xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 5a244f48)
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 
605 static inline void fib_add_weight(struct fib_info *fi,
606 				  const struct fib_nh *nh)
607 {
608 	fi->fib_weight += nh->nh_weight;
609 }
610 
611 #else /* CONFIG_IP_ROUTE_MULTIPATH */
612 
613 #define fib_rebalance(fi) do { } while (0)
614 #define fib_add_weight(fi, nh) do { } while (0)
615 
616 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
617 
618 static int fib_encap_match(u16 encap_type,
619 			   struct nlattr *encap,
620 			   const struct fib_nh *nh,
621 			   const struct fib_config *cfg,
622 			   struct netlink_ext_ack *extack)
623 {
624 	struct lwtunnel_state *lwtstate;
625 	int ret, result = 0;
626 
627 	if (encap_type == LWTUNNEL_ENCAP_NONE)
628 		return 0;
629 
630 	ret = lwtunnel_build_state(encap_type, encap, AF_INET,
631 				   cfg, &lwtstate, extack);
632 	if (!ret) {
633 		result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
634 		lwtstate_free(lwtstate);
635 	}
636 
637 	return result;
638 }
639 
640 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
641 		 struct netlink_ext_ack *extack)
642 {
643 #ifdef CONFIG_IP_ROUTE_MULTIPATH
644 	struct rtnexthop *rtnh;
645 	int remaining;
646 #endif
647 
648 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
649 		return 1;
650 
651 	if (cfg->fc_oif || cfg->fc_gw) {
652 		if (cfg->fc_encap) {
653 			if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
654 					    fi->fib_nh, cfg, extack))
655 				return 1;
656 		}
657 		if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
658 		    (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
659 			return 0;
660 		return 1;
661 	}
662 
663 #ifdef CONFIG_IP_ROUTE_MULTIPATH
664 	if (!cfg->fc_mp)
665 		return 0;
666 
667 	rtnh = cfg->fc_mp;
668 	remaining = cfg->fc_mp_len;
669 
670 	for_nexthops(fi) {
671 		int attrlen;
672 
673 		if (!rtnh_ok(rtnh, remaining))
674 			return -EINVAL;
675 
676 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
677 			return 1;
678 
679 		attrlen = rtnh_attrlen(rtnh);
680 		if (attrlen > 0) {
681 			struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
682 
683 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
684 			if (nla && nla_get_in_addr(nla) != nh->nh_gw)
685 				return 1;
686 #ifdef CONFIG_IP_ROUTE_CLASSID
687 			nla = nla_find(attrs, attrlen, RTA_FLOW);
688 			if (nla && nla_get_u32(nla) != nh->nh_tclassid)
689 				return 1;
690 #endif
691 		}
692 
693 		rtnh = rtnh_next(rtnh, &remaining);
694 	} endfor_nexthops(fi);
695 #endif
696 	return 0;
697 }
698 
699 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
700 {
701 	struct nlattr *nla;
702 	int remaining;
703 
704 	if (!cfg->fc_mx)
705 		return true;
706 
707 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
708 		int type = nla_type(nla);
709 		u32 val;
710 
711 		if (!type)
712 			continue;
713 		if (type > RTAX_MAX)
714 			return false;
715 
716 		if (type == RTAX_CC_ALGO) {
717 			char tmp[TCP_CA_NAME_MAX];
718 			bool ecn_ca = false;
719 
720 			nla_strlcpy(tmp, nla, sizeof(tmp));
721 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
722 		} else {
723 			val = nla_get_u32(nla);
724 		}
725 
726 		if (fi->fib_metrics->metrics[type - 1] != val)
727 			return false;
728 	}
729 
730 	return true;
731 }
732 
733 
734 /*
735  * Picture
736  * -------
737  *
738  * Semantics of nexthop is very messy by historical reasons.
739  * We have to take into account, that:
740  * a) gateway can be actually local interface address,
741  *    so that gatewayed route is direct.
742  * b) gateway must be on-link address, possibly
743  *    described not by an ifaddr, but also by a direct route.
744  * c) If both gateway and interface are specified, they should not
745  *    contradict.
746  * d) If we use tunnel routes, gateway could be not on-link.
747  *
748  * Attempt to reconcile all of these (alas, self-contradictory) conditions
749  * results in pretty ugly and hairy code with obscure logic.
750  *
751  * I chose to generalized it instead, so that the size
752  * of code does not increase practically, but it becomes
753  * much more general.
754  * Every prefix is assigned a "scope" value: "host" is local address,
755  * "link" is direct route,
756  * [ ... "site" ... "interior" ... ]
757  * and "universe" is true gateway route with global meaning.
