xref: /linux/net/ipv4/fib_frontend.c (revision a6cdeeb16bff89c8486324f53577db058cbe81ba)
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 
74 static bool fib4_has_custom_rules(struct net *net)
75 {
76 	return false;
77 }
78 #else
79 
80 struct fib_table *fib_new_table(struct net *net, u32 id)
81 {
82 	struct fib_table *tb, *alias = NULL;
83 	unsigned int h;
84 
85 	if (id == 0)
86 		id = RT_TABLE_MAIN;
87 	tb = fib_get_table(net, id);
88 	if (tb)
89 		return tb;
90 
91 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
92 		alias = fib_new_table(net, RT_TABLE_MAIN);
93 
94 	tb = fib_trie_table(id, alias);
95 	if (!tb)
96 		return NULL;
97 
98 	switch (id) {
99 	case RT_TABLE_MAIN:
100 		rcu_assign_pointer(net->ipv4.fib_main, tb);
101 		break;
102 	case RT_TABLE_DEFAULT:
103 		rcu_assign_pointer(net->ipv4.fib_default, tb);
104 		break;
105 	default:
106 		break;
107 	}
108 
109 	h = id & (FIB_TABLE_HASHSZ - 1);
110 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
111 	return tb;
112 }
113 EXPORT_SYMBOL_GPL(fib_new_table);
114 
115 /* caller must hold either rtnl or rcu read lock */
116 struct fib_table *fib_get_table(struct net *net, u32 id)
117 {
118 	struct fib_table *tb;
119 	struct hlist_head *head;
120 	unsigned int h;
121 
122 	if (id == 0)
123 		id = RT_TABLE_MAIN;
124 	h = id & (FIB_TABLE_HASHSZ - 1);
125 
126 	head = &net->ipv4.fib_table_hash[h];
127 	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
128 		if (tb->tb_id == id)
129 			return tb;
130 	}
131 	return NULL;
132 }
133 
134 static bool fib4_has_custom_rules(struct net *net)
135 {
136 	return net->ipv4.fib_has_custom_rules;
137 }
138 #endif /* CONFIG_IP_MULTIPLE_TABLES */
139 
140 static void fib_replace_table(struct net *net, struct fib_table *old,
141 			      struct fib_table *new)
142 {
143 #ifdef CONFIG_IP_MULTIPLE_TABLES
144 	switch (new->tb_id) {
145 	case RT_TABLE_MAIN:
146 		rcu_assign_pointer(net->ipv4.fib_main, new);
147 		break;
148 	case RT_TABLE_DEFAULT:
149 		rcu_assign_pointer(net->ipv4.fib_default, new);
150 		break;
151 	default:
152 		break;
153 	}
154 
155 #endif
156 	/* replace the old table in the hlist */
157 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
158 }
159 
160 int fib_unmerge(struct net *net)
161 {
162 	struct fib_table *old, *new, *main_table;
163 
164 	/* attempt to fetch local table if it has been allocated */
165 	old = fib_get_table(net, RT_TABLE_LOCAL);
166 	if (!old)
167 		return 0;
168 
169 	new = fib_trie_unmerge(old);
170 	if (!new)
171 		return -ENOMEM;
172 
173 	/* table is already unmerged */
174 	if (new == old)
175 		return 0;
176 
177 	/* replace merged table with clean table */
178 	fib_replace_table(net, old, new);
179 	fib_free_table(old);
180 
181 	/* attempt to fetch main table if it has been allocated */
182 	main_table = fib_get_table(net, RT_TABLE_MAIN);
183 	if (!main_table)
184 		return 0;
185 
186 	/* flush local entries from main table */
187 	fib_table_flush_external(main_table);
188 
189 	return 0;
190 }
191 
192 void fib_flush(struct net *net)
193 {
194 	int flushed = 0;
195 	unsigned int h;
196 
197 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
198 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
199 		struct hlist_node *tmp;
200 		struct fib_table *tb;
201 
202 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
203 			flushed += fib_table_flush(net, tb, false);
204 	}
205 
206 	if (flushed)
207 		rt_cache_flush(net);
208 }
209 
210 /*
211  * Find address type as if only "dev" was present in the system. If
212  * on_dev is NULL then all interfaces are taken into consideration.
