xref: /linux/net/ipv4/fib_frontend.c (revision 2638eb8b50cfc16240e0bb080b9afbf541a9b39d)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		IPv4 Forwarding Information Base: FIB frontend.
7  *
8  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9  *
10  *		This program is free software; you can redistribute it and/or
11  *		modify it under the terms of the GNU General Public License
12  *		as published by the Free Software Foundation; either version
13  *		2 of the License, or (at your option) any later version.
14  */
15 
16 #include <linux/module.h>
17 #include <linux/uaccess.h>
18 #include <linux/bitops.h>
19 #include <linux/capability.h>
20 #include <linux/types.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
27 #include <linux/in.h>
28 #include <linux/inet.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_addr.h>
32 #include <linux/if_arp.h>
33 #include <linux/skbuff.h>
34 #include <linux/cache.h>
35 #include <linux/init.h>
36 #include <linux/list.h>
37 #include <linux/slab.h>
38 
39 #include <net/ip.h>
40 #include <net/protocol.h>
41 #include <net/route.h>
42 #include <net/tcp.h>
43 #include <net/sock.h>
44 #include <net/arp.h>
45 #include <net/ip_fib.h>
46 #include <net/rtnetlink.h>
47 #include <net/xfrm.h>
48 #include <net/l3mdev.h>
49 #include <net/lwtunnel.h>
50 #include <trace/events/fib.h>
51 
52 #ifndef CONFIG_IP_MULTIPLE_TABLES
53 
54 static int __net_init fib4_rules_init(struct net *net)
55 {
56 	struct fib_table *local_table, *main_table;
57 
58 	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
59 	if (!main_table)
60 		return -ENOMEM;
61 
62 	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
63 	if (!local_table)
64 		goto fail;
65 
66 	hlist_add_head_rcu(&local_table->tb_hlist,
67 				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
68 	hlist_add_head_rcu(&main_table->tb_hlist,
69 				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
70 	return 0;
71 
72 fail:
73 	fib_free_table(main_table);
74 	return -ENOMEM;
75 }
76 
77 static bool fib4_has_custom_rules(struct net *net)
78 {
79 	return false;
80 }
81 #else
82 
83 struct fib_table *fib_new_table(struct net *net, u32 id)
84 {
85 	struct fib_table *tb, *alias = NULL;
86 	unsigned int h;
87 
88 	if (id == 0)
89 		id = RT_TABLE_MAIN;
90 	tb = fib_get_table(net, id);
91 	if (tb)
92 		return tb;
93 
94 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
95 		alias = fib_new_table(net, RT_TABLE_MAIN);
96 
97 	tb = fib_trie_table(id, alias);
98 	if (!tb)
99 		return NULL;
100 
101 	switch (id) {
102 	case RT_TABLE_MAIN:
103 		rcu_assign_pointer(net->ipv4.fib_main, tb);
104 		break;
105 	case RT_TABLE_DEFAULT:
106 		rcu_assign_pointer(net->ipv4.fib_default, tb);
107 		break;
108 	default:
109 		break;
110 	}
111 
112 	h = id & (FIB_TABLE_HASHSZ - 1);
113 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
114 	return tb;
115 }
116 EXPORT_SYMBOL_GPL(fib_new_table);
117 
118 /* caller must hold either rtnl or rcu read lock */
119 struct fib_table *fib_get_table(struct net *net, u32 id)
120 {
121 	struct fib_table *tb;
122 	struct hlist_head *head;
123 	unsigned int h;
124 
125 	if (id == 0)
126 		id = RT_TABLE_MAIN;
127 	h = id & (FIB_TABLE_HASHSZ - 1);
128 
129 	head = &net->ipv4.fib_table_hash[h];
130 	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
131 		if (tb->tb_id == id)
132 			return tb;
133 	}
134 	return NULL;
135 }
136 
137 static bool fib4_has_custom_rules(struct net *net)
138 {
139 	return net->ipv4.fib_has_custom_rules;
140 }
141 #endif /* CONFIG_IP_MULTIPLE_TABLES */
142 
143 static void fib_replace_table(struct net *net, struct fib_table *old,
144 			      struct fib_table *new)
145 {
146 #ifdef CONFIG_IP_MULTIPLE_TABLES
147 	switch (new->tb_id) {
148 	case RT_TABLE_MAIN:
149 		rcu_assign_pointer(net->ipv4.fib_main, new);
150 		break;
151 	case RT_TABLE_DEFAULT:
152 		rcu_assign_pointer(net->ipv4.fib_default, new);
153 		break;
154 	default:
155 		break;
156 	}
157 
158 #endif
159 	/* replace the old table in the hlist */
160 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
161 }
162 
163 int fib_unmerge(struct net *net)
164 {
165 	struct fib_table *old, *new, *main_table;
166 
167 	/* attempt to fetch local table if it has been allocated */
168 	old = fib_get_table(net, RT_TABLE_LOCAL);
169 	if (!old)
170 		return 0;
171 
172 	new = fib_trie_unmerge(old);
173 	if (!new)
174 		return -ENOMEM;
175 
176 	/* table is already unmerged */
177 	if (new == old)
178 		return 0;
179 
180 	/* replace merged table with clean table */
181 	fib_replace_table(net, old, new);
182 	fib_free_table(old);
183 
184 	/* attempt to fetch main table if it has been allocated */
185 	main_table = fib_get_table(net, RT_TABLE_MAIN);
186 	if (!main_table)
187 		return 0;
188 
189 	/* flush local entries from main table */
190 	fib_table_flush_external(main_table);
191 
192 	return 0;
193 }
194 
195 void fib_flush(struct net *net)
196 {
197 	int flushed = 0;
198 	unsigned int h;
199 
200 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
201 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
202 		struct hlist_node *tmp;
203 		struct fib_table *tb;
204 
205 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
206 			flushed += fib_table_flush(net, tb, false);
207 	}
208 
209 	if (flushed)
210 		rt_cache_flush(net);
211 }
212 
213 /*
214  * Find address type as if only "dev" was present in the system. If
215  * on_dev is NULL then all interfaces are taken into consideration.
