xref: /openbmc/linux/net/ipv4/fib_semantics.c (revision 69f397e6)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: semantics.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/uaccess.h>
13 #include <linux/bitops.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
17 #include <linux/mm.h>
18 #include <linux/string.h>
19 #include <linux/socket.h>
20 #include <linux/sockios.h>
21 #include <linux/errno.h>
22 #include <linux/in.h>
23 #include <linux/inet.h>
24 #include <linux/inetdevice.h>
25 #include <linux/netdevice.h>
26 #include <linux/if_arp.h>
27 #include <linux/proc_fs.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/netlink.h>
32 #include <linux/hash.h>
33 #include <linux/nospec.h>
34 
35 #include <net/arp.h>
36 #include <net/inet_dscp.h>
37 #include <net/ip.h>
38 #include <net/protocol.h>
39 #include <net/route.h>
40 #include <net/tcp.h>
41 #include <net/sock.h>
42 #include <net/ip_fib.h>
43 #include <net/ip6_fib.h>
44 #include <net/nexthop.h>
45 #include <net/netlink.h>
46 #include <net/rtnh.h>
47 #include <net/lwtunnel.h>
48 #include <net/fib_notifier.h>
49 #include <net/addrconf.h>
50 
51 #include "fib_lookup.h"
52 
53 static DEFINE_SPINLOCK(fib_info_lock);
54 static struct hlist_head *fib_info_hash;
55 static struct hlist_head *fib_info_laddrhash;
56 static unsigned int fib_info_hash_size;
57 static unsigned int fib_info_hash_bits;
58 static unsigned int fib_info_cnt;
59 
60 #define DEVINDEX_HASHBITS 8
61 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
62 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
63 
64 /* for_nexthops and change_nexthops only used when nexthop object
65  * is not set in a fib_info. The logic within can reference fib_nh.
66  */
67 #ifdef CONFIG_IP_ROUTE_MULTIPATH
68 
69 #define for_nexthops(fi) {						\
70 	int nhsel; const struct fib_nh *nh;				\
71 	for (nhsel = 0, nh = (fi)->fib_nh;				\
72 	     nhsel < fib_info_num_path((fi));				\
73 	     nh++, nhsel++)
74 
75 #define change_nexthops(fi) {						\
76 	int nhsel; struct fib_nh *nexthop_nh;				\
77 	for (nhsel = 0,	nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
78 	     nhsel < fib_info_num_path((fi));				\
79 	     nexthop_nh++, nhsel++)
80 
81 #else /* CONFIG_IP_ROUTE_MULTIPATH */
82 
83 /* Hope, that gcc will optimize it to get rid of dummy loop */
84 
85 #define for_nexthops(fi) {						\
86 	int nhsel; const struct fib_nh *nh = (fi)->fib_nh;		\
87 	for (nhsel = 0; nhsel < 1; nhsel++)
88 
89 #define change_nexthops(fi) {						\
90 	int nhsel;							\
91 	struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);	\
92 	for (nhsel = 0; nhsel < 1; nhsel++)
93 
94 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
95 
96 #define endfor_nexthops(fi) }
97 
98 
99 const struct fib_prop fib_props[RTN_MAX + 1] = {
100 	[RTN_UNSPEC] = {
101 		.error	= 0,
102 		.scope	= RT_SCOPE_NOWHERE,
103 	},
104 	[RTN_UNICAST] = {
105 		.error	= 0,
106 		.scope	= RT_SCOPE_UNIVERSE,
107 	},
108 	[RTN_LOCAL] = {
109 		.error	= 0,
110 		.scope	= RT_SCOPE_HOST,
111 	},
112 	[RTN_BROADCAST] = {
113 		.error	= 0,
114 		.scope	= RT_SCOPE_LINK,
115 	},
116 	[RTN_ANYCAST] = {
117 		.error	= 0,
118 		.scope	= RT_SCOPE_LINK,
119 	},
120 	[RTN_MULTICAST] = {
121 		.error	= 0,
122 		.scope	= RT_SCOPE_UNIVERSE,
123 	},
124 	[RTN_BLACKHOLE] = {
125 		.error	= -EINVAL,
126 		.scope	= RT_SCOPE_UNIVERSE,
127 	},
128 	[RTN_UNREACHABLE] = {
129 		.error	= -EHOSTUNREACH,
130 		.scope	= RT_SCOPE_UNIVERSE,
131 	},
132 	[RTN_PROHIBIT] = {
133 		.error	= -EACCES,
134 		.scope	= RT_SCOPE_UNIVERSE,
135 	},
136 	[RTN_THROW] = {
137 		.error	= -EAGAIN,
138 		.scope	= RT_SCOPE_UNIVERSE,
139 	},
140 	[RTN_NAT] = {
141 		.error	= -EINVAL,
142 		.scope	= RT_SCOPE_NOWHERE,
143 	},
144 	[RTN_XRESOLVE] = {
145 		.error	= -EINVAL,
146 		.scope	= RT_SCOPE_NOWHERE,
147 	},
148 };
149 
rt_fibinfo_free(struct rtable __rcu ** rtp)150 static void rt_fibinfo_free(struct rtable __rcu **rtp)
151 {
152 	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
153 
154 	if (!rt)
155 		return;
156 
157 	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
158 	 * because we waited an RCU grace period before calling
159 	 * free_fib_info_rcu()
160 	 */
161 
162 	dst_dev_put(&rt->dst);
163 	dst_release_immediate(&rt->dst);
164 }
165 
free_nh_exceptions(struct fib_nh_common * nhc)166 static void free_nh_exceptions(struct fib_nh_common *nhc)
167 {
168 	struct fnhe_hash_bucket *hash;
169 	int i;
170 
171 	hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
172 	if (!hash)
173 		return;
174 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
175 		struct fib_nh_exception *fnhe;
176 
177 		fnhe = rcu_dereference_protected(hash[i].chain, 1);
178 		while (fnhe) {
179 			struct fib_nh_exception *next;
180 
181 			next = rcu_dereference_protected(fnhe->fnhe_next, 1);
182 
183 			rt_fibinfo_free(&fnhe->fnhe_rth_input);
184 			rt_fibinfo_free(&fnhe->fnhe_rth_output);
185 
186 			kfree(fnhe);
187 
188 			fnhe = next;
189 		}
190 	}
191 	kfree(hash);
192 }
193 
rt_fibinfo_free_cpus(struct rtable __rcu * __percpu * rtp)194 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
195 {
196 	int cpu;
197 
198 	if (!rtp)
199 		return;
200 
201 	for_each_possible_cpu(cpu) {
202 		struct rtable *rt;
203 
204 		rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
205 		if (rt) {
206 			dst_dev_put(&rt->dst);
207 			dst_release_immediate(&rt->dst);
208 		}
209 	}
210 	free_percpu(rtp);
211 }
212 
fib_nh_common_release(struct fib_nh_common * nhc)213 void fib_nh_common_release(struct fib_nh_common *nhc)
214 {
215 	netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
216 	lwtstate_put(nhc->nhc_lwtstate);
217 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
218 	rt_fibinfo_free(&nhc->nhc_rth_input);
219 	free_nh_exceptions(nhc);
220 }
221 EXPORT_SYMBOL_GPL(fib_nh_common_release);
222 
fib_nh_release(struct net * net,struct fib_nh * fib_nh)223 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
224 {
225 #ifdef CONFIG_IP_ROUTE_CLASSID
226 	if (fib_nh->nh_tclassid)
227 		atomic_dec(&net->ipv4.fib_num_tclassid_users);
228 #endif
229 	fib_nh_common_release(&fib_nh->nh_common);
230 }
231 
232 /* Release a nexthop info record */
free_fib_info_rcu(struct rcu_head * head)233 static void free_fib_info_rcu(struct rcu_head *head)
234 {
235 	struct fib_info *fi = container_of(head, struct fib_info, rcu);
236 
237 	if (fi->nh) {
238 		nexthop_put(fi->nh);
239 	} else {
240 		change_nexthops(fi) {
241 			fib_nh_release(fi->fib_net, nexthop_nh);
242 		} endfor_nexthops(fi);
243 	}
244 
245 	ip_fib_metrics_put(fi->fib_metrics);
246 
247 	kfree(fi);
248 }
249 
free_fib_info(struct fib_info * fi)250 void free_fib_info(struct fib_info *fi)
251 {
252 	if (fi->fib_dead == 0) {
253 		pr_warn("Freeing alive fib_info %p\n", fi);
254 		return;
255 	}
256 
257 	call_rcu(&fi->rcu, free_fib_info_rcu);
258 }
259 EXPORT_SYMBOL_GPL(free_fib_info);
260 
fib_release_info(struct fib_info * fi)261 void fib_release_info(struct fib_info *fi)
262 {
263 	spin_lock_bh(&fib_info_lock);
264 	if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
265 		hlist_del(&fi->fib_hash);
266 
267 		/* Paired with READ_ONCE() in fib_create_info(). */
268 		WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
269 
270 		if (fi->fib_prefsrc)
271 			hlist_del(&fi->fib_lhash);
272 		if (fi->nh) {
273 			list_del(&fi->nh_list);
274 		} else {
275 			change_nexthops(fi) {
276 				if (!nexthop_nh->fib_nh_dev)
277 					continue;
278 				hlist_del(&nexthop_nh->nh_hash);
279 			} endfor_nexthops(fi)
280 		}
281 		/* Paired with READ_ONCE() from fib_table_lookup() */
282 		WRITE_ONCE(fi->fib_dead, 1);
283 		fib_info_put(fi);
284 	}
285 	spin_unlock_bh(&fib_info_lock);
286 }
287 
nh_comp(struct fib_info * fi,struct fib_info * ofi)288 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
289 {
290 	const struct fib_nh *onh;
291 
292 	if (fi->nh || ofi->nh)
293 		return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
294 
295 	if (ofi->fib_nhs == 0)
296 		return 0;
297 
298 	for_nexthops(fi) {
299 		onh = fib_info_nh(ofi, nhsel);
300 
301 		if (nh->fib_nh_oif != onh->fib_nh_oif ||
302 		    nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
303 		    nh->fib_nh_scope != onh->fib_nh_scope ||
304 #ifdef CONFIG_IP_ROUTE_MULTIPATH
