xref: /linux/net/ipv4/fib_semantics.c (revision 722ecdbce68a87de2d9296f91308f44ea900a039)
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.dscp = 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;
893 
894 		/* cannot match on nexthop object attributes */
895 		if (fi->nh)
896 			return 1;
897 
898 		nh = fib_info_nh(fi, 0);
899 		if (cfg->fc_encap) {
900 			if (fib_encap_match(net, cfg->fc_encap_type,
901 					    cfg->fc_encap, nh, cfg, extack))
902 				return 1;
903 		}
904 #ifdef CONFIG_IP_ROUTE_CLASSID
905 		if (cfg->fc_flow &&
906 		    cfg->fc_flow != nh->nh_tclassid)
907 			return 1;
908 #endif
909 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
910 		    (cfg->fc_gw_family &&
911 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
912 			return 1;
913 
914 		if (cfg->fc_gw_family == AF_INET &&
915 		    cfg->fc_gw4 != nh->fib_nh_gw4)
916 			return 1;
917 
918 		if (cfg->fc_gw_family == AF_INET6 &&
919 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
920 			return 1;
921 
922 		return 0;
923 	}
924 
925 #ifdef CONFIG_IP_ROUTE_MULTIPATH
926 	if (!cfg->fc_mp)
927 		return 0;
928 
929 	rtnh = cfg->fc_mp;
930 	remaining = cfg->fc_mp_len;
931 
932 	for_nexthops(fi) {
933 		int attrlen;
934 
935 		if (!rtnh_ok(rtnh, remaining))
936 			return -EINVAL;
937 
938 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
939 			return 1;
940 
941 		attrlen = rtnh_attrlen(rtnh);
942 		if (attrlen > 0) {
943 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
944 			int err;
945 
946 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
947 			nlav = nla_find(attrs, attrlen, RTA_VIA);
948 			if (nla && nlav) {
949 				NL_SET_ERR_MSG(extack,
950 					       "Nexthop configuration can not contain both GATEWAY and VIA");
951 				return -EINVAL;
952 			}
953 
954 			if (nla) {
955 				__be32 gw;
956 
957 				err = fib_gw_from_attr(&gw, nla, extack);
958 				if (err)
959 					return err;
960 
961 				if (nh->fib_nh_gw_family != AF_INET ||
962 				    gw != nh->fib_nh_gw4)
963 					return 1;
964 			} else if (nlav) {
965 				struct fib_config cfg2;
966 
967 				err = fib_gw_from_via(&cfg2, nlav, extack);
968 				if (err)
969 					return err;
970 
971 				switch (nh->fib_nh_gw_family) {
972 				case AF_INET:
973 					if (cfg2.fc_gw_family != AF_INET ||
974 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
975 						return 1;
976 					break;
977 				case AF_INET6:
978 					if (cfg2.fc_gw_family != AF_INET6 ||
979 					    ipv6_addr_cmp(&cfg2.fc_gw6,
980 							  &nh->fib_nh_gw6))
981 						return 1;
982 					break;
983 				}
984 			}
985 
986 #ifdef CONFIG_IP_ROUTE_CLASSID
987 			nla = nla_find(attrs, attrlen, RTA_FLOW);
988 			if (nla) {
989 				if (nla_len(nla) < sizeof(u32)) {
990 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
991 					return -EINVAL;
992 				}
993 				if (nla_get_u32(nla) != nh->nh_tclassid)
994 					return 1;
995 			}
996 #endif
997 		}
998 
999 		rtnh = rtnh_next(rtnh, &remaining);
1000 	} endfor_nexthops(fi);
1001 #endif
1002 	return 0;
1003 }
1004 
1005 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1006 {
1007 	struct nlattr *nla;
1008 	int remaining;
1009 
1010 	if (!cfg->fc_mx)
1011 		return true;
1012 
1013 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1014 		int type = nla_type(nla);
1015 		u32 fi_val, val;
1016 
1017 		if (!type)
1018 			continue;
1019 		if (type > RTAX_MAX)
1020 			return false;
1021 
1022 		if (type == RTAX_CC_ALGO) {
1023 			char tmp[TCP_CA_NAME_MAX];
1024 			bool ecn_ca = false;
1025 
1026 			nla_strscpy(tmp, nla, sizeof(tmp));
1027 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1028 		} else {
1029 			if (nla_len(nla) != sizeof(u32))
1030 				return false;
1031 			val = nla_get_u32(nla);
1032 		}
1033 
1034 		fi_val = fi->fib_metrics->metrics[type - 1];
1035 		if (type == RTAX_FEATURES)
1036 			fi_val &= ~DST_FEATURE_ECN_CA;
1037 
1038 		if (fi_val != val)
1039 			return false;
1040 	}
1041 
1042 	return true;
1043 }
1044 
1045 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1046 			      u32 table, struct netlink_ext_ack *extack)
1047 {
1048 	struct fib6_config cfg = {
1049 		.fc_table = table,
1050 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1051 		.fc_ifindex = nh->fib_nh_oif,
1052 		.fc_gateway = nh->fib_nh_gw6,
1053 	};
1054 	struct fib6_nh fib6_nh = {};
1055 	int err;
1056 
1057 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1058 	if (!err) {
1059 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1060 		dev_hold_track(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_KERNEL);
1061 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1062 		nh->fib_nh_scope = RT_SCOPE_LINK;
1063 
1064 		ipv6_stub->fib6_nh_release(&fib6_nh);
1065 	}
1066 
1067 	return err;
1068 }
1069 
1070 /*
1071  * Picture
1072  * -------
1073  *
1074  * Semantics of nexthop is very messy by historical reasons.