213  */
214 static inline unsigned int __inet_dev_addr_type(struct net *net,
215 						const struct net_device *dev,
216 						__be32 addr, u32 tb_id)
217 {
218 	struct flowi4		fl4 = { .daddr = addr };
219 	struct fib_result	res;
220 	unsigned int ret = RTN_BROADCAST;
221 	struct fib_table *table;
222 
223 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
224 		return RTN_BROADCAST;
225 	if (ipv4_is_multicast(addr))
226 		return RTN_MULTICAST;
227 
228 	rcu_read_lock();
229 
230 	table = fib_get_table(net, tb_id);
231 	if (table) {
232 		ret = RTN_UNICAST;
233 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
234 			struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
235 
236 			if (!dev || dev == nhc->nhc_dev)
237 				ret = res.type;
238 		}
239 	}
240 
241 	rcu_read_unlock();
242 	return ret;
243 }
244 
245 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
246 {
247 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
248 }
249 EXPORT_SYMBOL(inet_addr_type_table);
250 
251 unsigned int inet_addr_type(struct net *net, __be32 addr)
252 {
253 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
254 }
255 EXPORT_SYMBOL(inet_addr_type);
256 
257 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
258 				__be32 addr)
259 {
260 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
261 
262 	return __inet_dev_addr_type(net, dev, addr, rt_table);
263 }
264 EXPORT_SYMBOL(inet_dev_addr_type);
265 
266 /* inet_addr_type with dev == NULL but using the table from a dev
267  * if one is associated
268  */
269 unsigned int inet_addr_type_dev_table(struct net *net,
270 				      const struct net_device *dev,
271 				      __be32 addr)
272 {
273 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
274 
275 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
276 }
277 EXPORT_SYMBOL(inet_addr_type_dev_table);
278 
279 __be32 fib_compute_spec_dst(struct sk_buff *skb)
280 {
281 	struct net_device *dev = skb->dev;
282 	struct in_device *in_dev;
283 	struct fib_result res;
284 	struct rtable *rt;
285 	struct net *net;
286 	int scope;
287 
288 	rt = skb_rtable(skb);
289 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
290 	    RTCF_LOCAL)
291 		return ip_hdr(skb)->daddr;
292 
293 	in_dev = __in_dev_get_rcu(dev);
294 
295 	net = dev_net(dev);
296 
297 	scope = RT_SCOPE_UNIVERSE;
298 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
299 		bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
300 		struct flowi4 fl4 = {
301 			.flowi4_iif = LOOPBACK_IFINDEX,
302 			.flowi4_oif = l3mdev_master_ifindex_rcu(dev),
303 			.daddr = ip_hdr(skb)->saddr,
304 			.flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
305 			.flowi4_scope = scope,
306 			.flowi4_mark = vmark ? skb->mark : 0,
307 		};
308 		if (!fib_lookup(net, &fl4, &res, 0))
309 			return fib_result_prefsrc(net, &res);
310 	} else {
311 		scope = RT_SCOPE_LINK;
312 	}
313 
314 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
315 }
316 
317 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
318 {
319 	bool dev_match = false;
320 #ifdef CONFIG_IP_ROUTE_MULTIPATH
321 	int ret;
322 
323 	for (ret = 0; ret < fib_info_num_path(fi); ret++) {
324 		const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
325 
326 		if (nhc->nhc_dev == dev) {
327 			dev_match = true;
328 			break;
329 		} else if (l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex) {
330 			dev_match = true;
331 			break;
332 		}
333 	}
334 #else
335 	if (fib_info_nhc(fi, 0)->nhc_dev == dev)
336 		dev_match = true;
337 #endif
338 
339 	return dev_match;
340 }
341 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
342 
343 /* Given (packet source, input interface) and optional (dst, oif, tos):
344  * - (main) check, that source is valid i.e. not broadcast or our local
345  *   address.
346  * - figure out what "logical" interface this packet arrived
347  *   and calculate "specific destination" address.
348  * - check, that packet arrived from expected physical interface.
349  * called with rcu_read_lock()
350  */
351 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
352 				 u8 tos, int oif, struct net_device *dev,
353 				 int rpf, struct in_device *idev, u32 *itag)
354 {
355 	struct net *net = dev_net(dev);
356 	struct flow_keys flkeys;
357 	int ret, no_addr;
358 	struct fib_result res;
359 	struct flowi4 fl4;
360 	bool dev_match;
361 
362 	fl4.flowi4_oif = 0;
363 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
364 	if (!fl4.flowi4_iif)
365 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
366 	fl4.daddr = src;
367 	fl4.saddr = dst;
368 	fl4.flowi4_tos = tos;
369 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
370 	fl4.flowi4_tun_key.tun_id = 0;
371 	fl4.flowi4_flags = 0;
372 	fl4.flowi4_uid = sock_net_uid(net, NULL);
373 
374 	no_addr = idev->ifa_list == NULL;
375 
376 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
377 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
378 		fl4.flowi4_proto = 0;
379 		fl4.fl4_sport = 0;
380 		fl4.fl4_dport = 0;
381 	}
382 
383 	if (fib_lookup(net, &fl4, &res, 0))
384 		goto last_resort;
385 	if (res.type != RTN_UNICAST &&
386 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
387 		goto e_inval;
388 	fib_combine_itag(itag, &res);
389 
390 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
391 	if (dev_match) {
392 		ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
393 		return ret;
394 	}
395 	if (no_addr)
396 		goto last_resort;
397 	if (rpf == 1)
398 		goto e_rpf;
399 	fl4.flowi4_oif = dev->ifindex;
400 
401 	ret = 0;
402 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
403 		if (res.type == RTN_UNICAST)
404 			ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
405 	}
406 	return ret;
407 
408 last_resort:
409 	if (rpf)
410 		goto e_rpf;
411 	*itag = 0;
412 	return 0;
413 
414 e_inval:
415 	return -EINVAL;
416 e_rpf:
417 	return -EXDEV;
418 }
419 
420 /* Ignore rp_filter for packets protected by IPsec. */
421 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
422 			u8 tos, int oif, struct net_device *dev,
423 			struct in_device *idev, u32 *itag)
424 {
425 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
426 	struct net *net = dev_net(dev);
427 
428 	if (!r && !fib_num_tclassid_users(net) &&
429 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
430 		if (IN_DEV_ACCEPT_LOCAL(idev))
431 			goto ok;
432 		/* with custom local routes in place, checking local addresses
433 		 * only will be too optimistic, with custom rules, checking
434 		 * local addresses only can be too strict, e.g. due to vrf
435 		 */
436 		if (net->ipv4.fib_has_custom_local_routes ||
437 		    fib4_has_custom_rules(net))
438 			goto full_check;
439 		if (inet_lookup_ifaddr_rcu(net, src))
440 			return -EINVAL;
441 
442 ok:
443 		*itag = 0;
444 		return 0;
445 	}
446 
447 full_check:
448 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
449 }
450 
451 static inline __be32 sk_extract_addr(struct sockaddr *addr)
452 {
453 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
454 }
455 
456 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
457 {
458 	struct nlattr *nla;
459 
460 	nla = (struct nlattr *) ((char *) mx + len);
461 	nla->nla_type = type;
462 	nla->nla_len = nla_attr_size(4);
463 	*(u32 *) nla_data(nla) = value;
464 
465 	return len + nla_total_size(4);
466 }
467 
468 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
469 				 struct fib_config *cfg)
470 {
471 	__be32 addr;
472 	int plen;
473 
474 	memset(cfg, 0, sizeof(*cfg));
475 	cfg->fc_nlinfo.nl_net = net;
476 
477 	if (rt->rt_dst.sa_family != AF_INET)
478 		return -EAFNOSUPPORT;
479 
480 	/*
481 	 * Check mask for validity:
482 	 * a) it must be contiguous.
