              force Flags           Description
1961  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1962  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1963  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1964  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1965  *
1966  * only used when fib_nh is built into fib_info
1967  */
1968 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1969 {
1970 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1971 	struct fib_info *prev_fi = NULL;
1972 	int scope = RT_SCOPE_NOWHERE;
1973 	struct fib_nh *nh;
1974 	int ret = 0;
1975 
1976 	if (force)
1977 		scope = -1;
1978 
1979 	hlist_for_each_entry(nh, head, nh_hash) {
1980 		struct fib_info *fi = nh->nh_parent;
1981 		int dead;
1982 
1983 		BUG_ON(!fi->fib_nhs);
1984 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1985 			continue;
1986 		prev_fi = fi;
1987 		dead = 0;
1988 		change_nexthops(fi) {
1989 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1990 				dead++;
1991 			else if (nexthop_nh->fib_nh_dev == dev &&
1992 				 nexthop_nh->fib_nh_scope != scope) {
1993 				switch (event) {
1994 				case NETDEV_DOWN:
1995 				case NETDEV_UNREGISTER:
1996 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1997 					fallthrough;
1998 				case NETDEV_CHANGE:
1999 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2000 					break;
2001 				}
2002 				call_fib_nh_notifiers(nexthop_nh,
2003 						      FIB_EVENT_NH_DEL);
2004 				dead++;
2005 			}
2006 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2007 			if (event == NETDEV_UNREGISTER &&
2008 			    nexthop_nh->fib_nh_dev == dev) {
2009 				dead = fi->fib_nhs;
2010 				break;
2011 			}
2012 #endif
2013 		} endfor_nexthops(fi)
2014 		if (dead == fi->fib_nhs) {
2015 			switch (event) {
2016 			case NETDEV_DOWN:
2017 			case NETDEV_UNREGISTER:
2018 				fi->fib_flags |= RTNH_F_DEAD;
2019 				fallthrough;
2020 			case NETDEV_CHANGE:
2021 				fi->fib_flags |= RTNH_F_LINKDOWN;
2022 				break;
2023 			}
2024 			ret++;
2025 		}
2026 
2027 		fib_rebalance(fi);
2028 	}
2029 
2030 	return ret;
2031 }
2032 
2033 /* Must be invoked inside of an RCU protected region.  */
2034 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2035 {
2036 	struct fib_info *fi = NULL, *last_resort = NULL;
2037 	struct hlist_head *fa_head = res->fa_head;
2038 	struct fib_table *tb = res->table;
2039 	u8 slen = 32 - res->prefixlen;
2040 	int order = -1, last_idx = -1;
2041 	struct fib_alias *fa, *fa1 = NULL;
2042 	u32 last_prio = res->fi->fib_priority;
2043 	dscp_t last_dscp = 0;
2044 
2045 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2046 		struct fib_info *next_fi = fa->fa_info;
2047 		struct fib_nh_common *nhc;
2048 
2049 		if (fa->fa_slen != slen)
2050 			continue;
2051 		if (fa->fa_dscp &&
2052 		    fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2053 			continue;
2054 		if (fa->tb_id != tb->tb_id)
2055 			continue;
2056 		if (next_fi->fib_priority > last_prio &&
2057 		    fa->fa_dscp == last_dscp) {
2058 			if (last_dscp)
2059 				continue;
2060 			break;
2061 		}
2062 		if (next_fi->fib_flags & RTNH_F_DEAD)
2063 			continue;
2064 		last_dscp = fa->fa_dscp;
2065 		last_prio = next_fi->fib_priority;
2066 
2067 		if (next_fi->fib_scope != res->scope ||
2068 		    fa->fa_type != RTN_UNICAST)
2069 			continue;
2070 
2071 		nhc = fib_info_nhc(next_fi, 0);
2072 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2073 			continue;
2074 
2075 		fib_alias_accessed(fa);
2076 
2077 		if (!fi) {
2078 			if (next_fi != res->fi)
2079 				break;
2080 			fa1 = fa;
2081 		} else if (!fib_detect_death(fi, order, &last_resort,
2082 					     &last_idx, fa1->fa_default)) {
2083 			fib_result_assign(res, fi);
2084 			fa1->fa_default = order;
2085 			goto out;
2086 		}
2087 		fi = next_fi;
2088 		order++;
2089 	}
2090 
2091 	if (order <= 0 || !fi) {
2092 		if (fa1)
2093 			fa1->fa_default = -1;
2094 		goto out;
2095 	}
2096 
2097 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2098 			      fa1->fa_default)) {
2099 		fib_result_assign(res, fi);
2100 		fa1->fa_default = order;
2101 		goto out;
2102 	}
2103 
2104 	if (last_idx >= 0)
2105 		fib_result_assign(res, last_resort);
2106 	fa1->fa_default = last_idx;
2107 out:
2108 	return;
2109 }
2110 
2111 /*
2112  * Dead device goes up. We wake up dead nexthops.