1075  * We have to take into account, that:
1076  * a) gateway can be actually local interface address,
1077  *    so that gatewayed route is direct.
1078  * b) gateway must be on-link address, possibly
1079  *    described not by an ifaddr, but also by a direct route.
1080  * c) If both gateway and interface are specified, they should not
1081  *    contradict.
1082  * d) If we use tunnel routes, gateway could be not on-link.
1083  *
1084  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1085  * results in pretty ugly and hairy code with obscure logic.
1086  *
1087  * I chose to generalized it instead, so that the size
1088  * of code does not increase practically, but it becomes
1089  * much more general.
1090  * Every prefix is assigned a "scope" value: "host" is local address,
1091  * "link" is direct route,
1092  * [ ... "site" ... "interior" ... ]
1093  * and "universe" is true gateway route with global meaning.
1094  *
1095  * Every prefix refers to a set of "nexthop"s (gw, oif),
1096  * where gw must have narrower scope. This recursion stops
1097  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1098  * which means that gw is forced to be on link.
1099  *
1100  * Code is still hairy, but now it is apparently logically
1101  * consistent and very flexible. F.e. as by-product it allows
1102  * to co-exists in peace independent exterior and interior
1103  * routing processes.
1104  *
1105  * Normally it looks as following.
1106  *
1107  * {universe prefix}  -> (gw, oif) [scope link]
1108  *		  |
1109  *		  |-> {link prefix} -> (gw, oif) [scope local]
1110  *					|
1111  *					|-> {local prefix} (terminal node)
1112  */
1113 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1114 			      u8 scope, struct netlink_ext_ack *extack)
1115 {
1116 	struct net_device *dev;
1117 	struct fib_result res;
1118 	int err = 0;
1119 
1120 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1121 		unsigned int addr_type;
1122 
1123 		if (scope >= RT_SCOPE_LINK) {
1124 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1125 			return -EINVAL;
1126 		}
1127 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1128 		if (!dev) {
1129 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1130 			return -ENODEV;
1131 		}
1132 		if (!(dev->flags & IFF_UP)) {
1133 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1134 			return -ENETDOWN;
1135 		}
1136 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1137 		if (addr_type != RTN_UNICAST) {
1138 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1139 			return -EINVAL;
1140 		}
1141 		if (!netif_carrier_ok(dev))
1142 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1143 		nh->fib_nh_dev = dev;
1144 		dev_hold_track(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1145 		nh->fib_nh_scope = RT_SCOPE_LINK;
1146 		return 0;
1147 	}
1148 	rcu_read_lock();
1149 	{
1150 		struct fib_table *tbl = NULL;
1151 		struct flowi4 fl4 = {
1152 			.daddr = nh->fib_nh_gw4,
1153 			.flowi4_scope = scope + 1,
1154 			.flowi4_oif = nh->fib_nh_oif,
1155 			.flowi4_iif = LOOPBACK_IFINDEX,
1156 		};
1157 
1158 		/* It is not necessary, but requires a bit of thinking */
1159 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1160 			fl4.flowi4_scope = RT_SCOPE_LINK;
1161 
1162 		if (table && table != RT_TABLE_MAIN)
1163 			tbl = fib_get_table(net, table);
1164 
1165 		if (tbl)
1166 			err = fib_table_lookup(tbl, &fl4, &res,
1167 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1168 					       FIB_LOOKUP_NOREF);
1169 
1170 		/* on error or if no table given do full lookup. This
1171 		 * is needed for example when nexthops are in the local
1172 		 * table rather than the given table
1173 		 */
1174 		if (!tbl || err) {
1175 			err = fib_lookup(net, &fl4, &res,
1176 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1177 		}
1178 
1179 		if (err) {
1180 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1181 			goto out;
1182 		}
1183 	}
1184 
1185 	err = -EINVAL;
1186 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1187 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1188 		goto out;
1189 	}
1190 	nh->fib_nh_scope = res.scope;
1191 	nh->fib_nh_oif = FIB_RES_OIF(res);
1192 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1193 	if (!dev) {
1194 		NL_SET_ERR_MSG(extack,
1195 			       "No egress device for nexthop gateway");
1196 		goto out;
1197 	}
1198 	dev_hold_track(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1199 	if (!netif_carrier_ok(dev))
1200 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1201 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1202 out:
1203 	rcu_read_unlock();
1204 	return err;
1205 }
1206 
1207 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1208 			      struct netlink_ext_ack *extack)
1209 {
1210 	struct in_device *in_dev;
1211 	int err;
1212 
1213 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1214 		NL_SET_ERR_MSG(extack,
1215 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1216 		return -EINVAL;
1217 	}
1218 
1219 	rcu_read_lock();
1220 
1221 	err = -ENODEV;
1222 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1223 	if (!in_dev)