483 	 * b) destination must have all host bits clear.
484 	 * c) if application forgot to set correct family (AF_INET),
485 	 *    reject request unless it is absolutely clear i.e.
486 	 *    both family and mask are zero.
487 	 */
488 	plen = 32;
489 	addr = sk_extract_addr(&rt->rt_dst);
490 	if (!(rt->rt_flags & RTF_HOST)) {
491 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
492 
493 		if (rt->rt_genmask.sa_family != AF_INET) {
494 			if (mask || rt->rt_genmask.sa_family)
495 				return -EAFNOSUPPORT;
496 		}
497 
498 		if (bad_mask(mask, addr))
499 			return -EINVAL;
500 
501 		plen = inet_mask_len(mask);
502 	}
503 
504 	cfg->fc_dst_len = plen;
505 	cfg->fc_dst = addr;
506 
507 	if (cmd != SIOCDELRT) {
508 		cfg->fc_nlflags = NLM_F_CREATE;
509 		cfg->fc_protocol = RTPROT_BOOT;
510 	}
511 
512 	if (rt->rt_metric)
513 		cfg->fc_priority = rt->rt_metric - 1;
514 
515 	if (rt->rt_flags & RTF_REJECT) {
516 		cfg->fc_scope = RT_SCOPE_HOST;
517 		cfg->fc_type = RTN_UNREACHABLE;
518 		return 0;
519 	}
520 
521 	cfg->fc_scope = RT_SCOPE_NOWHERE;
522 	cfg->fc_type = RTN_UNICAST;
523 
524 	if (rt->rt_dev) {
525 		char *colon;
526 		struct net_device *dev;
527 		char devname[IFNAMSIZ];
528 
529 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
530 			return -EFAULT;
531 
532 		devname[IFNAMSIZ-1] = 0;
533 		colon = strchr(devname, ':');
534 		if (colon)
535 			*colon = 0;
536 		dev = __dev_get_by_name(net, devname);
537 		if (!dev)
538 			return -ENODEV;
539 		cfg->fc_oif = dev->ifindex;
540 		cfg->fc_table = l3mdev_fib_table(dev);
541 		if (colon) {
542 			const struct in_ifaddr *ifa;
543 			struct in_device *in_dev;
544 
545 			in_dev = __in_dev_get_rtnl(dev);
546 			if (!in_dev)
547 				return -ENODEV;
548 
549 			*colon = ':';
550 
551 			rcu_read_lock();
552 			in_dev_for_each_ifa_rcu(ifa, in_dev) {
553 				if (strcmp(ifa->ifa_label, devname) == 0)
554 					break;
555 			}
556 			rcu_read_unlock();
557 
558 			if (!ifa)
559 				return -ENODEV;
560 			cfg->fc_prefsrc = ifa->ifa_local;
561 		}
562 	}
563 
564 	addr = sk_extract_addr(&rt->rt_gateway);
565 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
566 		unsigned int addr_type;
567 
568 		cfg->fc_gw4 = addr;
569 		cfg->fc_gw_family = AF_INET;
570 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
571 		if (rt->rt_flags & RTF_GATEWAY &&
572 		    addr_type == RTN_UNICAST)
573 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
574 	}
575 
576 	if (cmd == SIOCDELRT)
577 		return 0;
578 
579 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
580 		return -EINVAL;
581 
582 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
583 		cfg->fc_scope = RT_SCOPE_LINK;
584 
585 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
586 		struct nlattr *mx;
587 		int len = 0;
588 
589 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
590 		if (!mx)
591 			return -ENOMEM;
592 
593 		if (rt->rt_flags & RTF_MTU)
594 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
595 
596 		if (rt->rt_flags & RTF_WINDOW)
597 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
598 
599 		if (rt->rt_flags & RTF_IRTT)
600 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
601 
602 		cfg->fc_mx = mx;
603 		cfg->fc_mx_len = len;
604 	}
605 
606 	return 0;
607 }
608 
609 /*
610  * Handle IP routing ioctl calls.
