xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision feac8c8b)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		IPv4 Forwarding Information Base: semantics.
7  *
8  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9  *
10  *		This program is free software; you can redistribute it and/or
11  *		modify it under the terms of the GNU General Public License
12  *		as published by the Free Software Foundation; either version
13  *		2 of the License, or (at your option) any later version.
14  */
15 
16 #include <linux/uaccess.h>
17 #include <linux/bitops.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/jiffies.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/errno.h>
26 #include <linux/in.h>
27 #include <linux/inet.h>
28 #include <linux/inetdevice.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_arp.h>
31 #include <linux/proc_fs.h>
32 #include <linux/skbuff.h>
33 #include <linux/init.h>
34 #include <linux/slab.h>
35 #include <linux/netlink.h>
36 
37 #include <net/arp.h>
38 #include <net/ip.h>
39 #include <net/protocol.h>
40 #include <net/route.h>
41 #include <net/tcp.h>
42 #include <net/sock.h>
43 #include <net/ip_fib.h>
44 #include <net/netlink.h>
45 #include <net/nexthop.h>
46 #include <net/lwtunnel.h>
47 #include <net/fib_notifier.h>
48 
49 #include "fib_lookup.h"
50 
51 static DEFINE_SPINLOCK(fib_info_lock);
52 static struct hlist_head *fib_info_hash;
53 static struct hlist_head *fib_info_laddrhash;
54 static unsigned int fib_info_hash_size;
55 static unsigned int fib_info_cnt;
56 
57 #define DEVINDEX_HASHBITS 8
58 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
59 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
60 
61 #ifdef CONFIG_IP_ROUTE_MULTIPATH
62 
63 #define for_nexthops(fi) {						\
64 	int nhsel; const struct fib_nh *nh;				\
65 	for (nhsel = 0, nh = (fi)->fib_nh;				\
66 	     nhsel < (fi)->fib_nhs;					\
67 	     nh++, nhsel++)
68 
69 #define change_nexthops(fi) {						\
70 	int nhsel; struct fib_nh *nexthop_nh;				\
71 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
72 	     nhsel < (fi)->fib_nhs;					\
73 	     nexthop_nh++, nhsel++)
74 
75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
76 
77 /* Hope, that gcc will optimize it to get rid of dummy loop */
78 
79 #define for_nexthops(fi) {						\
80 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
81 	for (nhsel = 0; nhsel < 1; nhsel++)
82 
83 #define change_nexthops(fi) {						\
84 	int nhsel;							\
85 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
86 	for (nhsel = 0; nhsel < 1; nhsel++)
87 
88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
89 
90 #define endfor_nexthops(fi) }
91 
92 
93 const struct fib_prop fib_props[RTN_MAX + 1] = {
94 	[RTN_UNSPEC] = {
95 		.error	= 0,
96 		.scope	= RT_SCOPE_NOWHERE,
97 	},
98 	[RTN_UNICAST] = {
99 		.error	= 0,
100 		.scope	= RT_SCOPE_UNIVERSE,
101 	},
102 	[RTN_LOCAL] = {
103 		.error	= 0,
104 		.scope	= RT_SCOPE_HOST,
105 	},
106 	[RTN_BROADCAST] = {
107 		.error	= 0,
108 		.scope	= RT_SCOPE_LINK,
109 	},
110 	[RTN_ANYCAST] = {
111 		.error	= 0,
112 		.scope	= RT_SCOPE_LINK,
113 	},
114 	[RTN_MULTICAST] = {
115 		.error	= 0,
116 		.scope	= RT_SCOPE_UNIVERSE,
117 	},
118 	[RTN_BLACKHOLE] = {
119 		.error	= -EINVAL,
120 		.scope	= RT_SCOPE_UNIVERSE,
121 	},
122 	[RTN_UNREACHABLE] = {
123 		.error	= -EHOSTUNREACH,
124 		.scope	= RT_SCOPE_UNIVERSE,
125 	},
126 	[RTN_PROHIBIT] = {
127 		.error	= -EACCES,
128 		.scope	= RT_SCOPE_UNIVERSE,
129 	},
130 	[RTN_THROW] = {
131 		.error	= -EAGAIN,
132 		.scope	= RT_SCOPE_UNIVERSE,
133 	},
134 	[RTN_NAT] = {
135 		.error	= -EINVAL,
136 		.scope	= RT_SCOPE_NOWHERE,
137 	},
138 	[RTN_XRESOLVE] = {
139 		.error	= -EINVAL,
140 		.scope	= RT_SCOPE_NOWHERE,
141 	},
142 };
143 
144 static void rt_fibinfo_free(struct rtable __rcu **rtp)
145 {
146 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
147 
148 	if (!rt)
149 		return;
150 
151 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
152 	 * because we waited an RCU grace period before calling
153 	 * free_fib_info_rcu()
154 	 */
155 
156 	dst_dev_put(&rt->dst);
157 	dst_release_immediate(&rt->dst);
158 }
159 
160 static void free_nh_exceptions(struct fib_nh *nh)
161 {
162 	struct fnhe_hash_bucket *hash;
163 	int i;
164 
165 	hash = rcu_dereference_protected(nh->nh_exceptions, 1);
166 	if (!hash)
167 		return;
168 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
169 		struct fib_nh_exception *fnhe;
170 
171 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
172 		while (fnhe) {
173 			struct fib_nh_exception *next;
174 
175 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
176 
177 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
178 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
179 
180 			kfree(fnhe);
181 
182 			fnhe = next;
183 		}
184 	}
185 	kfree(hash);
186 }
187 
188 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
189 {
190 	int cpu;
191 
192 	if (!rtp)
193 		return;
194 
195 	for_each_possible_cpu(cpu) {
196 		struct rtable *rt;
197 
198 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
199 		if (rt) {
200 			dst_dev_put(&rt->dst);
201 			dst_release_immediate(&rt->dst);
202 		}
203 	}
204 	free_percpu(rtp);
205 }
206 
207 /* Release a nexthop info record */
208 static void free_fib_info_rcu(struct rcu_head *head)
209 {
210 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
211 	struct dst_metrics *m;
212 
213 	change_nexthops(fi) {
214 		if (nexthop_nh->nh_dev)
215 			dev_put(nexthop_nh->nh_dev);
216 		lwtstate_put(nexthop_nh->nh_lwtstate);
217 		free_nh_exceptions(nexthop_nh);
218 		rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
219 		rt_fibinfo_free(&nexthop_nh->nh_rth_input);
220 	} endfor_nexthops(fi);
221 
222 	m = fi->fib_metrics;
223 	if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
224 		kfree(m);
225 	kfree(fi);
226 }
227 
228 void free_fib_info(struct fib_info *fi)
229 {
230 	if (fi->fib_dead == 0) {
231 		pr_warn("Freeing alive fib_info %p\n", fi);
232 		return;
233 	}
234 	fib_info_cnt--;
235 #ifdef CONFIG_IP_ROUTE_CLASSID
236 	change_nexthops(fi) {
237 		if (nexthop_nh->nh_tclassid)
238 			fi->fib_net->ipv4.fib_num_tclassid_users--;
239 	} endfor_nexthops(fi);
240 #endif
241 	call_rcu(&fi->rcu, free_fib_info_rcu);
242 }
243 EXPORT_SYMBOL_GPL(free_fib_info);
244 
245 void fib_release_info(struct fib_info *fi)
246 {
247 	spin_lock_bh(&fib_info_lock);
248 	if (fi && --fi->fib_treeref == 0) {
249 		hlist_del(&fi->fib_hash);
250 		if (fi->fib_prefsrc)
251 			hlist_del(&fi->fib_lhash);
252 		change_nexthops(fi) {
253 			if (!nexthop_nh->nh_dev)