1224 		goto out;
1225 	err = -ENETDOWN;
1226 	if (!(in_dev->dev->flags & IFF_UP)) {
1227 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1228 		goto out;
1229 	}
1230 
1231 	nh->fib_nh_dev = in_dev->dev;
1232 	dev_hold_track(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1233 	nh->fib_nh_scope = RT_SCOPE_HOST;
1234 	if (!netif_carrier_ok(nh->fib_nh_dev))
1235 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1236 	err = 0;
1237 out:
1238 	rcu_read_unlock();
1239 	return err;
1240 }
1241 
1242 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1243 		 struct netlink_ext_ack *extack)
1244 {
1245 	int err;
1246 
1247 	if (nh->fib_nh_gw_family == AF_INET)
1248 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1249 	else if (nh->fib_nh_gw_family == AF_INET6)
1250 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1251 	else
1252 		err = fib_check_nh_nongw(net, nh, extack);
1253 
1254 	return err;
1255 }
1256 
1257 static struct hlist_head *
1258 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1259 {
1260 	u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1261 			   fib_info_hash_bits);
1262 
1263 	return &fib_info_laddrhash[slot];
1264 }
1265 
1266 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1267 			       struct hlist_head *new_laddrhash,
1268 			       unsigned int new_size)
1269 {
1270 	struct hlist_head *old_info_hash, *old_laddrhash;
1271 	unsigned int old_size = fib_info_hash_size;
1272 	unsigned int i;
1273 
1274 	spin_lock_bh(&fib_info_lock);
1275 	old_info_hash = fib_info_hash;
1276 	old_laddrhash = fib_info_laddrhash;
1277 	fib_info_hash_size = new_size;
1278 	fib_info_hash_bits = ilog2(new_size);
1279 
1280 	for (i = 0; i < old_size; i++) {
1281 		struct hlist_head *head = &fib_info_hash[i];
1282 		struct hlist_node *n;
1283 		struct fib_info *fi;
1284 
1285 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1286 			struct hlist_head *dest;
1287 			unsigned int new_hash;
1288 
1289 			new_hash = fib_info_hashfn(fi);
1290 			dest = &new_info_hash[new_hash];
1291 			hlist_add_head(&fi->fib_hash, dest);
1292 		}
1293 	}
1294 	fib_info_hash = new_info_hash;
1295 
1296 	fib_info_laddrhash = new_laddrhash;
1297 	for (i = 0; i < old_size; i++) {
1298 		struct hlist_head *lhead = &old_laddrhash[i];
1299 		struct hlist_node *n;
1300 		struct fib_info *fi;
1301 
1302 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1303 			struct hlist_head *ldest;
1304 
1305 			ldest = fib_info_laddrhash_bucket(fi->fib_net,
1306 							  fi->fib_prefsrc);
1307 			hlist_add_head(&fi->fib_lhash, ldest);
1308 		}
1309 	}
1310 
1311 	spin_unlock_bh(&fib_info_lock);
1312 
1313 	kvfree(old_info_hash);
1314 	kvfree(old_laddrhash);
1315 }
1316 
1317 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1318 				 unsigned char scope)
1319 {
1320 	struct fib_nh *nh;
1321 
1322 	if (nhc->nhc_family != AF_INET)
1323 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1324 
1325 	nh = container_of(nhc, struct fib_nh, nh_common);
1326 	nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1327 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1328 
1329 	return nh->nh_saddr;
1330 }
1331 
1332 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1333 {
1334 	struct fib_nh_common *nhc = res->nhc;
1335 
1336 	if (res->fi->fib_prefsrc)
1337 		return res->fi->fib_prefsrc;
1338 
1339 	if (nhc->nhc_family == AF_INET) {
1340 		struct fib_nh *nh;
1341 
1342 		nh = container_of(nhc, struct fib_nh, nh_common);
1343 		if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1344 			return nh->nh_saddr;
1345 	}
1346 
1347 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1348 }
1349 
1350 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1351 {
1352 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1353 	    fib_prefsrc != cfg->fc_dst) {
1354 		u32 tb_id = cfg->fc_table;
1355 		int rc;
1356 
1357 		if (tb_id == RT_TABLE_MAIN)
1358 			tb_id = RT_TABLE_LOCAL;
1359 
1360 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1361 					  fib_prefsrc, tb_id);
1362 
1363 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1364 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1365 						  fib_prefsrc, RT_TABLE_LOCAL);
1366 		}
1367 
1368 		if (rc != RTN_LOCAL)
1369 			return false;
1370 	}
1371 	return true;
1372 }
1373 
1374 struct fib_info *fib_create_info(struct fib_config *cfg,
1375 				 struct netlink_ext_ack *extack)
1376 {
1377 	int err;
1378 	struct fib_info *fi = NULL;
1379 	struct nexthop *nh = NULL;
1380 	struct fib_info *ofi;
1381 	int nhs = 1;
1382 	struct net *net = cfg->fc_nlinfo.nl_net;
1383 
1384 	if (cfg->fc_type > RTN_MAX)
1385 		goto err_inval;
1386 
1387 	/* Fast check to catch the most weird cases */
1388 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1389 		NL_SET_ERR_MSG(extack, "Invalid scope");
1390 		goto err_inval;
1391 	}
1392 
1393 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1394 		NL_SET_ERR_MSG(extack,
1395 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1396 		goto err_inval;
1397 	}
1398 
1399 	if (cfg->fc_nh_id) {
1400 		if (!cfg->fc_mx) {
1401 			fi = fib_find_info_nh(net, cfg);
1402 			if (fi) {