758  *
759  * Every prefix refers to a set of "nexthop"s (gw, oif),
760  * where gw must have narrower scope. This recursion stops
761  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
762  * which means that gw is forced to be on link.
763  *
764  * Code is still hairy, but now it is apparently logically
765  * consistent and very flexible. F.e. as by-product it allows
766  * to co-exists in peace independent exterior and interior
767  * routing processes.
768  *
769  * Normally it looks as following.
770  *
771  * {universe prefix}  -> (gw, oif) [scope link]
772  *		  |
773  *		  |-> {link prefix} -> (gw, oif) [scope local]
774  *					|
775  *					|-> {local prefix} (terminal node)
776  */
777 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
778 			struct fib_nh *nh, struct netlink_ext_ack *extack)
779 {
780 	int err = 0;
781 	struct net *net;
782 	struct net_device *dev;
783 
784 	net = cfg->fc_nlinfo.nl_net;
785 	if (nh->nh_gw) {
786 		struct fib_result res;
787 
788 		if (nh->nh_flags & RTNH_F_ONLINK) {
789 			unsigned int addr_type;
790 
791 			if (cfg->fc_scope >= RT_SCOPE_LINK) {
792 				NL_SET_ERR_MSG(extack,
793 					       "Nexthop has invalid scope");
794 				return -EINVAL;
795 			}
796 			dev = __dev_get_by_index(net, nh->nh_oif);
797 			if (!dev)
798 				return -ENODEV;
799 			if (!(dev->flags & IFF_UP)) {
800 				NL_SET_ERR_MSG(extack,
801 					       "Nexthop device is not up");
802 				return -ENETDOWN;
803 			}
804 			addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
805 			if (addr_type != RTN_UNICAST) {
806 				NL_SET_ERR_MSG(extack,
807 					       "Nexthop has invalid gateway");
808 				return -EINVAL;
809 			}
810 			if (!netif_carrier_ok(dev))
811 				nh->nh_flags |= RTNH_F_LINKDOWN;
812 			nh->nh_dev = dev;
813 			dev_hold(dev);
814 			nh->nh_scope = RT_SCOPE_LINK;
815 			return 0;
816 		}
817 		rcu_read_lock();
818 		{
819 			struct fib_table *tbl = NULL;
820 			struct flowi4 fl4 = {
821 				.daddr = nh->nh_gw,
822 				.flowi4_scope = cfg->fc_scope + 1,
823 				.flowi4_oif = nh->nh_oif,
824 				.flowi4_iif = LOOPBACK_IFINDEX,
825 			};
826 
827 			/* It is not necessary, but requires a bit of thinking */
828 			if (fl4.flowi4_scope < RT_SCOPE_LINK)
829 				fl4.flowi4_scope = RT_SCOPE_LINK;
830 
831 			if (cfg->fc_table)
832 				tbl = fib_get_table(net, cfg->fc_table);
833 
834 			if (tbl)
835 				err = fib_table_lookup(tbl, &fl4, &res,
836 						       FIB_LOOKUP_IGNORE_LINKSTATE |
837 						       FIB_LOOKUP_NOREF);
838 
839 			/* on error or if no table given do full lookup. This
840 			 * is needed for example when nexthops are in the local
841 			 * table rather than the given table
842 			 */
843 			if (!tbl || err) {
844 				err = fib_lookup(net, &fl4, &res,
845 						 FIB_LOOKUP_IGNORE_LINKSTATE);
846 			}
847 
848 			if (err) {
849 				NL_SET_ERR_MSG(extack,
850 					       "Nexthop has invalid gateway");
851 				rcu_read_unlock();
852 				return err;
853 			}
854 		}
855 		err = -EINVAL;
856 		if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
857 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
858 			goto out;
859 		}
860 		nh->nh_scope = res.scope;
861 		nh->nh_oif = FIB_RES_OIF(res);
862 		nh->nh_dev = dev = FIB_RES_DEV(res);
863 		if (!dev) {
864 			NL_SET_ERR_MSG(extack,
865 				       "No egress device for nexthop gateway");
866 			goto out;
867 		}
868 		dev_hold(dev);
869 		if (!netif_carrier_ok(dev))
870 			nh->nh_flags |= RTNH_F_LINKDOWN;
871 		err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
872 	} else {
873 		struct in_device *in_dev;
874 
875 		if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
876 			NL_SET_ERR_MSG(extack,
877 				       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
878 			return -EINVAL;
879 		}
880 		rcu_read_lock();
881 		err = -ENODEV;
882 		in_dev = inetdev_by_index(net, nh->nh_oif);
883 		if (!in_dev)
884 			goto out;
885 		err = -ENETDOWN;
886 		if (!(in_dev->dev->flags & IFF_UP)) {