254 				continue;
255 			hlist_del(&nexthop_nh->nh_hash);
256 		} endfor_nexthops(fi)
257 		fi->fib_dead = 1;
258 		fib_info_put(fi);
259 	}
260 	spin_unlock_bh(&fib_info_lock);
261 }
262 
263 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
264 {
265 	const struct fib_nh *onh = ofi->fib_nh;
266 
267 	for_nexthops(fi) {
268 		if (nh->nh_oif != onh->nh_oif ||
269 		    nh->nh_gw  != onh->nh_gw ||
270 		    nh->nh_scope != onh->nh_scope ||
271 #ifdef CONFIG_IP_ROUTE_MULTIPATH
272 		    nh->nh_weight != onh->nh_weight ||
273 #endif
274 #ifdef CONFIG_IP_ROUTE_CLASSID
275 		    nh->nh_tclassid != onh->nh_tclassid ||
276 #endif
277 		    lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
278 		    ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
279 			return -1;
280 		onh++;
281 	} endfor_nexthops(fi);
282 	return 0;
283 }
284 
285 static inline unsigned int fib_devindex_hashfn(unsigned int val)
286 {
287 	unsigned int mask = DEVINDEX_HASHSIZE - 1;
288 
289 	return (val ^
290 		(val >> DEVINDEX_HASHBITS) ^
291 		(val >> (DEVINDEX_HASHBITS * 2))) & mask;
292 }
293 
294 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
295 {
296 	unsigned int mask = (fib_info_hash_size - 1);
297 	unsigned int val = fi->fib_nhs;
298 
299 	val ^= (fi->fib_protocol << 8) | fi->fib_scope;
300 	val ^= (__force u32)fi->fib_prefsrc;
301 	val ^= fi->fib_priority;
302 	for_nexthops(fi) {
303 		val ^= fib_devindex_hashfn(nh->nh_oif);
304 	} endfor_nexthops(fi)
305 
306 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
307 }
308 
309 static struct fib_info *fib_find_info(const struct fib_info *nfi)
310 {
311 	struct hlist_head *head;
312 	struct fib_info *fi;
313 	unsigned int hash;
314 
315 	hash = fib_info_hashfn(nfi);
316 	head = &fib_info_hash[hash];
317 
318 	hlist_for_each_entry(fi, head, fib_hash) {
319 		if (!net_eq(fi->fib_net, nfi->fib_net))
320 			continue;
321 		if (fi->fib_nhs != nfi->fib_nhs)
322 			continue;
323 		if (nfi->fib_protocol == fi->fib_protocol &&
324 		    nfi->fib_scope == fi->fib_scope &&
325 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
326 		    nfi->fib_priority == fi->fib_priority &&
327 		    nfi->fib_type == fi->fib_type &&
328 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
329 			   sizeof(u32) * RTAX_MAX) == 0 &&
330 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
331 		    (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
332 			return fi;
333 	}
334 
335 	return NULL;
336 }
337 
338 /* Check, that the gateway is already configured.
339  * Used only by redirect accept routine.
340  */
341 int ip_fib_check_default(__be32 gw, struct net_device *dev)
342 {
343 	struct hlist_head *head;
344 	struct fib_nh *nh;
345 	unsigned int hash;
346 
347 	spin_lock(&fib_info_lock);
348 
349 	hash = fib_devindex_hashfn(dev->ifindex);
350 	head = &fib_info_devhash[hash];
351 	hlist_for_each_entry(nh, head, nh_hash) {
352 		if (nh->nh_dev == dev &&
353 		    nh->nh_gw == gw &&
354 		    !(nh->nh_flags & RTNH_F_DEAD)) {
355 			spin_unlock(&fib_info_lock);
356 			return 0;
357 		}
358 	}
359 
360 	spin_unlock(&fib_info_lock);
361 
362 	return -1;
363 }
364 
365 static inline size_t fib_nlmsg_size(struct fib_info *fi)
366 {
367 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
368 			 + nla_total_size(4) /* RTA_TABLE */
369 			 + nla_total_size(4) /* RTA_DST */
370 			 + nla_total_size(4) /* RTA_PRIORITY */
371 			 + nla_total_size(4) /* RTA_PREFSRC */
372 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
373 
374 	/* space for nested metrics */
375 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
376 
377 	if (fi->fib_nhs) {
378 		size_t nh_encapsize = 0;
379 		/* Also handles the special case fib_nhs == 1 */
380 
381 		/* each nexthop is packed in an attribute */
382 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
383 
384 		/* may contain flow and gateway attribute */
385 		nhsize += 2 * nla_total_size(4);
386 
387 		/* grab encap info */
388 		for_nexthops(fi) {
389 			if (nh->nh_lwtstate) {
390 				/* RTA_ENCAP_TYPE */
391 				nh_encapsize += lwtunnel_get_encap_size(
392 						nh->nh_lwtstate);
393 				/* RTA_ENCAP */
394 				nh_encapsize +=  nla_total_size(2);
395 			}
396 		} endfor_nexthops(fi);
397 
398 		/* all nexthops are packed in a nested attribute */
399 		payload += nla_total_size((fi->fib_nhs * nhsize) +
400 					  nh_encapsize);
401 
402 	}
403 
404 	return payload;
405 }
406 
407 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
408 	       int dst_len, u32 tb_id, const struct nl_info *info,
409 	       unsigned int nlm_flags)
410 {
411 	struct sk_buff *skb;
412 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
413 	int err = -ENOBUFS;
414 
415 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
416 	if (!skb)
417 		goto errout;
418 
419 	err = fib_dump_info(skb, info->portid, seq, event, tb_id,
420 			    fa->fa_type, key, dst_len,
421 			    fa->fa_tos, fa->fa_info, nlm_flags);
422 	if (err < 0) {
423 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
424 		WARN_ON(err == -EMSGSIZE);
425 		kfree_skb(skb);
426 		goto errout;
427 	}
428 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
429 		    info->nlh, GFP_KERNEL);
430 	return;
431 errout:
432 	if (err < 0)
433 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
434 }
435 
436 static int fib_detect_death(struct fib_info *fi, int order,
437 			    struct fib_info **last_resort, int *last_idx,
438 			    int dflt)
439 {
440 	struct neighbour *n;
441 	int state = NUD_NONE;
442 
443 	n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
444 	if (n) {
445 		state = n->nud_state;
446 		neigh_release(n);
447 	} else {
448 		return 0;
449 	}
450 	if (state == NUD_REACHABLE)
451 		return 0;
452 	if ((state & NUD_VALID) && order != dflt)
453 		return 0;
454 	if ((state & NUD_VALID) ||
455 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
456 		*last_resort = fi;
457 		*last_idx = order;
458 	}
459 	return 1;
460 }
461 
462 #ifdef CONFIG_IP_ROUTE_MULTIPATH
463 
464 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
465 			      struct netlink_ext_ack *extack)
466 {
467 	int nhs = 0;
468 
469 	while (rtnh_ok(rtnh, remaining)) {
470 		nhs++;
471 		rtnh = rtnh_next(rtnh, &remaining);
472 	}
473 
474 	/* leftover implies invalid nexthop configuration, discard it */
475 	if (remaining > 0) {
476 		NL_SET_ERR_MSG(extack,
477 			       "Invalid nexthop configuration - extra data after nexthops");
478 		nhs = 0;
479 	}
480 
481 	return nhs;
482 }
483 
484 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
485 		       int remaining, struct fib_config *cfg,