1403 				refcount_inc(&fi->fib_treeref);
1404 				return fi;
1405 			}
1406 		}
1407 
1408 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1409 		if (!nh) {
1410 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1411 			goto err_inval;
1412 		}
1413 		nhs = 0;
1414 	}
1415 
1416 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1417 	if (cfg->fc_mp) {
1418 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1419 		if (nhs == 0)
1420 			goto err_inval;
1421 	}
1422 #endif
1423 
1424 	err = -ENOBUFS;
1425 
1426 	/* Paired with WRITE_ONCE() in fib_release_info() */
1427 	if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1428 		unsigned int new_size = fib_info_hash_size << 1;
1429 		struct hlist_head *new_info_hash;
1430 		struct hlist_head *new_laddrhash;
1431 		size_t bytes;
1432 
1433 		if (!new_size)
1434 			new_size = 16;
1435 		bytes = (size_t)new_size * sizeof(struct hlist_head *);
1436 		new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1437 		new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1438 		if (!new_info_hash || !new_laddrhash) {
1439 			kvfree(new_info_hash);
1440 			kvfree(new_laddrhash);
1441 		} else {
1442 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1443 		}
1444 		if (!fib_info_hash_size)
1445 			goto failure;
1446 	}
1447 
1448 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1449 	if (!fi)
1450 		goto failure;
1451 	fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1452 					      cfg->fc_mx_len, extack);
1453 	if (IS_ERR(fi->fib_metrics)) {
1454 		err = PTR_ERR(fi->fib_metrics);
1455 		kfree(fi);
1456 		return ERR_PTR(err);
1457 	}
1458 
1459 	fi->fib_net = net;
1460 	fi->fib_protocol = cfg->fc_protocol;
1461 	fi->fib_scope = cfg->fc_scope;
1462 	fi->fib_flags = cfg->fc_flags;
1463 	fi->fib_priority = cfg->fc_priority;
1464 	fi->fib_prefsrc = cfg->fc_prefsrc;
1465 	fi->fib_type = cfg->fc_type;
1466 	fi->fib_tb_id = cfg->fc_table;
1467 
1468 	fi->fib_nhs = nhs;
1469 	if (nh) {
1470 		if (!nexthop_get(nh)) {
1471 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1472 			err = -EINVAL;
1473 		} else {
1474 			err = 0;
1475 			fi->nh = nh;
1476 		}
1477 	} else {
1478 		change_nexthops(fi) {
1479 			nexthop_nh->nh_parent = fi;
1480 		} endfor_nexthops(fi)
1481 
1482 		if (cfg->fc_mp)
1483 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1484 					  extack);
1485 		else
1486 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1487 	}
1488 
1489 	if (err != 0)
1490 		goto failure;
1491 
1492 	if (fib_props[cfg->fc_type].error) {
1493 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1494 			NL_SET_ERR_MSG(extack,
1495 				       "Gateway, device and multipath can not be specified for this route type");
1496 			goto err_inval;
1497 		}
1498 		goto link_it;
1499 	} else {
1500 		switch (cfg->fc_type) {
1501 		case RTN_UNICAST:
1502 		case RTN_LOCAL:
1503 		case RTN_BROADCAST:
1504 		case RTN_ANYCAST:
1505 		case RTN_MULTICAST:
1506 			break;
1507 		default:
1508 			NL_SET_ERR_MSG(extack, "Invalid route type");
1509 			goto err_inval;
1510 		}
1511 	}
1512 
1513 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1514 		NL_SET_ERR_MSG(extack, "Invalid scope");
1515 		goto err_inval;
1516 	}
1517 
1518 	if (fi->nh) {
1519 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1520 		if (err)
1521 			goto failure;
1522 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1523 		struct fib_nh *nh = fi->fib_nh;
1524 
1525 		/* Local address is added. */
1526 		if (nhs != 1) {
1527 			NL_SET_ERR_MSG(extack,
1528 				       "Route with host scope can not have multiple nexthops");
1529 			goto err_inval;
1530 		}
1531 		if (nh->fib_nh_gw_family) {
1532 			NL_SET_ERR_MSG(extack,
1533 				       "Route with host scope can not have a gateway");
1534 			goto err_inval;
1535 		}
1536 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1537 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1538 		err = -ENODEV;
1539 		if (!nh->fib_nh_dev)
1540 			goto failure;
1541 		netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1542 				     GFP_KERNEL);
1543 	} else {
1544 		int linkdown = 0;
1545 
1546 		change_nexthops(fi) {
1547 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1548 					   cfg->fc_table, cfg->fc_scope,
1549 					   extack);
1550 			if (err != 0)
1551 				goto failure;
1552 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1553 				linkdown++;
1554 		} endfor_nexthops(fi)
1555 		if (linkdown == fi->fib_nhs)
1556 			fi->fib_flags |= RTNH_F_LINKDOWN;
1557 	}
1558 
1559 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1560 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1561 		goto err_inval;
1562 	}
1563 
1564 	if (!fi->nh) {
1565 		change_nexthops(fi) {
1566 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1567 						  fi->fib_scope);
1568 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1569 				fi->fib_nh_is_v6 = true;
1570 		} endfor_nexthops(fi)
1571 
1572 		fib_rebalance(fi);
1573 	}
1574 
1575 link_it:
1576 	ofi = fib_find_info(fi);
1577 	if (ofi) {
1578 		fi->fib_dead = 1;
1579 		free_fib_info(fi);
1580 		refcount_inc(&ofi->fib_treeref);
