xref: /linux/net/ipv4/fib_frontend.c (revision c79c3c34f75d72a066e292b10aa50fc758c97c89)
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: FIB frontend.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/bitops.h>
15 #include <linux/capability.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/errno.h>
23 #include <linux/in.h>
24 #include <linux/inet.h>
25 #include <linux/inetdevice.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_addr.h>
28 #include <linux/if_arp.h>
29 #include <linux/skbuff.h>
30 #include <linux/cache.h>
31 #include <linux/init.h>
32 #include <linux/list.h>
33 #include <linux/slab.h>
34 
35 #include <net/ip.h>
36 #include <net/protocol.h>
37 #include <net/route.h>
38 #include <net/tcp.h>
39 #include <net/sock.h>
40 #include <net/arp.h>
41 #include <net/ip_fib.h>
42 #include <net/nexthop.h>
43 #include <net/rtnetlink.h>
44 #include <net/xfrm.h>
45 #include <net/l3mdev.h>
46 #include <net/lwtunnel.h>
47 #include <trace/events/fib.h>
48 
49 #ifndef CONFIG_IP_MULTIPLE_TABLES
50 
51 static int __net_init fib4_rules_init(struct net *net)
52 {
53 	struct fib_table *local_table, *main_table;
54 
55 	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
56 	if (!main_table)
57 		return -ENOMEM;
58 
59 	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
60 	if (!local_table)
61 		goto fail;
62 
63 	hlist_add_head_rcu(&local_table->tb_hlist,
64 				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
65 	hlist_add_head_rcu(&main_table->tb_hlist,
66 				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
67 	return 0;
68 
69 fail:
70 	fib_free_table(main_table);
71 	return -ENOMEM;
72 }
73 #else
74 
75 struct fib_table *fib_new_table(struct net *net, u32 id)
76 {
77 	struct fib_table *tb, *alias = NULL;
78 	unsigned int h;
79 
80 	if (id == 0)
81 		id = RT_TABLE_MAIN;
82 	tb = fib_get_table(net, id);
83 	if (tb)
84 		return tb;
85 
86 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
87 		alias = fib_new_table(net, RT_TABLE_MAIN);
88 
89 	tb = fib_trie_table(id, alias);
90 	if (!tb)
91 		return NULL;
92 
93 	switch (id) {
94 	case RT_TABLE_MAIN:
95 		rcu_assign_pointer(net->ipv4.fib_main, tb);
96 		break;
97 	case RT_TABLE_DEFAULT:
98 		rcu_assign_pointer(net->ipv4.fib_default, tb);
99 		break;
100 	default:
101 		break;
102 	}
103 
104 	h = id & (FIB_TABLE_HASHSZ - 1);
105 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
106 	return tb;
107 }
108 EXPORT_SYMBOL_GPL(fib_new_table);
109 
110 /* caller must hold either rtnl or rcu read lock */
111 struct fib_table *fib_get_table(struct net *net, u32 id)
112 {
113 	struct fib_table *tb;
114 	struct hlist_head *head;
115 	unsigned int h;
116 
117 	if (id == 0)
118 		id = RT_TABLE_MAIN;
119 	h = id & (FIB_TABLE_HASHSZ - 1);
120 
121 	head = &net->ipv4.fib_table_hash[h];
122 	hlist_for_each_entry_rcu(tb, head, tb_hlist,
123 				 lockdep_rtnl_is_held()) {
124 		if (tb->tb_id == id)
125 			return tb;
126 	}
127 	return NULL;
128 }
129 #endif /* CONFIG_IP_MULTIPLE_TABLES */
130 
131 static void fib_replace_table(struct net *net, struct fib_table *old,
132 			      struct fib_table *new)
133 {
134 #ifdef CONFIG_IP_MULTIPLE_TABLES
135 	switch (new->tb_id) {
136 	case RT_TABLE_MAIN:
137 		rcu_assign_pointer(net->ipv4.fib_main, new);
138 		break;
139 	case RT_TABLE_DEFAULT:
140 		rcu_assign_pointer(net->ipv4.fib_default, new);
141 		break;
142 	default:
143 		break;
144 	}
145 
146 #endif
147 	/* replace the old table in the hlist */
148 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
149 }
150 
151 int fib_unmerge(struct net *net)
152 {
153 	struct fib_table *old, *new, *main_table;
154 
155 	/* attempt to fetch local table if it has been allocated */
156 	old = fib_get_table(net, RT_TABLE_LOCAL);
157 	if (!old)
158 		return 0;
159 
160 	new = fib_trie_unmerge(old);
161 	if (!new)
162 		return -ENOMEM;
163 
164 	/* table is already unmerged */
165 	if (new == old)
166 		return 0;
167 
168 	/* replace merged table with clean table */
169 	fib_replace_table(net, old, new);
170 	fib_free_table(old);
171 
172 	/* attempt to fetch main table if it has been allocated */
173 	main_table = fib_get_table(net, RT_TABLE_MAIN);
174 	if (!main_table)
175 		return 0;
176 
177 	/* flush local entries from main table */
178 	fib_table_flush_external(main_table);
179 
180 	return 0;
181 }
182 
183 void fib_flush(struct net *net)
184 {
185 	int flushed = 0;
186 	unsigned int h;
187 
188 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
189 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
190 		struct hlist_node *tmp;
191 		struct fib_table *tb;
192 
193 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
194 			flushed += fib_table_flush(net, tb, false);
195 	}
196 
197 	if (flushed)
198 		rt_cache_flush(net);
199 }
200 
201 /*
202  * Find address type as if only "dev" was present in the system. If
203  * on_dev is NULL then all interfaces are taken into consideration.
