xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision 6c33a6f4)
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 = RT_TOS(ip_hdr(skb)->tos),
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 	int ret;
313 
314 	for (ret = 0; ret < fib_info_num_path(fi); ret++) {
315 		const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
316 
317 		if (nhc->nhc_dev == dev) {
318 			dev_match = true;
319 			break;
320 		} else if (l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex) {
321 			dev_match = true;
322 			break;
323 		}
324 	}
325 #else
326 	if (fib_info_nhc(fi, 0)->nhc_dev == dev)
327 		dev_match = true;
328 #endif
329 
330 	return dev_match;
331 }
332 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
333 
334 /* Given (packet source, input interface) and optional (dst, oif, tos):
335  * - (main) check, that source is valid i.e. not broadcast or our local
336  *   address.
337  * - figure out what "logical" interface this packet arrived
338  *   and calculate "specific destination" address.
339  * - check, that packet arrived from expected physical interface.
340  * called with rcu_read_lock()
341  */
342 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
343 				 u8 tos, int oif, struct net_device *dev,
344 				 int rpf, struct in_device *idev, u32 *itag)
345 {
346 	struct net *net = dev_net(dev);
347 	struct flow_keys flkeys;
348 	int ret, no_addr;
349 	struct fib_result res;
350 	struct flowi4 fl4;
351 	bool dev_match;
352 
353 	fl4.flowi4_oif = 0;
354 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
355 	if (!fl4.flowi4_iif)
356 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
357 	fl4.daddr = src;
358 	fl4.saddr = dst;
359 	fl4.flowi4_tos = tos;
360 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
361 	fl4.flowi4_tun_key.tun_id = 0;
362 	fl4.flowi4_flags = 0;
363 	fl4.flowi4_uid = sock_net_uid(net, NULL);
364 
365 	no_addr = idev->ifa_list == NULL;
366 
367 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
368 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
369 		fl4.flowi4_proto = 0;
370 		fl4.fl4_sport = 0;
371 		fl4.fl4_dport = 0;
372 	}
373 
374 	if (fib_lookup(net, &fl4, &res, 0))
375 		goto last_resort;
376 	if (res.type != RTN_UNICAST &&
377 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
378 		goto e_inval;
379 	fib_combine_itag(itag, &res);
380 
381 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
382 	/* This is not common, loopback packets retain skb_dst so normally they
383 	 * would not even hit this slow path.
384 	 */
385 	dev_match = dev_match || (res.type == RTN_LOCAL &&
386 				  dev == net->loopback_dev);
387 	if (dev_match) {
388 		ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
389 		return ret;
390 	}
391 	if (no_addr)
392 		goto last_resort;
393 	if (rpf == 1)
394 		goto e_rpf;
395 	fl4.flowi4_oif = dev->ifindex;
396 
397 	ret = 0;
398 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
399 		if (res.type == RTN_UNICAST)
400 			ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
401 	}
402 	return ret;
403 
404 last_resort:
405 	if (rpf)
406 		goto e_rpf;
407 	*itag = 0;
408 	return 0;
409 
410 e_inval:
411 	return -EINVAL;
412 e_rpf:
413 	return -EXDEV;
414 }
415 
416 /* Ignore rp_filter for packets protected by IPsec. */
417 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
418 			u8 tos, int oif, struct net_device *dev,
419 			struct in_device *idev, u32 *itag)
420 {
421 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
422 	struct net *net = dev_net(dev);
423 
424 	if (!r && !fib_num_tclassid_users(net) &&
425 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
426 		if (IN_DEV_ACCEPT_LOCAL(idev))
427 			goto ok;
428 		/* with custom local routes in place, checking local addresses
429 		 * only will be too optimistic, with custom rules, checking
430 		 * local addresses only can be too strict, e.g. due to vrf
431 		 */
432 		if (net->ipv4.fib_has_custom_local_routes ||
433 		    fib4_has_custom_rules(net))
434 			goto full_check;
435 		if (inet_lookup_ifaddr_rcu(net, src))
436 			return -EINVAL;
437 
438 ok:
439 		*itag = 0;
440 		return 0;
441 	}
442 
443 full_check:
444 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
445 }
446 
447 static inline __be32 sk_extract_addr(struct sockaddr *addr)
448 {
449 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
450 }
451 
452 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
453 {
454 	struct nlattr *nla;
455 
456 	nla = (struct nlattr *) ((char *) mx + len);
457 	nla->nla_type = type;
458 	nla->nla_len = nla_attr_size(4);
459 	*(u32 *) nla_data(nla) = value;
460 
461 	return len + nla_total_size(4);
462 }
463 
464 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
465 				 struct fib_config *cfg)
466 {
467 	__be32 addr;
468 	int plen;
469 
470 	memset(cfg, 0, sizeof(*cfg));
471 	cfg->fc_nlinfo.nl_net = net;
472 
473 	if (rt->rt_dst.sa_family != AF_INET)
474 		return -EAFNOSUPPORT;
475 
476 	/*
477 	 * Check mask for validity:
478 	 * a) it must be contiguous.
479 	 * b) destination must have all host bits clear.