1581 		return ofi;
1582 	}
1583 
1584 	refcount_set(&fi->fib_treeref, 1);
1585 	refcount_set(&fi->fib_clntref, 1);
1586 	spin_lock_bh(&fib_info_lock);
1587 	fib_info_cnt++;
1588 	hlist_add_head(&fi->fib_hash,
1589 		       &fib_info_hash[fib_info_hashfn(fi)]);
1590 	if (fi->fib_prefsrc) {
1591 		struct hlist_head *head;
1592 
1593 		head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1594 		hlist_add_head(&fi->fib_lhash, head);
1595 	}
1596 	if (fi->nh) {
1597 		list_add(&fi->nh_list, &nh->fi_list);
1598 	} else {
1599 		change_nexthops(fi) {
1600 			struct hlist_head *head;
1601 
1602 			if (!nexthop_nh->fib_nh_dev)
1603 				continue;
1604 			head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1605 			hlist_add_head(&nexthop_nh->nh_hash, head);
1606 		} endfor_nexthops(fi)
1607 	}
1608 	spin_unlock_bh(&fib_info_lock);
1609 	return fi;
1610 
1611 err_inval:
1612 	err = -EINVAL;
1613 
1614 failure:
1615 	if (fi) {
1616 		fi->fib_dead = 1;
1617 		free_fib_info(fi);
1618 	}
1619 
1620 	return ERR_PTR(err);
1621 }
1622 
1623 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1624 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1625 {
1626 	if (nhc->nhc_flags & RTNH_F_DEAD)
1627 		*flags |= RTNH_F_DEAD;
1628 
1629 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1630 		*flags |= RTNH_F_LINKDOWN;
1631 
1632 		rcu_read_lock();
1633 		switch (nhc->nhc_family) {
1634 		case AF_INET:
1635 			if (ip_ignore_linkdown(nhc->nhc_dev))
1636 				*flags |= RTNH_F_DEAD;
1637 			break;
1638 		case AF_INET6:
1639 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1640 				*flags |= RTNH_F_DEAD;
1641 			break;
1642 		}
1643 		rcu_read_unlock();
1644 	}
1645 
1646 	switch (nhc->nhc_gw_family) {
1647 	case AF_INET:
1648 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1649 			goto nla_put_failure;
1650 		break;
1651 	case AF_INET6:
1652 		/* if gateway family does not match nexthop family
1653 		 * gateway is encoded as RTA_VIA
1654 		 */
1655 		if (rt_family != nhc->nhc_gw_family) {
1656 			int alen = sizeof(struct in6_addr);
1657 			struct nlattr *nla;
1658 			struct rtvia *via;
1659 
1660 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1661 			if (!nla)
1662 				goto nla_put_failure;
1663 
1664 			via = nla_data(nla);
1665 			via->rtvia_family = AF_INET6;
1666 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1667 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1668 					    &nhc->nhc_gw.ipv6) < 0) {
1669 			goto nla_put_failure;
1670 		}
1671 		break;
1672 	}
1673 
1674 	*flags |= (nhc->nhc_flags &
1675 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1676 
1677 	if (!skip_oif && nhc->nhc_dev &&
1678 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1679 		goto nla_put_failure;
1680 
1681 	if (nhc->nhc_lwtstate &&
1682 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1683 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1684 		goto nla_put_failure;
1685 
1686 	return 0;
1687 
1688 nla_put_failure:
1689 	return -EMSGSIZE;
1690 }
1691 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1692 
1693 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1694 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1695 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1696 {
1697 	const struct net_device *dev = nhc->nhc_dev;
1698 	struct rtnexthop *rtnh;
1699 	unsigned char flags = 0;
1700 
1701 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1702 	if (!rtnh)
1703 		goto nla_put_failure;
1704 
1705 	rtnh->rtnh_hops = nh_weight - 1;
1706 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1707 
1708 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1709 		goto nla_put_failure;
1710 
1711 	rtnh->rtnh_flags = flags;
1712 
1713 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1714 		goto nla_put_failure;
1715 
1716 	/* length of rtnetlink header + attributes */
1717 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1718 
1719 	return 0;
1720 
1721 nla_put_failure:
1722 	return -EMSGSIZE;
1723 }
1724 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1725 #endif
1726 
1727 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1728 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1729 {
1730 	struct nlattr *mp;
1731 
1732 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1733 	if (!mp)
1734 		goto nla_put_failure;
1735 
1736 	if (unlikely(fi->nh)) {
1737 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1738 			goto nla_put_failure;
1739 		goto mp_end;
1740 	}
1741 
1742 	for_nexthops(fi) {
1743 		u32 nh_tclassid = 0;
1744 #ifdef CONFIG_IP_ROUTE_CLASSID
1745 		nh_tclassid = nh->nh_tclassid;
1746 #endif
1747 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1748 				    AF_INET, nh_tclassid) < 0)
1749 			goto nla_put_failure;
1750 	} endfor_nexthops(fi);
1751 
1752 mp_end:
1753 	nla_nest_end(skb, mp);
1754 
1755 	return 0;
1756 
1757 nla_put_failure:
1758 	return -EMSGSIZE;
1759 }
1760 #else
1761 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1762 {
1763 	return 0;
1764 }
1765 #endif
1766 
1767 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1768 		  const struct fib_rt_info *fri, unsigned int flags)