887 			NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
888 			goto out;
889 		}
890 		nh->nh_dev = in_dev->dev;
891 		dev_hold(nh->nh_dev);
892 		nh->nh_scope = RT_SCOPE_HOST;
893 		if (!netif_carrier_ok(nh->nh_dev))
894 			nh->nh_flags |= RTNH_F_LINKDOWN;
895 		err = 0;
896 	}
897 out:
898 	rcu_read_unlock();
899 	return err;
900 }
901 
902 static inline unsigned int fib_laddr_hashfn(__be32 val)
903 {
904 	unsigned int mask = (fib_info_hash_size - 1);
905 
906 	return ((__force u32)val ^
907 		((__force u32)val >> 7) ^
908 		((__force u32)val >> 14)) & mask;
909 }
910 
911 static struct hlist_head *fib_info_hash_alloc(int bytes)
912 {
913 	if (bytes <= PAGE_SIZE)
914 		return kzalloc(bytes, GFP_KERNEL);
915 	else
916 		return (struct hlist_head *)
917 			__get_free_pages(GFP_KERNEL | __GFP_ZERO,
918 					 get_order(bytes));
919 }
920 
921 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
922 {
923 	if (!hash)
924 		return;
925 
926 	if (bytes <= PAGE_SIZE)
927 		kfree(hash);
928 	else
929 		free_pages((unsigned long) hash, get_order(bytes));
930 }
931 
932 static void fib_info_hash_move(struct hlist_head *new_info_hash,
933 			       struct hlist_head *new_laddrhash,
934 			       unsigned int new_size)
935 {
936 	struct hlist_head *old_info_hash, *old_laddrhash;
937 	unsigned int old_size = fib_info_hash_size;
938 	unsigned int i, bytes;
939 
940 	spin_lock_bh(&fib_info_lock);
941 	old_info_hash = fib_info_hash;
942 	old_laddrhash = fib_info_laddrhash;
943 	fib_info_hash_size = new_size;
944 
945 	for (i = 0; i < old_size; i++) {
946 		struct hlist_head *head = &fib_info_hash[i];
947 		struct hlist_node *n;
948 		struct fib_info *fi;
949 
950 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
951 			struct hlist_head *dest;
952 			unsigned int new_hash;
953 
954 			new_hash = fib_info_hashfn(fi);
955 			dest = &new_info_hash[new_hash];
956 			hlist_add_head(&fi->fib_hash, dest);
957 		}
958 	}
959 	fib_info_hash = new_info_hash;
960 
961 	for (i = 0; i < old_size; i++) {
962 		struct hlist_head *lhead = &fib_info_laddrhash[i];
963 		struct hlist_node *n;
964 		struct fib_info *fi;
965 
966 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
967 			struct hlist_head *ldest;
968 			unsigned int new_hash;
969 
970 			new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
971 			ldest = &new_laddrhash[new_hash];
972 			hlist_add_head(&fi->fib_lhash, ldest);
973 		}
974 	}
975 	fib_info_laddrhash = new_laddrhash;
976 
977 	spin_unlock_bh(&fib_info_lock);
978 
979 	bytes = old_size * sizeof(struct hlist_head *);
980 	fib_info_hash_free(old_info_hash, bytes);
981 	fib_info_hash_free(old_laddrhash, bytes);
982 }
983 
984 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
985 {
986 	nh->nh_saddr = inet_select_addr(nh->nh_dev,
987 					nh->nh_gw,
988 					nh->nh_parent->fib_scope);
989 	nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
990 
991 	return nh->nh_saddr;
992 }
993 
994 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
995 {
996 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
997 	    fib_prefsrc != cfg->fc_dst) {
998 		u32 tb_id = cfg->fc_table;
999 		int rc;
1000 
1001 		if (tb_id == RT_TABLE_MAIN)
1002 			tb_id = RT_TABLE_LOCAL;
1003 
1004 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1005 					  fib_prefsrc, tb_id);
1006 
1007 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1008 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1009 						  fib_prefsrc, RT_TABLE_LOCAL);
1010 		}
1011 
1012 		if (rc != RTN_LOCAL)
1013 			return false;
1014 	}
1015 	return true;
1016 }
1017 
1018 static int
1019 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
1020 {
1021 	bool ecn_ca = false;
1022 	struct nlattr *nla;
1023 	int remaining;
1024 
1025 	if (!cfg->fc_mx)
1026 		return 0;
1027 
1028 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1029 		int type = nla_type(nla);
1030 		u32 val;
1031 
1032 		if (!type)
1033 			continue;
1034 		if (type > RTAX_MAX)