204  */
205 static inline unsigned int __inet_dev_addr_type(struct net *net,
206 						const struct net_device *dev,
207 						__be32 addr, u32 tb_id)
208 {
209 	struct flowi4		fl4 = { .daddr = addr };
210 	struct fib_result	res;
211 	unsigned int ret = RTN_BROADCAST;
212 	struct fib_table *table;
213 
214 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
215 		return RTN_BROADCAST;
216 	if (ipv4_is_multicast(addr))
217 		return RTN_MULTICAST;
218 
219 	rcu_read_lock();
220 
221 	table = fib_get_table(net, tb_id);
222 	if (table) {
223 		ret = RTN_UNICAST;
224 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
225 			struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
226 
227 			if (!dev || dev == nhc->nhc_dev)
228 				ret = res.type;
229 		}
230 	}
231 
232 	rcu_read_unlock();
233 	return ret;
234 }
235 
236 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
237 {
238 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
239 }
240 EXPORT_SYMBOL(inet_addr_type_table);
241 
242 unsigned int inet_addr_type(struct net *net, __be32 addr)
243 {
244 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
245 }
246 EXPORT_SYMBOL(inet_addr_type);
247 
248 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
249 				__be32 addr)
250 {
251 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
252 
253 	return __inet_dev_addr_type(net, dev, addr, rt_table);
254 }
255 EXPORT_SYMBOL(inet_dev_addr_type);
256 
257 /* inet_addr_type with dev == NULL but using the table from a dev
258  * if one is associated
259  */
260 unsigned int inet_addr_type_dev_table(struct net *net,
261 				      const struct net_device *dev,
262 				      __be32 addr)
263 {
264 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
265 
266 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
267 }
268 EXPORT_SYMBOL(inet_addr_type_dev_table);
269 
270 __be32 fib_compute_spec_dst(struct sk_buff *skb)
271 {
272 	struct net_device *dev = skb->dev;
273 	struct in_device *in_dev;
274 	struct fib_result res;
275 	struct rtable *rt;
276 	struct net *net;
277 	int scope;
278 
279 	rt = skb_rtable(skb);
280 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
281 	    RTCF_LOCAL)
282 		return ip_hdr(skb)->daddr;
283 
284 	in_dev = __in_dev_get_rcu(dev);
285 
286 	net = dev_net(dev);
287 
288 	scope = RT_SCOPE_UNIVERSE;
289 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
290 		bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
291 		struct flowi4 fl4 = {
292 			.flowi4_iif = LOOPBACK_IFINDEX,
293 			.flowi4_oif = l3mdev_master_ifindex_rcu(dev),
294 			.daddr = ip_hdr(skb)->saddr,
295 			.flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
296 			.flowi4_scope = scope,
297 			.flowi4_mark = vmark ? skb->mark : 0,
298 		};
299 		if (!fib_lookup(net, &fl4, &res, 0))
300 			return fib_result_prefsrc(net, &res);
301 	} else {
302 		scope = RT_SCOPE_LINK;
303 	}
304 
305 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
306 }
307 
308 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
309 {
310 	bool dev_match = false;
311 #ifdef CONFIG_IP_ROUTE_MULTIPATH
312 	if (unlikely(fi->nh)) {
313 		dev_match = nexthop_uses_dev(fi->nh, dev);
314 	} else {
315 		int ret;
316 
317 		for (ret = 0; ret < fib_info_num_path(fi); ret++) {
318 			const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
319 
320 			if (nhc_l3mdev_matches_dev(nhc, dev)) {
321 				dev_match = true;
322 				break;
323 			}
324 		}
325 	}
326 #else
327 	if (fib_info_nhc(fi, 0)->nhc_dev == dev)
328 		dev_match = true;
329 #endif
330 
331 	return dev_match;
332 }
333 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
334 
335 /* Given (packet source, input interface) and optional (dst, oif, tos):
336  * - (main) check, that source is valid i.e. not broadcast or our local
337  *   address.
338  * - figure out what "logical" interface this packet arrived
339  *   and calculate "specific destination" address.
340  * - check, that packet arrived from expected physical interface.
341  * called with rcu_read_lock()
342  */
343 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
344 				 u8 tos, int oif, struct net_device *dev,
345 				 int rpf, struct in_device *idev, u32 *itag)
346 {
347 	struct net *net = dev_net(dev);
348 	struct flow_keys flkeys;
349 	int ret, no_addr;
350 	struct fib_result res;
351 	struct flowi4 fl4;
352 	bool dev_match;
353 
354 	fl4.flowi4_oif = 0;
355 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
356 	if (!fl4.flowi4_iif)
357 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
358 	fl4.daddr = src;
359 	fl4.saddr = dst;
360 	fl4.flowi4_tos = tos;
361 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
362 	fl4.flowi4_tun_key.tun_id = 0;
363 	fl4.flowi4_flags = 0;
364 	fl4.flowi4_uid = sock_net_uid(net, NULL);
365 	fl4.flowi4_multipath_hash = 0;
366 
367 	no_addr = idev->ifa_list == NULL;
368 
369 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
370 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
371 		fl4.flowi4_proto = 0;
372 		fl4.fl4_sport = 0;
373 		fl4.fl4_dport = 0;
374 	}
375 
376 	if (fib_lookup(net, &fl4, &res, 0))
377 		goto last_resort;
378 	if (res.type != RTN_UNICAST &&
379 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
380 		goto e_inval;
381 	fib_combine_itag(itag, &res);
382 
383 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
384 	/* This is not common, loopback packets retain skb_dst so normally they
385 	 * would not even hit this slow path.