216  */
217 static inline unsigned int __inet_dev_addr_type(struct net *net,
218 						const struct net_device *dev,
219 						__be32 addr, u32 tb_id)
220 {
221 	struct flowi4		fl4 = { .daddr = addr };
222 	struct fib_result	res;
223 	unsigned int ret = RTN_BROADCAST;
224 	struct fib_table *table;
225 
226 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
227 		return RTN_BROADCAST;
228 	if (ipv4_is_multicast(addr))
229 		return RTN_MULTICAST;
230 
231 	rcu_read_lock();
232 
233 	table = fib_get_table(net, tb_id);
234 	if (table) {
235 		ret = RTN_UNICAST;
236 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
237 			if (!dev || dev == res.fi->fib_dev)
238 				ret = res.type;
239 		}
240 	}
241 
242 	rcu_read_unlock();
243 	return ret;
244 }
245 
246 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
247 {
248 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
249 }
250 EXPORT_SYMBOL(inet_addr_type_table);
251 
252 unsigned int inet_addr_type(struct net *net, __be32 addr)
253 {
254 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
255 }
256 EXPORT_SYMBOL(inet_addr_type);
257 
258 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
259 				__be32 addr)
260 {
261 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
262 
263 	return __inet_dev_addr_type(net, dev, addr, rt_table);
264 }
265 EXPORT_SYMBOL(inet_dev_addr_type);
266 
267 /* inet_addr_type with dev == NULL but using the table from a dev
268  * if one is associated
269  */
270 unsigned int inet_addr_type_dev_table(struct net *net,
271 				      const struct net_device *dev,
272 				      __be32 addr)
273 {
274 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
275 
276 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
277 }
278 EXPORT_SYMBOL(inet_addr_type_dev_table);
279 
280 __be32 fib_compute_spec_dst(struct sk_buff *skb)
281 {
282 	struct net_device *dev = skb->dev;
283 	struct in_device *in_dev;
284 	struct fib_result res;
285 	struct rtable *rt;
286 	struct net *net;
287 	int scope;
288 
289 	rt = skb_rtable(skb);
290 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
291 	    RTCF_LOCAL)
292 		return ip_hdr(skb)->daddr;
293 
294 	in_dev = __in_dev_get_rcu(dev);
295 
296 	net = dev_net(dev);
297 
298 	scope = RT_SCOPE_UNIVERSE;
299 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
300 		bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
301 		struct flowi4 fl4 = {
302 			.flowi4_iif = LOOPBACK_IFINDEX,
303 			.flowi4_oif = l3mdev_master_ifindex_rcu(dev),
304 			.daddr = ip_hdr(skb)->saddr,
305 			.flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
306 			.flowi4_scope = scope,
307 			.flowi4_mark = vmark ? skb->mark : 0,
308 		};
309 		if (!fib_lookup(net, &fl4, &res, 0))
310 			return fib_result_prefsrc(net, &res);
311 	} else {
312 		scope = RT_SCOPE_LINK;
313 	}
314 
315 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
316 }
317 
318 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
319 {
320 	bool dev_match = false;
321 #ifdef CONFIG_IP_ROUTE_MULTIPATH
322 	int ret;
323 
324 	for (ret = 0; ret < fi->fib_nhs; ret++) {
325 		struct fib_nh *nh = &fi->fib_nh[ret];
326 
327 		if (nh->fib_nh_dev == dev) {
328 			dev_match = true;
329 			break;
330 		} else if (l3mdev_master_ifindex_rcu(nh->fib_nh_dev) == dev->ifindex) {
331 			dev_match = true;
332 			break;
333 		}
334 	}
335 #else
336 	if (fi->fib_nh[0].fib_nh_dev == dev)
337 		dev_match = true;
338 #endif
339 
340 	return dev_match;
341 }
342 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
343 
344 /* Given (packet source, input interface) and optional (dst, oif, tos):
345  * - (main) check, that source is valid i.e. not broadcast or our local
346  *   address.
347  * - figure out what "logical" interface this packet arrived
348  *   and calculate "specific destination" address.
349  * - check, that packet arrived from expected physical interface.
350  * called with rcu_read_lock()
351  */
352 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
353 				 u8 tos, int oif, struct net_device *dev,
354 				 int rpf, struct in_device *idev, u32 *itag)
355 {
356 	struct net *net = dev_net(dev);
357 	struct flow_keys flkeys;
358 	int ret, no_addr;
359 	struct fib_result res;
360 	struct flowi4 fl4;
361 	bool dev_match;
362 
363 	fl4.flowi4_oif = 0;
364 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
365 	if (!fl4.flowi4_iif)
366 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
367 	fl4.daddr = src;
368 	fl4.saddr = dst;
369 	fl4.flowi4_tos = tos;
370 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
371 	fl4.flowi4_tun_key.tun_id = 0;
372 	fl4.flowi4_flags = 0;
373 	fl4.flowi4_uid = sock_net_uid(net, NULL);
374 
375 	no_addr = idev->ifa_list == NULL;
376 
377 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
378 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
379 		fl4.flowi4_proto = 0;
380 		fl4.fl4_sport = 0;
381 		fl4.fl4_dport = 0;
382 	}
383 
384 	if (fib_lookup(net, &fl4, &res, 0))
385 		goto last_resort;
386 	if (res.type != RTN_UNICAST &&
387 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
388 		goto e_inval;
389 	fib_combine_itag(itag, &res);
390 
391 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
392 	if (dev_match) {
393 		ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
394 		return ret;
395 	}
396 	if (no_addr)
397 		goto last_resort;
398 	if (rpf == 1)
399 		goto e_rpf;
400 	fl4.flowi4_oif = dev->ifindex;
401 
402 	ret = 0;
403 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
404 		if (res.type == RTN_UNICAST)
405 			ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
406 	}
407 	return ret;
408 
409 last_resort:
410 	if (rpf)
411 		goto e_rpf;
412 	*itag = 0;
413 	return 0;
414 
415 e_inval:
416 	return -EINVAL;
417 e_rpf:
418 	return -EXDEV;
419 }
420 
421 /* Ignore rp_filter for packets protected by IPsec. */
422 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
423 			u8 tos, int oif, struct net_device *dev,
424 			struct in_device *idev, u32 *itag)
425 {
426 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
427 	struct net *net = dev_net(dev);
428 
429 	if (!r && !fib_num_tclassid_users(net) &&
430 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
431 		if (IN_DEV_ACCEPT_LOCAL(idev))
432 			goto ok;
433 		/* with custom local routes in place, checking local addresses
434 		 * only will be too optimistic, with custom rules, checking
435 		 * local addresses only can be too strict, e.g. due to vrf
436 		 */
437 		if (net->ipv4.fib_has_custom_local_routes ||
438 		    fib4_has_custom_rules(net))
439 			goto full_check;
440 		if (inet_lookup_ifaddr_rcu(net, src))
441 			return -EINVAL;
442 
443 ok:
444 		*itag = 0;
445 		return 0;
446 	}
447 
448 full_check:
449 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
450 }
451 
452 static inline __be32 sk_extract_addr(struct sockaddr *addr)
453 {
454 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
455 }
456 
457 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
458 {
459 	struct nlattr *nla;
460 
461 	nla = (struct nlattr *) ((char *) mx + len);
462 	nla->nla_type = type;
463 	nla->nla_len = nla_attr_size(4);
464 	*(u32 *) nla_data(nla) = value;
465 
466 	return len + nla_total_size(4);
467 }
468 
469 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
470 				 struct fib_config *cfg)
471 {
472 	__be32 addr;
473 	int plen;
474 
475 	memset(cfg, 0, sizeof(*cfg));
476 	cfg->fc_nlinfo.nl_net = net;
477 
478 	if (rt->rt_dst.sa_family != AF_INET)
479 		return -EAFNOSUPPORT;
480 
481 	/*
482 	 * Check mask for validity:
483 	 * a) it must be contiguous.