1769 {
1770 	unsigned int nhs = fib_info_num_path(fri->fi);
1771 	struct fib_info *fi = fri->fi;
1772 	u32 tb_id = fri->tb_id;
1773 	struct nlmsghdr *nlh;
1774 	struct rtmsg *rtm;
1775 
1776 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1777 	if (!nlh)
1778 		return -EMSGSIZE;
1779 
1780 	rtm = nlmsg_data(nlh);
1781 	rtm->rtm_family = AF_INET;
1782 	rtm->rtm_dst_len = fri->dst_len;
1783 	rtm->rtm_src_len = 0;
1784 	rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1785 	if (tb_id < 256)
1786 		rtm->rtm_table = tb_id;
1787 	else
1788 		rtm->rtm_table = RT_TABLE_COMPAT;
1789 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1790 		goto nla_put_failure;
1791 	rtm->rtm_type = fri->type;
1792 	rtm->rtm_flags = fi->fib_flags;
1793 	rtm->rtm_scope = fi->fib_scope;
1794 	rtm->rtm_protocol = fi->fib_protocol;
1795 
1796 	if (rtm->rtm_dst_len &&
1797 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1798 		goto nla_put_failure;
1799 	if (fi->fib_priority &&
1800 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1801 		goto nla_put_failure;
1802 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1803 		goto nla_put_failure;
1804 
1805 	if (fi->fib_prefsrc &&
1806 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1807 		goto nla_put_failure;
1808 
1809 	if (fi->nh) {
1810 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1811 			goto nla_put_failure;
1812 		if (nexthop_is_blackhole(fi->nh))
1813 			rtm->rtm_type = RTN_BLACKHOLE;
1814 		if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode)
1815 			goto offload;
1816 	}
1817 
1818 	if (nhs == 1) {
1819 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1820 		unsigned char flags = 0;
1821 
1822 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1823 			goto nla_put_failure;
1824 
1825 		rtm->rtm_flags = flags;
1826 #ifdef CONFIG_IP_ROUTE_CLASSID
1827 		if (nhc->nhc_family == AF_INET) {
1828 			struct fib_nh *nh;
1829 
1830 			nh = container_of(nhc, struct fib_nh, nh_common);
1831 			if (nh->nh_tclassid &&
1832 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1833 				goto nla_put_failure;
1834 		}
1835 #endif
1836 	} else {
1837 		if (fib_add_multipath(skb, fi) < 0)
1838 			goto nla_put_failure;
1839 	}
1840 
1841 offload:
1842 	if (fri->offload)
1843 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1844 	if (fri->trap)
1845 		rtm->rtm_flags |= RTM_F_TRAP;
1846 	if (fri->offload_failed)
1847 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1848 
1849 	nlmsg_end(skb, nlh);
1850 	return 0;
1851 
1852 nla_put_failure:
1853 	nlmsg_cancel(skb, nlh);
1854 	return -EMSGSIZE;
1855 }
1856 
1857 /*
1858  * Update FIB if:
1859  * - local address disappeared -> we must delete all the entries
1860  *   referring to it.
1861  * - device went down -> we must shutdown all nexthops going via it.
1862  */
1863 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1864 {
1865 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1866 	struct net *net = dev_net(dev);
1867 	struct hlist_head *head;
1868 	struct fib_info *fi;
1869 	int ret = 0;
1870 
1871 	if (!fib_info_laddrhash || local == 0)
1872 		return 0;
1873 
1874 	head = fib_info_laddrhash_bucket(net, local);
1875 	hlist_for_each_entry(fi, head, fib_lhash) {
1876 		if (!net_eq(fi->fib_net, net) ||
1877 		    fi->fib_tb_id != tb_id)
1878 			continue;
1879 		if (fi->fib_prefsrc == local) {
1880 			fi->fib_flags |= RTNH_F_DEAD;
1881 			ret++;
1882 		}
1883 	}
1884 	return ret;
1885 }
1886 
1887 static int call_fib_nh_notifiers(struct fib_nh *nh,
1888 				 enum fib_event_type event_type)
1889 {
1890 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1891 	struct fib_nh_notifier_info info = {
1892 		.fib_nh = nh,
1893 	};
1894 
1895 	switch (event_type) {
1896 	case FIB_EVENT_NH_ADD:
1897 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1898 			break;
1899 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1900 			break;
1901 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1902 					   &info.info);
1903 	case FIB_EVENT_NH_DEL:
1904 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1905 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1906 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1907 						   event_type, &info.info);
1908 		break;
1909 	default:
1910 		break;
1911 	}
1912 
1913 	return NOTIFY_DONE;
1914 }
1915 
1916 /* Update the PMTU of exceptions when:
1917  * - the new MTU of the first hop becomes smaller than the PMTU
1918  * - the old MTU was the same as the PMTU, and it limited discovery of
1919  *   larger MTUs on the path. With that limit raised, we can now
1920  *   discover larger MTUs
1921  * A special case is locked exceptions, for which the PMTU is smaller
1922  * than the minimal accepted PMTU:
1923  * - if the new MTU is greater than the PMTU, don't make any change
1924  * - otherwise, unlock and set PMTU
1925  */
1926 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1927 {
1928 	struct fnhe_hash_bucket *bucket;
1929 	int i;
1930 
1931 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1932 	if (!bucket)
1933 		return;
1934 
1935 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1936 		struct fib_nh_exception *fnhe;