611  * These are used to manipulate the routing tables
612  */
613 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
614 {
615 	struct fib_config cfg;
616 	int err;
617 
618 	switch (cmd) {
619 	case SIOCADDRT:		/* Add a route */
620 	case SIOCDELRT:		/* Delete a route */
621 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
622 			return -EPERM;
623 
624 		rtnl_lock();
625 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
626 		if (err == 0) {
627 			struct fib_table *tb;
628 
629 			if (cmd == SIOCDELRT) {
630 				tb = fib_get_table(net, cfg.fc_table);
631 				if (tb)
632 					err = fib_table_delete(net, tb, &cfg,
633 							       NULL);
634 				else
635 					err = -ESRCH;
636 			} else {
637 				tb = fib_new_table(net, cfg.fc_table);
638 				if (tb)
639 					err = fib_table_insert(net, tb,
640 							       &cfg, NULL);
641 				else
642 					err = -ENOBUFS;
643 			}
644 
645 			/* allocated by rtentry_to_fib_config() */
646 			kfree(cfg.fc_mx);
647 		}
648 		rtnl_unlock();
649 		return err;
650 	}
651 	return -EINVAL;
652 }
653 
654 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
655 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
656 	[RTA_DST]		= { .type = NLA_U32 },
657 	[RTA_SRC]		= { .type = NLA_U32 },
658 	[RTA_IIF]		= { .type = NLA_U32 },
659 	[RTA_OIF]		= { .type = NLA_U32 },
660 	[RTA_GATEWAY]		= { .type = NLA_U32 },
661 	[RTA_PRIORITY]		= { .type = NLA_U32 },
662 	[RTA_PREFSRC]		= { .type = NLA_U32 },
663 	[RTA_METRICS]		= { .type = NLA_NESTED },
664 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
665 	[RTA_FLOW]		= { .type = NLA_U32 },
666 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
667 	[RTA_ENCAP]		= { .type = NLA_NESTED },
668 	[RTA_UID]		= { .type = NLA_U32 },
669 	[RTA_MARK]		= { .type = NLA_U32 },
670 	[RTA_TABLE]		= { .type = NLA_U32 },
671 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
672 	[RTA_SPORT]		= { .type = NLA_U16 },
673 	[RTA_DPORT]		= { .type = NLA_U16 },
674 };
675 
676 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
677 		    struct netlink_ext_ack *extack)
678 {
679 	struct rtvia *via;
680 	int alen;
681 
682 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
683 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
684 		return -EINVAL;
685 	}
686 
687 	via = nla_data(nla);
688 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
689 
690 	switch (via->rtvia_family) {
691 	case AF_INET:
692 		if (alen != sizeof(__be32)) {
693 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
694 			return -EINVAL;
695 		}
696 		cfg->fc_gw_family = AF_INET;
697 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
698 		break;
699 	case AF_INET6:
700 #ifdef CONFIG_IPV6
701 		if (alen != sizeof(struct in6_addr)) {
702 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
703 			return -EINVAL;
704 		}
705 		cfg->fc_gw_family = AF_INET6;
706 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
707 #else
708 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
709 		return -EINVAL;
710 #endif
711 		break;
712 	default:
713 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
714 		return -EINVAL;
715 	}
716 
717 	return 0;
718 }
719 
720 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
721 			     struct nlmsghdr *nlh, struct fib_config *cfg,
722 			     struct netlink_ext_ack *extack)
723 {
724 	bool has_gw = false, has_via = false;
725 	struct nlattr *attr;
726 	int err, remaining;
727 	struct rtmsg *rtm;
728 
729 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
730 					rtm_ipv4_policy, extack);
731 	if (err < 0)
732 		goto errout;
733 
734 	memset(cfg, 0, sizeof(*cfg));
735 
736 	rtm = nlmsg_data(nlh);
737 	cfg->fc_dst_len = rtm->rtm_dst_len;
738 	cfg->fc_tos = rtm->rtm_tos;
739 	cfg->fc_table = rtm->rtm_table;
740 	cfg->fc_protocol = rtm->rtm_protocol;
741 	cfg->fc_scope = rtm->rtm_scope;
742 	cfg->fc_type = rtm->rtm_type;
743 	cfg->fc_flags = rtm->rtm_flags;
744 	cfg->fc_nlflags = nlh->nlmsg_flags;
745 
746 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
747 	cfg->fc_nlinfo.nlh = nlh;
748 	cfg->fc_nlinfo.nl_net = net;
749 
750 	if (cfg->fc_type > RTN_MAX) {
751 		NL_SET_ERR_MSG(extack, "Invalid route type");
752 		err = -EINVAL;
753 		goto errout;
754 	}
755 
756 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
757 		switch (nla_type(attr)) {
758 		case RTA_DST:
759 			cfg->fc_dst = nla_get_be32(attr);
760 			break;
761 		case RTA_OIF:
762 			cfg->fc_oif = nla_get_u32(attr);
763 			break;
764 		case RTA_GATEWAY:
765 			has_gw = true;
766 			cfg->fc_gw4 = nla_get_be32(attr);
767 			if (cfg->fc_gw4)
768 				cfg->fc_gw_family = AF_INET;
769 			break;
770 		case RTA_VIA:
771 			has_via = true;
772 			err = fib_gw_from_via(cfg, attr, extack);
773 			if (err)
774 				goto errout;
775 			break;
776 		case RTA_PRIORITY:
777 			cfg->fc_priority = nla_get_u32(attr);
778 			break;
779 		case RTA_PREFSRC:
780 			cfg->fc_prefsrc = nla_get_be32(attr);
781 			break;
782 		case RTA_METRICS:
783 			cfg->fc_mx = nla_data(attr);
784 			cfg->fc_mx_len = nla_len(attr);
785 			break;
786 		case RTA_MULTIPATH:
787 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
788 							     nla_len(attr),
789 							     extack);
790 			if (err < 0)
791 				goto errout;
792 			cfg->fc_mp = nla_data(attr);
793 			cfg->fc_mp_len = nla_len(attr);
794 			break;
795 		case RTA_FLOW:
796 			cfg->fc_flow = nla_get_u32(attr);
797 			break;
798 		case RTA_TABLE:
799 			cfg->fc_table = nla_get_u32(attr);
800 			break;
801 		case RTA_ENCAP:
802 			cfg->fc_encap = attr;
803 			break;
804 		case RTA_ENCAP_TYPE:
805 			cfg->fc_encap_type = nla_get_u16(attr);
806 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