484 	 * b) destination must have all host bits clear.
485 	 * c) if application forgot to set correct family (AF_INET),
486 	 *    reject request unless it is absolutely clear i.e.
487 	 *    both family and mask are zero.
488 	 */
489 	plen = 32;
490 	addr = sk_extract_addr(&rt->rt_dst);
491 	if (!(rt->rt_flags & RTF_HOST)) {
492 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
493 
494 		if (rt->rt_genmask.sa_family != AF_INET) {
495 			if (mask || rt->rt_genmask.sa_family)
496 				return -EAFNOSUPPORT;
497 		}
498 
499 		if (bad_mask(mask, addr))
500 			return -EINVAL;
501 
502 		plen = inet_mask_len(mask);
503 	}
504 
505 	cfg->fc_dst_len = plen;
506 	cfg->fc_dst = addr;
507 
508 	if (cmd != SIOCDELRT) {
509 		cfg->fc_nlflags = NLM_F_CREATE;
510 		cfg->fc_protocol = RTPROT_BOOT;
511 	}
512 
513 	if (rt->rt_metric)
514 		cfg->fc_priority = rt->rt_metric - 1;
515 
516 	if (rt->rt_flags & RTF_REJECT) {
517 		cfg->fc_scope = RT_SCOPE_HOST;
518 		cfg->fc_type = RTN_UNREACHABLE;
519 		return 0;
520 	}
521 
522 	cfg->fc_scope = RT_SCOPE_NOWHERE;
523 	cfg->fc_type = RTN_UNICAST;
524 
525 	if (rt->rt_dev) {
526 		char *colon;
527 		struct net_device *dev;
528 		char devname[IFNAMSIZ];
529 
530 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
531 			return -EFAULT;
532 
533 		devname[IFNAMSIZ-1] = 0;
534 		colon = strchr(devname, ':');
535 		if (colon)
536 			*colon = 0;
537 		dev = __dev_get_by_name(net, devname);
538 		if (!dev)
539 			return -ENODEV;
540 		cfg->fc_oif = dev->ifindex;
541 		cfg->fc_table = l3mdev_fib_table(dev);
542 		if (colon) {
543 			const struct in_ifaddr *ifa;
544 			struct in_device *in_dev;
545 
546 			in_dev = __in_dev_get_rtnl(dev);
547 			if (!in_dev)
548 				return -ENODEV;
549 
550 			*colon = ':';
551 
552 			rcu_read_lock();
553 			in_dev_for_each_ifa_rcu(ifa, in_dev) {
554 				if (strcmp(ifa->ifa_label, devname) == 0)
555 					break;
556 			}
557 			rcu_read_unlock();
558 
559 			if (!ifa)
560 				return -ENODEV;
561 			cfg->fc_prefsrc = ifa->ifa_local;
562 		}
563 	}
564 
565 	addr = sk_extract_addr(&rt->rt_gateway);
566 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
567 		unsigned int addr_type;
568 
569 		cfg->fc_gw4 = addr;
570 		cfg->fc_gw_family = AF_INET;
571 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
572 		if (rt->rt_flags & RTF_GATEWAY &&
573 		    addr_type == RTN_UNICAST)
574 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
575 	}
576 
577 	if (cmd == SIOCDELRT)
578 		return 0;
579 
580 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
581 		return -EINVAL;
582 
583 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
584 		cfg->fc_scope = RT_SCOPE_LINK;
585 
586 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
587 		struct nlattr *mx;
588 		int len = 0;
589 
590 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
591 		if (!mx)
592 			return -ENOMEM;
593 
594 		if (rt->rt_flags & RTF_MTU)
595 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
596 
597 		if (rt->rt_flags & RTF_WINDOW)
598 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
599 
600 		if (rt->rt_flags & RTF_IRTT)
601 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
602 
603 		cfg->fc_mx = mx;
604 		cfg->fc_mx_len = len;
605 	}
606 
607 	return 0;
608 }
609 
610 /*
611  * Handle IP routing ioctl calls.