1937 
1938 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1939 		     fnhe;
1940 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1941 			if (fnhe->fnhe_mtu_locked) {
1942 				if (new <= fnhe->fnhe_pmtu) {
1943 					fnhe->fnhe_pmtu = new;
1944 					fnhe->fnhe_mtu_locked = false;
1945 				}
1946 			} else if (new < fnhe->fnhe_pmtu ||
1947 				   orig == fnhe->fnhe_pmtu) {
1948 				fnhe->fnhe_pmtu = new;
1949 			}
1950 		}
1951 	}
1952 }
1953 
1954 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1955 {
1956 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1957 	struct fib_nh *nh;
1958 
1959 	hlist_for_each_entry(nh, head, nh_hash) {
1960 		if (nh->fib_nh_dev == dev)
1961 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1962 	}
1963 }
1964 
1965 /* Event              force Flags           Description
1966  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1967  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1968  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1969  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1970  *
1971  * only used when fib_nh is built into fib_info
1972  */
1973 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1974 {
1975 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1976 	struct fib_info *prev_fi = NULL;
1977 	int scope = RT_SCOPE_NOWHERE;
1978 	struct fib_nh *nh;
1979 	int ret = 0;
1980 
1981 	if (force)
1982 		scope = -1;
1983 
1984 	hlist_for_each_entry(nh, head, nh_hash) {
1985 		struct fib_info *fi = nh->nh_parent;
1986 		int dead;
1987 
1988 		BUG_ON(!fi->fib_nhs);
1989 		if (nh->fib_nh_dev != dev || fi == prev_fi)
1990 			continue;
1991 		prev_fi = fi;
1992 		dead = 0;
1993 		change_nexthops(fi) {
1994 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1995 				dead++;
1996 			else if (nexthop_nh->fib_nh_dev == dev &&
1997 				 nexthop_nh->fib_nh_scope != scope) {
1998 				switch (event) {
1999 				case NETDEV_DOWN:
2000 				case NETDEV_UNREGISTER:
2001 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2002 					fallthrough;
2003 				case NETDEV_CHANGE:
2004 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2005 					break;
2006 				}
2007 				call_fib_nh_notifiers(nexthop_nh,
2008 						      FIB_EVENT_NH_DEL);
2009 				dead++;
2010 			}
2011 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2012 			if (event == NETDEV_UNREGISTER &&
2013 			    nexthop_nh->fib_nh_dev == dev) {
2014 				dead = fi->fib_nhs;
2015 				break;
2016 			}
2017 #endif
2018 		} endfor_nexthops(fi)
2019 		if (dead == fi->fib_nhs) {
2020 			switch (event) {
2021 			case NETDEV_DOWN:
2022 			case NETDEV_UNREGISTER:
2023 				fi->fib_flags |= RTNH_F_DEAD;
2024 				fallthrough;
2025 			case NETDEV_CHANGE:
2026 				fi->fib_flags |= RTNH_F_LINKDOWN;
2027 				break;
2028 			}
2029 			ret++;
2030 		}
2031 
2032 		fib_rebalance(fi);
2033 	}
2034 
2035 	return ret;
2036 }
2037 
2038 /* Must be invoked inside of an RCU protected region.  */
2039 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2040 {
2041 	struct fib_info *fi = NULL, *last_resort = NULL;
2042 	struct hlist_head *fa_head = res->fa_head;
2043 	struct fib_table *tb = res->table;
2044 	u8 slen = 32 - res->prefixlen;
2045 	int order = -1, last_idx = -1;
2046 	struct fib_alias *fa, *fa1 = NULL;
2047 	u32 last_prio = res->fi->fib_priority;
2048 	dscp_t last_dscp = 0;
2049 
2050 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2051 		struct fib_info *next_fi = fa->fa_info;
2052 		struct fib_nh_common *nhc;
2053 
2054 		if (fa->fa_slen != slen)
2055 			continue;
2056 		if (fa->fa_dscp &&
2057 		    fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2058 			continue;
2059 		if (fa->tb_id != tb->tb_id)
2060 			continue;
2061 		if (next_fi->fib_priority > last_prio &&
2062 		    fa->fa_dscp == last_dscp) {
2063 			if (last_dscp)
2064 				continue;
2065 			break;
2066 		}
2067 		if (next_fi->fib_flags & RTNH_F_DEAD)
2068 			continue;
2069 		last_dscp = fa->fa_dscp;
2070 		last_prio = next_fi->fib_priority;
2071 
2072 		if (next_fi->fib_scope != res->scope ||
2073 		    fa->fa_type != RTN_UNICAST)
2074 			continue;
2075 
2076 		nhc = fib_info_nhc(next_fi, 0);
2077 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2078 			continue;
2079 
2080 		fib_alias_accessed(fa);
2081 
2082 		if (!fi) {
2083 			if (next_fi != res->fi)
2084 				break;
2085 			fa1 = fa;
2086 		} else if (!fib_detect_death(fi, order, &last_resort,
2087 					     &last_idx, fa1->fa_default)) {
2088 			fib_result_assign(res, fi);
2089 			fa1->fa_default = order;
2090 			goto out;
2091 		}
2092 		fi = next_fi;
2093 		order++;
2094 	}
2095 
2096 	if (order <= 0 || !fi) {
2097 		if (fa1)
2098 			fa1->fa_default = -1;
2099 		goto out;
2100 	}
2101 
2102 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2103 			      fa1->fa_default)) {
2104 		fib_result_assign(res, fi);
2105 		fa1->fa_default = order;
2106 		goto out;
2107 	}
2108 
2109 	if (last_idx >= 0)
2110 		fib_result_assign(res, last_resort);
2111 	fa1->fa_default = last_idx;
2112 out:
2113 	return;
2114 }
2115 
2116 /*
2117  * Dead device goes up. We wake up dead nexthops.