386 	 */
387 	dev_match = dev_match || (res.type == RTN_LOCAL &&
388 				  dev == net->loopback_dev);
389 	if (dev_match) {
390 		ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
391 		return ret;
392 	}
393 	if (no_addr)
394 		goto last_resort;
395 	if (rpf == 1)
396 		goto e_rpf;
397 	fl4.flowi4_oif = dev->ifindex;
398 
399 	ret = 0;
400 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
401 		if (res.type == RTN_UNICAST)
402 			ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
403 	}
404 	return ret;
405 
406 last_resort:
407 	if (rpf)
408 		goto e_rpf;
409 	*itag = 0;
410 	return 0;
411 
412 e_inval:
413 	return -EINVAL;
414 e_rpf:
415 	return -EXDEV;
416 }
417 
418 /* Ignore rp_filter for packets protected by IPsec. */
419 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
420 			u8 tos, int oif, struct net_device *dev,
421 			struct in_device *idev, u32 *itag)
422 {
423 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
424 	struct net *net = dev_net(dev);
425 
426 	if (!r && !fib_num_tclassid_users(net) &&
427 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
428 		if (IN_DEV_ACCEPT_LOCAL(idev))
429 			goto ok;
430 		/* with custom local routes in place, checking local addresses
431 		 * only will be too optimistic, with custom rules, checking
432 		 * local addresses only can be too strict, e.g. due to vrf
433 		 */
434 		if (net->ipv4.fib_has_custom_local_routes ||
435 		    fib4_has_custom_rules(net))
436 			goto full_check;
437 		if (inet_lookup_ifaddr_rcu(net, src))
438 			return -EINVAL;
439 
440 ok:
441 		*itag = 0;
442 		return 0;
443 	}
444 
445 full_check:
446 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
447 }
448 
449 static inline __be32 sk_extract_addr(struct sockaddr *addr)
450 {
451 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
452 }
453 
454 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
455 {
456 	struct nlattr *nla;
457 
458 	nla = (struct nlattr *) ((char *) mx + len);
459 	nla->nla_type = type;
460 	nla->nla_len = nla_attr_size(4);
461 	*(u32 *) nla_data(nla) = value;
462 
463 	return len + nla_total_size(4);
464 }
465 
466 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
467 				 struct fib_config *cfg)
468 {
469 	__be32 addr;
470 	int plen;
471 
472 	memset(cfg, 0, sizeof(*cfg));
473 	cfg->fc_nlinfo.nl_net = net;
474 
475 	if (rt->rt_dst.sa_family != AF_INET)
476 		return -EAFNOSUPPORT;
477 
478 	/*
479 	 * Check mask for validity:
480 	 * a) it must be contiguous.
481 	 * b) destination must have all host bits clear.
482 	 * c) if application forgot to set correct family (AF_INET),
483 	 *    reject request unless it is absolutely clear i.e.
484 	 *    both family and mask are zero.
485 	 */
486 	plen = 32;
487 	addr = sk_extract_addr(&rt->rt_dst);
488 	if (!(rt->rt_flags & RTF_HOST)) {
489 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
490 
491 		if (rt->rt_genmask.sa_family != AF_INET) {
492 			if (mask || rt->rt_genmask.sa_family)
493 				return -EAFNOSUPPORT;
494 		}
495 
496 		if (bad_mask(mask, addr))
497 			return -EINVAL;
498 
499 		plen = inet_mask_len(mask);
500 	}
501 
502 	cfg->fc_dst_len = plen;
503 	cfg->fc_dst = addr;
504 
505 	if (cmd != SIOCDELRT) {
506 		cfg->fc_nlflags = NLM_F_CREATE;
507 		cfg->fc_protocol = RTPROT_BOOT;
508 	}
509 
510 	if (rt->rt_metric)
511 		cfg->fc_priority = rt->rt_metric - 1;
512 
513 	if (rt->rt_flags & RTF_REJECT) {
514 		cfg->fc_scope = RT_SCOPE_HOST;
515 		cfg->fc_type = RTN_UNREACHABLE;
516 		return 0;
517 	}
518 
519 	cfg->fc_scope = RT_SCOPE_NOWHERE;
520 	cfg->fc_type = RTN_UNICAST;
521 
522 	if (rt->rt_dev) {
523 		char *colon;
524 		struct net_device *dev;
525 		char devname[IFNAMSIZ];
526 
527 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
528 			return -EFAULT;
529 
530 		devname[IFNAMSIZ-1] = 0;
531 		colon = strchr(devname, ':');
532 		if (colon)
533 			*colon = 0;
534 		dev = __dev_get_by_name(net, devname);
535 		if (!dev)
536 			return -ENODEV;
537 		cfg->fc_oif = dev->ifindex;
538 		cfg->fc_table = l3mdev_fib_table(dev);
539 		if (colon) {
540 			const struct in_ifaddr *ifa;
541 			struct in_device *in_dev;
542 
543 			in_dev = __in_dev_get_rtnl(dev);
544 			if (!in_dev)
545 				return -ENODEV;
546 
547 			*colon = ':';
548 
549 			rcu_read_lock();
550 			in_dev_for_each_ifa_rcu(ifa, in_dev) {
551 				if (strcmp(ifa->ifa_label, devname) == 0)
552 					break;
553 			}
554 			rcu_read_unlock();
555 
556 			if (!ifa)
557 				return -ENODEV;
558 			cfg->fc_prefsrc = ifa->ifa_local;
559 		}
560 	}
561 
562 	addr = sk_extract_addr(&rt->rt_gateway);
563 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
564 		unsigned int addr_type;
565 
566 		cfg->fc_gw4 = addr;
567 		cfg->fc_gw_family = AF_INET;
568 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
569 		if (rt->rt_flags & RTF_GATEWAY &&
570 		    addr_type == RTN_UNICAST)
571 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
572 	}
573 
574 	if (cmd == SIOCDELRT)
575 		return 0;
576 
577 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
578 		return -EINVAL;
579 
580 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
581 		cfg->fc_scope = RT_SCOPE_LINK;
582 
583 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
584 		struct nlattr *mx;
585 		int len = 0;
586 
587 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
588 		if (!mx)
589 			return -ENOMEM;
590 
591 		if (rt->rt_flags & RTF_MTU)
592 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
593 
594 		if (rt->rt_flags & RTF_WINDOW)
595 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
596 
597 		if (rt->rt_flags & RTF_IRTT)
598 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
599 
600 		cfg->fc_mx = mx;
601 		cfg->fc_mx_len = len;
602 	}
603 
604 	return 0;
605 }
606 
607 /*
608  * Handle IP routing ioctl calls.