807 							extack);
808 			if (err < 0)
809 				goto errout;
810 			break;
811 		}
812 	}
813 
814 	if (has_gw && has_via) {
815 		NL_SET_ERR_MSG(extack,
816 			       "Nexthop configuration can not contain both GATEWAY and VIA");
817 		goto errout;
818 	}
819 
820 	return 0;
821 errout:
822 	return err;
823 }
824 
825 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
826 			     struct netlink_ext_ack *extack)
827 {
828 	struct net *net = sock_net(skb->sk);
829 	struct fib_config cfg;
830 	struct fib_table *tb;
831 	int err;
832 
833 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
834 	if (err < 0)
835 		goto errout;
836 
837 	tb = fib_get_table(net, cfg.fc_table);
838 	if (!tb) {
839 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
840 		err = -ESRCH;
841 		goto errout;
842 	}
843 
844 	err = fib_table_delete(net, tb, &cfg, extack);
845 errout:
846 	return err;
847 }
848 
849 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
850 			     struct netlink_ext_ack *extack)
851 {
852 	struct net *net = sock_net(skb->sk);
853 	struct fib_config cfg;
854 	struct fib_table *tb;
855 	int err;
856 
857 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
858 	if (err < 0)
859 		goto errout;
860 
861 	tb = fib_new_table(net, cfg.fc_table);
862 	if (!tb) {
863 		err = -ENOBUFS;
864 		goto errout;
865 	}
866 
867 	err = fib_table_insert(net, tb, &cfg, extack);
868 	if (!err && cfg.fc_type == RTN_LOCAL)
869 		net->ipv4.fib_has_custom_local_routes = true;
870 errout:
871 	return err;
872 }
873 
874 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
875 			  struct fib_dump_filter *filter,
876 			  struct netlink_callback *cb)
877 {
878 	struct netlink_ext_ack *extack = cb->extack;
879 	struct nlattr *tb[RTA_MAX + 1];
880 	struct rtmsg *rtm;
881 	int err, i;
882 
883 	ASSERT_RTNL();
884 
885 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
886 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
887 		return -EINVAL;
888 	}
889 
890 	rtm = nlmsg_data(nlh);
891 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
892 	    rtm->rtm_scope) {
893 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
894 		return -EINVAL;
895 	}
896 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
897 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
898 		return -EINVAL;
899 	}
900 
901 	filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
902 	filter->flags    = rtm->rtm_flags;
903 	filter->protocol = rtm->rtm_protocol;
904 	filter->rt_type  = rtm->rtm_type;
905 	filter->table_id = rtm->rtm_table;
906 
907 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
908 					    rtm_ipv4_policy, extack);
909 	if (err < 0)
910 		return err;
911 
912 	for (i = 0; i <= RTA_MAX; ++i) {
913 		int ifindex;
914 
915 		if (!tb[i])
916 			continue;
917 
918 		switch (i) {
919 		case RTA_TABLE:
920 			filter->table_id = nla_get_u32(tb[i]);
921 			break;
922 		case RTA_OIF:
923 			ifindex = nla_get_u32(tb[i]);
924 			filter->dev = __dev_get_by_index(net, ifindex);
925 			if (!filter->dev)
926 				return -ENODEV;
927 			break;
928 		default:
929 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
930 			return -EINVAL;
931 		}
932 	}
933 
934 	if (filter->flags || filter->protocol || filter->rt_type ||
935 	    filter->table_id || filter->dev) {
936 		filter->filter_set = 1;
937 		cb->answer_flags = NLM_F_DUMP_FILTERED;
938 	}
939 
940 	return 0;
941 }
942 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
943 
944 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
945 {
946 	const struct nlmsghdr *nlh = cb->nlh;
947 	struct net *net = sock_net(skb->sk);
948 	struct fib_dump_filter filter = {};
949 	unsigned int h, s_h;
950 	unsigned int e = 0, s_e;
951 	struct fib_table *tb;
952 	struct hlist_head *head;
953 	int dumped = 0, err;
954 
955 	if (cb->strict_check) {
956 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
957 		if (err < 0)
958 			return err;
959 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
960 		struct rtmsg *rtm = nlmsg_data(nlh);
961 
962 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
963 	}
964 
965 	/* fib entries are never clones and ipv4 does not use prefix flag */
966 	if (filter.flags & (RTM_F_PREFIX | RTM_F_CLONED))
967 		return skb->len;
968 
969 	if (filter.table_id) {
970 		tb = fib_get_table(net, filter.table_id);
971 		if (!tb) {
972 			if (filter.dump_all_families)
973 				return skb->len;
974 
975 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
976 			return -ENOENT;
977 		}
978 
979 		err = fib_table_dump(tb, skb, cb, &filter);
980 		return skb->len ? : err;
981 	}
982 
983 	s_h = cb->args[0];
984 	s_e = cb->args[1];
985 
986 	rcu_read_lock();
987 
988 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
989 		e = 0;
990 		head = &net->ipv4.fib_table_hash[h];
991 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
992 			if (e < s_e)
993 				goto next;
994 			if (dumped)
995 				memset(&cb->args[2], 0, sizeof(cb->args) -
996 						 2 * sizeof(cb->args[0]));
997 			err = fib_table_dump(tb, skb, cb, &filter);
998 			if (err < 0) {
999 				if (likely(skb->len))
1000 					goto out;
1001 
1002 				goto out_err;
1003 			}
1004 			dumped = 1;
1005 next:
1006 			e++;
1007 		}
1008 	}
1009 out:
1010 	err = skb->len;
1011 out_err:
1012 	rcu_read_unlock();
1013 
1014 	cb->args[1] = e;
1015 	cb->args[0] = h;
1016 
1017 	return err;
1018 }
1019 
1020 /* Prepare and feed intra-kernel routing request.
1021  * Really, it should be netlink message, but :-( netlink
1022  * can be not configured, so that we feed it directly
1023  * to fib engine. It is legal, because all events occur
1024  * only when netlink is already locked.