305 		    nh->fib_nh_weight != onh->fib_nh_weight ||
306 #endif
307 #ifdef CONFIG_IP_ROUTE_CLASSID
308 		    nh->nh_tclassid != onh->nh_tclassid ||
309 #endif
310 		    lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
311 		    ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
312 			return -1;
313 
314 		if (nh->fib_nh_gw_family == AF_INET &&
315 		    nh->fib_nh_gw4 != onh->fib_nh_gw4)
316 			return -1;
317 
318 		if (nh->fib_nh_gw_family == AF_INET6 &&
319 		    ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
320 			return -1;
321 	} endfor_nexthops(fi);
322 	return 0;
323 }
324 
fib_devindex_hashfn(unsigned int val)325 static inline unsigned int fib_devindex_hashfn(unsigned int val)
326 {
327 	return hash_32(val, DEVINDEX_HASHBITS);
328 }
329 
330 static struct hlist_head *
fib_info_devhash_bucket(const struct net_device * dev)331 fib_info_devhash_bucket(const struct net_device *dev)
332 {
333 	u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
334 
335 	return &fib_info_devhash[fib_devindex_hashfn(val)];
336 }
337 
fib_info_hashfn_1(int init_val,u8 protocol,u8 scope,u32 prefsrc,u32 priority)338 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
339 				      u32 prefsrc, u32 priority)
340 {
341 	unsigned int val = init_val;
342 
343 	val ^= (protocol << 8) | scope;
344 	val ^= prefsrc;
345 	val ^= priority;
346 
347 	return val;
348 }
349 
fib_info_hashfn_result(unsigned int val)350 static unsigned int fib_info_hashfn_result(unsigned int val)
351 {
352 	unsigned int mask = (fib_info_hash_size - 1);
353 
354 	return (val ^ (val >> 7) ^ (val >> 12)) & mask;
355 }
356 
fib_info_hashfn(struct fib_info * fi)357 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
358 {
359 	unsigned int val;
360 
361 	val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
362 				fi->fib_scope, (__force u32)fi->fib_prefsrc,
363 				fi->fib_priority);
364 
365 	if (fi->nh) {
366 		val ^= fib_devindex_hashfn(fi->nh->id);
367 	} else {
368 		for_nexthops(fi) {
369 			val ^= fib_devindex_hashfn(nh->fib_nh_oif);
370 		} endfor_nexthops(fi)
371 	}
372 
373 	return fib_info_hashfn_result(val);
374 }
375 
376 /* no metrics, only nexthop id */
fib_find_info_nh(struct net * net,const struct fib_config * cfg)377 static struct fib_info *fib_find_info_nh(struct net *net,
378 					 const struct fib_config *cfg)
379 {
380 	struct hlist_head *head;
381 	struct fib_info *fi;
382 	unsigned int hash;
383 
384 	hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
385 				 cfg->fc_protocol, cfg->fc_scope,
386 				 (__force u32)cfg->fc_prefsrc,
387 				 cfg->fc_priority);
388 	hash = fib_info_hashfn_result(hash);
389 	head = &fib_info_hash[hash];
390 
391 	hlist_for_each_entry(fi, head, fib_hash) {
392 		if (!net_eq(fi->fib_net, net))
393 			continue;
394 		if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
395 			continue;
396 		if (cfg->fc_protocol == fi->fib_protocol &&
397 		    cfg->fc_scope == fi->fib_scope &&
398 		    cfg->fc_prefsrc == fi->fib_prefsrc &&
399 		    cfg->fc_priority == fi->fib_priority &&
400 		    cfg->fc_type == fi->fib_type &&
401 		    cfg->fc_table == fi->fib_tb_id &&
402 		    !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
403 			return fi;
404 	}
405 
406 	return NULL;
407 }
408 
fib_find_info(struct fib_info * nfi)409 static struct fib_info *fib_find_info(struct fib_info *nfi)
410 {
411 	struct hlist_head *head;
412 	struct fib_info *fi;
413 	unsigned int hash;
414 
415 	hash = fib_info_hashfn(nfi);
416 	head = &fib_info_hash[hash];
417 
418 	hlist_for_each_entry(fi, head, fib_hash) {
419 		if (!net_eq(fi->fib_net, nfi->fib_net))
420 			continue;
421 		if (fi->fib_nhs != nfi->fib_nhs)
422 			continue;
423 		if (nfi->fib_protocol == fi->fib_protocol &&
424 		    nfi->fib_scope == fi->fib_scope &&
425 		    nfi->fib_prefsrc == fi->fib_prefsrc &&
426 		    nfi->fib_priority == fi->fib_priority &&
427 		    nfi->fib_type == fi->fib_type &&
428 		    nfi->fib_tb_id == fi->fib_tb_id &&
429 		    memcmp(nfi->fib_metrics, fi->fib_metrics,
430 			   sizeof(u32) * RTAX_MAX) == 0 &&
431 		    !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
432 		    nh_comp(fi, nfi) == 0)
433 			return fi;
434 	}
435 
436 	return NULL;
437 }
438 
439 /* Check, that the gateway is already configured.
440  * Used only by redirect accept routine.
441  */
ip_fib_check_default(__be32 gw,struct net_device * dev)442 int ip_fib_check_default(__be32 gw, struct net_device *dev)
443 {
444 	struct hlist_head *head;
445 	struct fib_nh *nh;
446 
447 	spin_lock(&fib_info_lock);
448 
449 	head = fib_info_devhash_bucket(dev);
450 
451 	hlist_for_each_entry(nh, head, nh_hash) {
452 		if (nh->fib_nh_dev == dev &&
453 		    nh->fib_nh_gw4 == gw &&
454 		    !(nh->fib_nh_flags & RTNH_F_DEAD)) {
455 			spin_unlock(&fib_info_lock);
456 			return 0;
457 		}
458 	}
459 
460 	spin_unlock(&fib_info_lock);
461 
462 	return -1;
463 }
464 
fib_nlmsg_size(struct fib_info * fi)465 size_t fib_nlmsg_size(struct fib_info *fi)
466 {
467 	size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
468 			 + nla_total_size(4) /* RTA_TABLE */
469 			 + nla_total_size(4) /* RTA_DST */
470 			 + nla_total_size(4) /* RTA_PRIORITY */
471 			 + nla_total_size(4) /* RTA_PREFSRC */
472 			 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
473 	unsigned int nhs = fib_info_num_path(fi);
474 
475 	/* space for nested metrics */
476 	payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
477 
478 	if (fi->nh)
479 		payload += nla_total_size(4); /* RTA_NH_ID */
480 
481 	if (nhs) {
482 		size_t nh_encapsize = 0;
483 		/* Also handles the special case nhs == 1 */
484 
485 		/* each nexthop is packed in an attribute */
486 		size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
487 		unsigned int i;
488 
489 		/* may contain flow and gateway attribute */
490 		nhsize += 2 * nla_total_size(4);
491 
492 		/* grab encap info */
493 		for (i = 0; i < fib_info_num_path(fi); i++) {
494 			struct fib_nh_common *nhc = fib_info_nhc(fi, i);
495 
496 			if (nhc->nhc_lwtstate) {
497 				/* RTA_ENCAP_TYPE */
498 				nh_encapsize += lwtunnel_get_encap_size(
499 						nhc->nhc_lwtstate);
500 				/* RTA_ENCAP */
501 				nh_encapsize +=  nla_total_size(2);
502 			}
503 		}
504 
505 		/* all nexthops are packed in a nested attribute */
506 		payload += nla_total_size((nhs * nhsize) + nh_encapsize);
507 
508 	}
509 
510 	return payload;
511 }
512 
rtmsg_fib(int event,__be32 key,struct fib_alias * fa,int dst_len,u32 tb_id,const struct nl_info * info,unsigned int nlm_flags)513 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
514 	       int dst_len, u32 tb_id, const struct nl_info *info,
515 	       unsigned int nlm_flags)
516 {
517 	struct fib_rt_info fri;
518 	struct sk_buff *skb;
519 	u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
520 	int err = -ENOBUFS;
521 
522 	skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
523 	if (!skb)
524 		goto errout;
525 
526 	fri.fi = fa->fa_info;
527 	fri.tb_id = tb_id;
528 	fri.dst = key;
529 	fri.dst_len = dst_len;
530 	fri.dscp = fa->fa_dscp;
531 	fri.type = fa->fa_type;
532 	fri.offload = READ_ONCE(fa->offload);
533 	fri.trap = READ_ONCE(fa->trap);
534 	fri.offload_failed = READ_ONCE(fa->offload_failed);
535 	err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
536 	if (err < 0) {
537 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
538 		WARN_ON(err == -EMSGSIZE);
539 		kfree_skb(skb);
540 		goto errout;
541 	}
542 	rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
543 		    info->nlh, GFP_KERNEL);
544 	return;
545 errout:
546 	if (err < 0)
547 		rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
548 }
549 
fib_detect_death(struct fib_info * fi,int order,struct fib_info ** last_resort,int * last_idx,int dflt)550 static int fib_detect_death(struct fib_info *fi, int order,
551 			    struct fib_info **last_resort, int *last_idx,
552 			    int dflt)
553 {
554 	const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
555 	struct neighbour *n;
556 	int state = NUD_NONE;
557 
558 	if (likely(nhc->nhc_gw_family == AF_INET))
559 		n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
560 	else if (nhc->nhc_gw_family == AF_INET6)
561 		n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
562 				 nhc->nhc_dev);
563 	else