1035 			return -EINVAL;
1036 
1037 		if (type == RTAX_CC_ALGO) {
1038 			char tmp[TCP_CA_NAME_MAX];
1039 
1040 			nla_strlcpy(tmp, nla, sizeof(tmp));
1041 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
1042 			if (val == TCP_CA_UNSPEC)
1043 				return -EINVAL;
1044 		} else {
1045 			val = nla_get_u32(nla);
1046 		}
1047 		if (type == RTAX_ADVMSS && val > 65535 - 40)
1048 			val = 65535 - 40;
1049 		if (type == RTAX_MTU && val > 65535 - 15)
1050 			val = 65535 - 15;
1051 		if (type == RTAX_HOPLIMIT && val > 255)
1052 			val = 255;
1053 		if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
1054 			return -EINVAL;
1055 		fi->fib_metrics->metrics[type - 1] = val;
1056 	}
1057 
1058 	if (ecn_ca)
1059 		fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
1060 
1061 	return 0;
1062 }
1063 
1064 struct fib_info *fib_create_info(struct fib_config *cfg,
1065 				 struct netlink_ext_ack *extack)
1066 {
1067 	int err;
1068 	struct fib_info *fi = NULL;
1069 	struct fib_info *ofi;
1070 	int nhs = 1;
1071 	struct net *net = cfg->fc_nlinfo.nl_net;
1072 
1073 	if (cfg->fc_type > RTN_MAX)
1074 		goto err_inval;
1075 
1076 	/* Fast check to catch the most weird cases */
1077 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1078 		NL_SET_ERR_MSG(extack, "Invalid scope");
1079 		goto err_inval;
1080 	}
1081 
1082 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1083 		NL_SET_ERR_MSG(extack,
1084 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1085 		goto err_inval;
1086 	}
1087 
1088 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1089 	if (cfg->fc_mp) {
1090 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1091 		if (nhs == 0)
1092 			goto err_inval;
1093 	}
1094 #endif
1095 
1096 	err = -ENOBUFS;
1097 	if (fib_info_cnt >= fib_info_hash_size) {
1098 		unsigned int new_size = fib_info_hash_size << 1;
1099 		struct hlist_head *new_info_hash;
1100 		struct hlist_head *new_laddrhash;
1101 		unsigned int bytes;
1102 
1103 		if (!new_size)
1104 			new_size = 16;
1105 		bytes = new_size * sizeof(struct hlist_head *);
1106 		new_info_hash = fib_info_hash_alloc(bytes);
1107 		new_laddrhash = fib_info_hash_alloc(bytes);
1108 		if (!new_info_hash || !new_laddrhash) {
1109 			fib_info_hash_free(new_info_hash, bytes);
1110 			fib_info_hash_free(new_laddrhash, bytes);
1111 		} else
1112 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1113 
1114 		if (!fib_info_hash_size)
1115 			goto failure;
1116 	}
1117 
1118 	fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1119 	if (!fi)
1120 		goto failure;
1121 	if (cfg->fc_mx) {
1122 		fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1123 		if (unlikely(!fi->fib_metrics)) {
1124 			kfree(fi);
1125 			return ERR_PTR(err);
1126 		}
1127 		refcount_set(&fi->fib_metrics->refcnt, 1);
1128 	} else {
1129 		fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1130 	}
1131 	fib_info_cnt++;
1132 	fi->fib_net = net;
1133 	fi->fib_protocol = cfg->fc_protocol;
1134 	fi->fib_scope = cfg->fc_scope;
1135 	fi->fib_flags = cfg->fc_flags;
1136 	fi->fib_priority = cfg->fc_priority;
1137 	fi->fib_prefsrc = cfg->fc_prefsrc;
1138 	fi->fib_type = cfg->fc_type;
1139 	fi->fib_tb_id = cfg->fc_table;
1140 
1141 	fi->fib_nhs = nhs;
1142 	change_nexthops(fi) {
1143 		nexthop_nh->nh_parent = fi;
1144 		nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1145 		if (!nexthop_nh->nh_pcpu_rth_output)
1146 			goto failure;
1147 	} endfor_nexthops(fi)
1148 
1149 	err = fib_convert_metrics(fi, cfg);
1150 	if (err)
1151 		goto failure;
1152 
1153 	if (cfg->fc_mp) {
1154 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1155 		err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1156 		if (err != 0)
1157 			goto failure;
1158 		if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1159 			NL_SET_ERR_MSG(extack,
1160 				       "Nexthop device index does not match RTA_OIF");
1161 			goto err_inval;
1162 		}
1163 		if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1164 			NL_SET_ERR_MSG(extack,