486 		       struct netlink_ext_ack *extack)
487 {
488 	int ret;
489 
490 	change_nexthops(fi) {
491 		int attrlen;
492 
493 		if (!rtnh_ok(rtnh, remaining)) {
494 			NL_SET_ERR_MSG(extack,
495 				       "Invalid nexthop configuration - extra data after nexthop");
496 			return -EINVAL;
497 		}
498 
499 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
500 			NL_SET_ERR_MSG(extack,
501 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
502 			return -EINVAL;
503 		}
504 
505 		nexthop_nh->nh_flags =
506 			(cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
507 		nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
508 		nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
509 
510 		attrlen = rtnh_attrlen(rtnh);
511 		if (attrlen > 0) {
512 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
513 
514 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
515 			nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
516 #ifdef CONFIG_IP_ROUTE_CLASSID
517 			nla = nla_find(attrs, attrlen, RTA_FLOW);
518 			nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
519 			if (nexthop_nh->nh_tclassid)
520 				fi->fib_net->ipv4.fib_num_tclassid_users++;
521 #endif
522 			nla = nla_find(attrs, attrlen, RTA_ENCAP);
523 			if (nla) {
524 				struct lwtunnel_state *lwtstate;
525 				struct nlattr *nla_entype;
526 
527 				nla_entype = nla_find(attrs, attrlen,
528 						      RTA_ENCAP_TYPE);
529 				if (!nla_entype) {
530 					NL_SET_BAD_ATTR(extack, nla);
531 					NL_SET_ERR_MSG(extack,
532 						       "Encap type is missing");
533 					goto err_inval;
534 				}
535 
536 				ret = lwtunnel_build_state(nla_get_u16(
537 							   nla_entype),
538 							   nla,  AF_INET, cfg,
539 							   &lwtstate, extack);
540 				if (ret)
541 					goto errout;
542 				nexthop_nh->nh_lwtstate =
543 					lwtstate_get(lwtstate);
544 			}
545 		}
546 
547 		rtnh = rtnh_next(rtnh, &remaining);
548 	} endfor_nexthops(fi);
549 
550 	return 0;
551 
552 err_inval:
553 	ret = -EINVAL;
554 
555 errout:
556 	return ret;
557 }
558 
559 static void fib_rebalance(struct fib_info *fi)
560 {
561 	int total;
562 	int w;
563 	struct in_device *in_dev;
564 
565 	if (fi->fib_nhs < 2)
566 		return;
567 
568 	total = 0;
569 	for_nexthops(fi) {
570 		if (nh->nh_flags & RTNH_F_DEAD)
571 			continue;
572 
573 		in_dev = __in_dev_get_rtnl(nh->nh_dev);
574 
575 		if (in_dev &&
576 		    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
577 		    nh->nh_flags & RTNH_F_LINKDOWN)
578 			continue;
579 
580 		total += nh->nh_weight;
581 	} endfor_nexthops(fi);
582 
583 	w = 0;
584 	change_nexthops(fi) {
585 		int upper_bound;
586 
587 		in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
588 
589 		if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
590 			upper_bound = -1;
591 		} else if (in_dev &&
592 			   IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
593 			   nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
594 			upper_bound = -1;
595 		} else {
596 			w += nexthop_nh->nh_weight;
597 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
598 							    total) - 1;
599 		}
600 
601 		atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
602 	} endfor_nexthops(fi);
603 }
604 #else /* CONFIG_IP_ROUTE_MULTIPATH */
605 
606 #define fib_rebalance(fi) do { } while (0)
607 
608 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
609 
610 static int fib_encap_match(u16 encap_type,
611 			   struct nlattr *encap,
612 			   const struct fib_nh *nh,
613 			   const struct fib_config *cfg,
614 			   struct netlink_ext_ack *extack)
615 {
616 	struct lwtunnel_state *lwtstate;
617 	int ret, result = 0;
618 
619 	if (encap_type == LWTUNNEL_ENCAP_NONE)
620 		return 0;
621 
622 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
623 				   cfg, &lwtstate, extack);
624 	if (!ret) {
625 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
626 		lwtstate_free(lwtstate);
627 	}
628 
629 	return result;
630 }
631 
632 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
633 		 struct netlink_ext_ack *extack)
634 {
635 #ifdef CONFIG_IP_ROUTE_MULTIPATH
636 	struct rtnexthop *rtnh;
637 	int remaining;
638 #endif
639 
640 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
641 		return 1;
642 
643 	if (cfg->fc_oif || cfg->fc_gw) {
644 		if (cfg->fc_encap) {
645 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
646 					    fi->fib_nh, cfg, extack))
647 				return 1;
648 		}
649 #ifdef CONFIG_IP_ROUTE_CLASSID
650 		if (cfg->fc_flow &&
651 		    cfg->fc_flow != fi->fib_nh->nh_tclassid)
652 			return 1;
653 #endif
654 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
655 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
656 			return 0;
657 		return 1;
658 	}
659 
660 #ifdef CONFIG_IP_ROUTE_MULTIPATH
661 	if (!cfg->fc_mp)
662 		return 0;
663 
664 	rtnh = cfg->fc_mp;
665 	remaining = cfg->fc_mp_len;
666 
667 	for_nexthops(fi) {
668 		int attrlen;
669 
670 		if (!rtnh_ok(rtnh, remaining))
671 			return -EINVAL;
672 
673 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
674 			return 1;
675 
676 		attrlen = rtnh_attrlen(rtnh);
677 		if (attrlen > 0) {
678 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
679 
680 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
681 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
682 				return 1;
683 #ifdef CONFIG_IP_ROUTE_CLASSID
684 			nla = nla_find(attrs, attrlen, RTA_FLOW);
685 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
686 				return 1;
687 #endif
688 		}
689 
690 		rtnh = rtnh_next(rtnh, &remaining);
691 	} endfor_nexthops(fi);
692 #endif
693 	return 0;
694 }
695 
696 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
697 {
698 	struct nlattr *nla;
699 	int remaining;
700 
701 	if (!cfg->fc_mx)
702 		return true;
703 
704 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
705 		int type = nla_type(nla);
706 		u32 fi_val, val;
707 
708 		if (!type)
709 			continue;
710 		if (type > RTAX_MAX)
711 			return false;
712 
713 		if (type == RTAX_CC_ALGO) {
714 			char tmp[TCP_CA_NAME_MAX];
715 			bool ecn_ca = false;
716 
717 			nla_strlcpy(tmp, nla, sizeof(tmp));
718 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
719 		} else {
720 			val = nla_get_u32(nla);
721 		}
722 
723 		fi_val = fi->fib_metrics->metrics[type - 1];
724 		if (type == RTAX_FEATURES)
725 			fi_val &= ~DST_FEATURE_ECN_CA;
726 
727 		if (fi_val != val)
728 			return false;
729 	}
730 
731 	return true;
732 }
733 
734 
735 /*
736  * Picture
737  * -------
738  *
739  * Semantics of nexthop is very messy by historical reasons.