609  * These are used to manipulate the routing tables
610  */
611 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
612 {
613 	struct fib_config cfg;
614 	int err;
615 
616 	switch (cmd) {
617 	case SIOCADDRT:		/* Add a route */
618 	case SIOCDELRT:		/* Delete a route */
619 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
620 			return -EPERM;
621 
622 		rtnl_lock();
623 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
624 		if (err == 0) {
625 			struct fib_table *tb;
626 
627 			if (cmd == SIOCDELRT) {
628 				tb = fib_get_table(net, cfg.fc_table);
629 				if (tb)
630 					err = fib_table_delete(net, tb, &cfg,
631 							       NULL);
632 				else
633 					err = -ESRCH;
634 			} else {
635 				tb = fib_new_table(net, cfg.fc_table);
636 				if (tb)
637 					err = fib_table_insert(net, tb,
638 							       &cfg, NULL);
639 				else
640 					err = -ENOBUFS;
641 			}
642 
643 			/* allocated by rtentry_to_fib_config() */
644 			kfree(cfg.fc_mx);
645 		}
646 		rtnl_unlock();
647 		return err;
648 	}
649 	return -EINVAL;
650 }
651 
652 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
653 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
654 	[RTA_DST]		= { .type = NLA_U32 },
655 	[RTA_SRC]		= { .type = NLA_U32 },
656 	[RTA_IIF]		= { .type = NLA_U32 },
657 	[RTA_OIF]		= { .type = NLA_U32 },
658 	[RTA_GATEWAY]		= { .type = NLA_U32 },
659 	[RTA_PRIORITY]		= { .type = NLA_U32 },
660 	[RTA_PREFSRC]		= { .type = NLA_U32 },
661 	[RTA_METRICS]		= { .type = NLA_NESTED },
662 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
663 	[RTA_FLOW]		= { .type = NLA_U32 },
664 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
665 	[RTA_ENCAP]		= { .type = NLA_NESTED },
666 	[RTA_UID]		= { .type = NLA_U32 },
667 	[RTA_MARK]		= { .type = NLA_U32 },
668 	[RTA_TABLE]		= { .type = NLA_U32 },
669 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
670 	[RTA_SPORT]		= { .type = NLA_U16 },
671 	[RTA_DPORT]		= { .type = NLA_U16 },
672 	[RTA_NH_ID]		= { .type = NLA_U32 },
673 };
674 
675 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
676 		    struct netlink_ext_ack *extack)
677 {
678 	struct rtvia *via;
679 	int alen;
680 
681 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
682 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
683 		return -EINVAL;
684 	}
685 
686 	via = nla_data(nla);
687 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
688 
689 	switch (via->rtvia_family) {
690 	case AF_INET:
691 		if (alen != sizeof(__be32)) {
692 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
693 			return -EINVAL;
694 		}
695 		cfg->fc_gw_family = AF_INET;
696 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
697 		break;
698 	case AF_INET6:
699 #if IS_ENABLED(CONFIG_IPV6)
700 		if (alen != sizeof(struct in6_addr)) {
701 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
702 			return -EINVAL;
703 		}
704 		cfg->fc_gw_family = AF_INET6;
705 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
706 #else
707 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
708 		return -EINVAL;
709 #endif
710 		break;
711 	default:
712 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
713 		return -EINVAL;
714 	}
715 
716 	return 0;
717 }
718 
719 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
720 			     struct nlmsghdr *nlh, struct fib_config *cfg,
721 			     struct netlink_ext_ack *extack)
722 {
723 	bool has_gw = false, has_via = false;
724 	struct nlattr *attr;
725 	int err, remaining;
726 	struct rtmsg *rtm;
727 
728 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
729 					rtm_ipv4_policy, extack);
730 	if (err < 0)
731 		goto errout;
732 
733 	memset(cfg, 0, sizeof(*cfg));
734 
735 	rtm = nlmsg_data(nlh);
736 	cfg->fc_dst_len = rtm->rtm_dst_len;
737 	cfg->fc_tos = rtm->rtm_tos;
738 	cfg->fc_table = rtm->rtm_table;
739 	cfg->fc_protocol = rtm->rtm_protocol;
740 	cfg->fc_scope = rtm->rtm_scope;
741 	cfg->fc_type = rtm->rtm_type;
742 	cfg->fc_flags = rtm->rtm_flags;
743 	cfg->fc_nlflags = nlh->nlmsg_flags;
744 
745 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
746 	cfg->fc_nlinfo.nlh = nlh;
747 	cfg->fc_nlinfo.nl_net = net;
748 
749 	if (cfg->fc_type > RTN_MAX) {
750 		NL_SET_ERR_MSG(extack, "Invalid route type");
751 		err = -EINVAL;
752 		goto errout;
753 	}
754 
755 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
756 		switch (nla_type(attr)) {
757 		case RTA_DST:
758 			cfg->fc_dst = nla_get_be32(attr);
759 			break;
760 		case RTA_OIF:
761 			cfg->fc_oif = nla_get_u32(attr);
762 			break;
763 		case RTA_GATEWAY:
764 			has_gw = true;
765 			cfg->fc_gw4 = nla_get_be32(attr);
766 			if (cfg->fc_gw4)
767 				cfg->fc_gw_family = AF_INET;
768 			break;
769 		case RTA_VIA:
770 			has_via = true;
771 			err = fib_gw_from_via(cfg, attr, extack);
772 			if (err)
773 				goto errout;
774 			break;
775 		case RTA_PRIORITY:
776 			cfg->fc_priority = nla_get_u32(attr);
777 			break;
778 		case RTA_PREFSRC:
779 			cfg->fc_prefsrc = nla_get_be32(attr);
780 			break;
781 		case RTA_METRICS:
782 			cfg->fc_mx = nla_data(attr);
783 			cfg->fc_mx_len = nla_len(attr);
784 			break;
785 		case RTA_MULTIPATH:
786 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
787 							     nla_len(attr),
788 							     extack);
789 			if (err < 0)
790 				goto errout;
791 			cfg->fc_mp = nla_data(attr);
792 			cfg->fc_mp_len = nla_len(attr);
793 			break;
794 		case RTA_FLOW:
795 			cfg->fc_flow = nla_get_u32(attr);
796 			break;
797 		case RTA_TABLE:
798 			cfg->fc_table = nla_get_u32(attr);
799 			break;
800 		case RTA_ENCAP:
801 			cfg->fc_encap = attr;
802 			break;
803 		case RTA_ENCAP_TYPE:
804 			cfg->fc_encap_type = nla_get_u16(attr);
805 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
806 							extack);
807 			if (err < 0)
808 				goto errout;
809 			break;
810 		case RTA_NH_ID:
811 			cfg->fc_nh_id = nla_get_u32(attr);
812 			break;
813 		}
814 	}
815 
816 	if (cfg->fc_nh_id) {