1025  */
1026 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1027 		      struct in_ifaddr *ifa, u32 rt_priority)
1028 {
1029 	struct net *net = dev_net(ifa->ifa_dev->dev);
1030 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1031 	struct fib_table *tb;
1032 	struct fib_config cfg = {
1033 		.fc_protocol = RTPROT_KERNEL,
1034 		.fc_type = type,
1035 		.fc_dst = dst,
1036 		.fc_dst_len = dst_len,
1037 		.fc_priority = rt_priority,
1038 		.fc_prefsrc = ifa->ifa_local,
1039 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1040 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1041 		.fc_nlinfo = {
1042 			.nl_net = net,
1043 		},
1044 	};
1045 
1046 	if (!tb_id)
1047 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1048 
1049 	tb = fib_new_table(net, tb_id);
1050 	if (!tb)
1051 		return;
1052 
1053 	cfg.fc_table = tb->tb_id;
1054 
1055 	if (type != RTN_LOCAL)
1056 		cfg.fc_scope = RT_SCOPE_LINK;
1057 	else
1058 		cfg.fc_scope = RT_SCOPE_HOST;
1059 
1060 	if (cmd == RTM_NEWROUTE)
1061 		fib_table_insert(net, tb, &cfg, NULL);
1062 	else
1063 		fib_table_delete(net, tb, &cfg, NULL);
1064 }
1065 
1066 void fib_add_ifaddr(struct in_ifaddr *ifa)
1067 {
1068 	struct in_device *in_dev = ifa->ifa_dev;
1069 	struct net_device *dev = in_dev->dev;
1070 	struct in_ifaddr *prim = ifa;
1071 	__be32 mask = ifa->ifa_mask;
1072 	__be32 addr = ifa->ifa_local;
1073 	__be32 prefix = ifa->ifa_address & mask;
1074 
1075 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1076 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1077 		if (!prim) {
1078 			pr_warn("%s: bug: prim == NULL\n", __func__);
1079 			return;
1080 		}
1081 	}
1082 
1083 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1084 
1085 	if (!(dev->flags & IFF_UP))
1086 		return;
1087 
1088 	/* Add broadcast address, if it is explicitly assigned. */
1089 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1090 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1091 			  prim, 0);
1092 
1093 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1094 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1095 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1096 			fib_magic(RTM_NEWROUTE,
1097 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1098 				  prefix, ifa->ifa_prefixlen, prim,
1099 				  ifa->ifa_rt_priority);
1100 
1101 		/* Add network specific broadcasts, when it takes a sense */
1102 		if (ifa->ifa_prefixlen < 31) {
1103 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1104 				  prim, 0);
1105 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1106 				  32, prim, 0);
1107 		}
1108 	}
1109 }
1110 
1111 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1112 {
1113 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1114 	struct in_device *in_dev = ifa->ifa_dev;
1115 	struct net_device *dev = in_dev->dev;
1116 
1117 	if (!(dev->flags & IFF_UP) ||
1118 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1119 	    ipv4_is_zeronet(prefix) ||
1120 	    prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1121 		return;
1122 
1123 	/* add the new */
1124 	fib_magic(RTM_NEWROUTE,
1125 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1126 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1127 
1128 	/* delete the old */
1129 	fib_magic(RTM_DELROUTE,
1130 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1131 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1132 }
1133 
1134 /* Delete primary or secondary address.
1135  * Optionally, on secondary address promotion consider the addresses
1136  * from subnet iprim as deleted, even if they are in device list.
1137  * In this case the secondary ifa can be in device list.
1138  */
1139 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1140 {
1141 	struct in_device *in_dev = ifa->ifa_dev;
1142 	struct net_device *dev = in_dev->dev;
1143 	struct in_ifaddr *ifa1;
1144 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1145 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1146 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1147 #define LOCAL_OK	1
1148 #define BRD_OK		2
1149 #define BRD0_OK		4
1150 #define BRD1_OK		8
1151 	unsigned int ok = 0;
1152 	int subnet = 0;		/* Primary network */
1153 	int gone = 1;		/* Address is missing */
1154 	int same_prefsrc = 0;	/* Another primary with same IP */
1155 
1156 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1157 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1158 		if (!prim) {
1159 			/* if the device has been deleted, we don't perform
1160 			 * address promotion
1161 			 */
1162 			if (!in_dev->dead)
1163 				pr_warn("%s: bug: prim == NULL\n", __func__);
1164 			return;
1165 		}
1166 		if (iprim && iprim != prim) {
1167 			pr_warn("%s: bug: iprim != prim\n", __func__);
1168 			return;
1169 		}
1170 	} else if (!ipv4_is_zeronet(any) &&
1171 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1172 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1173 			fib_magic(RTM_DELROUTE,
1174 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1175 				  any, ifa->ifa_prefixlen, prim, 0);
1176 		subnet = 1;
1177 	}
1178 
1179 	if (in_dev->dead)
1180 		goto no_promotions;
1181 
1182 	/* Deletion is more complicated than add.
1183 	 * We should take care of not to delete too much :-)
1184 	 *
1185 	 * Scan address list to be sure that addresses are really gone.
1186 	 */
1187 	rcu_read_lock();
1188 	in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1189 		if (ifa1 == ifa) {
1190 			/* promotion, keep the IP */
1191 			gone = 0;
1192 			continue;
1193 		}
1194 		/* Ignore IFAs from our subnet */
1195 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1196 		    inet_ifa_match(ifa1->ifa_address, iprim))
1197 			continue;
1198 
1199 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1200 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1201 			/* Another address from our subnet? */
1202 			if (ifa1->ifa_mask == prim->ifa_mask &&
1203 			    inet_ifa_match(ifa1->ifa_address, prim))
1204 				prim1 = prim;
1205 			else {
1206 				/* We reached the secondaries, so
1207 				 * same_prefsrc should be determined.