564 		n = NULL;
565 
566 	if (n) {
567 		state = READ_ONCE(n->nud_state);
568 		neigh_release(n);
569 	} else {
570 		return 0;
571 	}
572 	if (state == NUD_REACHABLE)
573 		return 0;
574 	if ((state & NUD_VALID) && order != dflt)
575 		return 0;
576 	if ((state & NUD_VALID) ||
577 	    (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
578 		*last_resort = fi;
579 		*last_idx = order;
580 	}
581 	return 1;
582 }
583 
fib_nh_common_init(struct net * net,struct fib_nh_common * nhc,struct nlattr * encap,u16 encap_type,void * cfg,gfp_t gfp_flags,struct netlink_ext_ack * extack)584 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
585 		       struct nlattr *encap, u16 encap_type,
586 		       void *cfg, gfp_t gfp_flags,
587 		       struct netlink_ext_ack *extack)
588 {
589 	int err;
590 
591 	nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
592 						    gfp_flags);
593 	if (!nhc->nhc_pcpu_rth_output)
594 		return -ENOMEM;
595 
596 	if (encap) {
597 		struct lwtunnel_state *lwtstate;
598 
599 		if (encap_type == LWTUNNEL_ENCAP_NONE) {
600 			NL_SET_ERR_MSG(extack, "LWT encap type not specified");
601 			err = -EINVAL;
602 			goto lwt_failure;
603 		}
604 		err = lwtunnel_build_state(net, encap_type, encap,
605 					   nhc->nhc_family, cfg, &lwtstate,
606 					   extack);
607 		if (err)
608 			goto lwt_failure;
609 
610 		nhc->nhc_lwtstate = lwtstate_get(lwtstate);
611 	}
612 
613 	return 0;
614 
615 lwt_failure:
616 	rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
617 	nhc->nhc_pcpu_rth_output = NULL;
618 	return err;
619 }
620 EXPORT_SYMBOL_GPL(fib_nh_common_init);
621 
fib_nh_init(struct net * net,struct fib_nh * nh,struct fib_config * cfg,int nh_weight,struct netlink_ext_ack * extack)622 int fib_nh_init(struct net *net, struct fib_nh *nh,
623 		struct fib_config *cfg, int nh_weight,
624 		struct netlink_ext_ack *extack)
625 {
626 	int err;
627 
628 	nh->fib_nh_family = AF_INET;
629 
630 	err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
631 				 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
632 	if (err)
633 		return err;
634 
635 	nh->fib_nh_oif = cfg->fc_oif;
636 	nh->fib_nh_gw_family = cfg->fc_gw_family;
637 	if (cfg->fc_gw_family == AF_INET)
638 		nh->fib_nh_gw4 = cfg->fc_gw4;
639 	else if (cfg->fc_gw_family == AF_INET6)
640 		nh->fib_nh_gw6 = cfg->fc_gw6;
641 
642 	nh->fib_nh_flags = cfg->fc_flags;
643 
644 #ifdef CONFIG_IP_ROUTE_CLASSID
645 	nh->nh_tclassid = cfg->fc_flow;
646 	if (nh->nh_tclassid)
647 		atomic_inc(&net->ipv4.fib_num_tclassid_users);
648 #endif
649 #ifdef CONFIG_IP_ROUTE_MULTIPATH
650 	nh->fib_nh_weight = nh_weight;
651 #endif
652 	return 0;
653 }
654 
655 #ifdef CONFIG_IP_ROUTE_MULTIPATH
656 
fib_count_nexthops(struct rtnexthop * rtnh,int remaining,struct netlink_ext_ack * extack)657 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
658 			      struct netlink_ext_ack *extack)
659 {
660 	int nhs = 0;
661 
662 	while (rtnh_ok(rtnh, remaining)) {
663 		nhs++;
664 		rtnh = rtnh_next(rtnh, &remaining);
665 	}
666 
667 	/* leftover implies invalid nexthop configuration, discard it */
668 	if (remaining > 0) {
669 		NL_SET_ERR_MSG(extack,
670 			       "Invalid nexthop configuration - extra data after nexthops");
671 		nhs = 0;
672 	}
673 
674 	return nhs;
675 }
676 
fib_gw_from_attr(__be32 * gw,struct nlattr * nla,struct netlink_ext_ack * extack)677 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
678 			    struct netlink_ext_ack *extack)
679 {
680 	if (nla_len(nla) < sizeof(*gw)) {
681 		NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
682 		return -EINVAL;
683 	}
684 
685 	*gw = nla_get_in_addr(nla);
686 
687 	return 0;
688 }
689 
690 /* only called when fib_nh is integrated into fib_info */
fib_get_nhs(struct fib_info * fi,struct rtnexthop * rtnh,int remaining,struct fib_config * cfg,struct netlink_ext_ack * extack)691 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
692 		       int remaining, struct fib_config *cfg,
693 		       struct netlink_ext_ack *extack)
694 {
695 	struct net *net = fi->fib_net;
696 	struct fib_config fib_cfg;
697 	struct fib_nh *nh;
698 	int ret;
699 
700 	change_nexthops(fi) {
701 		int attrlen;
702 
703 		memset(&fib_cfg, 0, sizeof(fib_cfg));
704 
705 		if (!rtnh_ok(rtnh, remaining)) {
706 			NL_SET_ERR_MSG(extack,
707 				       "Invalid nexthop configuration - extra data after nexthop");
708 			return -EINVAL;
709 		}
710 
711 		if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
712 			NL_SET_ERR_MSG(extack,
713 				       "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
714 			return -EINVAL;
715 		}
716 
717 		fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
718 		fib_cfg.fc_oif = rtnh->rtnh_ifindex;
719 
720 		attrlen = rtnh_attrlen(rtnh);
721 		if (attrlen > 0) {
722 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
723 
724 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
725 			nlav = nla_find(attrs, attrlen, RTA_VIA);
726 			if (nla && nlav) {
727 				NL_SET_ERR_MSG(extack,
728 					       "Nexthop configuration can not contain both GATEWAY and VIA");
729 				return -EINVAL;
730 			}
731 			if (nla) {
732 				ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
733 						       extack);
734 				if (ret)
735 					goto errout;
736 
737 				if (fib_cfg.fc_gw4)
738 					fib_cfg.fc_gw_family = AF_INET;
739 			} else if (nlav) {
740 				ret = fib_gw_from_via(&fib_cfg, nlav, extack);
741 				if (ret)
742 					goto errout;
743 			}
744 
745 			nla = nla_find(attrs, attrlen, RTA_FLOW);
746 			if (nla) {
747 				if (nla_len(nla) < sizeof(u32)) {
748 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
749 					return -EINVAL;
750 				}
751 				fib_cfg.fc_flow = nla_get_u32(nla);
752 			}
753 
754 			fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
755 			/* RTA_ENCAP_TYPE length checked in
756 			 * lwtunnel_valid_encap_type_attr
757 			 */
758 			nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
759 			if (nla)
760 				fib_cfg.fc_encap_type = nla_get_u16(nla);
761 		}
762 
763 		ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
764 				  rtnh->rtnh_hops + 1, extack);
765 		if (ret)
766 			goto errout;
767 
768 		rtnh = rtnh_next(rtnh, &remaining);
769 	} endfor_nexthops(fi);
770 
771 	ret = -EINVAL;
772 	nh = fib_info_nh(fi, 0);
773 	if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
774 		NL_SET_ERR_MSG(extack,
775 			       "Nexthop device index does not match RTA_OIF");
776 		goto errout;
777 	}
778 	if (cfg->fc_gw_family) {
779 		if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
780 		    (cfg->fc_gw_family == AF_INET &&
781 		     nh->fib_nh_gw4 != cfg->fc_gw4) ||
782 		    (cfg->fc_gw_family == AF_INET6 &&
783 		     ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
784 			NL_SET_ERR_MSG(extack,
785 				       "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
786 			goto errout;
787 		}
788 	}
789 #ifdef CONFIG_IP_ROUTE_CLASSID
790 	if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
791 		NL_SET_ERR_MSG(extack,
792 			       "Nexthop class id does not match RTA_FLOW");
793 		goto errout;
794 	}
795 #endif
796 	ret = 0;
797 errout:
798 	return ret;
799 }
800 
801 /* only called when fib_nh is integrated into fib_info */
fib_rebalance(struct fib_info * fi)802 static void fib_rebalance(struct fib_info *fi)
803 {
804 	int total;
805 	int w;
806 
807 	if (fib_info_num_path(fi) < 2)
808 		return;
809 
810 	total = 0;
811 	for_nexthops(fi) {
812 		if (nh->fib_nh_flags & RTNH_F_DEAD)
813 			continue;
814 
815 		if (ip_ignore_linkdown(nh->fib_nh_dev) &&
816 		    nh->fib_nh_flags & RTNH_F_LINKDOWN)
817 			continue;
818 
819 		total += nh->fib_nh_weight;
820 	} endfor_nexthops(fi);
821 
822 	w = 0;
823 	change_nexthops(fi) {
824 		int upper_bound;
825 
826 		if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
827 			upper_bound = -1;
828 		} else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
829 			   nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
830 			upper_bound = -1;
831 		} else {
832 			w += nexthop_nh->fib_nh_weight;
833 			upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
834 							    total) - 1;
835 		}
836 
837 		atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
838 	} endfor_nexthops(fi);
839 }
840 #else /* CONFIG_IP_ROUTE_MULTIPATH */