740  * We have to take into account, that:
741  * a) gateway can be actually local interface address,
742  *    so that gatewayed route is direct.
743  * b) gateway must be on-link address, possibly
744  *    described not by an ifaddr, but also by a direct route.
745  * c) If both gateway and interface are specified, they should not
746  *    contradict.
747  * d) If we use tunnel routes, gateway could be not on-link.
748  *
749  * Attempt to reconcile all of these (alas, self-contradictory) conditions
750  * results in pretty ugly and hairy code with obscure logic.
751  *
752  * I chose to generalized it instead, so that the size
753  * of code does not increase practically, but it becomes
754  * much more general.
755  * Every prefix is assigned a "scope" value: "host" is local address,
756  * "link" is direct route,
757  * [ ... "site" ... "interior" ... ]
758  * and "universe" is true gateway route with global meaning.
759  *
760  * Every prefix refers to a set of "nexthop"s (gw, oif),
761  * where gw must have narrower scope. This recursion stops
762  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
763  * which means that gw is forced to be on link.
764  *
765  * Code is still hairy, but now it is apparently logically
766  * consistent and very flexible. F.e. as by-product it allows
767  * to co-exists in peace independent exterior and interior
768  * routing processes.
769  *
770  * Normally it looks as following.
771  *
772  * {universe prefix}  -> (gw, oif) [scope link]
773  *		  |
774  *		  |-> {link prefix} -> (gw, oif) [scope local]
775  *					|
776  *					|-> {local prefix} (terminal node)
777  */
778 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh,
779 			struct netlink_ext_ack *extack)
780 {
781 	int err = 0;
782 	struct net *net;
783 	struct net_device *dev;
784 
785 	net = cfg->fc_nlinfo.nl_net;
786 	if (nh->nh_gw) {
787 		struct fib_result res;
788 
789 		if (nh->nh_flags & RTNH_F_ONLINK) {
790 			unsigned int addr_type;
791 
792 			if (cfg->fc_scope >= RT_SCOPE_LINK) {
793 				NL_SET_ERR_MSG(extack,
794 					       "Nexthop has invalid scope");
795 				return -EINVAL;
796 			}
797 			dev = __dev_get_by_index(net, nh->nh_oif);
798 			if (!dev)
799 				return -ENODEV;
800 			if (!(dev->flags & IFF_UP)) {
801 				NL_SET_ERR_MSG(extack,
802 					       "Nexthop device is not up");
803 				return -ENETDOWN;
804 			}
805 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
806 			if (addr_type != RTN_UNICAST) {
807 				NL_SET_ERR_MSG(extack,
808 					       "Nexthop has invalid gateway");
809 				return -EINVAL;
810 			}
811 			if (!netif_carrier_ok(dev))
812 				nh->nh_flags |= RTNH_F_LINKDOWN;
813 			nh->nh_dev = dev;
814 			dev_hold(dev);
815 			nh->nh_scope = RT_SCOPE_LINK;
816 			return 0;
817 		}
818 		rcu_read_lock();
819 		{
820 			struct fib_table *tbl = NULL;
821 			struct flowi4 fl4 = {
822 				.daddr = nh->nh_gw,
823 				.flowi4_scope = cfg->fc_scope + 1,
824 				.flowi4_oif = nh->nh_oif,
825 				.flowi4_iif = LOOPBACK_IFINDEX,
826 			};
827 
828 			/* It is not necessary, but requires a bit of thinking */
829 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
830 				fl4.flowi4_scope = RT_SCOPE_LINK;
831 
832 			if (cfg->fc_table)
833 				tbl = fib_get_table(net, cfg->fc_table);
834 
835 			if (tbl)
836 				err = fib_table_lookup(tbl, &fl4, &res,
837 						       FIB_LOOKUP_IGNORE_LINKSTATE |
838 						       FIB_LOOKUP_NOREF);
839 
840 			/* on error or if no table given do full lookup. This
841 			 * is needed for example when nexthops are in the local
842 			 * table rather than the given table
843 			 */
844 			if (!tbl || err) {
845 				err = fib_lookup(net, &fl4, &res,
846 						 FIB_LOOKUP_IGNORE_LINKSTATE);
847 			}
848 
849 			if (err) {
850 				NL_SET_ERR_MSG(extack,
851 					       "Nexthop has invalid gateway");
852 				rcu_read_unlock();
853 				return err;
854 			}
855 		}
856 		err = -EINVAL;
857 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
858 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
859 			goto out;
860 		}
861 		nh->nh_scope = res.scope;
862 		nh->nh_oif = FIB_RES_OIF(res);
863 		nh->nh_dev = dev = FIB_RES_DEV(res);
864 		if (!dev) {
865 			NL_SET_ERR_MSG(extack,
866 				       "No egress device for nexthop gateway");
867 			goto out;
868 		}
869 		dev_hold(dev);
870 		if (!netif_carrier_ok(dev))
871 			nh->nh_flags |= RTNH_F_LINKDOWN;
872 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
873 	} else {
874 		struct in_device *in_dev;
875 
876 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
877 			NL_SET_ERR_MSG(extack,
878 				       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
879 			return -EINVAL;
880 		}
881 		rcu_read_lock();
882 		err = -ENODEV;
883 		in_dev = inetdev_by_index(net, nh->nh_oif);
884 		if (!in_dev)
885 			goto out;
886 		err = -ENETDOWN;
887 		if (!(in_dev->dev->flags & IFF_UP)) {
888 			NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
889 			goto out;
890 		}
891 		nh->nh_dev = in_dev->dev;
892 		dev_hold(nh->nh_dev);
893 		nh->nh_scope = RT_SCOPE_HOST;
894 		if (!netif_carrier_ok(nh->nh_dev))
895 			nh->nh_flags |= RTNH_F_LINKDOWN;
896 		err = 0;
897 	}
898 out:
899 	rcu_read_unlock();
900 	return err;
901 }
902 
903 static inline unsigned int fib_laddr_hashfn(__be32 val)
904 {
905 	unsigned int mask = (fib_info_hash_size - 1);
906 
907 	return ((__force u32)val ^
908 		((__force u32)val >> 7) ^
909 		((__force u32)val >> 14)) & mask;
910 }
911 
912 static struct hlist_head *fib_info_hash_alloc(int bytes)
913 {
914 	if (bytes <= PAGE_SIZE)
915 		return kzalloc(bytes, GFP_KERNEL);
916 	else
917 		return (struct hlist_head *)
918 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
919 					 get_order(bytes));
920 }
921 
922 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
923 {
924 	if (!hash)
925 		return;
926 
927 	if (bytes <= PAGE_SIZE)
928 		kfree(hash);
929 	else
930 		free_pages((unsigned long) hash, get_order(bytes));
931 }
932 
933 static void fib_info_hash_move(struct hlist_head *new_info_hash,
934 			       struct hlist_head *new_laddrhash,
935 			       unsigned int new_size)
936 {
937 	struct hlist_head *old_info_hash, *old_laddrhash;