480 	 * c) if application forgot to set correct family (AF_INET),
481 	 *    reject request unless it is absolutely clear i.e.
482 	 *    both family and mask are zero.
483 	 */
484 	plen = 32;
485 	addr = sk_extract_addr(&rt->rt_dst);
486 	if (!(rt->rt_flags & RTF_HOST)) {
487 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
488 
489 		if (rt->rt_genmask.sa_family != AF_INET) {
490 			if (mask || rt->rt_genmask.sa_family)
491 				return -EAFNOSUPPORT;
492 		}
493 
494 		if (bad_mask(mask, addr))
495 			return -EINVAL;
496 
497 		plen = inet_mask_len(mask);
498 	}
499 
500 	cfg->fc_dst_len = plen;
501 	cfg->fc_dst = addr;
502 
503 	if (cmd != SIOCDELRT) {
504 		cfg->fc_nlflags = NLM_F_CREATE;
505 		cfg->fc_protocol = RTPROT_BOOT;
506 	}
507 
508 	if (rt->rt_metric)
509 		cfg->fc_priority = rt->rt_metric - 1;
510 
511 	if (rt->rt_flags & RTF_REJECT) {
512 		cfg->fc_scope = RT_SCOPE_HOST;
513 		cfg->fc_type = RTN_UNREACHABLE;
514 		return 0;
515 	}
516 
517 	cfg->fc_scope = RT_SCOPE_NOWHERE;
518 	cfg->fc_type = RTN_UNICAST;
519 
520 	if (rt->rt_dev) {
521 		char *colon;
522 		struct net_device *dev;
523 		char devname[IFNAMSIZ];
524 
525 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
526 			return -EFAULT;
527 
528 		devname[IFNAMSIZ-1] = 0;
529 		colon = strchr(devname, ':');
530 		if (colon)
531 			*colon = 0;
532 		dev = __dev_get_by_name(net, devname);
533 		if (!dev)
534 			return -ENODEV;
535 		cfg->fc_oif = dev->ifindex;
536 		cfg->fc_table = l3mdev_fib_table(dev);
537 		if (colon) {
538 			const struct in_ifaddr *ifa;
539 			struct in_device *in_dev;
540 
541 			in_dev = __in_dev_get_rtnl(dev);
542 			if (!in_dev)
543 				return -ENODEV;
544 
545 			*colon = ':';
546 
547 			rcu_read_lock();
548 			in_dev_for_each_ifa_rcu(ifa, in_dev) {
549 				if (strcmp(ifa->ifa_label, devname) == 0)
550 					break;
551 			}
552 			rcu_read_unlock();
553 
554 			if (!ifa)
555 				return -ENODEV;
556 			cfg->fc_prefsrc = ifa->ifa_local;
557 		}
558 	}
559 
560 	addr = sk_extract_addr(&rt->rt_gateway);
561 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
562 		unsigned int addr_type;
563 
564 		cfg->fc_gw4 = addr;
565 		cfg->fc_gw_family = AF_INET;
566 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
567 		if (rt->rt_flags & RTF_GATEWAY &&
568 		    addr_type == RTN_UNICAST)
569 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
570 	}
571 
572 	if (cmd == SIOCDELRT)
573 		return 0;
574 
575 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
576 		return -EINVAL;
577 
578 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
579 		cfg->fc_scope = RT_SCOPE_LINK;
580 
581 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
582 		struct nlattr *mx;
583 		int len = 0;
584 
585 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
586 		if (!mx)
587 			return -ENOMEM;
588 
589 		if (rt->rt_flags & RTF_MTU)
590 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
591 
592 		if (rt->rt_flags & RTF_WINDOW)
593 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
594 
595 		if (rt->rt_flags & RTF_IRTT)
596 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
597 
598 		cfg->fc_mx = mx;
599 		cfg->fc_mx_len = len;
600 	}
601 
602 	return 0;
603 }
604 
605 /*
606  * Handle IP routing ioctl calls.