817 		if (cfg->fc_oif || cfg->fc_gw_family ||
818 		    cfg->fc_encap || cfg->fc_mp) {
819 			NL_SET_ERR_MSG(extack,
820 				       "Nexthop specification and nexthop id are mutually exclusive");
821 			return -EINVAL;
822 		}
823 	}
824 
825 	if (has_gw && has_via) {
826 		NL_SET_ERR_MSG(extack,
827 			       "Nexthop configuration can not contain both GATEWAY and VIA");
828 		return -EINVAL;
829 	}
830 
831 	return 0;
832 errout:
833 	return err;
834 }
835 
836 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
837 			     struct netlink_ext_ack *extack)
838 {
839 	struct net *net = sock_net(skb->sk);
840 	struct fib_config cfg;
841 	struct fib_table *tb;
842 	int err;
843 
844 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
845 	if (err < 0)
846 		goto errout;
847 
848 	if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
849 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
850 		err = -EINVAL;
851 		goto errout;
852 	}
853 
854 	tb = fib_get_table(net, cfg.fc_table);
855 	if (!tb) {
856 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
857 		err = -ESRCH;
858 		goto errout;
859 	}
860 
861 	err = fib_table_delete(net, tb, &cfg, extack);
862 errout:
863 	return err;
864 }
865 
866 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
867 			     struct netlink_ext_ack *extack)
868 {
869 	struct net *net = sock_net(skb->sk);
870 	struct fib_config cfg;
871 	struct fib_table *tb;
872 	int err;
873 
874 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
875 	if (err < 0)
876 		goto errout;
877 
878 	tb = fib_new_table(net, cfg.fc_table);
879 	if (!tb) {
880 		err = -ENOBUFS;
881 		goto errout;
882 	}
883 
884 	err = fib_table_insert(net, tb, &cfg, extack);
885 	if (!err && cfg.fc_type == RTN_LOCAL)
886 		net->ipv4.fib_has_custom_local_routes = true;
887 errout:
888 	return err;
889 }
890 
891 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
892 			  struct fib_dump_filter *filter,
893 			  struct netlink_callback *cb)
894 {
895 	struct netlink_ext_ack *extack = cb->extack;
896 	struct nlattr *tb[RTA_MAX + 1];
897 	struct rtmsg *rtm;
898 	int err, i;
899 
900 	ASSERT_RTNL();
901 
902 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
903 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
904 		return -EINVAL;
905 	}
906 
907 	rtm = nlmsg_data(nlh);
908 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
909 	    rtm->rtm_scope) {
910 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
911 		return -EINVAL;
912 	}
913 
914 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
915 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
916 		return -EINVAL;
917 	}
918 	if (rtm->rtm_flags & RTM_F_CLONED)
919 		filter->dump_routes = false;
920 	else
921 		filter->dump_exceptions = false;
922 
923 	filter->flags    = rtm->rtm_flags;
924 	filter->protocol = rtm->rtm_protocol;
925 	filter->rt_type  = rtm->rtm_type;
926 	filter->table_id = rtm->rtm_table;
927 
928 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
929 					    rtm_ipv4_policy, extack);
930 	if (err < 0)
931 		return err;
932 
933 	for (i = 0; i <= RTA_MAX; ++i) {
934 		int ifindex;
935 
936 		if (!tb[i])
937 			continue;
938 
939 		switch (i) {
940 		case RTA_TABLE:
941 			filter->table_id = nla_get_u32(tb[i]);
942 			break;
943 		case RTA_OIF:
944 			ifindex = nla_get_u32(tb[i]);
945 			filter->dev = __dev_get_by_index(net, ifindex);
946 			if (!filter->dev)
947 				return -ENODEV;
948 			break;
949 		default:
950 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
951 			return -EINVAL;
952 		}
953 	}
954 
955 	if (filter->flags || filter->protocol || filter->rt_type ||
956 	    filter->table_id || filter->dev) {
957 		filter->filter_set = 1;
958 		cb->answer_flags = NLM_F_DUMP_FILTERED;
959 	}
960 
961 	return 0;
962 }
963 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
964 
965 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
966 {
967 	struct fib_dump_filter filter = { .dump_routes = true,
968 					  .dump_exceptions = true };
969 	const struct nlmsghdr *nlh = cb->nlh;
970 	struct net *net = sock_net(skb->sk);
971 	unsigned int h, s_h;
972 	unsigned int e = 0, s_e;
973 	struct fib_table *tb;
974 	struct hlist_head *head;
975 	int dumped = 0, err;
976 
977 	if (cb->strict_check) {
978 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
979 		if (err < 0)
980 			return err;
981 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
982 		struct rtmsg *rtm = nlmsg_data(nlh);
983 
984 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
985 	}
986 
987 	/* ipv4 does not use prefix flag */
988 	if (filter.flags & RTM_F_PREFIX)
989 		return skb->len;
990 
991 	if (filter.table_id) {
992 		tb = fib_get_table(net, filter.table_id);
993 		if (!tb) {
994 			if (rtnl_msg_family(cb->nlh) != PF_INET)
995 				return skb->len;
996 
997 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
998 			return -ENOENT;
999 		}
1000 
1001 		rcu_read_lock();
1002 		err = fib_table_dump(tb, skb, cb, &filter);
1003 		rcu_read_unlock();
1004 		return skb->len ? : err;
1005 	}
1006 
1007 	s_h = cb->args[0];
1008 	s_e = cb->args[1];
1009 
1010 	rcu_read_lock();
1011 
1012 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1013 		e = 0;
1014 		head = &net->ipv4.fib_table_hash[h];
1015 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1016 			if (e < s_e)
1017 				goto next;
1018 			if (dumped)
1019 				memset(&cb->args[2], 0, sizeof(cb->args) -
1020 						 2 * sizeof(cb->args[0]));
1021 			err = fib_table_dump(tb, skb, cb, &filter);
1022 			if (err < 0) {
1023 				if (likely(skb->len))
1024 					goto out;
1025 
1026 				goto out_err;
1027 			}
1028 			dumped = 1;
1029 next:
1030 			e++;
1031 		}
1032 	}
1033 out:
1034 	err = skb->len;
1035 out_err:
1036 	rcu_read_unlock();
1037 
1038 	cb->args[1] = e;
1039 	cb->args[0] = h;
1040 
1041 	return err;
1042 }
1043 
1044 /* Prepare and feed intra-kernel routing request.