841 
fib_get_nhs(struct fib_info * fi,struct rtnexthop * rtnh,int remaining,struct fib_config * cfg,struct netlink_ext_ack * extack)842 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
843 		       int remaining, struct fib_config *cfg,
844 		       struct netlink_ext_ack *extack)
845 {
846 	NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
847 
848 	return -EINVAL;
849 }
850 
851 #define fib_rebalance(fi) do { } while (0)
852 
853 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
854 
fib_encap_match(struct net * net,u16 encap_type,struct nlattr * encap,const struct fib_nh * nh,const struct fib_config * cfg,struct netlink_ext_ack * extack)855 static int fib_encap_match(struct net *net, u16 encap_type,
856 			   struct nlattr *encap,
857 			   const struct fib_nh *nh,
858 			   const struct fib_config *cfg,
859 			   struct netlink_ext_ack *extack)
860 {
861 	struct lwtunnel_state *lwtstate;
862 	int ret, result = 0;
863 
864 	if (encap_type == LWTUNNEL_ENCAP_NONE)
865 		return 0;
866 
867 	ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
868 				   cfg, &lwtstate, extack);
869 	if (!ret) {
870 		result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
871 		lwtstate_free(lwtstate);
872 	}
873 
874 	return result;
875 }
876 
fib_nh_match(struct net * net,struct fib_config * cfg,struct fib_info * fi,struct netlink_ext_ack * extack)877 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
878 		 struct netlink_ext_ack *extack)
879 {
880 #ifdef CONFIG_IP_ROUTE_MULTIPATH
881 	struct rtnexthop *rtnh;
882 	int remaining;
883 #endif
884 
885 	if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
886 		return 1;
887 
888 	if (cfg->fc_nh_id) {
889 		if (fi->nh && cfg->fc_nh_id == fi->nh->id)
890 			return 0;
891 		return 1;
892 	}
893 
894 	if (fi->nh) {
895 		if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
896 			return 1;
897 		return 0;
898 	}
899 
900 	if (cfg->fc_oif || cfg->fc_gw_family) {
901 		struct fib_nh *nh;
902 
903 		nh = fib_info_nh(fi, 0);
904 		if (cfg->fc_encap) {
905 			if (fib_encap_match(net, cfg->fc_encap_type,
906 					    cfg->fc_encap, nh, cfg, extack))
907 				return 1;
908 		}
909 #ifdef CONFIG_IP_ROUTE_CLASSID
910 		if (cfg->fc_flow &&
911 		    cfg->fc_flow != nh->nh_tclassid)
912 			return 1;
913 #endif
914 		if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
915 		    (cfg->fc_gw_family &&
916 		     cfg->fc_gw_family != nh->fib_nh_gw_family))
917 			return 1;
918 
919 		if (cfg->fc_gw_family == AF_INET &&
920 		    cfg->fc_gw4 != nh->fib_nh_gw4)
921 			return 1;
922 
923 		if (cfg->fc_gw_family == AF_INET6 &&
924 		    ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
925 			return 1;
926 
927 		return 0;
928 	}
929 
930 #ifdef CONFIG_IP_ROUTE_MULTIPATH
931 	if (!cfg->fc_mp)
932 		return 0;
933 
934 	rtnh = cfg->fc_mp;
935 	remaining = cfg->fc_mp_len;
936 
937 	for_nexthops(fi) {
938 		int attrlen;
939 
940 		if (!rtnh_ok(rtnh, remaining))
941 			return -EINVAL;
942 
943 		if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
944 			return 1;
945 
946 		attrlen = rtnh_attrlen(rtnh);
947 		if (attrlen > 0) {
948 			struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
949 			int err;
950 
951 			nla = nla_find(attrs, attrlen, RTA_GATEWAY);
952 			nlav = nla_find(attrs, attrlen, RTA_VIA);
953 			if (nla && nlav) {
954 				NL_SET_ERR_MSG(extack,
955 					       "Nexthop configuration can not contain both GATEWAY and VIA");
956 				return -EINVAL;
957 			}
958 
959 			if (nla) {
960 				__be32 gw;
961 
962 				err = fib_gw_from_attr(&gw, nla, extack);
963 				if (err)
964 					return err;
965 
966 				if (nh->fib_nh_gw_family != AF_INET ||
967 				    gw != nh->fib_nh_gw4)
968 					return 1;
969 			} else if (nlav) {
970 				struct fib_config cfg2;
971 
972 				err = fib_gw_from_via(&cfg2, nlav, extack);
973 				if (err)
974 					return err;
975 
976 				switch (nh->fib_nh_gw_family) {
977 				case AF_INET:
978 					if (cfg2.fc_gw_family != AF_INET ||
979 					    cfg2.fc_gw4 != nh->fib_nh_gw4)
980 						return 1;
981 					break;
982 				case AF_INET6:
983 					if (cfg2.fc_gw_family != AF_INET6 ||
984 					    ipv6_addr_cmp(&cfg2.fc_gw6,
985 							  &nh->fib_nh_gw6))
986 						return 1;
987 					break;
988 				}
989 			}
990 
991 #ifdef CONFIG_IP_ROUTE_CLASSID
992 			nla = nla_find(attrs, attrlen, RTA_FLOW);
993 			if (nla) {
994 				if (nla_len(nla) < sizeof(u32)) {
995 					NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
996 					return -EINVAL;
997 				}
998 				if (nla_get_u32(nla) != nh->nh_tclassid)
999 					return 1;
1000 			}
1001 #endif
1002 		}
1003 
1004 		rtnh = rtnh_next(rtnh, &remaining);
1005 	} endfor_nexthops(fi);
1006 #endif
1007 	return 0;
1008 }
1009 
fib_metrics_match(struct fib_config * cfg,struct fib_info * fi)1010 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1011 {
1012 	struct nlattr *nla;
1013 	int remaining;
1014 
1015 	if (!cfg->fc_mx)
1016 		return true;
1017 
1018 	nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1019 		int type = nla_type(nla);
1020 		u32 fi_val, val;
1021 
1022 		if (!type)
1023 			continue;
1024 		if (type > RTAX_MAX)
1025 			return false;
1026 
1027 		type = array_index_nospec(type, RTAX_MAX + 1);
1028 		if (type == RTAX_CC_ALGO) {
1029 			char tmp[TCP_CA_NAME_MAX];
1030 			bool ecn_ca = false;
1031 
1032 			nla_strscpy(tmp, nla, sizeof(tmp));
1033 			val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
1034 		} else {
1035 			if (nla_len(nla) != sizeof(u32))
1036 				return false;
1037 			val = nla_get_u32(nla);
1038 		}
1039 
1040 		fi_val = fi->fib_metrics->metrics[type - 1];
1041 		if (type == RTAX_FEATURES)
1042 			fi_val &= ~DST_FEATURE_ECN_CA;
1043 
1044 		if (fi_val != val)
1045 			return false;
1046 	}
1047 
1048 	return true;
1049 }
1050 
fib_check_nh_v6_gw(struct net * net,struct fib_nh * nh,u32 table,struct netlink_ext_ack * extack)1051 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1052 			      u32 table, struct netlink_ext_ack *extack)
1053 {
1054 	struct fib6_config cfg = {
1055 		.fc_table = table,
1056 		.fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1057 		.fc_ifindex = nh->fib_nh_oif,
1058 		.fc_gateway = nh->fib_nh_gw6,
1059 	};
1060 	struct fib6_nh fib6_nh = {};
1061 	int err;
1062 
1063 	err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1064 	if (!err) {
1065 		nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1066 		netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1067 			    GFP_KERNEL);
1068 		nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1069 		nh->fib_nh_scope = RT_SCOPE_LINK;
1070 
1071 		ipv6_stub->fib6_nh_release(&fib6_nh);
1072 	}
1073 
1074 	return err;
1075 }
1076 
1077 /*
1078  * Picture
1079  * -------
1080  *
1081  * Semantics of nexthop is very messy by historical reasons.
1082  * We have to take into account, that:
1083  * a) gateway can be actually local interface address,
1084  *    so that gatewayed route is direct.
1085  * b) gateway must be on-link address, possibly
1086  *    described not by an ifaddr, but also by a direct route.
1087  * c) If both gateway and interface are specified, they should not
1088  *    contradict.
1089  * d) If we use tunnel routes, gateway could be not on-link.
1090  *
1091  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1092  * results in pretty ugly and hairy code with obscure logic.
1093  *
1094  * I chose to generalized it instead, so that the size
1095  * of code does not increase practically, but it becomes
1096  * much more general.
1097  * Every prefix is assigned a "scope" value: "host" is local address,
1098  * "link" is direct route,
1099  * [ ... "site" ... "interior" ... ]
1100  * and "universe" is true gateway route with global meaning.
1101  *
1102  * Every prefix refers to a set of "nexthop"s (gw, oif),
1103  * where gw must have narrower scope. This recursion stops
1104  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1105  * which means that gw is forced to be on link.
1106  *
1107  * Code is still hairy, but now it is apparently logically
1108  * consistent and very flexible. F.e. as by-product it allows
1109  * to co-exists in peace independent exterior and interior
1110  * routing processes.
1111  *
1112  * Normally it looks as following.