607  * These are used to manipulate the routing tables
608  */
609 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
610 {
611 	struct fib_config cfg;
612 	int err;
613 
614 	switch (cmd) {
615 	case SIOCADDRT:		/* Add a route */
616 	case SIOCDELRT:		/* Delete a route */
617 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
618 			return -EPERM;
619 
620 		rtnl_lock();
621 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
622 		if (err == 0) {
623 			struct fib_table *tb;
624 
625 			if (cmd == SIOCDELRT) {
626 				tb = fib_get_table(net, cfg.fc_table);
627 				if (tb)
628 					err = fib_table_delete(net, tb, &cfg,
629 							       NULL);
630 				else
631 					err = -ESRCH;
632 			} else {
633 				tb = fib_new_table(net, cfg.fc_table);
634 				if (tb)
635 					err = fib_table_insert(net, tb,
636 							       &cfg, NULL);
637 				else
638 					err = -ENOBUFS;
639 			}
640 
641 			/* allocated by rtentry_to_fib_config() */
642 			kfree(cfg.fc_mx);
643 		}
644 		rtnl_unlock();
645 		return err;
646 	}
647 	return -EINVAL;
648 }
649 
650 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
651 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
652 	[RTA_DST]		= { .type = NLA_U32 },
653 	[RTA_SRC]		= { .type = NLA_U32 },
654 	[RTA_IIF]		= { .type = NLA_U32 },
655 	[RTA_OIF]		= { .type = NLA_U32 },
656 	[RTA_GATEWAY]		= { .type = NLA_U32 },
657 	[RTA_PRIORITY]		= { .type = NLA_U32 },
658 	[RTA_PREFSRC]		= { .type = NLA_U32 },
659 	[RTA_METRICS]		= { .type = NLA_NESTED },
660 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
661 	[RTA_FLOW]		= { .type = NLA_U32 },
662 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
663 	[RTA_ENCAP]		= { .type = NLA_NESTED },
664 	[RTA_UID]		= { .type = NLA_U32 },
665 	[RTA_MARK]		= { .type = NLA_U32 },
666 	[RTA_TABLE]		= { .type = NLA_U32 },
667 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
668 	[RTA_SPORT]		= { .type = NLA_U16 },
669 	[RTA_DPORT]		= { .type = NLA_U16 },
670 	[RTA_NH_ID]		= { .type = NLA_U32 },
671 };
672 
673 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
674 		    struct netlink_ext_ack *extack)
675 {
676 	struct rtvia *via;
677 	int alen;
678 
679 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
680 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
681 		return -EINVAL;
682 	}
683 
684 	via = nla_data(nla);
685 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
686 
687 	switch (via->rtvia_family) {
688 	case AF_INET:
689 		if (alen != sizeof(__be32)) {
690 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
691 			return -EINVAL;
692 		}
693 		cfg->fc_gw_family = AF_INET;
694 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
695 		break;
696 	case AF_INET6:
697 #ifdef CONFIG_IPV6
698 		if (alen != sizeof(struct in6_addr)) {
699 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
700 			return -EINVAL;
701 		}
702 		cfg->fc_gw_family = AF_INET6;
703 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
704 #else
705 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
706 		return -EINVAL;
707 #endif
708 		break;
709 	default:
710 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
711 		return -EINVAL;
712 	}
713 
714 	return 0;
715 }
716 
717 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
718 			     struct nlmsghdr *nlh, struct fib_config *cfg,
719 			     struct netlink_ext_ack *extack)
720 {
721 	bool has_gw = false, has_via = false;
722 	struct nlattr *attr;
723 	int err, remaining;
724 	struct rtmsg *rtm;
725 
726 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
727 					rtm_ipv4_policy, extack);
728 	if (err < 0)
729 		goto errout;
730 
731 	memset(cfg, 0, sizeof(*cfg));
732 
733 	rtm = nlmsg_data(nlh);
734 	cfg->fc_dst_len = rtm->rtm_dst_len;
735 	cfg->fc_tos = rtm->rtm_tos;
736 	cfg->fc_table = rtm->rtm_table;
737 	cfg->fc_protocol = rtm->rtm_protocol;
738 	cfg->fc_scope = rtm->rtm_scope;
739 	cfg->fc_type = rtm->rtm_type;
740 	cfg->fc_flags = rtm->rtm_flags;
741 	cfg->fc_nlflags = nlh->nlmsg_flags;
742 
743 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
744 	cfg->fc_nlinfo.nlh = nlh;
745 	cfg->fc_nlinfo.nl_net = net;
746 
747 	if (cfg->fc_type > RTN_MAX) {
748 		NL_SET_ERR_MSG(extack, "Invalid route type");
749 		err = -EINVAL;
750 		goto errout;
751 	}
752 
753 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
754 		switch (nla_type(attr)) {
755 		case RTA_DST:
756 			cfg->fc_dst = nla_get_be32(attr);
757 			break;
758 		case RTA_OIF:
759 			cfg->fc_oif = nla_get_u32(attr);
760 			break;
761 		case RTA_GATEWAY:
762 			has_gw = true;
763 			cfg->fc_gw4 = nla_get_be32(attr);
764 			if (cfg->fc_gw4)
765 				cfg->fc_gw_family = AF_INET;
766 			break;
767 		case RTA_VIA:
768 			has_via = true;
769 			err = fib_gw_from_via(cfg, attr, extack);
770 			if (err)
771 				goto errout;
772 			break;
773 		case RTA_PRIORITY:
774 			cfg->fc_priority = nla_get_u32(attr);
775 			break;
776 		case RTA_PREFSRC:
777 			cfg->fc_prefsrc = nla_get_be32(attr);
778 			break;
779 		case RTA_METRICS:
780 			cfg->fc_mx = nla_data(attr);
781 			cfg->fc_mx_len = nla_len(attr);
782 			break;
783 		case RTA_MULTIPATH:
784 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