938 	unsigned int old_size = fib_info_hash_size;
939 	unsigned int i, bytes;
940 
941 	spin_lock_bh(&fib_info_lock);
942 	old_info_hash = fib_info_hash;
943 	old_laddrhash = fib_info_laddrhash;
944 	fib_info_hash_size = new_size;
945 
946 	for (i = 0; i < old_size; i++) {
947 		struct hlist_head *head = &fib_info_hash[i];
948 		struct hlist_node *n;
949 		struct fib_info *fi;
950 
951 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
952 			struct hlist_head *dest;
953 			unsigned int new_hash;
954 
955 			new_hash = fib_info_hashfn(fi);
956 			dest = &new_info_hash[new_hash];
957 			hlist_add_head(&fi->fib_hash, dest);
958 		}
959 	}
960 	fib_info_hash = new_info_hash;
961 
962 	for (i = 0; i < old_size; i++) {
963 		struct hlist_head *lhead = &fib_info_laddrhash[i];
964 		struct hlist_node *n;
965 		struct fib_info *fi;
966 
967 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
968 			struct hlist_head *ldest;
969 			unsigned int new_hash;
970 
971 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
972 			ldest = &new_laddrhash[new_hash];
973 			hlist_add_head(&fi->fib_lhash, ldest);
974 		}
975 	}
976 	fib_info_laddrhash = new_laddrhash;
977 
978 	spin_unlock_bh(&fib_info_lock);
979 
980 	bytes = old_size * sizeof(struct hlist_head *);
981 	fib_info_hash_free(old_info_hash, bytes);
982 	fib_info_hash_free(old_laddrhash, bytes);
983 }
984 
985 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
986 {
987 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
988 					nh->nh_gw,
989 					nh->nh_parent->fib_scope);
990 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
991 
992 	return nh->nh_saddr;
993 }
994 
995 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
996 {
997 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
998 	    fib_prefsrc != cfg->fc_dst) {
999 		u32 tb_id = cfg->fc_table;
1000 		int rc;
1001 
1002 		if (tb_id == RT_TABLE_MAIN)
1003 			tb_id = RT_TABLE_LOCAL;
1004 
1005 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1006 					  fib_prefsrc, tb_id);
1007 
1008 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1009 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1010 						  fib_prefsrc, RT_TABLE_LOCAL);
1011 		}
1012 
1013 		if (rc != RTN_LOCAL)
1014 			return false;
1015 	}
1016 	return true;
1017 }
1018 
1019 static int
1020 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
1021 {
1022 	bool ecn_ca = false;
1023 	struct nlattr *nla;
1024 	int remaining;
1025 
1026 	if (!cfg->fc_mx)
1027 		return 0;
1028 
1029 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1030 		int type = nla_type(nla);
1031 		u32 val;
1032 
1033 		if (!type)
1034 			continue;
1035 		if (type > RTAX_MAX)
1036 			return -EINVAL;
1037 
1038 		if (type == RTAX_CC_ALGO) {
1039 			char tmp[TCP_CA_NAME_MAX];
1040 
1041 			nla_strlcpy(tmp, nla, sizeof(tmp));
1042 			val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1043 			if (val == TCP_CA_UNSPEC)
1044 				return -EINVAL;
1045 		} else {
1046 			val = nla_get_u32(nla);
1047 		}
1048 		if (type == RTAX_ADVMSS && val > 65535 - 40)
1049 			val = 65535 - 40;
1050 		if (type == RTAX_MTU && val > 65535 - 15)
1051 			val = 65535 - 15;
1052 		if (type == RTAX_HOPLIMIT && val > 255)
1053 			val = 255;
1054 		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
1055 			return -EINVAL;
1056 		fi->fib_metrics->metrics[type - 1] = val;
1057 	}
1058 
1059 	if (ecn_ca)
1060 		fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
1061 
1062 	return 0;
1063 }
1064 
1065 struct fib_info *fib_create_info(struct fib_config *cfg,
1066 				 struct netlink_ext_ack *extack)
1067 {
1068 	int err;
1069 	struct fib_info *fi = NULL;
1070 	struct fib_info *ofi;
1071 	int nhs = 1;
1072 	struct net *net = cfg->fc_nlinfo.nl_net;
1073 
1074 	if (cfg->fc_type > RTN_MAX)
1075 		goto err_inval;
1076 
1077 	/* Fast check to catch the most weird cases */
1078 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1079 		NL_SET_ERR_MSG(extack, "Invalid scope");
1080 		goto err_inval;
1081 	}
1082 
1083 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1084 		NL_SET_ERR_MSG(extack,
1085 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1086 		goto err_inval;
1087 	}
1088 
1089 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1090 	if (cfg->fc_mp) {
1091 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1092 		if (nhs == 0)
1093 			goto err_inval;
1094 	}
1095 #endif
1096 
1097 	err = -ENOBUFS;
1098 	if (fib_info_cnt >= fib_info_hash_size) {
1099 		unsigned int new_size = fib_info_hash_size << 1;
1100 		struct hlist_head *new_info_hash;
1101 		struct hlist_head *new_laddrhash;
1102 		unsigned int bytes;
1103 
1104 		if (!new_size)
1105 			new_size = 16;
1106 		bytes = new_size * sizeof(struct hlist_head *);
1107 		new_info_hash = fib_info_hash_alloc(bytes);
1108 		new_laddrhash = fib_info_hash_alloc(bytes);
1109 		if (!new_info_hash || !new_laddrhash) {
1110 			fib_info_hash_free(new_info_hash, bytes);
1111 			fib_info_hash_free(new_laddrhash, bytes);
1112 		} else
1113 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1114 
1115 		if (!fib_info_hash_size)
1116 			goto failure;
1117 	}
1118 
1119 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1120 	if (!fi)
1121 		goto failure;
1122 	if (cfg->fc_mx) {
1123 		fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1124 		if (unlikely(!fi->fib_metrics)) {
1125 			kfree(fi);
1126 			return ERR_PTR(err);
1127 		}
1128 		refcount_set(&fi->fib_metrics->refcnt, 1);
1129 	} else {
1130 		fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1131 	}
1132 	fib_info_cnt++;
1133 	fi->fib_net = net;
1134 	fi->fib_protocol = cfg->fc_protocol;
1135 	fi->fib_scope = cfg->fc_scope;
1136 	fi->fib_flags = cfg->fc_flags;
1137 	fi->fib_priority = cfg->fc_priority;
1138 	fi->fib_prefsrc = cfg->fc_prefsrc;
1139 	fi->fib_type = cfg->fc_type;
1140 	fi->fib_tb_id = cfg->fc_table;
1141 
1142 	fi->fib_nhs = nhs;
1143 	change_nexthops(fi) {
1144 		nexthop_nh->nh_parent = fi;
1145 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1146 		if (!nexthop_nh->nh_pcpu_rth_output)
1147 			goto failure;
1148 	} endfor_nexthops(fi)
1149 