1165 				       "Nexthop gateway does not match RTA_GATEWAY");
1166 			goto err_inval;
1167 		}
1168 #ifdef CONFIG_IP_ROUTE_CLASSID
1169 		if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1170 			NL_SET_ERR_MSG(extack,
1171 				       "Nexthop class id does not match RTA_FLOW");
1172 			goto err_inval;
1173 		}
1174 #endif
1175 #else
1176 		NL_SET_ERR_MSG(extack,
1177 			       "Multipath support not enabled in kernel");
1178 		goto err_inval;
1179 #endif
1180 	} else {
1181 		struct fib_nh *nh = fi->fib_nh;
1182 
1183 		if (cfg->fc_encap) {
1184 			struct lwtunnel_state *lwtstate;
1185 
1186 			if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1187 				NL_SET_ERR_MSG(extack,
1188 					       "LWT encap type not specified");
1189 				goto err_inval;
1190 			}
1191 			err = lwtunnel_build_state(cfg->fc_encap_type,
1192 						   cfg->fc_encap, AF_INET, cfg,
1193 						   &lwtstate, extack);
1194 			if (err)
1195 				goto failure;
1196 
1197 			nh->nh_lwtstate = lwtstate_get(lwtstate);
1198 		}
1199 		nh->nh_oif = cfg->fc_oif;
1200 		nh->nh_gw = cfg->fc_gw;
1201 		nh->nh_flags = cfg->fc_flags;
1202 #ifdef CONFIG_IP_ROUTE_CLASSID
1203 		nh->nh_tclassid = cfg->fc_flow;
1204 		if (nh->nh_tclassid)
1205 			fi->fib_net->ipv4.fib_num_tclassid_users++;
1206 #endif
1207 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1208 		nh->nh_weight = 1;
1209 #endif
1210 	}
1211 
1212 	if (fib_props[cfg->fc_type].error) {
1213 		if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1214 			NL_SET_ERR_MSG(extack,
1215 				       "Gateway, device and multipath can not be specified for this route type");
1216 			goto err_inval;
1217 		}
1218 		goto link_it;
1219 	} else {
1220 		switch (cfg->fc_type) {
1221 		case RTN_UNICAST:
1222 		case RTN_LOCAL:
1223 		case RTN_BROADCAST:
1224 		case RTN_ANYCAST:
1225 		case RTN_MULTICAST:
1226 			break;
1227 		default:
1228 			NL_SET_ERR_MSG(extack, "Invalid route type");
1229 			goto err_inval;
1230 		}
1231 	}
1232 
1233 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1234 		NL_SET_ERR_MSG(extack, "Invalid scope");
1235 		goto err_inval;
1236 	}
1237 
1238 	if (cfg->fc_scope == RT_SCOPE_HOST) {
1239 		struct fib_nh *nh = fi->fib_nh;
1240 
1241 		/* Local address is added. */
1242 		if (nhs != 1) {
1243 			NL_SET_ERR_MSG(extack,
1244 				       "Route with host scope can not have multiple nexthops");
1245 			goto err_inval;
1246 		}
1247 		if (nh->nh_gw) {
1248 			NL_SET_ERR_MSG(extack,
1249 				       "Route with host scope can not have a gateway");
1250 			goto err_inval;
1251 		}
1252 		nh->nh_scope = RT_SCOPE_NOWHERE;
1253 		nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1254 		err = -ENODEV;
1255 		if (!nh->nh_dev)
1256 			goto failure;
1257 	} else {
1258 		int linkdown = 0;
1259 
1260 		change_nexthops(fi) {
1261 			err = fib_check_nh(cfg, fi, nexthop_nh, extack);
1262 			if (err != 0)
1263 				goto failure;
1264 			if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1265 				linkdown++;
1266 		} endfor_nexthops(fi)
1267 		if (linkdown == fi->fib_nhs)
1268 			fi->fib_flags |= RTNH_F_LINKDOWN;
1269 	}
1270 
1271 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1272 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1273 		goto err_inval;
1274 	}
1275 
1276 	change_nexthops(fi) {
1277 		fib_info_update_nh_saddr(net, nexthop_nh);
1278 		fib_add_weight(fi, 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 		struct in_device *in_dev;
1369 
1370 		if (fi->fib_nh->nh_gw &&
1371 		    nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1372 			goto nla_put_failure;
1373 		if (fi->fib_nh->nh_oif &&
1374 		    nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1375 			goto nla_put_failure;
1376 		if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1377 			in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1378 			if (in_dev &&
1379 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1380 				rtm->rtm_flags |= RTNH_F_DEAD;