785 							     nla_len(attr),
786 							     extack);
787 			if (err < 0)
788 				goto errout;
789 			cfg->fc_mp = nla_data(attr);
790 			cfg->fc_mp_len = nla_len(attr);
791 			break;
792 		case RTA_FLOW:
793 			cfg->fc_flow = nla_get_u32(attr);
794 			break;
795 		case RTA_TABLE:
796 			cfg->fc_table = nla_get_u32(attr);
797 			break;
798 		case RTA_ENCAP:
799 			cfg->fc_encap = attr;
800 			break;
801 		case RTA_ENCAP_TYPE:
802 			cfg->fc_encap_type = nla_get_u16(attr);
803 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
804 							extack);
805 			if (err < 0)
806 				goto errout;
807 			break;
808 		case RTA_NH_ID:
809 			cfg->fc_nh_id = nla_get_u32(attr);
810 			break;
811 		}
812 	}
813 
814 	if (cfg->fc_nh_id) {
815 		if (cfg->fc_oif || cfg->fc_gw_family ||
816 		    cfg->fc_encap || cfg->fc_mp) {
817 			NL_SET_ERR_MSG(extack,
818 				       "Nexthop specification and nexthop id are mutually exclusive");
819 			return -EINVAL;
820 		}
821 	}
822 
823 	if (has_gw && has_via) {
824 		NL_SET_ERR_MSG(extack,
825 			       "Nexthop configuration can not contain both GATEWAY and VIA");
826 		goto errout;
827 	}
828 
829 	return 0;
830 errout:
831 	return err;
832 }
833 
834 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
835 			     struct netlink_ext_ack *extack)
836 {
837 	struct net *net = sock_net(skb->sk);
838 	struct fib_config cfg;
839 	struct fib_table *tb;
840 	int err;
841 
842 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
843 	if (err < 0)
844 		goto errout;
845 
846 	if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
847 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
848 		err = -EINVAL;
849 		goto errout;
850 	}
851 
852 	tb = fib_get_table(net, cfg.fc_table);
853 	if (!tb) {
854 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
855 		err = -ESRCH;
856 		goto errout;
857 	}
858 
859 	err = fib_table_delete(net, tb, &cfg, extack);
860 errout:
861 	return err;
862 }
863 
864 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
865 			     struct netlink_ext_ack *extack)
866 {
867 	struct net *net = sock_net(skb->sk);
868 	struct fib_config cfg;
869 	struct fib_table *tb;
870 	int err;
871 
872 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
873 	if (err < 0)
874 		goto errout;
875 
876 	tb = fib_new_table(net, cfg.fc_table);
877 	if (!tb) {
878 		err = -ENOBUFS;
879 		goto errout;
880 	}
881 
882 	err = fib_table_insert(net, tb, &cfg, extack);
883 	if (!err && cfg.fc_type == RTN_LOCAL)
884 		net->ipv4.fib_has_custom_local_routes = true;
885 errout:
886 	return err;
887 }
888 
889 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
890 			  struct fib_dump_filter *filter,
891 			  struct netlink_callback *cb)
892 {
893 	struct netlink_ext_ack *extack = cb->extack;
894 	struct nlattr *tb[RTA_MAX + 1];
895 	struct rtmsg *rtm;
896 	int err, i;
897 
898 	ASSERT_RTNL();
899 
900 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
901 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
902 		return -EINVAL;
903 	}
904 
905 	rtm = nlmsg_data(nlh);
906 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
907 	    rtm->rtm_scope) {
908 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
909 		return -EINVAL;
910 	}
911 
912 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
913 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
914 		return -EINVAL;
915 	}
916 	if (rtm->rtm_flags & RTM_F_CLONED)
917 		filter->dump_routes = false;
918 	else
919 		filter->dump_exceptions = false;
920 
921 	filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
922 	filter->flags    = rtm->rtm_flags;
923 	filter->protocol = rtm->rtm_protocol;
924 	filter->rt_type  = rtm->rtm_type;
925 	filter->table_id = rtm->rtm_table;
926 
927 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
928 					    rtm_ipv4_policy, extack);
929 	if (err < 0)
930 		return err;
931 
932 	for (i = 0; i <= RTA_MAX; ++i) {
933 		int ifindex;
934 
935 		if (!tb[i])
936 			continue;
937 
938 		switch (i) {
939 		case RTA_TABLE:
940 			filter->table_id = nla_get_u32(tb[i]);
941 			break;
942 		case RTA_OIF:
943 			ifindex = nla_get_u32(tb[i]);
944 			filter->dev = __dev_get_by_index(net, ifindex);
945 			if (!filter->dev)
946 				return -ENODEV;
947 			break;
948 		default:
949 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
950 			return -EINVAL;
951 		}
952 	}
953 
954 	if (filter->flags || filter->protocol || filter->rt_type ||
955 	    filter->table_id || filter->dev) {
956 		filter->filter_set = 1;
957 		cb->answer_flags = NLM_F_DUMP_FILTERED;
958 	}
959 
960 	return 0;
961 }
962 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
963 
964 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
965 {
966 	struct fib_dump_filter filter = { .dump_routes = true,
967 					  .dump_exceptions = true };
968 	const struct nlmsghdr *nlh = cb->nlh;
969 	struct net *net = sock_net(skb->sk);
970 	unsigned int h, s_h;
971 	unsigned int e = 0, s_e;
972 	struct fib_table *tb;
973 	struct hlist_head *head;
974 	int dumped = 0, err;
975 
976 	if (cb->strict_check) {
977 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
978 		if (err < 0)
979 			return err;
980 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