1208 				 */
1209 				if (!same_prefsrc)
1210 					continue;
1211 				/* Search new prim1 if ifa1 is not
1212 				 * using the current prim1
1213 				 */
1214 				if (!prim1 ||
1215 				    ifa1->ifa_mask != prim1->ifa_mask ||
1216 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1217 					prim1 = inet_ifa_byprefix(in_dev,
1218 							ifa1->ifa_address,
1219 							ifa1->ifa_mask);
1220 				if (!prim1)
1221 					continue;
1222 				if (prim1->ifa_local != prim->ifa_local)
1223 					continue;
1224 			}
1225 		} else {
1226 			if (prim->ifa_local != ifa1->ifa_local)
1227 				continue;
1228 			prim1 = ifa1;
1229 			if (prim != prim1)
1230 				same_prefsrc = 1;
1231 		}
1232 		if (ifa->ifa_local == ifa1->ifa_local)
1233 			ok |= LOCAL_OK;
1234 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1235 			ok |= BRD_OK;
1236 		if (brd == ifa1->ifa_broadcast)
1237 			ok |= BRD1_OK;
1238 		if (any == ifa1->ifa_broadcast)
1239 			ok |= BRD0_OK;
1240 		/* primary has network specific broadcasts */
1241 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1242 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1243 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1244 
1245 			if (!ipv4_is_zeronet(any1)) {
1246 				if (ifa->ifa_broadcast == brd1 ||
1247 				    ifa->ifa_broadcast == any1)
1248 					ok |= BRD_OK;
1249 				if (brd == brd1 || brd == any1)
1250 					ok |= BRD1_OK;
1251 				if (any == brd1 || any == any1)
1252 					ok |= BRD0_OK;
1253 			}
1254 		}
1255 	}
1256 	rcu_read_unlock();
1257 
1258 no_promotions:
1259 	if (!(ok & BRD_OK))
1260 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1261 			  prim, 0);
1262 	if (subnet && ifa->ifa_prefixlen < 31) {
1263 		if (!(ok & BRD1_OK))
1264 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1265 				  prim, 0);
1266 		if (!(ok & BRD0_OK))
1267 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1268 				  prim, 0);
1269 	}
1270 	if (!(ok & LOCAL_OK)) {
1271 		unsigned int addr_type;
1272 
1273 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1274 
1275 		/* Check, that this local address finally disappeared. */
1276 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1277 						     ifa->ifa_local);
1278 		if (gone && addr_type != RTN_LOCAL) {
1279 			/* And the last, but not the least thing.
1280 			 * We must flush stray FIB entries.
1281 			 *
1282 			 * First of all, we scan fib_info list searching
1283 			 * for stray nexthop entries, then ignite fib_flush.
1284 			 */
1285 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1286 				fib_flush(dev_net(dev));
1287 		}
1288 	}
1289 #undef LOCAL_OK
1290 #undef BRD_OK
1291 #undef BRD0_OK
1292 #undef BRD1_OK
1293 }
1294 
1295 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1296 {
1297 
1298 	struct fib_result       res;
1299 	struct flowi4           fl4 = {
1300 		.flowi4_mark = frn->fl_mark,
1301 		.daddr = frn->fl_addr,
1302 		.flowi4_tos = frn->fl_tos,
1303 		.flowi4_scope = frn->fl_scope,
1304 	};
1305 	struct fib_table *tb;
1306 
1307 	rcu_read_lock();
1308 
1309 	tb = fib_get_table(net, frn->tb_id_in);
1310 
1311 	frn->err = -ENOENT;
1312 	if (tb) {
1313 		local_bh_disable();
1314 
1315 		frn->tb_id = tb->tb_id;
1316 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1317 
1318 		if (!frn->err) {
1319 			frn->prefixlen = res.prefixlen;
1320 			frn->nh_sel = res.nh_sel;
1321 			frn->type = res.type;
1322 			frn->scope = res.scope;
1323 		}
1324 		local_bh_enable();
1325 	}
1326 
1327 	rcu_read_unlock();
1328 }
1329 
1330 static void nl_fib_input(struct sk_buff *skb)
1331 {
1332 	struct net *net;
1333 	struct fib_result_nl *frn;
1334 	struct nlmsghdr *nlh;
1335 	u32 portid;
1336 
1337 	net = sock_net(skb->sk);
1338 	nlh = nlmsg_hdr(skb);
1339 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1340 	    skb->len < nlh->nlmsg_len ||
1341 	    nlmsg_len(nlh) < sizeof(*frn))
1342 		return;
1343 
1344 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1345 	if (!skb)
1346 		return;
1347 	nlh = nlmsg_hdr(skb);
1348 
1349 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1350 	nl_fib_lookup(net, frn);
1351 
1352 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1353 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1354 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1355 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1356 }
1357 
1358 static int __net_init nl_fib_lookup_init(struct net *net)
1359 {
1360 	struct sock *sk;
1361 	struct netlink_kernel_cfg cfg = {
1362 		.input	= nl_fib_input,
1363 	};
1364 
1365 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1366 	if (!sk)
1367 		return -EAFNOSUPPORT;
1368 	net->ipv4.fibnl = sk;
1369 	return 0;
1370 }
1371 
1372 static void nl_fib_lookup_exit(struct net *net)
1373 {
1374 	netlink_kernel_release(net->ipv4.fibnl);
1375 	net->ipv4.fibnl = NULL;
1376 }
1377 
1378 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1379 			   bool force)
1380 {
1381 	if (fib_sync_down_dev(dev, event, force))
1382 		fib_flush(dev_net(dev));
1383 	else
1384 		rt_cache_flush(dev_net(dev));
1385 	arp_ifdown(dev);
1386 }
1387 
1388 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1389 {
1390 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1391 	struct net_device *dev = ifa->ifa_dev->dev;
1392 	struct net *net = dev_net(dev);
1393 
1394 	switch (event) {
1395 	case NETDEV_UP:
1396 		fib_add_ifaddr(ifa);
1397 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1398 		fib_sync_up(dev, RTNH_F_DEAD);
1399 #endif
1400 		atomic_inc(&net->ipv4.dev_addr_genid);
1401 		rt_cache_flush(dev_net(dev));
1402 		break;
1403 	case NETDEV_DOWN:
1404 		fib_del_ifaddr(ifa, NULL);
1405 		atomic_inc(&net->ipv4.dev_addr_genid);
1406 		if (!ifa->ifa_dev->ifa_list) {
1407 			/* Last address was deleted from this interface.
1408 			 * Disable IP.