1045  * Really, it should be netlink message, but :-( netlink
1046  * can be not configured, so that we feed it directly
1047  * to fib engine. It is legal, because all events occur
1048  * only when netlink is already locked.
1049  */
1050 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1051 		      struct in_ifaddr *ifa, u32 rt_priority)
1052 {
1053 	struct net *net = dev_net(ifa->ifa_dev->dev);
1054 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1055 	struct fib_table *tb;
1056 	struct fib_config cfg = {
1057 		.fc_protocol = RTPROT_KERNEL,
1058 		.fc_type = type,
1059 		.fc_dst = dst,
1060 		.fc_dst_len = dst_len,
1061 		.fc_priority = rt_priority,
1062 		.fc_prefsrc = ifa->ifa_local,
1063 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1064 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1065 		.fc_nlinfo = {
1066 			.nl_net = net,
1067 		},
1068 	};
1069 
1070 	if (!tb_id)
1071 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1072 
1073 	tb = fib_new_table(net, tb_id);
1074 	if (!tb)
1075 		return;
1076 
1077 	cfg.fc_table = tb->tb_id;
1078 
1079 	if (type != RTN_LOCAL)
1080 		cfg.fc_scope = RT_SCOPE_LINK;
1081 	else
1082 		cfg.fc_scope = RT_SCOPE_HOST;
1083 
1084 	if (cmd == RTM_NEWROUTE)
1085 		fib_table_insert(net, tb, &cfg, NULL);
1086 	else
1087 		fib_table_delete(net, tb, &cfg, NULL);
1088 }
1089 
1090 void fib_add_ifaddr(struct in_ifaddr *ifa)
1091 {
1092 	struct in_device *in_dev = ifa->ifa_dev;
1093 	struct net_device *dev = in_dev->dev;
1094 	struct in_ifaddr *prim = ifa;
1095 	__be32 mask = ifa->ifa_mask;
1096 	__be32 addr = ifa->ifa_local;
1097 	__be32 prefix = ifa->ifa_address & mask;
1098 
1099 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1100 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1101 		if (!prim) {
1102 			pr_warn("%s: bug: prim == NULL\n", __func__);
1103 			return;
1104 		}
1105 	}
1106 
1107 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1108 
1109 	if (!(dev->flags & IFF_UP))
1110 		return;
1111 
1112 	/* Add broadcast address, if it is explicitly assigned. */
1113 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1114 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1115 			  prim, 0);
1116 
1117 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1118 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1119 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1120 			fib_magic(RTM_NEWROUTE,
1121 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1122 				  prefix, ifa->ifa_prefixlen, prim,
1123 				  ifa->ifa_rt_priority);
1124 
1125 		/* Add network specific broadcasts, when it takes a sense */
1126 		if (ifa->ifa_prefixlen < 31) {
1127 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1128 				  prim, 0);
1129 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1130 				  32, prim, 0);
1131 		}
1132 	}
1133 }
1134 
1135 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1136 {
1137 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1138 	struct in_device *in_dev = ifa->ifa_dev;
1139 	struct net_device *dev = in_dev->dev;
1140 
1141 	if (!(dev->flags & IFF_UP) ||
1142 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1143 	    ipv4_is_zeronet(prefix) ||
1144 	    (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1145 		return;
1146 
1147 	/* add the new */
1148 	fib_magic(RTM_NEWROUTE,
1149 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1150 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1151 
1152 	/* delete the old */
1153 	fib_magic(RTM_DELROUTE,
1154 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1155 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1156 }
1157 
1158 /* Delete primary or secondary address.
1159  * Optionally, on secondary address promotion consider the addresses
1160  * from subnet iprim as deleted, even if they are in device list.
1161  * In this case the secondary ifa can be in device list.
1162  */
1163 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1164 {
1165 	struct in_device *in_dev = ifa->ifa_dev;
1166 	struct net_device *dev = in_dev->dev;
1167 	struct in_ifaddr *ifa1;
1168 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1169 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1170 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1171 #define LOCAL_OK	1
1172 #define BRD_OK		2
1173 #define BRD0_OK		4
1174 #define BRD1_OK		8
1175 	unsigned int ok = 0;
1176 	int subnet = 0;		/* Primary network */
1177 	int gone = 1;		/* Address is missing */
1178 	int same_prefsrc = 0;	/* Another primary with same IP */
1179 
1180 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1181 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1182 		if (!prim) {
1183 			/* if the device has been deleted, we don't perform
1184 			 * address promotion
1185 			 */
1186 			if (!in_dev->dead)
1187 				pr_warn("%s: bug: prim == NULL\n", __func__);
1188 			return;
1189 		}
1190 		if (iprim && iprim != prim) {
1191 			pr_warn("%s: bug: iprim != prim\n", __func__);
1192 			return;
1193 		}
1194 	} else if (!ipv4_is_zeronet(any) &&
1195 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1196 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1197 			fib_magic(RTM_DELROUTE,
1198 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1199 				  any, ifa->ifa_prefixlen, prim, 0);
1200 		subnet = 1;
1201 	}
1202 
1203 	if (in_dev->dead)
1204 		goto no_promotions;
1205 
1206 	/* Deletion is more complicated than add.
1207 	 * We should take care of not to delete too much :-)
1208 	 *
1209 	 * Scan address list to be sure that addresses are really gone.
1210 	 */
1211 	rcu_read_lock();
1212 	in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1213 		if (ifa1 == ifa) {
1214 			/* promotion, keep the IP */
1215 			gone = 0;
1216 			continue;
1217 		}
1218 		/* Ignore IFAs from our subnet */
1219 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1220 		    inet_ifa_match(ifa1->ifa_address, iprim))
1221 			continue;
1222 
1223 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1224 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1225 			/* Another address from our subnet? */
1226 			if (ifa1->ifa_mask == prim->ifa_mask &&
1227 			    inet_ifa_match(ifa1->ifa_address, prim))
1228 				prim1 = prim;
1229 			else {
1230 				/* We reached the secondaries, so
1231 				 * same_prefsrc should be determined.