981 		struct rtmsg *rtm = nlmsg_data(nlh);
982 
983 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
984 	}
985 
986 	/* ipv4 does not use prefix flag */
987 	if (filter.flags & RTM_F_PREFIX)
988 		return skb->len;
989 
990 	if (filter.table_id) {
991 		tb = fib_get_table(net, filter.table_id);
992 		if (!tb) {
993 			if (filter.dump_all_families)
994 				return skb->len;
995 
996 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
997 			return -ENOENT;
998 		}
999 
1000 		rcu_read_lock();
1001 		err = fib_table_dump(tb, skb, cb, &filter);
1002 		rcu_read_unlock();
1003 		return skb->len ? : err;
1004 	}
1005 
1006 	s_h = cb->args[0];
1007 	s_e = cb->args[1];
1008 
1009 	rcu_read_lock();
1010 
1011 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1012 		e = 0;
1013 		head = &net->ipv4.fib_table_hash[h];
1014 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1015 			if (e < s_e)
1016 				goto next;
1017 			if (dumped)
1018 				memset(&cb->args[2], 0, sizeof(cb->args) -
1019 						 2 * sizeof(cb->args[0]));
1020 			err = fib_table_dump(tb, skb, cb, &filter);
1021 			if (err < 0) {
1022 				if (likely(skb->len))
1023 					goto out;
1024 
1025 				goto out_err;
1026 			}
1027 			dumped = 1;
1028 next:
1029 			e++;
1030 		}
1031 	}
1032 out:
1033 	err = skb->len;
1034 out_err:
1035 	rcu_read_unlock();
1036 
1037 	cb->args[1] = e;
1038 	cb->args[0] = h;
1039 
1040 	return err;
1041 }
1042 
1043 /* Prepare and feed intra-kernel routing request.
1044  * Really, it should be netlink message, but :-( netlink
1045  * can be not configured, so that we feed it directly
1046  * to fib engine. It is legal, because all events occur
1047  * only when netlink is already locked.
1048  */
1049 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1050 		      struct in_ifaddr *ifa, u32 rt_priority)
1051 {
1052 	struct net *net = dev_net(ifa->ifa_dev->dev);
1053 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1054 	struct fib_table *tb;
1055 	struct fib_config cfg = {
1056 		.fc_protocol = RTPROT_KERNEL,
1057 		.fc_type = type,
1058 		.fc_dst = dst,
1059 		.fc_dst_len = dst_len,
1060 		.fc_priority = rt_priority,
1061 		.fc_prefsrc = ifa->ifa_local,
1062 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1063 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1064 		.fc_nlinfo = {
1065 			.nl_net = net,
1066 		},
1067 	};
1068 
1069 	if (!tb_id)
1070 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1071 
1072 	tb = fib_new_table(net, tb_id);
1073 	if (!tb)
1074 		return;
1075 
1076 	cfg.fc_table = tb->tb_id;
1077 
1078 	if (type != RTN_LOCAL)
1079 		cfg.fc_scope = RT_SCOPE_LINK;
1080 	else
1081 		cfg.fc_scope = RT_SCOPE_HOST;
1082 
1083 	if (cmd == RTM_NEWROUTE)
1084 		fib_table_insert(net, tb, &cfg, NULL);
1085 	else
1086 		fib_table_delete(net, tb, &cfg, NULL);
1087 }
1088 
1089 void fib_add_ifaddr(struct in_ifaddr *ifa)
1090 {
1091 	struct in_device *in_dev = ifa->ifa_dev;
1092 	struct net_device *dev = in_dev->dev;
1093 	struct in_ifaddr *prim = ifa;
1094 	__be32 mask = ifa->ifa_mask;
1095 	__be32 addr = ifa->ifa_local;
1096 	__be32 prefix = ifa->ifa_address & mask;
1097 
1098 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1099 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1100 		if (!prim) {
1101 			pr_warn("%s: bug: prim == NULL\n", __func__);
1102 			return;
1103 		}
1104 	}
1105 
1106 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1107 
1108 	if (!(dev->flags & IFF_UP))
1109 		return;
1110 
1111 	/* Add broadcast address, if it is explicitly assigned. */
1112 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1113 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1114 			  prim, 0);
1115 
1116 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1117 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1118 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1119 			fib_magic(RTM_NEWROUTE,
1120 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1121 				  prefix, ifa->ifa_prefixlen, prim,
1122 				  ifa->ifa_rt_priority);
1123 
1124 		/* Add network specific broadcasts, when it takes a sense */
1125 		if (ifa->ifa_prefixlen < 31) {
1126 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1127 				  prim, 0);
1128 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1129 				  32, prim, 0);
1130 		}
1131 	}
1132 }
1133 
1134 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1135 {
1136 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1137 	struct in_device *in_dev = ifa->ifa_dev;
1138 	struct net_device *dev = in_dev->dev;
1139 
1140 	if (!(dev->flags & IFF_UP) ||
1141 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1142 	    ipv4_is_zeronet(prefix) ||
1143 	    (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1144 		return;
1145 
1146 	/* add the new */
1147 	fib_magic(RTM_NEWROUTE,
1148 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1149 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1150 
1151 	/* delete the old */
1152 	fib_magic(RTM_DELROUTE,
1153 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1154 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1155 }
1156 
1157 /* Delete primary or secondary address.