1381 		}
1382 		if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1383 			rtm->rtm_flags |= RTNH_F_OFFLOAD;
1384 #ifdef CONFIG_IP_ROUTE_CLASSID
1385 		if (fi->fib_nh[0].nh_tclassid &&
1386 		    nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1387 			goto nla_put_failure;
1388 #endif
1389 		if (fi->fib_nh->nh_lwtstate &&
1390 		    lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1391 			goto nla_put_failure;
1392 	}
1393 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1394 	if (fi->fib_nhs > 1) {
1395 		struct rtnexthop *rtnh;
1396 		struct nlattr *mp;
1397 
1398 		mp = nla_nest_start(skb, RTA_MULTIPATH);
1399 		if (!mp)
1400 			goto nla_put_failure;
1401 
1402 		for_nexthops(fi) {
1403 			struct in_device *in_dev;
1404 
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 				in_dev = __in_dev_get_rtnl(nh->nh_dev);
1412 				if (in_dev &&
1413 				    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1414 					rtnh->rtnh_flags |= RTNH_F_DEAD;
1415 			}
1416 			rtnh->rtnh_hops = nh->nh_weight - 1;
1417 			rtnh->rtnh_ifindex = nh->nh_oif;
1418 
1419 			if (nh->nh_gw &&
1420 			    nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1421 				goto nla_put_failure;
1422 #ifdef CONFIG_IP_ROUTE_CLASSID
1423 			if (nh->nh_tclassid &&
1424 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1425 				goto nla_put_failure;
1426 #endif
1427 			if (nh->nh_lwtstate &&
1428 			    lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1429 				goto nla_put_failure;
1430 
1431 			/* length of rtnetlink header + attributes */
1432 			rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1433 		} endfor_nexthops(fi);
1434 
1435 		nla_nest_end(skb, mp);
1436 	}
1437 #endif
1438 	nlmsg_end(skb, nlh);
1439 	return 0;
1440 
1441 nla_put_failure:
1442 	nlmsg_cancel(skb, nlh);
1443 	return -EMSGSIZE;
1444 }
1445 
1446 /*
1447  * Update FIB if:
1448  * - local address disappeared -> we must delete all the entries
1449  *   referring to it.
1450  * - device went down -> we must shutdown all nexthops going via it.
1451  */
1452 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1453 {
1454 	int ret = 0;
1455 	unsigned int hash = fib_laddr_hashfn(local);
1456 	struct hlist_head *head = &fib_info_laddrhash[hash];
1457 	struct net *net = dev_net(dev);
1458 	int tb_id = l3mdev_fib_table(dev);
1459 	struct fib_info *fi;
1460 
1461 	if (!fib_info_laddrhash || local == 0)
1462 		return 0;
1463 
1464 	hlist_for_each_entry(fi, head, fib_lhash) {
1465 		if (!net_eq(fi->fib_net, net) ||
1466 		    fi->fib_tb_id != tb_id)
1467 			continue;
1468 		if (fi->fib_prefsrc == local) {
1469 			fi->fib_flags |= RTNH_F_DEAD;
1470 			ret++;
1471 		}
1472 	}
1473 	return ret;
1474 }
1475 
1476 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1477 				 enum fib_event_type event_type)
1478 {
1479 	struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1480 	struct fib_nh_notifier_info info = {
1481 		.fib_nh = fib_nh,
1482 	};
1483 
1484 	switch (event_type) {
1485 	case FIB_EVENT_NH_ADD:
1486 		if (fib_nh->nh_flags & RTNH_F_DEAD)
1487 			break;
1488 		if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1489 		    fib_nh->nh_flags & RTNH_F_LINKDOWN)
1490 			break;
1491 		return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1492 					   &info.info);
1493 	case FIB_EVENT_NH_DEL:
1494 		if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1495 		     fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1496 		    (fib_nh->nh_flags & RTNH_F_DEAD))
1497 			return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1498 						   event_type, &info.info);
1499 	default:
1500 		break;
1501 	}
1502 
1503 	return NOTIFY_DONE;
1504 }
1505 
1506 /* Event              force Flags           Description
1507  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1508  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1509  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1510  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1511  */