1150 	err = fib_convert_metrics(fi, cfg);
1151 	if (err)
1152 		goto failure;
1153 
1154 	if (cfg->fc_mp) {
1155 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1156 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1157 		if (err != 0)
1158 			goto failure;
1159 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1160 			NL_SET_ERR_MSG(extack,
1161 				       "Nexthop device index does not match RTA_OIF");
1162 			goto err_inval;
1163 		}
1164 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1165 			NL_SET_ERR_MSG(extack,
1166 				       "Nexthop gateway does not match RTA_GATEWAY");
1167 			goto err_inval;
1168 		}
1169 #ifdef CONFIG_IP_ROUTE_CLASSID
1170 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1171 			NL_SET_ERR_MSG(extack,
1172 				       "Nexthop class id does not match RTA_FLOW");
1173 			goto err_inval;
1174 		}
1175 #endif
1176 #else
1177 		NL_SET_ERR_MSG(extack,
1178 			       "Multipath support not enabled in kernel");
1179 		goto err_inval;
1180 #endif
1181 	} else {
1182 		struct fib_nh *nh = fi->fib_nh;
1183 
1184 		if (cfg->fc_encap) {
1185 			struct lwtunnel_state *lwtstate;
1186 
1187 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1188 				NL_SET_ERR_MSG(extack,
1189 					       "LWT encap type not specified");
1190 				goto err_inval;
1191 			}
1192 			err = lwtunnel_build_state(cfg->fc_encap_type,
1193 						   cfg->fc_encap, AF_INET, cfg,
1194 						   &lwtstate, extack);
1195 			if (err)
1196 				goto failure;
1197 
1198 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1199 		}
1200 		nh->nh_oif = cfg->fc_oif;
1201 		nh->nh_gw = cfg->fc_gw;
1202 		nh->nh_flags = cfg->fc_flags;
1203 #ifdef CONFIG_IP_ROUTE_CLASSID
1204 		nh->nh_tclassid = cfg->fc_flow;
1205 		if (nh->nh_tclassid)
1206 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1207 #endif
1208 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1209 		nh->nh_weight = 1;
1210 #endif
1211 	}
1212 
1213 	if (fib_props[cfg->fc_type].error) {
1214 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1215 			NL_SET_ERR_MSG(extack,
1216 				       "Gateway, device and multipath can not be specified for this route type");
1217 			goto err_inval;
1218 		}
1219 		goto link_it;
1220 	} else {
1221 		switch (cfg->fc_type) {
1222 		case RTN_UNICAST:
1223 		case RTN_LOCAL:
1224 		case RTN_BROADCAST:
1225 		case RTN_ANYCAST:
1226 		case RTN_MULTICAST:
1227 			break;
1228 		default:
1229 			NL_SET_ERR_MSG(extack, "Invalid route type");
1230 			goto err_inval;
1231 		}
1232 	}
1233 
1234 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1235 		NL_SET_ERR_MSG(extack, "Invalid scope");
1236 		goto err_inval;
1237 	}
1238 
1239 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1240 		struct fib_nh *nh = fi->fib_nh;
1241 
1242 		/* Local address is added. */
1243 		if (nhs != 1) {
1244 			NL_SET_ERR_MSG(extack,
1245 				       "Route with host scope can not have multiple nexthops");
1246 			goto err_inval;
1247 		}
1248 		if (nh->nh_gw) {
1249 			NL_SET_ERR_MSG(extack,
1250 				       "Route with host scope can not have a gateway");
1251 			goto err_inval;
1252 		}
1253 		nh->nh_scope = RT_SCOPE_NOWHERE;
1254 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1255 		err = -ENODEV;
1256 		if (!nh->nh_dev)
1257 			goto failure;
1258 	} else {
1259 		int linkdown = 0;
1260 
1261 		change_nexthops(fi) {
1262 			err = fib_check_nh(cfg, nexthop_nh, extack);
1263 			if (err != 0)
1264 				goto failure;
1265 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1266 				linkdown++;
1267 		} endfor_nexthops(fi)
1268 		if (linkdown == fi->fib_nhs)
1269 			fi->fib_flags |= RTNH_F_LINKDOWN;
1270 	}
1271 
1272 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1273 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1274 		goto err_inval;
1275 	}
1276 
1277 	change_nexthops(fi) {
1278 		fib_info_update_nh_saddr(net, nexthop_nh);
1279 	} endfor_nexthops(fi)
1280 
1281 	fib_rebalance(fi);
1282 
1283 link_it:
1284 	ofi = fib_find_info(fi);
1285 	if (ofi) {
1286 		fi->fib_dead = 1;
1287 		free_fib_info(fi);
1288 		ofi->fib_treeref++;
1289 		return ofi;
1290 	}
1291 
1292 	fi->fib_treeref++;
1293 	refcount_set(&fi->fib_clntref, 1);
1294 	spin_lock_bh(&fib_info_lock);
1295 	hlist_add_head(&fi->fib_hash,
1296 		       &fib_info_hash[fib_info_hashfn(fi)]);
1297 	if (fi->fib_prefsrc) {
1298 		struct hlist_head *head;
1299 
1300 		head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1301 		hlist_add_head(&fi->fib_lhash, head);
1302 	}
1303 	change_nexthops(fi) {
1304 		struct hlist_head *head;
1305 		unsigned int hash;
1306 
1307 		if (!nexthop_nh->nh_dev)
1308 			continue;
1309 		hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1310 		head = &fib_info_devhash[hash];
1311 		hlist_add_head(&nexthop_nh->nh_hash, head);
1312 	} endfor_nexthops(fi)
1313 	spin_unlock_bh(&fib_info_lock);
1314 	return fi;
1315 
1316 err_inval:
1317 	err = -EINVAL;
1318 
1319 failure:
1320 	if (fi) {
1321 		fi->fib_dead = 1;
1322 		free_fib_info(fi);
1323 	}
1324 
1325 	return ERR_PTR(err);
1326 }
1327 
1328 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1329 		  u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1330 		  struct fib_info *fi, unsigned int flags)
1331 {
1332 	struct nlmsghdr *nlh;
1333 	struct rtmsg *rtm;
1334 
1335 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1336 	if (!nlh)
1337 		return -EMSGSIZE;
1338 
1339 	rtm = nlmsg_data(nlh);
1340 	rtm->rtm_family = AF_INET;
1341 	rtm->rtm_dst_len = dst_len;
1342 	rtm->rtm_src_len = 0;
1343 	rtm->rtm_tos = tos;
1344 	if (tb_id < 256)
1345 		rtm->rtm_table = tb_id;
1346 	else
1347 		rtm->rtm_table = RT_TABLE_COMPAT;
1348 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1349 		goto nla_put_failure;
1350 	rtm->rtm_type = type;
1351 	rtm->rtm_flags = fi->fib_flags;
1352 	rtm->rtm_scope = fi->fib_scope;
1353 	rtm->rtm_protocol = fi->fib_protocol;
1354 
1355 	if (rtm->rtm_dst_len &&
1356 	    nla_put_in_addr(skb, RTA_DST, dst))
1357 		goto nla_put_failure;
1358 	if (fi->fib_priority &&
1359 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1360 		goto nla_put_failure;
1361 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1362 		goto nla_put_failure;