1232 				 */
1233 				if (!same_prefsrc)
1234 					continue;
1235 				/* Search new prim1 if ifa1 is not
1236 				 * using the current prim1
1237 				 */
1238 				if (!prim1 ||
1239 				    ifa1->ifa_mask != prim1->ifa_mask ||
1240 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1241 					prim1 = inet_ifa_byprefix(in_dev,
1242 							ifa1->ifa_address,
1243 							ifa1->ifa_mask);
1244 				if (!prim1)
1245 					continue;
1246 				if (prim1->ifa_local != prim->ifa_local)
1247 					continue;
1248 			}
1249 		} else {
1250 			if (prim->ifa_local != ifa1->ifa_local)
1251 				continue;
1252 			prim1 = ifa1;
1253 			if (prim != prim1)
1254 				same_prefsrc = 1;
1255 		}
1256 		if (ifa->ifa_local == ifa1->ifa_local)
1257 			ok |= LOCAL_OK;
1258 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1259 			ok |= BRD_OK;
1260 		if (brd == ifa1->ifa_broadcast)
1261 			ok |= BRD1_OK;
1262 		if (any == ifa1->ifa_broadcast)
1263 			ok |= BRD0_OK;
1264 		/* primary has network specific broadcasts */
1265 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1266 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1267 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1268 
1269 			if (!ipv4_is_zeronet(any1)) {
1270 				if (ifa->ifa_broadcast == brd1 ||
1271 				    ifa->ifa_broadcast == any1)
1272 					ok |= BRD_OK;
1273 				if (brd == brd1 || brd == any1)
1274 					ok |= BRD1_OK;
1275 				if (any == brd1 || any == any1)
1276 					ok |= BRD0_OK;
1277 			}
1278 		}
1279 	}
1280 	rcu_read_unlock();
1281 
1282 no_promotions:
1283 	if (!(ok & BRD_OK))
1284 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1285 			  prim, 0);
1286 	if (subnet && ifa->ifa_prefixlen < 31) {
1287 		if (!(ok & BRD1_OK))
1288 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1289 				  prim, 0);
1290 		if (!(ok & BRD0_OK))
1291 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1292 				  prim, 0);
1293 	}
1294 	if (!(ok & LOCAL_OK)) {
1295 		unsigned int addr_type;
1296 
1297 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1298 
1299 		/* Check, that this local address finally disappeared. */
1300 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1301 						     ifa->ifa_local);
1302 		if (gone && addr_type != RTN_LOCAL) {
1303 			/* And the last, but not the least thing.
1304 			 * We must flush stray FIB entries.
1305 			 *
1306 			 * First of all, we scan fib_info list searching
1307 			 * for stray nexthop entries, then ignite fib_flush.
1308 			 */
1309 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1310 				fib_flush(dev_net(dev));
1311 		}
1312 	}
1313 #undef LOCAL_OK
1314 #undef BRD_OK
1315 #undef BRD0_OK
1316 #undef BRD1_OK
1317 }
1318 
1319 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1320 {
1321 
1322 	struct fib_result       res;
1323 	struct flowi4           fl4 = {
1324 		.flowi4_mark = frn->fl_mark,
1325 		.daddr = frn->fl_addr,
1326 		.flowi4_tos = frn->fl_tos,
1327 		.flowi4_scope = frn->fl_scope,
1328 	};
1329 	struct fib_table *tb;
1330 
1331 	rcu_read_lock();
1332 
1333 	tb = fib_get_table(net, frn->tb_id_in);
1334 
1335 	frn->err = -ENOENT;
1336 	if (tb) {
1337 		local_bh_disable();
1338 
1339 		frn->tb_id = tb->tb_id;
1340 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1341 
1342 		if (!frn->err) {
1343 			frn->prefixlen = res.prefixlen;
1344 			frn->nh_sel = res.nh_sel;
1345 			frn->type = res.type;
1346 			frn->scope = res.scope;
1347 		}
1348 		local_bh_enable();
1349 	}
1350 
1351 	rcu_read_unlock();
1352 }
1353 
1354 static void nl_fib_input(struct sk_buff *skb)
1355 {
1356 	struct net *net;
1357 	struct fib_result_nl *frn;
1358 	struct nlmsghdr *nlh;
1359 	u32 portid;
1360 
1361 	net = sock_net(skb->sk);
1362 	nlh = nlmsg_hdr(skb);
1363 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1364 	    skb->len < nlh->nlmsg_len ||
1365 	    nlmsg_len(nlh) < sizeof(*frn))
1366 		return;
1367 
1368 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1369 	if (!skb)
1370 		return;
1371 	nlh = nlmsg_hdr(skb);
1372 
1373 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1374 	nl_fib_lookup(net, frn);
1375 
1376 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1377 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1378 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1379 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1380 }
1381 
1382 static int __net_init nl_fib_lookup_init(struct net *net)
1383 {
1384 	struct sock *sk;
1385 	struct netlink_kernel_cfg cfg = {
1386 		.input	= nl_fib_input,
1387 	};
1388 
1389 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1390 	if (!sk)
1391 		return -EAFNOSUPPORT;
1392 	net->ipv4.fibnl = sk;
1393 	return 0;
1394 }
1395 
1396 static void nl_fib_lookup_exit(struct net *net)
1397 {
1398 	netlink_kernel_release(net->ipv4.fibnl);
1399 	net->ipv4.fibnl = NULL;
1400 }
1401 
1402 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1403 			   bool force)
1404 {
1405 	if (fib_sync_down_dev(dev, event, force))
1406 		fib_flush(dev_net(dev));
1407 	else
1408 		rt_cache_flush(dev_net(dev));
1409 	arp_ifdown(dev);
1410 }
1411 
1412 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1413 {
1414 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1415 	struct net_device *dev = ifa->ifa_dev->dev;
1416 	struct net *net = dev_net(dev);
1417 
1418 	switch (event) {
1419 	case NETDEV_UP:
1420 		fib_add_ifaddr(ifa);
1421 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1422 		fib_sync_up(dev, RTNH_F_DEAD);
1423 #endif
1424 		atomic_inc(&net->ipv4.dev_addr_genid);
1425 		rt_cache_flush(dev_net(dev));
1426 		break;
1427 	case NETDEV_DOWN:
1428 		fib_del_ifaddr(ifa, NULL);
1429 		atomic_inc(&net->ipv4.dev_addr_genid);
1430 		if (!ifa->ifa_dev->ifa_list) {
1431 			/* Last address was deleted from this interface.