1113  *
1114  * {universe prefix}  -> (gw, oif) [scope link]
1115  *		  |
1116  *		  |-> {link prefix} -> (gw, oif) [scope local]
1117  *					|
1118  *					|-> {local prefix} (terminal node)
1119  */
fib_check_nh_v4_gw(struct net * net,struct fib_nh * nh,u32 table,u8 scope,struct netlink_ext_ack * extack)1120 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1121 			      u8 scope, struct netlink_ext_ack *extack)
1122 {
1123 	struct net_device *dev;
1124 	struct fib_result res;
1125 	int err = 0;
1126 
1127 	if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1128 		unsigned int addr_type;
1129 
1130 		if (scope >= RT_SCOPE_LINK) {
1131 			NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1132 			return -EINVAL;
1133 		}
1134 		dev = __dev_get_by_index(net, nh->fib_nh_oif);
1135 		if (!dev) {
1136 			NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1137 			return -ENODEV;
1138 		}
1139 		if (!(dev->flags & IFF_UP)) {
1140 			NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1141 			return -ENETDOWN;
1142 		}
1143 		addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1144 		if (addr_type != RTN_UNICAST) {
1145 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1146 			return -EINVAL;
1147 		}
1148 		if (!netif_carrier_ok(dev))
1149 			nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1150 		nh->fib_nh_dev = dev;
1151 		netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1152 		nh->fib_nh_scope = RT_SCOPE_LINK;
1153 		return 0;
1154 	}
1155 	rcu_read_lock();
1156 	{
1157 		struct fib_table *tbl = NULL;
1158 		struct flowi4 fl4 = {
1159 			.daddr = nh->fib_nh_gw4,
1160 			.flowi4_scope = scope + 1,
1161 			.flowi4_oif = nh->fib_nh_oif,
1162 			.flowi4_iif = LOOPBACK_IFINDEX,
1163 		};
1164 
1165 		/* It is not necessary, but requires a bit of thinking */
1166 		if (fl4.flowi4_scope < RT_SCOPE_LINK)
1167 			fl4.flowi4_scope = RT_SCOPE_LINK;
1168 
1169 		if (table && table != RT_TABLE_MAIN)
1170 			tbl = fib_get_table(net, table);
1171 
1172 		if (tbl)
1173 			err = fib_table_lookup(tbl, &fl4, &res,
1174 					       FIB_LOOKUP_IGNORE_LINKSTATE |
1175 					       FIB_LOOKUP_NOREF);
1176 
1177 		/* on error or if no table given do full lookup. This
1178 		 * is needed for example when nexthops are in the local
1179 		 * table rather than the given table
1180 		 */
1181 		if (!tbl || err) {
1182 			err = fib_lookup(net, &fl4, &res,
1183 					 FIB_LOOKUP_IGNORE_LINKSTATE);
1184 		}
1185 
1186 		if (err) {
1187 			NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1188 			goto out;
1189 		}
1190 	}
1191 
1192 	err = -EINVAL;
1193 	if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1194 		NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1195 		goto out;
1196 	}
1197 	nh->fib_nh_scope = res.scope;
1198 	nh->fib_nh_oif = FIB_RES_OIF(res);
1199 	nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1200 	if (!dev) {
1201 		NL_SET_ERR_MSG(extack,
1202 			       "No egress device for nexthop gateway");
1203 		goto out;
1204 	}
1205 	netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1206 	if (!netif_carrier_ok(dev))
1207 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1208 	err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1209 out:
1210 	rcu_read_unlock();
1211 	return err;
1212 }
1213 
fib_check_nh_nongw(struct net * net,struct fib_nh * nh,struct netlink_ext_ack * extack)1214 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1215 			      struct netlink_ext_ack *extack)
1216 {
1217 	struct in_device *in_dev;
1218 	int err;
1219 
1220 	if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1221 		NL_SET_ERR_MSG(extack,
1222 			       "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1223 		return -EINVAL;
1224 	}
1225 
1226 	rcu_read_lock();
1227 
1228 	err = -ENODEV;
1229 	in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1230 	if (!in_dev)
1231 		goto out;
1232 	err = -ENETDOWN;
1233 	if (!(in_dev->dev->flags & IFF_UP)) {
1234 		NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1235 		goto out;
1236 	}
1237 
1238 	nh->fib_nh_dev = in_dev->dev;
1239 	netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1240 	nh->fib_nh_scope = RT_SCOPE_HOST;
1241 	if (!netif_carrier_ok(nh->fib_nh_dev))
1242 		nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1243 	err = 0;
1244 out:
1245 	rcu_read_unlock();
1246 	return err;
1247 }
1248 
fib_check_nh(struct net * net,struct fib_nh * nh,u32 table,u8 scope,struct netlink_ext_ack * extack)1249 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1250 		 struct netlink_ext_ack *extack)
1251 {
1252 	int err;
1253 
1254 	if (nh->fib_nh_gw_family == AF_INET)
1255 		err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1256 	else if (nh->fib_nh_gw_family == AF_INET6)
1257 		err = fib_check_nh_v6_gw(net, nh, table, extack);
1258 	else
1259 		err = fib_check_nh_nongw(net, nh, extack);
1260 
1261 	return err;
1262 }
1263 
1264 static struct hlist_head *
fib_info_laddrhash_bucket(const struct net * net,__be32 val)1265 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1266 {
1267 	u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1268 			   fib_info_hash_bits);
1269 
1270 	return &fib_info_laddrhash[slot];
1271 }
1272 
fib_info_hash_move(struct hlist_head * new_info_hash,struct hlist_head * new_laddrhash,unsigned int new_size)1273 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1274 			       struct hlist_head *new_laddrhash,
1275 			       unsigned int new_size)
1276 {
1277 	struct hlist_head *old_info_hash, *old_laddrhash;
1278 	unsigned int old_size = fib_info_hash_size;
1279 	unsigned int i;
1280 
1281 	spin_lock_bh(&fib_info_lock);
1282 	old_info_hash = fib_info_hash;
1283 	old_laddrhash = fib_info_laddrhash;
1284 	fib_info_hash_size = new_size;
1285 	fib_info_hash_bits = ilog2(new_size);
1286 
1287 	for (i = 0; i < old_size; i++) {
1288 		struct hlist_head *head = &fib_info_hash[i];
1289 		struct hlist_node *n;
1290 		struct fib_info *fi;
1291 
1292 		hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1293 			struct hlist_head *dest;
1294 			unsigned int new_hash;
1295 
1296 			new_hash = fib_info_hashfn(fi);
1297 			dest = &new_info_hash[new_hash];
1298 			hlist_add_head(&fi->fib_hash, dest);
1299 		}
1300 	}
1301 	fib_info_hash = new_info_hash;
1302 
1303 	fib_info_laddrhash = new_laddrhash;
1304 	for (i = 0; i < old_size; i++) {
1305 		struct hlist_head *lhead = &old_laddrhash[i];
1306 		struct hlist_node *n;
1307 		struct fib_info *fi;
1308 
1309 		hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1310 			struct hlist_head *ldest;
1311 
1312 			ldest = fib_info_laddrhash_bucket(fi->fib_net,
1313 							  fi->fib_prefsrc);
1314 			hlist_add_head(&fi->fib_lhash, ldest);
1315 		}
1316 	}
1317 
1318 	spin_unlock_bh(&fib_info_lock);
1319 
1320 	kvfree(old_info_hash);
1321 	kvfree(old_laddrhash);
1322 }
1323 
fib_info_update_nhc_saddr(struct net * net,struct fib_nh_common * nhc,unsigned char scope)1324 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1325 				 unsigned char scope)
1326 {
1327 	struct fib_nh *nh;
1328 	__be32 saddr;
1329 
1330 	if (nhc->nhc_family != AF_INET)
1331 		return inet_select_addr(nhc->nhc_dev, 0, scope);
1332 
1333 	nh = container_of(nhc, struct fib_nh, nh_common);
1334 	saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1335 
1336 	WRITE_ONCE(nh->nh_saddr, saddr);
1337 	WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid));
1338 
1339 	return saddr;
1340 }
1341 
fib_result_prefsrc(struct net * net,struct fib_result * res)1342 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1343 {
1344 	struct fib_nh_common *nhc = res->nhc;
1345 
1346 	if (res->fi->fib_prefsrc)
1347 		return res->fi->fib_prefsrc;
1348 
1349 	if (nhc->nhc_family == AF_INET) {
1350 		struct fib_nh *nh;
1351 
1352 		nh = container_of(nhc, struct fib_nh, nh_common);
1353 		if (READ_ONCE(nh->nh_saddr_genid) ==
1354 		    atomic_read(&net->ipv4.dev_addr_genid))
1355 			return READ_ONCE(nh->nh_saddr);
1356 	}
1357 
1358 	return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1359 }
1360 
fib_valid_prefsrc(struct fib_config * cfg,__be32 fib_prefsrc)1361 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1362 {
1363 	if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1364 	    fib_prefsrc != cfg->fc_dst) {
1365 		u32 tb_id = cfg->fc_table;
1366 		int rc;
1367 
1368 		if (tb_id == RT_TABLE_MAIN)
1369 			tb_id = RT_TABLE_LOCAL;
1370 
1371 		rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1372 					  fib_prefsrc, tb_id);
1373 
1374 		if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1375 			rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1376 						  fib_prefsrc, RT_TABLE_LOCAL);
1377 		}
1378 
1379 		if (rc != RTN_LOCAL)
1380 			return false;
1381 	}
1382 	return true;
1383 }
1384 
fib_create_info(struct fib_config * cfg,struct netlink_ext_ack * extack)1385 struct fib_info *fib_create_info(struct fib_config *cfg,