612  * These are used to manipulate the routing tables
613  */
614 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
615 {
616 	struct fib_config cfg;
617 	int err;
618 
619 	switch (cmd) {
620 	case SIOCADDRT:		/* Add a route */
621 	case SIOCDELRT:		/* Delete a route */
622 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
623 			return -EPERM;
624 
625 		rtnl_lock();
626 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
627 		if (err == 0) {
628 			struct fib_table *tb;
629 
630 			if (cmd == SIOCDELRT) {
631 				tb = fib_get_table(net, cfg.fc_table);
632 				if (tb)
633 					err = fib_table_delete(net, tb, &cfg,
634 							       NULL);
635 				else
636 					err = -ESRCH;
637 			} else {
638 				tb = fib_new_table(net, cfg.fc_table);
639 				if (tb)
640 					err = fib_table_insert(net, tb,
641 							       &cfg, NULL);
642 				else
643 					err = -ENOBUFS;
644 			}
645 
646 			/* allocated by rtentry_to_fib_config() */
647 			kfree(cfg.fc_mx);
648 		}
649 		rtnl_unlock();
650 		return err;
651 	}
652 	return -EINVAL;
653 }
654 
655 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
656 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
657 	[RTA_DST]		= { .type = NLA_U32 },
658 	[RTA_SRC]		= { .type = NLA_U32 },
659 	[RTA_IIF]		= { .type = NLA_U32 },
660 	[RTA_OIF]		= { .type = NLA_U32 },
661 	[RTA_GATEWAY]		= { .type = NLA_U32 },
662 	[RTA_PRIORITY]		= { .type = NLA_U32 },
663 	[RTA_PREFSRC]		= { .type = NLA_U32 },
664 	[RTA_METRICS]		= { .type = NLA_NESTED },
665 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
666 	[RTA_FLOW]		= { .type = NLA_U32 },
667 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
668 	[RTA_ENCAP]		= { .type = NLA_NESTED },
669 	[RTA_UID]		= { .type = NLA_U32 },
670 	[RTA_MARK]		= { .type = NLA_U32 },
671 	[RTA_TABLE]		= { .type = NLA_U32 },
672 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
673 	[RTA_SPORT]		= { .type = NLA_U16 },
674 	[RTA_DPORT]		= { .type = NLA_U16 },
675 };
676 
677 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
678 		    struct netlink_ext_ack *extack)
679 {
680 	struct rtvia *via;
681 	int alen;
682 
683 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
684 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
685 		return -EINVAL;
686 	}
687 
688 	via = nla_data(nla);
689 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
690 
691 	switch (via->rtvia_family) {
692 	case AF_INET:
693 		if (alen != sizeof(__be32)) {
694 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
695 			return -EINVAL;
696 		}
697 		cfg->fc_gw_family = AF_INET;
698 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
699 		break;
700 	case AF_INET6:
701 #ifdef CONFIG_IPV6
702 		if (alen != sizeof(struct in6_addr)) {
703 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
704 			return -EINVAL;
705 		}
706 		cfg->fc_gw_family = AF_INET6;
707 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
708 #else
709 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
710 		return -EINVAL;
711 #endif
712 		break;
713 	default:
714 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
715 		return -EINVAL;
716 	}
717 
718 	return 0;
719 }
720 
721 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
722 			     struct nlmsghdr *nlh, struct fib_config *cfg,
723 			     struct netlink_ext_ack *extack)
724 {
725 	bool has_gw = false, has_via = false;
726 	struct nlattr *attr;
727 	int err, remaining;
728 	struct rtmsg *rtm;
729 
730 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
731 					rtm_ipv4_policy, extack);
732 	if (err < 0)
733 		goto errout;
734 
735 	memset(cfg, 0, sizeof(*cfg));
736 
737 	rtm = nlmsg_data(nlh);
738 	cfg->fc_dst_len = rtm->rtm_dst_len;
739 	cfg->fc_tos = rtm->rtm_tos;
740 	cfg->fc_table = rtm->rtm_table;
741 	cfg->fc_protocol = rtm->rtm_protocol;
742 	cfg->fc_scope = rtm->rtm_scope;
743 	cfg->fc_type = rtm->rtm_type;
744 	cfg->fc_flags = rtm->rtm_flags;
745 	cfg->fc_nlflags = nlh->nlmsg_flags;
746 
747 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
748 	cfg->fc_nlinfo.nlh = nlh;
749 	cfg->fc_nlinfo.nl_net = net;
750 
751 	if (cfg->fc_type > RTN_MAX) {
752 		NL_SET_ERR_MSG(extack, "Invalid route type");
753 		err = -EINVAL;
754 		goto errout;
755 	}
756 
757 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
758 		switch (nla_type(attr)) {
759 		case RTA_DST:
760 			cfg->fc_dst = nla_get_be32(attr);
761 			break;
762 		case RTA_OIF:
763 			cfg->fc_oif = nla_get_u32(attr);
764 			break;
765 		case RTA_GATEWAY:
766 			has_gw = true;
767 			cfg->fc_gw4 = nla_get_be32(attr);
768 			if (cfg->fc_gw4)
769 				cfg->fc_gw_family = AF_INET;
770 			break;
771 		case RTA_VIA:
772 			has_via = true;
773 			err = fib_gw_from_via(cfg, attr, extack);
774 			if (err)
775 				goto errout;
776 			break;
777 		case RTA_PRIORITY:
778 			cfg->fc_priority = nla_get_u32(attr);
779 			break;
780 		case RTA_PREFSRC:
781 			cfg->fc_prefsrc = nla_get_be32(attr);
782 			break;
783 		case RTA_METRICS:
784 			cfg->fc_mx = nla_data(attr);
785 			cfg->fc_mx_len = nla_len(attr);
786 			break;
787 		case RTA_MULTIPATH:
788 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
789 							     nla_len(attr),
790 							     extack);
791 			if (err < 0)
792 				goto errout;
793 			cfg->fc_mp = nla_data(attr);
794 			cfg->fc_mp_len = nla_len(attr);
795 			break;
796 		case RTA_FLOW:
797 			cfg->fc_flow = nla_get_u32(attr);
798 			break;
799 		case RTA_TABLE:
800 			cfg->fc_table = nla_get_u32(attr);
801 			break;
802 		case RTA_ENCAP:
803 			cfg->fc_encap = attr;
804 			break;
805 		case RTA_ENCAP_TYPE:
806 			cfg->fc_encap_type = nla_get_u16(attr);
807 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