1363 
1364 	if (fi->fib_prefsrc &&
1365 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1366 		goto nla_put_failure;
1367 	if (fi->fib_nhs == 1) {
1368 		if (fi->fib_nh->nh_gw &&
1369 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1370 			goto nla_put_failure;
1371 		if (fi->fib_nh->nh_oif &&
1372 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1373 			goto nla_put_failure;
1374 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1375 			struct in_device *in_dev;
1376 
1377 			rcu_read_lock();
1378 			in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev);
1379 			if (in_dev &&
1380 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1381 				rtm->rtm_flags |= RTNH_F_DEAD;
1382 			rcu_read_unlock();
1383 		}
1384 		if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1385 			rtm->rtm_flags |= RTNH_F_OFFLOAD;
1386 #ifdef CONFIG_IP_ROUTE_CLASSID
1387 		if (fi->fib_nh[0].nh_tclassid &&
1388 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1389 			goto nla_put_failure;
1390 #endif
1391 		if (fi->fib_nh->nh_lwtstate &&
1392 		    lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1393 			goto nla_put_failure;
1394 	}
1395 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1396 	if (fi->fib_nhs > 1) {
1397 		struct rtnexthop *rtnh;
1398 		struct nlattr *mp;
1399 
1400 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1401 		if (!mp)
1402 			goto nla_put_failure;
1403 
1404 		for_nexthops(fi) {
1405 			rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1406 			if (!rtnh)
1407 				goto nla_put_failure;
1408 
1409 			rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1410 			if (nh->nh_flags & RTNH_F_LINKDOWN) {
1411 				struct in_device *in_dev;
1412 
1413 				rcu_read_lock();
1414 				in_dev = __in_dev_get_rcu(nh->nh_dev);
1415 				if (in_dev &&
1416 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1417 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1418 				rcu_read_unlock();
1419 			}
1420 			rtnh->rtnh_hops = nh->nh_weight - 1;
1421 			rtnh->rtnh_ifindex = nh->nh_oif;
1422 
1423 			if (nh->nh_gw &&
1424 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1425 				goto nla_put_failure;
1426 #ifdef CONFIG_IP_ROUTE_CLASSID
1427 			if (nh->nh_tclassid &&
1428 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1429 				goto nla_put_failure;
1430 #endif
1431 			if (nh->nh_lwtstate &&
1432 			    lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1433 				goto nla_put_failure;
1434 
1435 			/* length of rtnetlink header + attributes */
1436 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1437 		} endfor_nexthops(fi);
1438 
1439 		nla_nest_end(skb, mp);
1440 	}
1441 #endif
1442 	nlmsg_end(skb, nlh);
1443 	return 0;
1444 
1445 nla_put_failure:
1446 	nlmsg_cancel(skb, nlh);
1447 	return -EMSGSIZE;
1448 }
1449 
1450 /*
1451  * Update FIB if:
1452  * - local address disappeared -> we must delete all the entries
1453  *   referring to it.
1454  * - device went down -> we must shutdown all nexthops going via it.
1455  */
1456 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1457 {
1458 	int ret = 0;
1459 	unsigned int hash = fib_laddr_hashfn(local);
1460 	struct hlist_head *head = &fib_info_laddrhash[hash];
1461 	struct net *net = dev_net(dev);
1462 	int tb_id = l3mdev_fib_table(dev);
1463 	struct fib_info *fi;
1464 
1465 	if (!fib_info_laddrhash || local == 0)
1466 		return 0;
1467 
1468 	hlist_for_each_entry(fi, head, fib_lhash) {
1469 		if (!net_eq(fi->fib_net, net) ||
1470 		    fi->fib_tb_id != tb_id)
1471 			continue;
1472 		if (fi->fib_prefsrc == local) {
1473 			fi->fib_flags |= RTNH_F_DEAD;
1474 			ret++;
1475 		}
1476 	}
1477 	return ret;
1478 }
1479 
1480 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1481 				 enum fib_event_type event_type)
1482 {
1483 	struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1484 	struct fib_nh_notifier_info info = {
1485 		.fib_nh = fib_nh,
1486 	};
1487 
1488 	switch (event_type) {
1489 	case FIB_EVENT_NH_ADD:
1490 		if (fib_nh->nh_flags & RTNH_F_DEAD)
1491 			break;
1492 		if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1493 		    fib_nh->nh_flags & RTNH_F_LINKDOWN)
1494 			break;
1495 		return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1496 					   &info.info);
1497 	case FIB_EVENT_NH_DEL:
1498 		if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1499 		     fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1500 		    (fib_nh->nh_flags & RTNH_F_DEAD))
1501 			return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1502 						   event_type, &info.info);
1503 	default:
1504 		break;
1505 	}
1506 
1507 	return NOTIFY_DONE;
1508 }
1509 
1510 /* Event              force Flags           Description
1511  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1512  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1513  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1514  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1515  */
1516 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1517 {
1518 	int ret = 0;
1519 	int scope = RT_SCOPE_NOWHERE;
1520 	struct fib_info *prev_fi = NULL;
1521 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1522 	struct hlist_head *head = &fib_info_devhash[hash];
1523 	struct fib_nh *nh;
1524 
1525 	if (force)
1526 		scope = -1;
1527 
1528 	hlist_for_each_entry(nh, head, nh_hash) {
1529 		struct fib_info *fi = nh->nh_parent;
1530 		int dead;
1531 
1532 		BUG_ON(!fi->fib_nhs);
1533 		if (nh->nh_dev != dev || fi == prev_fi)
1534 			continue;
1535 		prev_fi = fi;
1536 		dead = 0;
1537 		change_nexthops(fi) {
1538 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1539 				dead++;
1540 			else if (nexthop_nh->nh_dev == dev &&
1541 				 nexthop_nh->nh_scope != scope) {
1542 				switch (event) {
1543 				case NETDEV_DOWN:
1544 				case NETDEV_UNREGISTER:
1545 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1546 					/* fall through */
1547 				case NETDEV_CHANGE:
1548 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1549 					break;
1550 				}
1551 				call_fib_nh_notifiers(nexthop_nh,
1552 						      FIB_EVENT_NH_DEL);
1553 				dead++;
1554 			}