1158  * Optionally, on secondary address promotion consider the addresses
1159  * from subnet iprim as deleted, even if they are in device list.
1160  * In this case the secondary ifa can be in device list.
1161  */
1162 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1163 {
1164 	struct in_device *in_dev = ifa->ifa_dev;
1165 	struct net_device *dev = in_dev->dev;
1166 	struct in_ifaddr *ifa1;
1167 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1168 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1169 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1170 #define LOCAL_OK	1
1171 #define BRD_OK		2
1172 #define BRD0_OK		4
1173 #define BRD1_OK		8
1174 	unsigned int ok = 0;
1175 	int subnet = 0;		/* Primary network */
1176 	int gone = 1;		/* Address is missing */
1177 	int same_prefsrc = 0;	/* Another primary with same IP */
1178 
1179 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1180 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1181 		if (!prim) {
1182 			/* if the device has been deleted, we don't perform
1183 			 * address promotion
1184 			 */
1185 			if (!in_dev->dead)
1186 				pr_warn("%s: bug: prim == NULL\n", __func__);
1187 			return;
1188 		}
1189 		if (iprim && iprim != prim) {
1190 			pr_warn("%s: bug: iprim != prim\n", __func__);
1191 			return;
1192 		}
1193 	} else if (!ipv4_is_zeronet(any) &&
1194 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1195 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1196 			fib_magic(RTM_DELROUTE,
1197 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1198 				  any, ifa->ifa_prefixlen, prim, 0);
1199 		subnet = 1;
1200 	}
1201 
1202 	if (in_dev->dead)
1203 		goto no_promotions;
1204 
1205 	/* Deletion is more complicated than add.
1206 	 * We should take care of not to delete too much :-)
1207 	 *
1208 	 * Scan address list to be sure that addresses are really gone.
1209 	 */
1210 	rcu_read_lock();
1211 	in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1212 		if (ifa1 == ifa) {
1213 			/* promotion, keep the IP */
1214 			gone = 0;
1215 			continue;
1216 		}
1217 		/* Ignore IFAs from our subnet */
1218 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1219 		    inet_ifa_match(ifa1->ifa_address, iprim))
1220 			continue;
1221 
1222 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1223 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1224 			/* Another address from our subnet? */
1225 			if (ifa1->ifa_mask == prim->ifa_mask &&
1226 			    inet_ifa_match(ifa1->ifa_address, prim))
1227 				prim1 = prim;
1228 			else {
1229 				/* We reached the secondaries, so
1230 				 * same_prefsrc should be determined.
1231 				 */
1232 				if (!same_prefsrc)
1233 					continue;
1234 				/* Search new prim1 if ifa1 is not
1235 				 * using the current prim1
1236 				 */
1237 				if (!prim1 ||
1238 				    ifa1->ifa_mask != prim1->ifa_mask ||
1239 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1240 					prim1 = inet_ifa_byprefix(in_dev,
1241 							ifa1->ifa_address,
1242 							ifa1->ifa_mask);
1243 				if (!prim1)
1244 					continue;
1245 				if (prim1->ifa_local != prim->ifa_local)
1246 					continue;
1247 			}
1248 		} else {
1249 			if (prim->ifa_local != ifa1->ifa_local)
1250 				continue;
1251 			prim1 = ifa1;
1252 			if (prim != prim1)
1253 				same_prefsrc = 1;
1254 		}
1255 		if (ifa->ifa_local == ifa1->ifa_local)
1256 			ok |= LOCAL_OK;
1257 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1258 			ok |= BRD_OK;
1259 		if (brd == ifa1->ifa_broadcast)
1260 			ok |= BRD1_OK;
1261 		if (any == ifa1->ifa_broadcast)
1262 			ok |= BRD0_OK;
1263 		/* primary has network specific broadcasts */
1264 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1265 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1266 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1267 
1268 			if (!ipv4_is_zeronet(any1)) {
1269 				if (ifa->ifa_broadcast == brd1 ||
1270 				    ifa->ifa_broadcast == any1)
1271 					ok |= BRD_OK;
1272 				if (brd == brd1 || brd == any1)
1273 					ok |= BRD1_OK;
1274 				if (any == brd1 || any == any1)
1275 					ok |= BRD0_OK;
1276 			}
1277 		}
1278 	}
1279 	rcu_read_unlock();
1280 
1281 no_promotions:
1282 	if (!(ok & BRD_OK))
1283 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1284 			  prim, 0);
1285 	if (subnet && ifa->ifa_prefixlen < 31) {
1286 		if (!(ok & BRD1_OK))
1287 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1288 				  prim, 0);
1289 		if (!(ok & BRD0_OK))
1290 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1291 				  prim, 0);
1292 	}
1293 	if (!(ok & LOCAL_OK)) {
1294 		unsigned int addr_type;
1295 
1296 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1297 
1298 		/* Check, that this local address finally disappeared. */
1299 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1300 						     ifa->ifa_local);
1301 		if (gone && addr_type != RTN_LOCAL) {
1302 			/* And the last, but not the least thing.