1512 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1513 {
1514 	int ret = 0;
1515 	int scope = RT_SCOPE_NOWHERE;
1516 	struct fib_info *prev_fi = NULL;
1517 	unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1518 	struct hlist_head *head = &fib_info_devhash[hash];
1519 	struct fib_nh *nh;
1520 
1521 	if (force)
1522 		scope = -1;
1523 
1524 	hlist_for_each_entry(nh, head, nh_hash) {
1525 		struct fib_info *fi = nh->nh_parent;
1526 		int dead;
1527 
1528 		BUG_ON(!fi->fib_nhs);
1529 		if (nh->nh_dev != dev || fi == prev_fi)
1530 			continue;
1531 		prev_fi = fi;
1532 		dead = 0;
1533 		change_nexthops(fi) {
1534 			if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1535 				dead++;
1536 			else if (nexthop_nh->nh_dev == dev &&
1537 				 nexthop_nh->nh_scope != scope) {
1538 				switch (event) {
1539 				case NETDEV_DOWN:
1540 				case NETDEV_UNREGISTER:
1541 					nexthop_nh->nh_flags |= RTNH_F_DEAD;
1542 					/* fall through */
1543 				case NETDEV_CHANGE:
1544 					nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1545 					break;
1546 				}
1547 				call_fib_nh_notifiers(nexthop_nh,
1548 						      FIB_EVENT_NH_DEL);
1549 				dead++;
1550 			}
1551 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1552 			if (event == NETDEV_UNREGISTER &&
1553 			    nexthop_nh->nh_dev == dev) {
1554 				dead = fi->fib_nhs;
1555 				break;
1556 			}
1557 #endif
1558 		} endfor_nexthops(fi)
1559 		if (dead == fi->fib_nhs) {
1560 			switch (event) {
1561 			case NETDEV_DOWN:
1562 			case NETDEV_UNREGISTER:
1563 				fi->fib_flags |= RTNH_F_DEAD;
1564 				/* fall through */
1565 			case NETDEV_CHANGE:
1566 				fi->fib_flags |= RTNH_F_LINKDOWN;
1567 				break;
1568 			}
1569 			ret++;
1570 		}
1571 
1572 		fib_rebalance(fi);
1573 	}
1574 
1575 	return ret;
1576 }
1577 
1578 /* Must be invoked inside of an RCU protected region.  */
1579 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1580 {
1581 	struct fib_info *fi = NULL, *last_resort = NULL;
1582 	struct hlist_head *fa_head = res->fa_head;
1583 	struct fib_table *tb = res->table;
1584 	u8 slen = 32 - res->prefixlen;
1585 	int order = -1, last_idx = -1;
1586 	struct fib_alias *fa, *fa1 = NULL;
1587 	u32 last_prio = res->fi->fib_priority;
1588 	u8 last_tos = 0;
1589 
1590 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1591 		struct fib_info *next_fi = fa->fa_info;
1592 
1593 		if (fa->fa_slen != slen)
1594 			continue;
1595 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1596 			continue;
1597 		if (fa->tb_id != tb->tb_id)
1598 			continue;
1599 		if (next_fi->fib_priority > last_prio &&
1600 		    fa->fa_tos == last_tos) {
1601 			if (last_tos)
1602 				continue;
1603 			break;
1604 		}
1605 		if (next_fi->fib_flags & RTNH_F_DEAD)
1606 			continue;
1607 		last_tos = fa->fa_tos;
1608 		last_prio = next_fi->fib_priority;
1609 
1610 		if (next_fi->fib_scope != res->scope ||
1611 		    fa->fa_type != RTN_UNICAST)
1612 			continue;
1613 		if (!next_fi->fib_nh[0].nh_gw ||
1614 		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1615 			continue;
1616 
1617 		fib_alias_accessed(fa);
1618 
1619 		if (!fi) {
1620 			if (next_fi != res->fi)
1621 				break;
1622 			fa1 = fa;
1623 		} else if (!fib_detect_death(fi, order, &last_resort,
1624 					     &last_idx, fa1->fa_default)) {
1625 			fib_result_assign(res, fi);
1626 			fa1->fa_default = order;
1627 			goto out;
1628 		}
1629 		fi = next_fi;
1630 		order++;
1631 	}
1632 
1633 	if (order <= 0 || !fi) {
1634 		if (fa1)
1635 			fa1->fa_default = -1;
1636 		goto out;
1637 	}
1638 
1639 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1640 			      fa1->fa_default)) {
1641 		fib_result_assign(res, fi);
1642 		fa1->fa_default = order;
1643 		goto out;
1644 	}
1645 
1646 	if (last_idx >= 0)
1647 		fib_result_assign(res, last_resort);
1648 	fa1->fa_default = last_idx;
1649 out:
1650 	return;
1651 }
1652 
1653 /*
1654  * Dead device goes up. We wake up dead nexthops.