1555 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1556 			if (event == NETDEV_UNREGISTER &&
1557 			    nexthop_nh->nh_dev == dev) {
1558 				dead = fi->fib_nhs;
1559 				break;
1560 			}
1561 #endif
1562 		} endfor_nexthops(fi)
1563 		if (dead == fi->fib_nhs) {
1564 			switch (event) {
1565 			case NETDEV_DOWN:
1566 			case NETDEV_UNREGISTER:
1567 				fi->fib_flags |= RTNH_F_DEAD;
1568 				/* fall through */
1569 			case NETDEV_CHANGE:
1570 				fi->fib_flags |= RTNH_F_LINKDOWN;
1571 				break;
1572 			}
1573 			ret++;
1574 		}
1575 
1576 		fib_rebalance(fi);
1577 	}
1578 
1579 	return ret;
1580 }
1581 
1582 /* Must be invoked inside of an RCU protected region.  */
1583 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1584 {
1585 	struct fib_info *fi = NULL, *last_resort = NULL;
1586 	struct hlist_head *fa_head = res->fa_head;
1587 	struct fib_table *tb = res->table;
1588 	u8 slen = 32 - res->prefixlen;
1589 	int order = -1, last_idx = -1;
1590 	struct fib_alias *fa, *fa1 = NULL;
1591 	u32 last_prio = res->fi->fib_priority;
1592 	u8 last_tos = 0;
1593 
1594 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1595 		struct fib_info *next_fi = fa->fa_info;
1596 
1597 		if (fa->fa_slen != slen)
1598 			continue;
1599 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1600 			continue;
1601 		if (fa->tb_id != tb->tb_id)
1602 			continue;
1603 		if (next_fi->fib_priority > last_prio &&
1604 		    fa->fa_tos == last_tos) {
1605 			if (last_tos)
1606 				continue;
1607 			break;
1608 		}
1609 		if (next_fi->fib_flags & RTNH_F_DEAD)
1610 			continue;
1611 		last_tos = fa->fa_tos;
1612 		last_prio = next_fi->fib_priority;
1613 
1614 		if (next_fi->fib_scope != res->scope ||
1615 		    fa->fa_type != RTN_UNICAST)
1616 			continue;
1617 		if (!next_fi->fib_nh[0].nh_gw ||
1618 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1619 			continue;
1620 
1621 		fib_alias_accessed(fa);
1622 
1623 		if (!fi) {
1624 			if (next_fi != res->fi)
1625 				break;
1626 			fa1 = fa;
1627 		} else if (!fib_detect_death(fi, order, &last_resort,
1628 					     &last_idx, fa1->fa_default)) {
1629 			fib_result_assign(res, fi);
1630 			fa1->fa_default = order;
1631 			goto out;
1632 		}
1633 		fi = next_fi;
1634 		order++;
1635 	}
1636 
1637 	if (order <= 0 || !fi) {
1638 		if (fa1)
1639 			fa1->fa_default = -1;
1640 		goto out;
1641 	}
1642 
1643 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1644 			      fa1->fa_default)) {
1645 		fib_result_assign(res, fi);
1646 		fa1->fa_default = order;
1647 		goto out;
1648 	}
1649 
1650 	if (last_idx >= 0)
1651 		fib_result_assign(res, last_resort);
1652 	fa1->fa_default = last_idx;
1653 out:
1654 	return;
1655 }
1656 
1657 /*
1658  * Dead device goes up. We wake up dead nexthops.
1659  * It takes sense only on multipath routes.
1660  */
1661 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1662 {
1663 	struct fib_info *prev_fi;
1664 	unsigned int hash;
1665 	struct hlist_head *head;
1666 	struct fib_nh *nh;
1667 	int ret;
1668 
1669 	if (!(dev->flags & IFF_UP))
1670 		return 0;
1671 
1672 	if (nh_flags & RTNH_F_DEAD) {
1673 		unsigned int flags = dev_get_flags(dev);
1674 
1675 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1676 			nh_flags |= RTNH_F_LINKDOWN;
1677 	}
1678 
1679 	prev_fi = NULL;
1680 	hash = fib_devindex_hashfn(dev->ifindex);
1681 	head = &fib_info_devhash[hash];
1682 	ret = 0;
1683 
1684 	hlist_for_each_entry(nh, head, nh_hash) {
1685 		struct fib_info *fi = nh->nh_parent;
1686 		int alive;
1687 
1688 		BUG_ON(!fi->fib_nhs);
1689 		if (nh->nh_dev != dev || fi == prev_fi)
1690 			continue;
1691 
1692 		prev_fi = fi;
1693 		alive = 0;
1694 		change_nexthops(fi) {
1695 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1696 				alive++;
1697 				continue;
1698 			}
1699 			if (!nexthop_nh->nh_dev ||
1700 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1701 				continue;
1702 			if (nexthop_nh->nh_dev != dev ||
1703 			    !__in_dev_get_rtnl(dev))
1704 				continue;
1705 			alive++;
1706 			nexthop_nh->nh_flags &= ~nh_flags;
1707 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1708 		} endfor_nexthops(fi)
1709 
1710 		if (alive > 0) {
1711 			fi->fib_flags &= ~nh_flags;
1712 			ret++;
1713 		}
1714 
1715 		fib_rebalance(fi);
1716 	}
1717 
1718 	return ret;
1719 }
1720 
1721 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1722 static bool fib_good_nh(const struct fib_nh *nh)
1723 {
1724 	int state = NUD_REACHABLE;
1725 
1726 	if (nh->nh_scope == RT_SCOPE_LINK) {
1727 		struct neighbour *n;
1728 
1729 		rcu_read_lock_bh();
1730 
1731 		n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1732 					      (__force u32)nh->nh_gw);
1733 		if (n)
1734 			state = n->nud_state;
1735 
1736 		rcu_read_unlock_bh();
1737 	}
1738 
1739 	return !!(state & NUD_VALID);
1740 }
1741 
1742 void fib_select_multipath(struct fib_result *res, int hash)
1743 {
1744 	struct fib_info *fi = res->fi;
1745 	struct net *net = fi->fib_net;
1746 	bool first = false;
1747 
1748 	for_nexthops(fi) {
1749 		if (hash > atomic_read(&nh->nh_upper_bound))
1750 			continue;
1751 
1752 		if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1753 		    fib_good_nh(nh)) {
1754 			res->nh_sel = nhsel;
1755 			return;
1756 		}
1757 		if (!first) {
1758 			res->nh_sel = nhsel;
1759 			first = true;
1760 		}
1761 	} endfor_nexthops(fi);
1762 }
1763 #endif
1764 
1765 void fib_select_path(struct net *net, struct fib_result *res,
1766 		     struct flowi4 *fl4, const struct sk_buff *skb)
1767 {
1768 	bool oif_check;
1769 
1770 	oif_check = (fl4->flowi4_oif == 0 ||
1771 		     fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1772 
1773 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1774 	if (res->fi->fib_nhs > 1 && oif_check) {
1775 		int h = fib_multipath_hash(res->fi, fl4, skb);
1776 
1777 		fib_select_multipath(res, h);
1778 	}
1779 	else
1780 #endif
1781 	if (!res->prefixlen &&
1782 	    res->table->tb_num_default > 1 &&
1783 	    res->type == RTN_UNICAST && oif_check)
1784 		fib_select_default(fl4, res);
1785 
1786 	if (!fl4->saddr)
1787 		fl4->saddr = FIB_RES_PREFSRC(net, *res);
1788 }
1789 EXPORT_SYMBOL_GPL(fib_select_path);
1790