808 							extack);
809 			if (err < 0)
810 				goto errout;
811 			break;
812 		}
813 	}
814 
815 	if (has_gw && has_via) {
816 		NL_SET_ERR_MSG(extack,
817 			       "Nexthop configuration can not contain both GATEWAY and VIA");
818 		goto errout;
819 	}
820 
821 	return 0;
822 errout:
823 	return err;
824 }
825 
826 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
827 			     struct netlink_ext_ack *extack)
828 {
829 	struct net *net = sock_net(skb->sk);
830 	struct fib_config cfg;
831 	struct fib_table *tb;
832 	int err;
833 
834 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
835 	if (err < 0)
836 		goto errout;
837 
838 	tb = fib_get_table(net, cfg.fc_table);
839 	if (!tb) {
840 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
841 		err = -ESRCH;
842 		goto errout;
843 	}
844 
845 	err = fib_table_delete(net, tb, &cfg, extack);
846 errout:
847 	return err;
848 }
849 
850 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
851 			     struct netlink_ext_ack *extack)
852 {
853 	struct net *net = sock_net(skb->sk);
854 	struct fib_config cfg;
855 	struct fib_table *tb;
856 	int err;
857 
858 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
859 	if (err < 0)
860 		goto errout;
861 
862 	tb = fib_new_table(net, cfg.fc_table);
863 	if (!tb) {
864 		err = -ENOBUFS;
865 		goto errout;
866 	}
867 
868 	err = fib_table_insert(net, tb, &cfg, extack);
869 	if (!err && cfg.fc_type == RTN_LOCAL)
870 		net->ipv4.fib_has_custom_local_routes = true;
871 errout:
872 	return err;
873 }
874 
875 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
876 			  struct fib_dump_filter *filter,
877 			  struct netlink_callback *cb)
878 {
879 	struct netlink_ext_ack *extack = cb->extack;
880 	struct nlattr *tb[RTA_MAX + 1];
881 	struct rtmsg *rtm;
882 	int err, i;
883 
884 	ASSERT_RTNL();
885 
886 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
887 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
888 		return -EINVAL;
889 	}
890 
891 	rtm = nlmsg_data(nlh);
892 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
893 	    rtm->rtm_scope) {
894 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
895 		return -EINVAL;
896 	}
897 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
898 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
899 		return -EINVAL;
900 	}
901 
902 	filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
903 	filter->flags    = rtm->rtm_flags;
904 	filter->protocol = rtm->rtm_protocol;
905 	filter->rt_type  = rtm->rtm_type;
906 	filter->table_id = rtm->rtm_table;
907 
908 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
909 					    rtm_ipv4_policy, extack);
910 	if (err < 0)
911 		return err;
912 
913 	for (i = 0; i <= RTA_MAX; ++i) {
914 		int ifindex;
915 
916 		if (!tb[i])
917 			continue;
918 
919 		switch (i) {
920 		case RTA_TABLE:
921 			filter->table_id = nla_get_u32(tb[i]);
922 			break;
923 		case RTA_OIF:
924 			ifindex = nla_get_u32(tb[i]);
925 			filter->dev = __dev_get_by_index(net, ifindex);
926 			if (!filter->dev)
927 				return -ENODEV;
928 			break;
929 		default:
930 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
931 			return -EINVAL;
932 		}
933 	}
934 
935 	if (filter->flags || filter->protocol || filter->rt_type ||
936 	    filter->table_id || filter->dev) {
937 		filter->filter_set = 1;
938 		cb->answer_flags = NLM_F_DUMP_FILTERED;
939 	}
940 
941 	return 0;
942 }
943 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
944 
945 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
946 {
947 	const struct nlmsghdr *nlh = cb->nlh;
948 	struct net *net = sock_net(skb->sk);
949 	struct fib_dump_filter filter = {};
950 	unsigned int h, s_h;
951 	unsigned int e = 0, s_e;
952 	struct fib_table *tb;
953 	struct hlist_head *head;
954 	int dumped = 0, err;
955 
956 	if (cb->strict_check) {
957 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
958 		if (err < 0)
959 			return err;
960 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
961 		struct rtmsg *rtm = nlmsg_data(nlh);
962 
963 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
964 	}
965 
966 	/* fib entries are never clones and ipv4 does not use prefix flag */
967 	if (filter.flags & (RTM_F_PREFIX | RTM_F_CLONED))
968 		return skb->len;
969 
970 	if (filter.table_id) {
971 		tb = fib_get_table(net, filter.table_id);
972 		if (!tb) {
973 			if (filter.dump_all_families)
974 				return skb->len;
975 
976 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
977 			return -ENOENT;
978 		}
979 
980 		err = fib_table_dump(tb, skb, cb, &filter);
981 		return skb->len ? : err;
982 	}
983 
984 	s_h = cb->args[0];
985 	s_e = cb->args[1];
986 
987 	rcu_read_lock();
988 
989 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
990 		e = 0;
991 		head = &net->ipv4.fib_table_hash[h];
992 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
993 			if (e < s_e)
994 				goto next;
995 			if (dumped)
996 				memset(&cb->args[2], 0, sizeof(cb->args) -
997 						 2 * sizeof(cb->args[0]));
998 			err = fib_table_dump(tb, skb, cb, &filter);
999 			if (err < 0) {
1000 				if (likely(skb->len))
1001 					goto out;
1002 
1003 				goto out_err;
1004 			}
1005 			dumped = 1;
1006 next:
1007 			e++;
1008 		}
1009 	}
1010 out:
1011 	err = skb->len;
1012 out_err:
1013 	rcu_read_unlock();
1014 
1015 	cb->args[1] = e;
1016 	cb->args[0] = h;
1017 
1018 	return err;
1019 }
1020 
1021 /* Prepare and feed intra-kernel routing request.
1022  * Really, it should be netlink message, but :-( netlink
1023  * can be not configured, so that we feed it directly
1024  * to fib engine. It is legal, because all events occur
1025  * only when netlink is already locked.