1409 			 */
1410 			fib_disable_ip(dev, event, true);
1411 		} else {
1412 			rt_cache_flush(dev_net(dev));
1413 		}
1414 		break;
1415 	}
1416 	return NOTIFY_DONE;
1417 }
1418 
1419 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1420 {
1421 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1422 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1423 	struct netdev_notifier_info_ext *info_ext = ptr;
1424 	struct in_device *in_dev;
1425 	struct net *net = dev_net(dev);
1426 	struct in_ifaddr *ifa;
1427 	unsigned int flags;
1428 
1429 	if (event == NETDEV_UNREGISTER) {
1430 		fib_disable_ip(dev, event, true);
1431 		rt_flush_dev(dev);
1432 		return NOTIFY_DONE;
1433 	}
1434 
1435 	in_dev = __in_dev_get_rtnl(dev);
1436 	if (!in_dev)
1437 		return NOTIFY_DONE;
1438 
1439 	switch (event) {
1440 	case NETDEV_UP:
1441 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1442 			fib_add_ifaddr(ifa);
1443 		}
1444 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1445 		fib_sync_up(dev, RTNH_F_DEAD);
1446 #endif
1447 		atomic_inc(&net->ipv4.dev_addr_genid);
1448 		rt_cache_flush(net);
1449 		break;
1450 	case NETDEV_DOWN:
1451 		fib_disable_ip(dev, event, false);
1452 		break;
1453 	case NETDEV_CHANGE:
1454 		flags = dev_get_flags(dev);
1455 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1456 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1457 		else
1458 			fib_sync_down_dev(dev, event, false);
1459 		rt_cache_flush(net);
1460 		break;
1461 	case NETDEV_CHANGEMTU:
1462 		fib_sync_mtu(dev, info_ext->ext.mtu);
1463 		rt_cache_flush(net);
1464 		break;
1465 	case NETDEV_CHANGEUPPER:
1466 		upper_info = ptr;
1467 		/* flush all routes if dev is linked to or unlinked from
1468 		 * an L3 master device (e.g., VRF)
1469 		 */
1470 		if (upper_info->upper_dev &&
1471 		    netif_is_l3_master(upper_info->upper_dev))
1472 			fib_disable_ip(dev, NETDEV_DOWN, true);
1473 		break;
1474 	}
1475 	return NOTIFY_DONE;
1476 }
1477 
1478 static struct notifier_block fib_inetaddr_notifier = {
1479 	.notifier_call = fib_inetaddr_event,
1480 };
1481 
1482 static struct notifier_block fib_netdev_notifier = {
1483 	.notifier_call = fib_netdev_event,
1484 };
1485 
1486 static int __net_init ip_fib_net_init(struct net *net)
1487 {
1488 	int err;
1489 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1490 
1491 	err = fib4_notifier_init(net);
1492 	if (err)
1493 		return err;
1494 
1495 	/* Avoid false sharing : Use at least a full cache line */
1496 	size = max_t(size_t, size, L1_CACHE_BYTES);
1497 
1498 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1499 	if (!net->ipv4.fib_table_hash) {
1500 		err = -ENOMEM;
1501 		goto err_table_hash_alloc;
1502 	}
1503 
1504 	err = fib4_rules_init(net);
1505 	if (err < 0)
1506 		goto err_rules_init;
1507 	return 0;
1508 
1509 err_rules_init:
1510 	kfree(net->ipv4.fib_table_hash);
1511 err_table_hash_alloc:
1512 	fib4_notifier_exit(net);
1513 	return err;
1514 }
1515 
1516 static void ip_fib_net_exit(struct net *net)
1517 {
1518 	int i;
1519 
1520 	rtnl_lock();
1521 #ifdef CONFIG_IP_MULTIPLE_TABLES
1522 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1523 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1524 #endif
1525 	/* Destroy the tables in reverse order to guarantee that the
1526 	 * local table, ID 255, is destroyed before the main table, ID
1527 	 * 254. This is necessary as the local table may contain
1528 	 * references to data contained in the main table.
1529 	 */
1530 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1531 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1532 		struct hlist_node *tmp;
1533 		struct fib_table *tb;
1534 
1535 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1536 			hlist_del(&tb->tb_hlist);
1537 			fib_table_flush(net, tb, true);
1538 			fib_free_table(tb);
1539 		}
1540 	}
1541 
1542 #ifdef CONFIG_IP_MULTIPLE_TABLES
1543 	fib4_rules_exit(net);
1544 #endif
1545 	rtnl_unlock();
1546 	kfree(net->ipv4.fib_table_hash);
1547 	fib4_notifier_exit(net);
1548 }
1549 
1550 static int __net_init fib_net_init(struct net *net)
1551 {
1552 	int error;
1553 
1554 #ifdef CONFIG_IP_ROUTE_CLASSID
1555 	net->ipv4.fib_num_tclassid_users = 0;
1556 #endif
1557 	error = ip_fib_net_init(net);
1558 	if (error < 0)
1559 		goto out;
1560 	error = nl_fib_lookup_init(net);
1561 	if (error < 0)
1562 		goto out_nlfl;
1563 	error = fib_proc_init(net);
1564 	if (error < 0)
1565 		goto out_proc;
1566 out:
1567 	return error;
1568 
1569 out_proc:
1570 	nl_fib_lookup_exit(net);
1571 out_nlfl:
1572 	ip_fib_net_exit(net);
1573 	goto out;
1574 }
1575 
1576 static void __net_exit fib_net_exit(struct net *net)
1577 {
1578 	fib_proc_exit(net);
1579 	nl_fib_lookup_exit(net);
1580 	ip_fib_net_exit(net);
1581 }
1582 
1583 static struct pernet_operations fib_net_ops = {
1584 	.init = fib_net_init,
1585 	.exit = fib_net_exit,
1586 };
1587 
1588 void __init ip_fib_init(void)
1589 {
1590 	fib_trie_init();
1591 
1592 	register_pernet_subsys(&fib_net_ops);
1593 
1594 	register_netdevice_notifier(&fib_netdev_notifier);
1595 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1596 
1597 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1598 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1599 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1600 }
1601