1303 			 * We must flush stray FIB entries.
1304 			 *
1305 			 * First of all, we scan fib_info list searching
1306 			 * for stray nexthop entries, then ignite fib_flush.
1307 			 */
1308 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1309 				fib_flush(dev_net(dev));
1310 		}
1311 	}
1312 #undef LOCAL_OK
1313 #undef BRD_OK
1314 #undef BRD0_OK
1315 #undef BRD1_OK
1316 }
1317 
1318 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1319 {
1320 
1321 	struct fib_result       res;
1322 	struct flowi4           fl4 = {
1323 		.flowi4_mark = frn->fl_mark,
1324 		.daddr = frn->fl_addr,
1325 		.flowi4_tos = frn->fl_tos,
1326 		.flowi4_scope = frn->fl_scope,
1327 	};
1328 	struct fib_table *tb;
1329 
1330 	rcu_read_lock();
1331 
1332 	tb = fib_get_table(net, frn->tb_id_in);
1333 
1334 	frn->err = -ENOENT;
1335 	if (tb) {
1336 		local_bh_disable();
1337 
1338 		frn->tb_id = tb->tb_id;
1339 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1340 
1341 		if (!frn->err) {
1342 			frn->prefixlen = res.prefixlen;
1343 			frn->nh_sel = res.nh_sel;
1344 			frn->type = res.type;
1345 			frn->scope = res.scope;
1346 		}
1347 		local_bh_enable();
1348 	}
1349 
1350 	rcu_read_unlock();
1351 }
1352 
1353 static void nl_fib_input(struct sk_buff *skb)
1354 {
1355 	struct net *net;
1356 	struct fib_result_nl *frn;
1357 	struct nlmsghdr *nlh;
1358 	u32 portid;
1359 
1360 	net = sock_net(skb->sk);
1361 	nlh = nlmsg_hdr(skb);
1362 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1363 	    skb->len < nlh->nlmsg_len ||
1364 	    nlmsg_len(nlh) < sizeof(*frn))
1365 		return;
1366 
1367 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1368 	if (!skb)
1369 		return;
1370 	nlh = nlmsg_hdr(skb);
1371 
1372 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1373 	nl_fib_lookup(net, frn);
1374 
1375 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1376 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1377 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1378 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1379 }
1380 
1381 static int __net_init nl_fib_lookup_init(struct net *net)
1382 {
1383 	struct sock *sk;
1384 	struct netlink_kernel_cfg cfg = {
1385 		.input	= nl_fib_input,
1386 	};
1387 
1388 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1389 	if (!sk)
1390 		return -EAFNOSUPPORT;
1391 	net->ipv4.fibnl = sk;
1392 	return 0;
1393 }
1394 
1395 static void nl_fib_lookup_exit(struct net *net)
1396 {
1397 	netlink_kernel_release(net->ipv4.fibnl);
1398 	net->ipv4.fibnl = NULL;
1399 }
1400 
1401 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1402 			   bool force)
1403 {
1404 	if (fib_sync_down_dev(dev, event, force))
1405 		fib_flush(dev_net(dev));
1406 	else
1407 		rt_cache_flush(dev_net(dev));
1408 	arp_ifdown(dev);
1409 }
1410 
1411 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1412 {
1413 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1414 	struct net_device *dev = ifa->ifa_dev->dev;
1415 	struct net *net = dev_net(dev);
1416 
1417 	switch (event) {
1418 	case NETDEV_UP:
1419 		fib_add_ifaddr(ifa);
1420 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1421 		fib_sync_up(dev, RTNH_F_DEAD);
1422 #endif
1423 		atomic_inc(&net->ipv4.dev_addr_genid);
1424 		rt_cache_flush(dev_net(dev));
1425 		break;
1426 	case NETDEV_DOWN:
1427 		fib_del_ifaddr(ifa, NULL);
1428 		atomic_inc(&net->ipv4.dev_addr_genid);
1429 		if (!ifa->ifa_dev->ifa_list) {
1430 			/* Last address was deleted from this interface.
1431 			 * Disable IP.