1386 				 struct netlink_ext_ack *extack)
1387 {
1388 	int err;
1389 	struct fib_info *fi = NULL;
1390 	struct nexthop *nh = NULL;
1391 	struct fib_info *ofi;
1392 	int nhs = 1;
1393 	struct net *net = cfg->fc_nlinfo.nl_net;
1394 
1395 	if (cfg->fc_type > RTN_MAX)
1396 		goto err_inval;
1397 
1398 	/* Fast check to catch the most weird cases */
1399 	if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1400 		NL_SET_ERR_MSG(extack, "Invalid scope");
1401 		goto err_inval;
1402 	}
1403 
1404 	if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1405 		NL_SET_ERR_MSG(extack,
1406 			       "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1407 		goto err_inval;
1408 	}
1409 
1410 	if (cfg->fc_nh_id) {
1411 		if (!cfg->fc_mx) {
1412 			fi = fib_find_info_nh(net, cfg);
1413 			if (fi) {
1414 				refcount_inc(&fi->fib_treeref);
1415 				return fi;
1416 			}
1417 		}
1418 
1419 		nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1420 		if (!nh) {
1421 			NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1422 			goto err_inval;
1423 		}
1424 		nhs = 0;
1425 	}
1426 
1427 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1428 	if (cfg->fc_mp) {
1429 		nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1430 		if (nhs == 0)
1431 			goto err_inval;
1432 	}
1433 #endif
1434 
1435 	err = -ENOBUFS;
1436 
1437 	/* Paired with WRITE_ONCE() in fib_release_info() */
1438 	if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1439 		unsigned int new_size = fib_info_hash_size << 1;
1440 		struct hlist_head *new_info_hash;
1441 		struct hlist_head *new_laddrhash;
1442 		size_t bytes;
1443 
1444 		if (!new_size)
1445 			new_size = 16;
1446 		bytes = (size_t)new_size * sizeof(struct hlist_head *);
1447 		new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1448 		new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1449 		if (!new_info_hash || !new_laddrhash) {
1450 			kvfree(new_info_hash);
1451 			kvfree(new_laddrhash);
1452 		} else {
1453 			fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1454 		}
1455 		if (!fib_info_hash_size)
1456 			goto failure;
1457 	}
1458 
1459 	fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1460 	if (!fi)
1461 		goto failure;
1462 	fi->fib_metrics = ip_fib_metrics_init(cfg->fc_mx, cfg->fc_mx_len, extack);
1463 	if (IS_ERR(fi->fib_metrics)) {
1464 		err = PTR_ERR(fi->fib_metrics);
1465 		kfree(fi);
1466 		return ERR_PTR(err);
1467 	}
1468 
1469 	fi->fib_net = net;
1470 	fi->fib_protocol = cfg->fc_protocol;
1471 	fi->fib_scope = cfg->fc_scope;
1472 	fi->fib_flags = cfg->fc_flags;
1473 	fi->fib_priority = cfg->fc_priority;
1474 	fi->fib_prefsrc = cfg->fc_prefsrc;
1475 	fi->fib_type = cfg->fc_type;
1476 	fi->fib_tb_id = cfg->fc_table;
1477 
1478 	fi->fib_nhs = nhs;
1479 	if (nh) {
1480 		if (!nexthop_get(nh)) {
1481 			NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1482 			err = -EINVAL;
1483 		} else {
1484 			err = 0;
1485 			fi->nh = nh;
1486 		}
1487 	} else {
1488 		change_nexthops(fi) {
1489 			nexthop_nh->nh_parent = fi;
1490 		} endfor_nexthops(fi)
1491 
1492 		if (cfg->fc_mp)
1493 			err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1494 					  extack);
1495 		else
1496 			err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1497 	}
1498 
1499 	if (err != 0)
1500 		goto failure;
1501 
1502 	if (fib_props[cfg->fc_type].error) {
1503 		if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1504 			NL_SET_ERR_MSG(extack,
1505 				       "Gateway, device and multipath can not be specified for this route type");
1506 			goto err_inval;
1507 		}
1508 		goto link_it;
1509 	} else {
1510 		switch (cfg->fc_type) {
1511 		case RTN_UNICAST:
1512 		case RTN_LOCAL:
1513 		case RTN_BROADCAST:
1514 		case RTN_ANYCAST:
1515 		case RTN_MULTICAST:
1516 			break;
1517 		default:
1518 			NL_SET_ERR_MSG(extack, "Invalid route type");
1519 			goto err_inval;
1520 		}
1521 	}
1522 
1523 	if (cfg->fc_scope > RT_SCOPE_HOST) {
1524 		NL_SET_ERR_MSG(extack, "Invalid scope");
1525 		goto err_inval;
1526 	}
1527 
1528 	if (fi->nh) {
1529 		err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1530 		if (err)
1531 			goto failure;
1532 	} else if (cfg->fc_scope == RT_SCOPE_HOST) {
1533 		struct fib_nh *nh = fi->fib_nh;
1534 
1535 		/* Local address is added. */
1536 		if (nhs != 1) {
1537 			NL_SET_ERR_MSG(extack,
1538 				       "Route with host scope can not have multiple nexthops");
1539 			goto err_inval;
1540 		}
1541 		if (nh->fib_nh_gw_family) {
1542 			NL_SET_ERR_MSG(extack,
1543 				       "Route with host scope can not have a gateway");
1544 			goto err_inval;
1545 		}
1546 		nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1547 		nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1548 		err = -ENODEV;
1549 		if (!nh->fib_nh_dev)
1550 			goto failure;
1551 		netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1552 				     GFP_KERNEL);
1553 	} else {
1554 		int linkdown = 0;
1555 
1556 		change_nexthops(fi) {
1557 			err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1558 					   cfg->fc_table, cfg->fc_scope,
1559 					   extack);
1560 			if (err != 0)
1561 				goto failure;
1562 			if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1563 				linkdown++;
1564 		} endfor_nexthops(fi)
1565 		if (linkdown == fi->fib_nhs)
1566 			fi->fib_flags |= RTNH_F_LINKDOWN;
1567 	}
1568 
1569 	if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1570 		NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1571 		goto err_inval;
1572 	}
1573 
1574 	if (!fi->nh) {
1575 		change_nexthops(fi) {
1576 			fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1577 						  fi->fib_scope);
1578 			if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1579 				fi->fib_nh_is_v6 = true;
1580 		} endfor_nexthops(fi)
1581 
1582 		fib_rebalance(fi);
1583 	}
1584 
1585 link_it:
1586 	ofi = fib_find_info(fi);
1587 	if (ofi) {
1588 		/* fib_table_lookup() should not see @fi yet. */
1589 		fi->fib_dead = 1;
1590 		free_fib_info(fi);
1591 		refcount_inc(&ofi->fib_treeref);
1592 		return ofi;
1593 	}
1594 
1595 	refcount_set(&fi->fib_treeref, 1);
1596 	refcount_set(&fi->fib_clntref, 1);
1597 	spin_lock_bh(&fib_info_lock);
1598 	fib_info_cnt++;
1599 	hlist_add_head(&fi->fib_hash,
1600 		       &fib_info_hash[fib_info_hashfn(fi)]);
1601 	if (fi->fib_prefsrc) {
1602 		struct hlist_head *head;
1603 
1604 		head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1605 		hlist_add_head(&fi->fib_lhash, head);
1606 	}
1607 	if (fi->nh) {
1608 		list_add(&fi->nh_list, &nh->fi_list);
1609 	} else {
1610 		change_nexthops(fi) {
1611 			struct hlist_head *head;
1612 
1613 			if (!nexthop_nh->fib_nh_dev)
1614 				continue;
1615 			head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1616 			hlist_add_head(&nexthop_nh->nh_hash, head);
1617 		} endfor_nexthops(fi)
1618 	}
1619 	spin_unlock_bh(&fib_info_lock);
1620 	return fi;
1621 
1622 err_inval:
1623 	err = -EINVAL;
1624 
1625 failure:
1626 	if (fi) {
1627 		/* fib_table_lookup() should not see @fi yet. */
1628 		fi->fib_dead = 1;
1629 		free_fib_info(fi);
1630 	}
1631 
1632 	return ERR_PTR(err);
1633 }
1634 
fib_nexthop_info(struct sk_buff * skb,const struct fib_nh_common * nhc,u8 rt_family,unsigned char * flags,bool skip_oif)1635 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1636 		     u8 rt_family, unsigned char *flags, bool skip_oif)
1637 {
1638 	if (nhc->nhc_flags & RTNH_F_DEAD)
1639 		*flags |= RTNH_F_DEAD;
1640 
1641 	if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1642 		*flags |= RTNH_F_LINKDOWN;
1643 
1644 		rcu_read_lock();
1645 		switch (nhc->nhc_family) {
1646 		case AF_INET:
1647 			if (ip_ignore_linkdown(nhc->nhc_dev))
1648 				*flags |= RTNH_F_DEAD;
1649 			break;
1650 		case AF_INET6:
1651 			if (ip6_ignore_linkdown(nhc->nhc_dev))
1652 				*flags |= RTNH_F_DEAD;
1653 			break;
1654 		}
1655 		rcu_read_unlock();
1656 	}
1657 
1658 	switch (nhc->nhc_gw_family) {
1659 	case AF_INET:
1660 		if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1661 			goto nla_put_failure;
1662 		break;
1663 	case AF_INET6:
1664 		/* if gateway family does not match nexthop family
1665 		 * gateway is encoded as RTA_VIA
1666 		 */
1667 		if (rt_family != nhc->nhc_gw_family) {
1668 			int alen = sizeof(struct in6_addr);
1669 			struct nlattr *nla;
1670 			struct rtvia *via;
1671 
1672 			nla = nla_reserve(skb, RTA_VIA, alen + 2);
1673 			if (!nla)
1674 				goto nla_put_failure;
1675 
1676 			via = nla_data(nla);
1677 			via->rtvia_family = AF_INET6;
1678 			memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1679 		} else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1680 					    &nhc->nhc_gw.ipv6) < 0) {
1681 			goto nla_put_failure;
1682 		}
1683 		break;
1684 	}
1685 
1686 	*flags |= (nhc->nhc_flags &
1687 		   (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1688 
1689 	if (!skip_oif && nhc->nhc_dev &&
1690 	    nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1691 		goto nla_put_failure;