1432 			 * Disable IP.
1433 			 */
1434 			fib_disable_ip(dev, event, true);
1435 		} else {
1436 			rt_cache_flush(dev_net(dev));
1437 		}
1438 		break;
1439 	}
1440 	return NOTIFY_DONE;
1441 }
1442 
1443 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1444 {
1445 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1446 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1447 	struct netdev_notifier_info_ext *info_ext = ptr;
1448 	struct in_device *in_dev;
1449 	struct net *net = dev_net(dev);
1450 	struct in_ifaddr *ifa;
1451 	unsigned int flags;
1452 
1453 	if (event == NETDEV_UNREGISTER) {
1454 		fib_disable_ip(dev, event, true);
1455 		rt_flush_dev(dev);
1456 		return NOTIFY_DONE;
1457 	}
1458 
1459 	in_dev = __in_dev_get_rtnl(dev);
1460 	if (!in_dev)
1461 		return NOTIFY_DONE;
1462 
1463 	switch (event) {
1464 	case NETDEV_UP:
1465 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1466 			fib_add_ifaddr(ifa);
1467 		}
1468 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1469 		fib_sync_up(dev, RTNH_F_DEAD);
1470 #endif
1471 		atomic_inc(&net->ipv4.dev_addr_genid);
1472 		rt_cache_flush(net);
1473 		break;
1474 	case NETDEV_DOWN:
1475 		fib_disable_ip(dev, event, false);
1476 		break;
1477 	case NETDEV_CHANGE:
1478 		flags = dev_get_flags(dev);
1479 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1480 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1481 		else
1482 			fib_sync_down_dev(dev, event, false);
1483 		rt_cache_flush(net);
1484 		break;
1485 	case NETDEV_CHANGEMTU:
1486 		fib_sync_mtu(dev, info_ext->ext.mtu);
1487 		rt_cache_flush(net);
1488 		break;
1489 	case NETDEV_CHANGEUPPER:
1490 		upper_info = ptr;
1491 		/* flush all routes if dev is linked to or unlinked from
1492 		 * an L3 master device (e.g., VRF)
1493 		 */
1494 		if (upper_info->upper_dev &&
1495 		    netif_is_l3_master(upper_info->upper_dev))
1496 			fib_disable_ip(dev, NETDEV_DOWN, true);
1497 		break;
1498 	}
1499 	return NOTIFY_DONE;
1500 }
1501 
1502 static struct notifier_block fib_inetaddr_notifier = {
1503 	.notifier_call = fib_inetaddr_event,
1504 };
1505 
1506 static struct notifier_block fib_netdev_notifier = {
1507 	.notifier_call = fib_netdev_event,
1508 };
1509 
1510 static int __net_init ip_fib_net_init(struct net *net)
1511 {
1512 	int err;
1513 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1514 
1515 	err = fib4_notifier_init(net);
1516 	if (err)
1517 		return err;
1518 
1519 	/* Avoid false sharing : Use at least a full cache line */
1520 	size = max_t(size_t, size, L1_CACHE_BYTES);
1521 
1522 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1523 	if (!net->ipv4.fib_table_hash) {
1524 		err = -ENOMEM;
1525 		goto err_table_hash_alloc;
1526 	}
1527 
1528 	err = fib4_rules_init(net);
1529 	if (err < 0)
1530 		goto err_rules_init;
1531 	return 0;
1532 
1533 err_rules_init:
1534 	kfree(net->ipv4.fib_table_hash);
1535 err_table_hash_alloc:
1536 	fib4_notifier_exit(net);
1537 	return err;
1538 }
1539 
1540 static void ip_fib_net_exit(struct net *net)
1541 {
1542 	int i;
1543 
1544 	rtnl_lock();
1545 #ifdef CONFIG_IP_MULTIPLE_TABLES
1546 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1547 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1548 #endif
1549 	/* Destroy the tables in reverse order to guarantee that the
1550 	 * local table, ID 255, is destroyed before the main table, ID
1551 	 * 254. This is necessary as the local table may contain
1552 	 * references to data contained in the main table.
1553 	 */
1554 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1555 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1556 		struct hlist_node *tmp;
1557 		struct fib_table *tb;
1558 
1559 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1560 			hlist_del(&tb->tb_hlist);
1561 			fib_table_flush(net, tb, true);
1562 			fib_free_table(tb);
1563 		}
1564 	}
1565 
1566 #ifdef CONFIG_IP_MULTIPLE_TABLES
1567 	fib4_rules_exit(net);
1568 #endif
1569 	rtnl_unlock();
1570 	kfree(net->ipv4.fib_table_hash);
1571 	fib4_notifier_exit(net);
1572 }
1573 
1574 static int __net_init fib_net_init(struct net *net)
1575 {
1576 	int error;
1577 
1578 #ifdef CONFIG_IP_ROUTE_CLASSID
1579 	net->ipv4.fib_num_tclassid_users = 0;
1580 #endif
1581 	error = ip_fib_net_init(net);
1582 	if (error < 0)
1583 		goto out;
1584 	error = nl_fib_lookup_init(net);
1585 	if (error < 0)
1586 		goto out_nlfl;
1587 	error = fib_proc_init(net);
1588 	if (error < 0)
1589 		goto out_proc;
1590 out:
1591 	return error;
1592 
1593 out_proc:
1594 	nl_fib_lookup_exit(net);
1595 out_nlfl:
1596 	ip_fib_net_exit(net);
1597 	goto out;
1598 }
1599 
1600 static void __net_exit fib_net_exit(struct net *net)
1601 {
1602 	fib_proc_exit(net);
1603 	nl_fib_lookup_exit(net);
1604 	ip_fib_net_exit(net);
1605 }
1606 
1607 static struct pernet_operations fib_net_ops = {
1608 	.init = fib_net_init,
1609 	.exit = fib_net_exit,
1610 };
1611 
1612 void __init ip_fib_init(void)
1613 {
1614 	fib_trie_init();
1615 
1616 	register_pernet_subsys(&fib_net_ops);
1617 
1618 	register_netdevice_notifier(&fib_netdev_notifier);
1619 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1620 
1621 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1622 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1623 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1624 }
1625