1655  * It takes sense only on multipath routes.
1656  */
1657 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1658 {
1659 	struct fib_info *prev_fi;
1660 	unsigned int hash;
1661 	struct hlist_head *head;
1662 	struct fib_nh *nh;
1663 	int ret;
1664 
1665 	if (!(dev->flags & IFF_UP))
1666 		return 0;
1667 
1668 	if (nh_flags & RTNH_F_DEAD) {
1669 		unsigned int flags = dev_get_flags(dev);
1670 
1671 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1672 			nh_flags |= RTNH_F_LINKDOWN;
1673 	}
1674 
1675 	prev_fi = NULL;
1676 	hash = fib_devindex_hashfn(dev->ifindex);
1677 	head = &fib_info_devhash[hash];
1678 	ret = 0;
1679 
1680 	hlist_for_each_entry(nh, head, nh_hash) {
1681 		struct fib_info *fi = nh->nh_parent;
1682 		int alive;
1683 
1684 		BUG_ON(!fi->fib_nhs);
1685 		if (nh->nh_dev != dev || fi == prev_fi)
1686 			continue;
1687 
1688 		prev_fi = fi;
1689 		alive = 0;
1690 		change_nexthops(fi) {
1691 			if (!(nexthop_nh->nh_flags & nh_flags)) {
1692 				alive++;
1693 				continue;
1694 			}
1695 			if (!nexthop_nh->nh_dev ||
1696 			    !(nexthop_nh->nh_dev->flags & IFF_UP))
1697 				continue;
1698 			if (nexthop_nh->nh_dev != dev ||
1699 			    !__in_dev_get_rtnl(dev))
1700 				continue;
1701 			alive++;
1702 			nexthop_nh->nh_flags &= ~nh_flags;
1703 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1704 		} endfor_nexthops(fi)
1705 
1706 		if (alive > 0) {
1707 			fi->fib_flags &= ~nh_flags;
1708 			ret++;
1709 		}
1710 
1711 		fib_rebalance(fi);
1712 	}
1713 
1714 	return ret;
1715 }
1716 
1717 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1718 static bool fib_good_nh(const struct fib_nh *nh)
1719 {
1720 	int state = NUD_REACHABLE;
1721 
1722 	if (nh->nh_scope == RT_SCOPE_LINK) {
1723 		struct neighbour *n;
1724 
1725 		rcu_read_lock_bh();
1726 
1727 		n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1728 					      (__force u32)nh->nh_gw);
1729 		if (n)
1730 			state = n->nud_state;
1731 
1732 		rcu_read_unlock_bh();
1733 	}
1734 
1735 	return !!(state & NUD_VALID);
1736 }
1737 
1738 void fib_select_multipath(struct fib_result *res, int hash)
1739 {
1740 	struct fib_info *fi = res->fi;
1741 	struct net *net = fi->fib_net;
1742 	bool first = false;
1743 
1744 	for_nexthops(fi) {
1745 		if (hash > atomic_read(&nh->nh_upper_bound))
1746 			continue;
1747 
1748 		if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1749 		    fib_good_nh(nh)) {
1750 			res->nh_sel = nhsel;
1751 			return;
1752 		}
1753 		if (!first) {
1754 			res->nh_sel = nhsel;
1755 			first = true;
1756 		}
1757 	} endfor_nexthops(fi);
1758 }
1759 #endif
1760 
1761 void fib_select_path(struct net *net, struct fib_result *res,
1762 		     struct flowi4 *fl4, const struct sk_buff *skb)
1763 {
1764 	bool oif_check;
1765 
1766 	oif_check = (fl4->flowi4_oif == 0 ||
1767 		     fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1768 
1769 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1770 	if (res->fi->fib_nhs > 1 && oif_check) {
1771 		int h = fib_multipath_hash(res->fi, fl4, skb);
1772 
1773 		fib_select_multipath(res, h);
1774 	}
1775 	else
1776 #endif
1777 	if (!res->prefixlen &&
1778 	    res->table->tb_num_default > 1 &&
1779 	    res->type == RTN_UNICAST && oif_check)
1780 		fib_select_default(fl4, res);
1781 
1782 	if (!fl4->saddr)
1783 		fl4->saddr = FIB_RES_PREFSRC(net, *res);
1784 }
1785 EXPORT_SYMBOL_GPL(fib_select_path);
1786