1432 			 */
1433 			fib_disable_ip(dev, event, true);
1434 		} else {
1435 			rt_cache_flush(dev_net(dev));
1436 		}
1437 		break;
1438 	}
1439 	return NOTIFY_DONE;
1440 }
1441 
1442 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1443 {
1444 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1445 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1446 	struct netdev_notifier_info_ext *info_ext = ptr;
1447 	struct in_device *in_dev;
1448 	struct net *net = dev_net(dev);
1449 	struct in_ifaddr *ifa;
1450 	unsigned int flags;
1451 
1452 	if (event == NETDEV_UNREGISTER) {
1453 		fib_disable_ip(dev, event, true);
1454 		rt_flush_dev(dev);
1455 		return NOTIFY_DONE;
1456 	}
1457 
1458 	in_dev = __in_dev_get_rtnl(dev);
1459 	if (!in_dev)
1460 		return NOTIFY_DONE;
1461 
1462 	switch (event) {
1463 	case NETDEV_UP:
1464 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1465 			fib_add_ifaddr(ifa);
1466 		}
1467 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1468 		fib_sync_up(dev, RTNH_F_DEAD);
1469 #endif
1470 		atomic_inc(&net->ipv4.dev_addr_genid);
1471 		rt_cache_flush(net);
1472 		break;
1473 	case NETDEV_DOWN:
1474 		fib_disable_ip(dev, event, false);
1475 		break;
1476 	case NETDEV_CHANGE:
1477 		flags = dev_get_flags(dev);
1478 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1479 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1480 		else
1481 			fib_sync_down_dev(dev, event, false);
1482 		rt_cache_flush(net);
1483 		break;
1484 	case NETDEV_CHANGEMTU:
1485 		fib_sync_mtu(dev, info_ext->ext.mtu);
1486 		rt_cache_flush(net);
1487 		break;
1488 	case NETDEV_CHANGEUPPER:
1489 		upper_info = ptr;
1490 		/* flush all routes if dev is linked to or unlinked from
1491 		 * an L3 master device (e.g., VRF)
1492 		 */
1493 		if (upper_info->upper_dev &&
1494 		    netif_is_l3_master(upper_info->upper_dev))
1495 			fib_disable_ip(dev, NETDEV_DOWN, true);
1496 		break;
1497 	}
1498 	return NOTIFY_DONE;
1499 }
1500 
1501 static struct notifier_block fib_inetaddr_notifier = {
1502 	.notifier_call = fib_inetaddr_event,
1503 };
1504 
1505 static struct notifier_block fib_netdev_notifier = {
1506 	.notifier_call = fib_netdev_event,
1507 };
1508 
1509 static int __net_init ip_fib_net_init(struct net *net)
1510 {
1511 	int err;
1512 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1513 
1514 	err = fib4_notifier_init(net);
1515 	if (err)
1516 		return err;
1517 
1518 	/* Avoid false sharing : Use at least a full cache line */
1519 	size = max_t(size_t, size, L1_CACHE_BYTES);
1520 
1521 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1522 	if (!net->ipv4.fib_table_hash) {
1523 		err = -ENOMEM;
1524 		goto err_table_hash_alloc;
1525 	}
1526 
1527 	err = fib4_rules_init(net);
1528 	if (err < 0)
1529 		goto err_rules_init;
1530 	return 0;
1531 
1532 err_rules_init:
1533 	kfree(net->ipv4.fib_table_hash);
1534 err_table_hash_alloc:
1535 	fib4_notifier_exit(net);
1536 	return err;
1537 }
1538 
1539 static void ip_fib_net_exit(struct net *net)
1540 {
1541 	int i;
1542 
1543 	rtnl_lock();
1544 #ifdef CONFIG_IP_MULTIPLE_TABLES
1545 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1546 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1547 #endif
1548 	/* Destroy the tables in reverse order to guarantee that the
1549 	 * local table, ID 255, is destroyed before the main table, ID
1550 	 * 254. This is necessary as the local table may contain
1551 	 * references to data contained in the main table.
1552 	 */
1553 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1554 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1555 		struct hlist_node *tmp;
1556 		struct fib_table *tb;
1557 
1558 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1559 			hlist_del(&tb->tb_hlist);
1560 			fib_table_flush(net, tb, true);
1561 			fib_free_table(tb);
1562 		}
1563 	}
1564 
1565 #ifdef CONFIG_IP_MULTIPLE_TABLES
1566 	fib4_rules_exit(net);
1567 #endif
1568 	rtnl_unlock();
1569 	kfree(net->ipv4.fib_table_hash);
1570 	fib4_notifier_exit(net);
1571 }
1572 
1573 static int __net_init fib_net_init(struct net *net)
1574 {
1575 	int error;
1576 
1577 #ifdef CONFIG_IP_ROUTE_CLASSID
1578 	net->ipv4.fib_num_tclassid_users = 0;
1579 #endif
1580 	error = ip_fib_net_init(net);
1581 	if (error < 0)
1582 		goto out;
1583 	error = nl_fib_lookup_init(net);
1584 	if (error < 0)
1585 		goto out_nlfl;
1586 	error = fib_proc_init(net);
1587 	if (error < 0)
1588 		goto out_proc;
1589 out:
1590 	return error;
1591 
1592 out_proc:
1593 	nl_fib_lookup_exit(net);
1594 out_nlfl:
1595 	ip_fib_net_exit(net);
1596 	goto out;
1597 }
1598 
1599 static void __net_exit fib_net_exit(struct net *net)
1600 {
1601 	fib_proc_exit(net);
1602 	nl_fib_lookup_exit(net);
1603 	ip_fib_net_exit(net);
1604 }
1605 
1606 static struct pernet_operations fib_net_ops = {
1607 	.init = fib_net_init,
1608 	.exit = fib_net_exit,
1609 };
1610 
1611 void __init ip_fib_init(void)
1612 {
1613 	fib_trie_init();
1614 
1615 	register_pernet_subsys(&fib_net_ops);
1616 
1617 	register_netdevice_notifier(&fib_netdev_notifier);
1618 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1619 
1620 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1621 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1622 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1623 }
1624