1026  */
1027 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1028 		      struct in_ifaddr *ifa, u32 rt_priority)
1029 {
1030 	struct net *net = dev_net(ifa->ifa_dev->dev);
1031 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1032 	struct fib_table *tb;
1033 	struct fib_config cfg = {
1034 		.fc_protocol = RTPROT_KERNEL,
1035 		.fc_type = type,
1036 		.fc_dst = dst,
1037 		.fc_dst_len = dst_len,
1038 		.fc_priority = rt_priority,
1039 		.fc_prefsrc = ifa->ifa_local,
1040 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1041 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1042 		.fc_nlinfo = {
1043 			.nl_net = net,
1044 		},
1045 	};
1046 
1047 	if (!tb_id)
1048 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1049 
1050 	tb = fib_new_table(net, tb_id);
1051 	if (!tb)
1052 		return;
1053 
1054 	cfg.fc_table = tb->tb_id;
1055 
1056 	if (type != RTN_LOCAL)
1057 		cfg.fc_scope = RT_SCOPE_LINK;
1058 	else
1059 		cfg.fc_scope = RT_SCOPE_HOST;
1060 
1061 	if (cmd == RTM_NEWROUTE)
1062 		fib_table_insert(net, tb, &cfg, NULL);
1063 	else
1064 		fib_table_delete(net, tb, &cfg, NULL);
1065 }
1066 
1067 void fib_add_ifaddr(struct in_ifaddr *ifa)
1068 {
1069 	struct in_device *in_dev = ifa->ifa_dev;
1070 	struct net_device *dev = in_dev->dev;
1071 	struct in_ifaddr *prim = ifa;
1072 	__be32 mask = ifa->ifa_mask;
1073 	__be32 addr = ifa->ifa_local;
1074 	__be32 prefix = ifa->ifa_address & mask;
1075 
1076 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1077 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1078 		if (!prim) {
1079 			pr_warn("%s: bug: prim == NULL\n", __func__);
1080 			return;
1081 		}
1082 	}
1083 
1084 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1085 
1086 	if (!(dev->flags & IFF_UP))
1087 		return;
1088 
1089 	/* Add broadcast address, if it is explicitly assigned. */
1090 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1091 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1092 			  prim, 0);
1093 
1094 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1095 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1096 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1097 			fib_magic(RTM_NEWROUTE,
1098 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1099 				  prefix, ifa->ifa_prefixlen, prim,
1100 				  ifa->ifa_rt_priority);
1101 
1102 		/* Add network specific broadcasts, when it takes a sense */
1103 		if (ifa->ifa_prefixlen < 31) {
1104 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1105 				  prim, 0);
1106 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1107 				  32, prim, 0);
1108 		}
1109 	}
1110 }
1111 
1112 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1113 {
1114 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1115 	struct in_device *in_dev = ifa->ifa_dev;
1116 	struct net_device *dev = in_dev->dev;
1117 
1118 	if (!(dev->flags & IFF_UP) ||
1119 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1120 	    ipv4_is_zeronet(prefix) ||
1121 	    prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1122 		return;
1123 
1124 	/* add the new */
1125 	fib_magic(RTM_NEWROUTE,
1126 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1127 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1128 
1129 	/* delete the old */
1130 	fib_magic(RTM_DELROUTE,
1131 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1132 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1133 }
1134 
1135 /* Delete primary or secondary address.
1136  * Optionally, on secondary address promotion consider the addresses
1137  * from subnet iprim as deleted, even if they are in device list.
1138  * In this case the secondary ifa can be in device list.
1139  */
1140 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1141 {
1142 	struct in_device *in_dev = ifa->ifa_dev;
1143 	struct net_device *dev = in_dev->dev;
1144 	struct in_ifaddr *ifa1;
1145 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1146 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1147 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1148 #define LOCAL_OK	1
1149 #define BRD_OK		2
1150 #define BRD0_OK		4
1151 #define BRD1_OK		8
1152 	unsigned int ok = 0;
1153 	int subnet = 0;		/* Primary network */
1154 	int gone = 1;		/* Address is missing */
1155 	int same_prefsrc = 0;	/* Another primary with same IP */
1156 
1157 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1158 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1159 		if (!prim) {
1160 			/* if the device has been deleted, we don't perform
1161 			 * address promotion
1162 			 */
1163 			if (!in_dev->dead)
1164 				pr_warn("%s: bug: prim == NULL\n", __func__);
1165 			return;
1166 		}
1167 		if (iprim && iprim != prim) {
1168 			pr_warn("%s: bug: iprim != prim\n", __func__);
1169 			return;
1170 		}
1171 	} else if (!ipv4_is_zeronet(any) &&
1172 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1173 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1174 			fib_magic(RTM_DELROUTE,
1175 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1176 				  any, ifa->ifa_prefixlen, prim, 0);
1177 		subnet = 1;
1178 	}
1179 
1180 	if (in_dev->dead)
1181 		goto no_promotions;
1182 
1183 	/* Deletion is more complicated than add.
1184 	 * We should take care of not to delete too much :-)
1185 	 *
1186 	 * Scan address list to be sure that addresses are really gone.
1187 	 */
1188 	rcu_read_lock();
1189 	in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1190 		if (ifa1 == ifa) {
1191 			/* promotion, keep the IP */
1192 			gone = 0;
1193 			continue;
1194 		}
1195 		/* Ignore IFAs from our subnet */
1196 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1197 		    inet_ifa_match(ifa1->ifa_address, iprim))
1198 			continue;
1199 
1200 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1201 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1202 			/* Another address from our subnet? */
1203 			if (ifa1->ifa_mask == prim->ifa_mask &&
1204 			    inet_ifa_match(ifa1->ifa_address, prim))
1205 				prim1 = prim;
1206 			else {
1207 				/* We reached the secondaries, so
1208 				 * same_prefsrc should be determined.