2113  * It takes sense only on multipath routes.
2114  *
2115  * only used when fib_nh is built into fib_info
2116  */
2117 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2118 {
2119 	struct fib_info *prev_fi;
2120 	struct hlist_head *head;
2121 	struct fib_nh *nh;
2122 	int ret;
2123 
2124 	if (!(dev->flags & IFF_UP))
2125 		return 0;
2126 
2127 	if (nh_flags & RTNH_F_DEAD) {
2128 		unsigned int flags = dev_get_flags(dev);
2129 
2130 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2131 			nh_flags |= RTNH_F_LINKDOWN;
2132 	}
2133 
2134 	prev_fi = NULL;
2135 	head = fib_info_devhash_bucket(dev);
2136 	ret = 0;
2137 
2138 	hlist_for_each_entry(nh, head, nh_hash) {
2139 		struct fib_info *fi = nh->nh_parent;
2140 		int alive;
2141 
2142 		BUG_ON(!fi->fib_nhs);
2143 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2144 			continue;
2145 
2146 		prev_fi = fi;
2147 		alive = 0;
2148 		change_nexthops(fi) {
2149 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2150 				alive++;
2151 				continue;
2152 			}
2153 			if (!nexthop_nh->fib_nh_dev ||
2154 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2155 				continue;
2156 			if (nexthop_nh->fib_nh_dev != dev ||
2157 			    !__in_dev_get_rtnl(dev))
2158 				continue;
2159 			alive++;
2160 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2161 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2162 		} endfor_nexthops(fi)
2163 
2164 		if (alive > 0) {
2165 			fi->fib_flags &= ~nh_flags;
2166 			ret++;
2167 		}
2168 
2169 		fib_rebalance(fi);
2170 	}
2171 
2172 	return ret;
2173 }
2174 
2175 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2176 static bool fib_good_nh(const struct fib_nh *nh)
2177 {
2178 	int state = NUD_REACHABLE;
2179 
2180 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2181 		struct neighbour *n;
2182 
2183 		rcu_read_lock_bh();
2184 
2185 		if (likely(nh->fib_nh_gw_family == AF_INET))
2186 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2187 						   (__force u32)nh->fib_nh_gw4);
2188 		else if (nh->fib_nh_gw_family == AF_INET6)
2189 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2190 							   &nh->fib_nh_gw6);
2191 		else
2192 			n = NULL;
2193 		if (n)
2194 			state = n->nud_state;
2195 
2196 		rcu_read_unlock_bh();
2197 	}
2198 
2199 	return !!(state & NUD_VALID);
2200 }
2201 
2202 void fib_select_multipath(struct fib_result *res, int hash)
2203 {
2204 	struct fib_info *fi = res->fi;
2205 	struct net *net = fi->fib_net;
2206 	bool first = false;
2207 
2208 	if (unlikely(res->fi->nh)) {
2209 		nexthop_path_fib_result(res, hash);
2210 		return;
2211 	}
2212 
2213 	change_nexthops(fi) {
2214 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2215 			if (!fib_good_nh(nexthop_nh))
2216 				continue;
2217 			if (!first) {
2218 				res->nh_sel = nhsel;
2219 				res->nhc = &nexthop_nh->nh_common;
2220 				first = true;
2221 			}
2222 		}
2223 
2224 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2225 			continue;
2226 
2227 		res->nh_sel = nhsel;
2228 		res->nhc = &nexthop_nh->nh_common;
2229 		return;
2230 	} endfor_nexthops(fi);
2231 }
2232 #endif
2233 
2234 void fib_select_path(struct net *net, struct fib_result *res,
2235 		     struct flowi4 *fl4, const struct sk_buff *skb)
2236 {
2237 	if (fl4->flowi4_oif)
2238 		goto check_saddr;
2239 
2240 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2241 	if (fib_info_num_path(res->fi) > 1) {
2242 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2243 
2244 		fib_select_multipath(res, h);
2245 	}
2246 	else
2247 #endif
2248 	if (!res->prefixlen &&
2249 	    res->table->tb_num_default > 1 &&
2250 	    res->type == RTN_UNICAST)
2251 		fib_select_default(fl4, res);
2252 
2253 check_saddr:
2254 	if (!fl4->saddr)
2255 		fl4->saddr = fib_result_prefsrc(net, res);
2256 }
2257