1692 
1693 	if (nhc->nhc_lwtstate &&
1694 	    lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1695 				RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1696 		goto nla_put_failure;
1697 
1698 	return 0;
1699 
1700 nla_put_failure:
1701 	return -EMSGSIZE;
1702 }
1703 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1704 
1705 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
fib_add_nexthop(struct sk_buff * skb,const struct fib_nh_common * nhc,int nh_weight,u8 rt_family,u32 nh_tclassid)1706 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1707 		    int nh_weight, u8 rt_family, u32 nh_tclassid)
1708 {
1709 	const struct net_device *dev = nhc->nhc_dev;
1710 	struct rtnexthop *rtnh;
1711 	unsigned char flags = 0;
1712 
1713 	rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1714 	if (!rtnh)
1715 		goto nla_put_failure;
1716 
1717 	rtnh->rtnh_hops = nh_weight - 1;
1718 	rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1719 
1720 	if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1721 		goto nla_put_failure;
1722 
1723 	rtnh->rtnh_flags = flags;
1724 
1725 	if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1726 		goto nla_put_failure;
1727 
1728 	/* length of rtnetlink header + attributes */
1729 	rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1730 
1731 	return 0;
1732 
1733 nla_put_failure:
1734 	return -EMSGSIZE;
1735 }
1736 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1737 #endif
1738 
1739 #ifdef CONFIG_IP_ROUTE_MULTIPATH
fib_add_multipath(struct sk_buff * skb,struct fib_info * fi)1740 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1741 {
1742 	struct nlattr *mp;
1743 
1744 	mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1745 	if (!mp)
1746 		goto nla_put_failure;
1747 
1748 	if (unlikely(fi->nh)) {
1749 		if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1750 			goto nla_put_failure;
1751 		goto mp_end;
1752 	}
1753 
1754 	for_nexthops(fi) {
1755 		u32 nh_tclassid = 0;
1756 #ifdef CONFIG_IP_ROUTE_CLASSID
1757 		nh_tclassid = nh->nh_tclassid;
1758 #endif
1759 		if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1760 				    AF_INET, nh_tclassid) < 0)
1761 			goto nla_put_failure;
1762 	} endfor_nexthops(fi);
1763 
1764 mp_end:
1765 	nla_nest_end(skb, mp);
1766 
1767 	return 0;
1768 
1769 nla_put_failure:
1770 	return -EMSGSIZE;
1771 }
1772 #else
fib_add_multipath(struct sk_buff * skb,struct fib_info * fi)1773 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1774 {
1775 	return 0;
1776 }
1777 #endif
1778 
fib_dump_info(struct sk_buff * skb,u32 portid,u32 seq,int event,const struct fib_rt_info * fri,unsigned int flags)1779 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1780 		  const struct fib_rt_info *fri, unsigned int flags)
1781 {
1782 	unsigned int nhs = fib_info_num_path(fri->fi);
1783 	struct fib_info *fi = fri->fi;
1784 	u32 tb_id = fri->tb_id;
1785 	struct nlmsghdr *nlh;
1786 	struct rtmsg *rtm;
1787 
1788 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1789 	if (!nlh)
1790 		return -EMSGSIZE;
1791 
1792 	rtm = nlmsg_data(nlh);
1793 	rtm->rtm_family = AF_INET;
1794 	rtm->rtm_dst_len = fri->dst_len;
1795 	rtm->rtm_src_len = 0;
1796 	rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1797 	if (tb_id < 256)
1798 		rtm->rtm_table = tb_id;
1799 	else
1800 		rtm->rtm_table = RT_TABLE_COMPAT;
1801 	if (nla_put_u32(skb, RTA_TABLE, tb_id))
1802 		goto nla_put_failure;
1803 	rtm->rtm_type = fri->type;
1804 	rtm->rtm_flags = fi->fib_flags;
1805 	rtm->rtm_scope = fi->fib_scope;
1806 	rtm->rtm_protocol = fi->fib_protocol;
1807 
1808 	if (rtm->rtm_dst_len &&
1809 	    nla_put_in_addr(skb, RTA_DST, fri->dst))
1810 		goto nla_put_failure;
1811 	if (fi->fib_priority &&
1812 	    nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1813 		goto nla_put_failure;
1814 	if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1815 		goto nla_put_failure;
1816 
1817 	if (fi->fib_prefsrc &&
1818 	    nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1819 		goto nla_put_failure;
1820 
1821 	if (fi->nh) {
1822 		if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1823 			goto nla_put_failure;
1824 		if (nexthop_is_blackhole(fi->nh))
1825 			rtm->rtm_type = RTN_BLACKHOLE;
1826 		if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1827 			goto offload;
1828 	}
1829 
1830 	if (nhs == 1) {
1831 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1832 		unsigned char flags = 0;
1833 
1834 		if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1835 			goto nla_put_failure;
1836 
1837 		rtm->rtm_flags = flags;
1838 #ifdef CONFIG_IP_ROUTE_CLASSID
1839 		if (nhc->nhc_family == AF_INET) {
1840 			struct fib_nh *nh;
1841 
1842 			nh = container_of(nhc, struct fib_nh, nh_common);
1843 			if (nh->nh_tclassid &&
1844 			    nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1845 				goto nla_put_failure;
1846 		}
1847 #endif
1848 	} else {
1849 		if (fib_add_multipath(skb, fi) < 0)
1850 			goto nla_put_failure;
1851 	}
1852 
1853 offload:
1854 	if (fri->offload)
1855 		rtm->rtm_flags |= RTM_F_OFFLOAD;
1856 	if (fri->trap)
1857 		rtm->rtm_flags |= RTM_F_TRAP;
1858 	if (fri->offload_failed)
1859 		rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1860 
1861 	nlmsg_end(skb, nlh);
1862 	return 0;
1863 
1864 nla_put_failure:
1865 	nlmsg_cancel(skb, nlh);
1866 	return -EMSGSIZE;
1867 }
1868 
1869 /*
1870  * Update FIB if:
1871  * - local address disappeared -> we must delete all the entries
1872  *   referring to it.
1873  * - device went down -> we must shutdown all nexthops going via it.
1874  */
fib_sync_down_addr(struct net_device * dev,__be32 local)1875 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1876 {
1877 	int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1878 	struct net *net = dev_net(dev);
1879 	struct hlist_head *head;
1880 	struct fib_info *fi;
1881 	int ret = 0;
1882 
1883 	if (!fib_info_laddrhash || local == 0)
1884 		return 0;
1885 
1886 	head = fib_info_laddrhash_bucket(net, local);
1887 	hlist_for_each_entry(fi, head, fib_lhash) {
1888 		if (!net_eq(fi->fib_net, net) ||
1889 		    fi->fib_tb_id != tb_id)
1890 			continue;
1891 		if (fi->fib_prefsrc == local) {
1892 			fi->fib_flags |= RTNH_F_DEAD;
1893 			fi->pfsrc_removed = true;
1894 			ret++;
1895 		}
1896 	}
1897 	return ret;
1898 }
1899 
call_fib_nh_notifiers(struct fib_nh * nh,enum fib_event_type event_type)1900 static int call_fib_nh_notifiers(struct fib_nh *nh,
1901 				 enum fib_event_type event_type)
1902 {
1903 	bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1904 	struct fib_nh_notifier_info info = {
1905 		.fib_nh = nh,
1906 	};
1907 
1908 	switch (event_type) {
1909 	case FIB_EVENT_NH_ADD:
1910 		if (nh->fib_nh_flags & RTNH_F_DEAD)
1911 			break;
1912 		if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1913 			break;
1914 		return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1915 					   &info.info);
1916 	case FIB_EVENT_NH_DEL:
1917 		if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1918 		    (nh->fib_nh_flags & RTNH_F_DEAD))
1919 			return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1920 						   event_type, &info.info);
1921 		break;
1922 	default:
1923 		break;
1924 	}
1925 
1926 	return NOTIFY_DONE;
1927 }
1928 
1929 /* Update the PMTU of exceptions when:
1930  * - the new MTU of the first hop becomes smaller than the PMTU
1931  * - the old MTU was the same as the PMTU, and it limited discovery of
1932  *   larger MTUs on the path. With that limit raised, we can now
1933  *   discover larger MTUs
1934  * A special case is locked exceptions, for which the PMTU is smaller
1935  * than the minimal accepted PMTU:
1936  * - if the new MTU is greater than the PMTU, don't make any change
1937  * - otherwise, unlock and set PMTU
1938  */
fib_nhc_update_mtu(struct fib_nh_common * nhc,u32 new,u32 orig)1939 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1940 {
1941 	struct fnhe_hash_bucket *bucket;
1942 	int i;
1943 
1944 	bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1945 	if (!bucket)
1946 		return;
1947 
1948 	for (i = 0; i < FNHE_HASH_SIZE; i++) {
1949 		struct fib_nh_exception *fnhe;
1950 
1951 		for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1952 		     fnhe;
1953 		     fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1954 			if (fnhe->fnhe_mtu_locked) {
1955 				if (new <= fnhe->fnhe_pmtu) {
1956 					fnhe->fnhe_pmtu = new;
1957 					fnhe->fnhe_mtu_locked = false;
1958 				}
1959 			} else if (new < fnhe->fnhe_pmtu ||
1960 				   orig == fnhe->fnhe_pmtu) {
1961 				fnhe->fnhe_pmtu = new;
1962 			}
1963 		}
1964 	}
1965 }
1966 
fib_sync_mtu(struct net_device * dev,u32 orig_mtu)1967 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1968 {
1969 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1970 	struct fib_nh *nh;
1971 
1972 	hlist_for_each_entry(nh, head, nh_hash) {
1973 		if (nh->fib_nh_dev == dev)