1209 				 */
1210 				if (!same_prefsrc)
1211 					continue;
1212 				/* Search new prim1 if ifa1 is not
1213 				 * using the current prim1
1214 				 */
1215 				if (!prim1 ||
1216 				    ifa1->ifa_mask != prim1->ifa_mask ||
1217 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1218 					prim1 = inet_ifa_byprefix(in_dev,
1219 							ifa1->ifa_address,
1220 							ifa1->ifa_mask);
1221 				if (!prim1)
1222 					continue;
1223 				if (prim1->ifa_local != prim->ifa_local)
1224 					continue;
1225 			}
1226 		} else {
1227 			if (prim->ifa_local != ifa1->ifa_local)
1228 				continue;
1229 			prim1 = ifa1;
1230 			if (prim != prim1)
1231 				same_prefsrc = 1;
1232 		}
1233 		if (ifa->ifa_local == ifa1->ifa_local)
1234 			ok |= LOCAL_OK;
1235 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1236 			ok |= BRD_OK;
1237 		if (brd == ifa1->ifa_broadcast)
1238 			ok |= BRD1_OK;
1239 		if (any == ifa1->ifa_broadcast)
1240 			ok |= BRD0_OK;
1241 		/* primary has network specific broadcasts */
1242 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1243 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1244 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1245 
1246 			if (!ipv4_is_zeronet(any1)) {
1247 				if (ifa->ifa_broadcast == brd1 ||
1248 				    ifa->ifa_broadcast == any1)
1249 					ok |= BRD_OK;
1250 				if (brd == brd1 || brd == any1)
1251 					ok |= BRD1_OK;
1252 				if (any == brd1 || any == any1)
1253 					ok |= BRD0_OK;
1254 			}
1255 		}
1256 	}
1257 	rcu_read_unlock();
1258 
1259 no_promotions:
1260 	if (!(ok & BRD_OK))
1261 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1262 			  prim, 0);
1263 	if (subnet && ifa->ifa_prefixlen < 31) {
1264 		if (!(ok & BRD1_OK))
1265 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1266 				  prim, 0);
1267 		if (!(ok & BRD0_OK))
1268 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1269 				  prim, 0);
1270 	}
1271 	if (!(ok & LOCAL_OK)) {
1272 		unsigned int addr_type;
1273 
1274 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1275 
1276 		/* Check, that this local address finally disappeared. */
1277 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1278 						     ifa->ifa_local);
1279 		if (gone && addr_type != RTN_LOCAL) {
1280 			/* And the last, but not the least thing.
1281 			 * We must flush stray FIB entries.
1282 			 *
1283 			 * First of all, we scan fib_info list searching
1284 			 * for stray nexthop entries, then ignite fib_flush.
1285 			 */
1286 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1287 				fib_flush(dev_net(dev));
1288 		}
1289 	}
1290 #undef LOCAL_OK
1291 #undef BRD_OK
1292 #undef BRD0_OK
1293 #undef BRD1_OK
1294 }
1295 
1296 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1297 {
1298 
1299 	struct fib_result       res;
1300 	struct flowi4           fl4 = {
1301 		.flowi4_mark = frn->fl_mark,
1302 		.daddr = frn->fl_addr,
1303 		.flowi4_tos = frn->fl_tos,
1304 		.flowi4_scope = frn->fl_scope,
1305 	};
1306 	struct fib_table *tb;
1307 
1308 	rcu_read_lock();
1309 
1310 	tb = fib_get_table(net, frn->tb_id_in);
1311 
1312 	frn->err = -ENOENT;
1313 	if (tb) {
1314 		local_bh_disable();
1315 
1316 		frn->tb_id = tb->tb_id;
1317 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1318 
1319 		if (!frn->err) {
1320 			frn->prefixlen = res.prefixlen;
1321 			frn->nh_sel = res.nh_sel;
1322 			frn->type = res.type;
1323 			frn->scope = res.scope;
1324 		}
1325 		local_bh_enable();
1326 	}
1327 
1328 	rcu_read_unlock();
1329 }
1330 
1331 static void nl_fib_input(struct sk_buff *skb)
1332 {
1333 	struct net *net;
1334 	struct fib_result_nl *frn;
1335 	struct nlmsghdr *nlh;
1336 	u32 portid;
1337 
1338 	net = sock_net(skb->sk);
1339 	nlh = nlmsg_hdr(skb);
1340 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1341 	    skb->len < nlh->nlmsg_len ||
1342 	    nlmsg_len(nlh) < sizeof(*frn))
1343 		return;
1344 
1345 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1346 	if (!skb)
1347 		return;
1348 	nlh = nlmsg_hdr(skb);
1349 
1350 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1351 	nl_fib_lookup(net, frn);
1352 
1353 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1354 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1355 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1356 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1357 }
1358 
1359 static int __net_init nl_fib_lookup_init(struct net *net)
1360 {
1361 	struct sock *sk;
1362 	struct netlink_kernel_cfg cfg = {
1363 		.input	= nl_fib_input,
1364 	};
1365 
1366 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1367 	if (!sk)
1368 		return -EAFNOSUPPORT;
1369 	net->ipv4.fibnl = sk;
1370 	return 0;
1371 }
1372 
1373 static void nl_fib_lookup_exit(struct net *net)
1374 {
1375 	netlink_kernel_release(net->ipv4.fibnl);
1376 	net->ipv4.fibnl = NULL;
1377 }
1378 
1379 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1380 			   bool force)
1381 {
1382 	if (fib_sync_down_dev(dev, event, force))
1383 		fib_flush(dev_net(dev));
1384 	else
1385 		rt_cache_flush(dev_net(dev));
1386 	arp_ifdown(dev);
1387 }
1388 
1389 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1390 {
1391 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1392 	struct net_device *dev = ifa->ifa_dev->dev;
1393 	struct net *net = dev_net(dev);
1394 
1395 	switch (event) {
1396 	case NETDEV_UP:
1397 		fib_add_ifaddr(ifa);
1398 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1399 		fib_sync_up(dev, RTNH_F_DEAD);
1400 #endif
1401 		atomic_inc(&net->ipv4.dev_addr_genid);
1402 		rt_cache_flush(dev_net(dev));
1403 		break;
1404 	case NETDEV_DOWN:
1405 		fib_del_ifaddr(ifa, NULL);
1406 		atomic_inc(&net->ipv4.dev_addr_genid);
1407 		if (!ifa->ifa_dev->ifa_list) {
1408 			/* Last address was deleted from this interface.
1409 			 * Disable IP.