2118  * It takes sense only on multipath routes.
2119  *
2120  * only used when fib_nh is built into fib_info
2121  */
2122 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2123 {
2124 	struct fib_info *prev_fi;
2125 	struct hlist_head *head;
2126 	struct fib_nh *nh;
2127 	int ret;
2128 
2129 	if (!(dev->flags & IFF_UP))
2130 		return 0;
2131 
2132 	if (nh_flags & RTNH_F_DEAD) {
2133 		unsigned int flags = dev_get_flags(dev);
2134 
2135 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2136 			nh_flags |= RTNH_F_LINKDOWN;
2137 	}
2138 
2139 	prev_fi = NULL;
2140 	head = fib_info_devhash_bucket(dev);
2141 	ret = 0;
2142 
2143 	hlist_for_each_entry(nh, head, nh_hash) {
2144 		struct fib_info *fi = nh->nh_parent;
2145 		int alive;
2146 
2147 		BUG_ON(!fi->fib_nhs);
2148 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2149 			continue;
2150 
2151 		prev_fi = fi;
2152 		alive = 0;
2153 		change_nexthops(fi) {
2154 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2155 				alive++;
2156 				continue;
2157 			}
2158 			if (!nexthop_nh->fib_nh_dev ||
2159 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2160 				continue;
2161 			if (nexthop_nh->fib_nh_dev != dev ||
2162 			    !__in_dev_get_rtnl(dev))
2163 				continue;
2164 			alive++;
2165 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2166 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2167 		} endfor_nexthops(fi)
2168 
2169 		if (alive > 0) {
2170 			fi->fib_flags &= ~nh_flags;
2171 			ret++;
2172 		}
2173 
2174 		fib_rebalance(fi);
2175 	}
2176 
2177 	return ret;
2178 }
2179 
2180 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2181 static bool fib_good_nh(const struct fib_nh *nh)
2182 {
2183 	int state = NUD_REACHABLE;
2184 
2185 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2186 		struct neighbour *n;
2187 
2188 		rcu_read_lock_bh();
2189 
2190 		if (likely(nh->fib_nh_gw_family == AF_INET))
2191 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2192 						   (__force u32)nh->fib_nh_gw4);
2193 		else if (nh->fib_nh_gw_family == AF_INET6)
2194 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2195 							   &nh->fib_nh_gw6);
2196 		else
2197 			n = NULL;
2198 		if (n)
2199 			state = n->nud_state;
2200 
2201 		rcu_read_unlock_bh();
2202 	}
2203 
2204 	return !!(state & NUD_VALID);
2205 }
2206 
2207 void fib_select_multipath(struct fib_result *res, int hash)
2208 {
2209 	struct fib_info *fi = res->fi;
2210 	struct net *net = fi->fib_net;
2211 	bool first = false;
2212 
2213 	if (unlikely(res->fi->nh)) {
2214 		nexthop_path_fib_result(res, hash);
2215 		return;
2216 	}
2217 
2218 	change_nexthops(fi) {
2219 		if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2220 			if (!fib_good_nh(nexthop_nh))
2221 				continue;
2222 			if (!first) {
2223 				res->nh_sel = nhsel;
2224 				res->nhc = &nexthop_nh->nh_common;
2225 				first = true;
2226 			}
2227 		}
2228 
2229 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2230 			continue;
2231 
2232 		res->nh_sel = nhsel;
2233 		res->nhc = &nexthop_nh->nh_common;
2234 		return;
2235 	} endfor_nexthops(fi);
2236 }
2237 #endif
2238 
2239 void fib_select_path(struct net *net, struct fib_result *res,
2240 		     struct flowi4 *fl4, const struct sk_buff *skb)
2241 {
2242 	if (fl4->flowi4_oif)
2243 		goto check_saddr;
2244 
2245 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2246 	if (fib_info_num_path(res->fi) > 1) {
2247 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2248 
2249 		fib_select_multipath(res, h);
2250 	}
2251 	else
2252 #endif
2253 	if (!res->prefixlen &&
2254 	    res->table->tb_num_default > 1 &&
2255 	    res->type == RTN_UNICAST)
2256 		fib_select_default(fl4, res);
2257 
2258 check_saddr:
2259 	if (!fl4->saddr)
2260 		fl4->saddr = fib_result_prefsrc(net, res);
2261 }
2262