1974 			fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1975 	}
1976 }
1977 
1978 /* Event              force Flags           Description
1979  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1980  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1981  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1982  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1983  *
1984  * only used when fib_nh is built into fib_info
1985  */
fib_sync_down_dev(struct net_device * dev,unsigned long event,bool force)1986 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1987 {
1988 	struct hlist_head *head = fib_info_devhash_bucket(dev);
1989 	struct fib_info *prev_fi = NULL;
1990 	int scope = RT_SCOPE_NOWHERE;
1991 	struct fib_nh *nh;
1992 	int ret = 0;
1993 
1994 	if (force)
1995 		scope = -1;
1996 
1997 	hlist_for_each_entry(nh, head, nh_hash) {
1998 		struct fib_info *fi = nh->nh_parent;
1999 		int dead;
2000 
2001 		BUG_ON(!fi->fib_nhs);
2002 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2003 			continue;
2004 		prev_fi = fi;
2005 		dead = 0;
2006 		change_nexthops(fi) {
2007 			if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2008 				dead++;
2009 			else if (nexthop_nh->fib_nh_dev == dev &&
2010 				 nexthop_nh->fib_nh_scope != scope) {
2011 				switch (event) {
2012 				case NETDEV_DOWN:
2013 				case NETDEV_UNREGISTER:
2014 					nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2015 					fallthrough;
2016 				case NETDEV_CHANGE:
2017 					nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2018 					break;
2019 				}
2020 				call_fib_nh_notifiers(nexthop_nh,
2021 						      FIB_EVENT_NH_DEL);
2022 				dead++;
2023 			}
2024 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2025 			if (event == NETDEV_UNREGISTER &&
2026 			    nexthop_nh->fib_nh_dev == dev) {
2027 				dead = fi->fib_nhs;
2028 				break;
2029 			}
2030 #endif
2031 		} endfor_nexthops(fi)
2032 		if (dead == fi->fib_nhs) {
2033 			switch (event) {
2034 			case NETDEV_DOWN:
2035 			case NETDEV_UNREGISTER:
2036 				fi->fib_flags |= RTNH_F_DEAD;
2037 				fallthrough;
2038 			case NETDEV_CHANGE:
2039 				fi->fib_flags |= RTNH_F_LINKDOWN;
2040 				break;
2041 			}
2042 			ret++;
2043 		}
2044 
2045 		fib_rebalance(fi);
2046 	}
2047 
2048 	return ret;
2049 }
2050 
2051 /* Must be invoked inside of an RCU protected region.  */
fib_select_default(const struct flowi4 * flp,struct fib_result * res)2052 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2053 {
2054 	struct fib_info *fi = NULL, *last_resort = NULL;
2055 	struct hlist_head *fa_head = res->fa_head;
2056 	struct fib_table *tb = res->table;
2057 	u8 slen = 32 - res->prefixlen;
2058 	int order = -1, last_idx = -1;
2059 	struct fib_alias *fa, *fa1 = NULL;
2060 	u32 last_prio = res->fi->fib_priority;
2061 	dscp_t last_dscp = 0;
2062 
2063 	hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2064 		struct fib_info *next_fi = fa->fa_info;
2065 		struct fib_nh_common *nhc;
2066 
2067 		if (fa->fa_slen != slen)
2068 			continue;
2069 		if (fa->fa_dscp &&
2070 		    fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2071 			continue;
2072 		if (fa->tb_id != tb->tb_id)
2073 			continue;
2074 		if (next_fi->fib_priority > last_prio &&
2075 		    fa->fa_dscp == last_dscp) {
2076 			if (last_dscp)
2077 				continue;
2078 			break;
2079 		}
2080 		if (next_fi->fib_flags & RTNH_F_DEAD)
2081 			continue;
2082 		last_dscp = fa->fa_dscp;
2083 		last_prio = next_fi->fib_priority;
2084 
2085 		if (next_fi->fib_scope != res->scope ||
2086 		    fa->fa_type != RTN_UNICAST)
2087 			continue;
2088 
2089 		nhc = fib_info_nhc(next_fi, 0);
2090 		if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2091 			continue;
2092 
2093 		fib_alias_accessed(fa);
2094 
2095 		if (!fi) {
2096 			if (next_fi != res->fi)
2097 				break;
2098 			fa1 = fa;
2099 		} else if (!fib_detect_death(fi, order, &last_resort,
2100 					     &last_idx, fa1->fa_default)) {
2101 			fib_result_assign(res, fi);
2102 			fa1->fa_default = order;
2103 			goto out;
2104 		}
2105 		fi = next_fi;
2106 		order++;
2107 	}
2108 
2109 	if (order <= 0 || !fi) {
2110 		if (fa1)
2111 			fa1->fa_default = -1;
2112 		goto out;
2113 	}
2114 
2115 	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2116 			      fa1->fa_default)) {
2117 		fib_result_assign(res, fi);
2118 		fa1->fa_default = order;
2119 		goto out;
2120 	}
2121 
2122 	if (last_idx >= 0)
2123 		fib_result_assign(res, last_resort);
2124 	fa1->fa_default = last_idx;
2125 out:
2126 	return;
2127 }
2128 
2129 /*
2130  * Dead device goes up. We wake up dead nexthops.
2131  * It takes sense only on multipath routes.
2132  *
2133  * only used when fib_nh is built into fib_info
2134  */
fib_sync_up(struct net_device * dev,unsigned char nh_flags)2135 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2136 {
2137 	struct fib_info *prev_fi;
2138 	struct hlist_head *head;
2139 	struct fib_nh *nh;
2140 	int ret;
2141 
2142 	if (!(dev->flags & IFF_UP))
2143 		return 0;
2144 
2145 	if (nh_flags & RTNH_F_DEAD) {
2146 		unsigned int flags = dev_get_flags(dev);
2147 
2148 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2149 			nh_flags |= RTNH_F_LINKDOWN;
2150 	}
2151 
2152 	prev_fi = NULL;
2153 	head = fib_info_devhash_bucket(dev);
2154 	ret = 0;
2155 
2156 	hlist_for_each_entry(nh, head, nh_hash) {
2157 		struct fib_info *fi = nh->nh_parent;
2158 		int alive;
2159 
2160 		BUG_ON(!fi->fib_nhs);
2161 		if (nh->fib_nh_dev != dev || fi == prev_fi)
2162 			continue;
2163 
2164 		prev_fi = fi;
2165 		alive = 0;
2166 		change_nexthops(fi) {
2167 			if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2168 				alive++;
2169 				continue;
2170 			}
2171 			if (!nexthop_nh->fib_nh_dev ||
2172 			    !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2173 				continue;
2174 			if (nexthop_nh->fib_nh_dev != dev ||
2175 			    !__in_dev_get_rtnl(dev))
2176 				continue;
2177 			alive++;
2178 			nexthop_nh->fib_nh_flags &= ~nh_flags;
2179 			call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2180 		} endfor_nexthops(fi)
2181 
2182 		if (alive > 0) {
2183 			fi->fib_flags &= ~nh_flags;
2184 			ret++;
2185 		}
2186 
2187 		fib_rebalance(fi);
2188 	}
2189 
2190 	return ret;
2191 }
2192 
2193 #ifdef CONFIG_IP_ROUTE_MULTIPATH
fib_good_nh(const struct fib_nh * nh)2194 static bool fib_good_nh(const struct fib_nh *nh)
2195 {
2196 	int state = NUD_REACHABLE;
2197 
2198 	if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2199 		struct neighbour *n;
2200 
2201 		rcu_read_lock();
2202 
2203 		if (likely(nh->fib_nh_gw_family == AF_INET))
2204 			n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2205 						   (__force u32)nh->fib_nh_gw4);
2206 		else if (nh->fib_nh_gw_family == AF_INET6)
2207 			n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2208 							   &nh->fib_nh_gw6);
2209 		else
2210 			n = NULL;
2211 		if (n)
2212 			state = READ_ONCE(n->nud_state);
2213 
2214 		rcu_read_unlock();
2215 	}
2216 
2217 	return !!(state & NUD_VALID);
2218 }
2219 
fib_select_multipath(struct fib_result * res,int hash)2220 void fib_select_multipath(struct fib_result *res, int hash)
2221 {
2222 	struct fib_info *fi = res->fi;
2223 	struct net *net = fi->fib_net;
2224 	bool first = false;
2225 
2226 	if (unlikely(res->fi->nh)) {
2227 		nexthop_path_fib_result(res, hash);
2228 		return;
2229 	}
2230 
2231 	change_nexthops(fi) {
2232 		if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2233 			if (!fib_good_nh(nexthop_nh))
2234 				continue;
2235 			if (!first) {
2236 				res->nh_sel = nhsel;
2237 				res->nhc = &nexthop_nh->nh_common;
2238 				first = true;
2239 			}
2240 		}
2241 
2242 		if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2243 			continue;
2244 
2245 		res->nh_sel = nhsel;
2246 		res->nhc = &nexthop_nh->nh_common;
2247 		return;
2248 	} endfor_nexthops(fi);
2249 }
2250 #endif
2251 
fib_select_path(struct net * net,struct fib_result * res,struct flowi4 * fl4,const struct sk_buff * skb)2252 void fib_select_path(struct net *net, struct fib_result *res,
2253 		     struct flowi4 *fl4, const struct sk_buff *skb)
2254 {
2255 	if (fl4->flowi4_oif)
2256 		goto check_saddr;
2257 
2258 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2259 	if (fib_info_num_path(res->fi) > 1) {
2260 		int h = fib_multipath_hash(net, fl4, skb, NULL);
2261 
2262 		fib_select_multipath(res, h);
2263 	}
2264 	else
2265 #endif
2266 	if (!res->prefixlen &&
2267 	    res->table->tb_num_default > 1 &&
2268 	    res->type == RTN_UNICAST)
2269 		fib_select_default(fl4, res);
2270 
2271 check_saddr:
2272 	if (!fl4->saddr) {
2273 		struct net_device *l3mdev;
2274 
2275 		l3mdev = dev_get_by_index_rcu(net, fl4->flowi4_l3mdev);
2276 
2277 		if (!l3mdev ||
2278 		    l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) == l3mdev)
2279 			fl4->saddr = fib_result_prefsrc(net, res);
2280 		else
2281 			fl4->saddr = inet_select_addr(l3mdev, 0, RT_SCOPE_LINK);
2282 	}
2283 }
2284