1410 			 */
1411 			fib_disable_ip(dev, event, true);
1412 		} else {
1413 			rt_cache_flush(dev_net(dev));
1414 		}
1415 		break;
1416 	}
1417 	return NOTIFY_DONE;
1418 }
1419 
1420 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1421 {
1422 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1423 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1424 	struct netdev_notifier_info_ext *info_ext = ptr;
1425 	struct in_device *in_dev;
1426 	struct net *net = dev_net(dev);
1427 	struct in_ifaddr *ifa;
1428 	unsigned int flags;
1429 
1430 	if (event == NETDEV_UNREGISTER) {
1431 		fib_disable_ip(dev, event, true);
1432 		rt_flush_dev(dev);
1433 		return NOTIFY_DONE;
1434 	}
1435 
1436 	in_dev = __in_dev_get_rtnl(dev);
1437 	if (!in_dev)
1438 		return NOTIFY_DONE;
1439 
1440 	switch (event) {
1441 	case NETDEV_UP:
1442 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1443 			fib_add_ifaddr(ifa);
1444 		}
1445 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1446 		fib_sync_up(dev, RTNH_F_DEAD);
1447 #endif
1448 		atomic_inc(&net->ipv4.dev_addr_genid);
1449 		rt_cache_flush(net);
1450 		break;
1451 	case NETDEV_DOWN:
1452 		fib_disable_ip(dev, event, false);
1453 		break;
1454 	case NETDEV_CHANGE:
1455 		flags = dev_get_flags(dev);
1456 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1457 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1458 		else
1459 			fib_sync_down_dev(dev, event, false);
1460 		rt_cache_flush(net);
1461 		break;
1462 	case NETDEV_CHANGEMTU:
1463 		fib_sync_mtu(dev, info_ext->ext.mtu);
1464 		rt_cache_flush(net);
1465 		break;
1466 	case NETDEV_CHANGEUPPER:
1467 		upper_info = ptr;
1468 		/* flush all routes if dev is linked to or unlinked from
1469 		 * an L3 master device (e.g., VRF)
1470 		 */
1471 		if (upper_info->upper_dev &&
1472 		    netif_is_l3_master(upper_info->upper_dev))
1473 			fib_disable_ip(dev, NETDEV_DOWN, true);
1474 		break;
1475 	}
1476 	return NOTIFY_DONE;
1477 }
1478 
1479 static struct notifier_block fib_inetaddr_notifier = {
1480 	.notifier_call = fib_inetaddr_event,
1481 };
1482 
1483 static struct notifier_block fib_netdev_notifier = {
1484 	.notifier_call = fib_netdev_event,
1485 };
1486 
1487 static int __net_init ip_fib_net_init(struct net *net)
1488 {
1489 	int err;
1490 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1491 
1492 	err = fib4_notifier_init(net);
1493 	if (err)
1494 		return err;
1495 
1496 	/* Avoid false sharing : Use at least a full cache line */
1497 	size = max_t(size_t, size, L1_CACHE_BYTES);
1498 
1499 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1500 	if (!net->ipv4.fib_table_hash) {
1501 		err = -ENOMEM;
1502 		goto err_table_hash_alloc;
1503 	}
1504 
1505 	err = fib4_rules_init(net);
1506 	if (err < 0)
1507 		goto err_rules_init;
1508 	return 0;
1509 
1510 err_rules_init:
1511 	kfree(net->ipv4.fib_table_hash);
1512 err_table_hash_alloc:
1513 	fib4_notifier_exit(net);
1514 	return err;
1515 }
1516 
1517 static void ip_fib_net_exit(struct net *net)
1518 {
1519 	int i;
1520 
1521 	rtnl_lock();
1522 #ifdef CONFIG_IP_MULTIPLE_TABLES
1523 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1524 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1525 #endif
1526 	/* Destroy the tables in reverse order to guarantee that the
1527 	 * local table, ID 255, is destroyed before the main table, ID
1528 	 * 254. This is necessary as the local table may contain
1529 	 * references to data contained in the main table.
1530 	 */
1531 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1532 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1533 		struct hlist_node *tmp;
1534 		struct fib_table *tb;
1535 
1536 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1537 			hlist_del(&tb->tb_hlist);
1538 			fib_table_flush(net, tb, true);
1539 			fib_free_table(tb);
1540 		}
1541 	}
1542 
1543 #ifdef CONFIG_IP_MULTIPLE_TABLES
1544 	fib4_rules_exit(net);
1545 #endif
1546 	rtnl_unlock();
1547 	kfree(net->ipv4.fib_table_hash);
1548 	fib4_notifier_exit(net);
1549 }
1550 
1551 static int __net_init fib_net_init(struct net *net)
1552 {
1553 	int error;
1554 
1555 #ifdef CONFIG_IP_ROUTE_CLASSID
1556 	net->ipv4.fib_num_tclassid_users = 0;
1557 #endif
1558 	error = ip_fib_net_init(net);
1559 	if (error < 0)
1560 		goto out;
1561 	error = nl_fib_lookup_init(net);
1562 	if (error < 0)
1563 		goto out_nlfl;
1564 	error = fib_proc_init(net);
1565 	if (error < 0)
1566 		goto out_proc;
1567 out:
1568 	return error;
1569 
1570 out_proc:
1571 	nl_fib_lookup_exit(net);
1572 out_nlfl:
1573 	ip_fib_net_exit(net);
1574 	goto out;
1575 }
1576 
1577 static void __net_exit fib_net_exit(struct net *net)
1578 {
1579 	fib_proc_exit(net);
1580 	nl_fib_lookup_exit(net);
1581 	ip_fib_net_exit(net);
1582 }
1583 
1584 static struct pernet_operations fib_net_ops = {
1585 	.init = fib_net_init,
1586 	.exit = fib_net_exit,
1587 };
1588 
1589 void __init ip_fib_init(void)
1590 {
1591 	fib_trie_init();
1592 
1593 	register_pernet_subsys(&fib_net_ops);
1594 
1595 	register_netdevice_notifier(&fib_netdev_notifier);
1596 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1597 
1598 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1599 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1600 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1601 }
1602