xref: /linux/net/ipv4/fib_frontend.c (revision 72bea132f3680ee51e7ed2cee62892b6f5121909)
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/inet_dscp.h>
36 #include <net/ip.h>
37 #include <net/protocol.h>
38 #include <net/route.h>
39 #include <net/tcp.h>
40 #include <net/sock.h>
41 #include <net/arp.h>
42 #include <net/ip_fib.h>
43 #include <net/nexthop.h>
44 #include <net/rtnetlink.h>
45 #include <net/xfrm.h>
46 #include <net/l3mdev.h>
47 #include <net/lwtunnel.h>
48 #include <trace/events/fib.h>
49 
50 #ifndef CONFIG_IP_MULTIPLE_TABLES
51 
52 static int __net_init fib4_rules_init(struct net *net)
53 {
54 	struct fib_table *local_table, *main_table;
55 
56 	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
57 	if (!main_table)
58 		return -ENOMEM;
59 
60 	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
61 	if (!local_table)
62 		goto fail;
63 
64 	hlist_add_head_rcu(&local_table->tb_hlist,
65 				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
66 	hlist_add_head_rcu(&main_table->tb_hlist,
67 				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
68 	return 0;
69 
70 fail:
71 	fib_free_table(main_table);
72 	return -ENOMEM;
73 }
74 #else
75 
76 struct fib_table *fib_new_table(struct net *net, u32 id)
77 {
78 	struct fib_table *tb, *alias = NULL;
79 	unsigned int h;
80 
81 	if (id == 0)
82 		id = RT_TABLE_MAIN;
83 	tb = fib_get_table(net, id);
84 	if (tb)
85 		return tb;
86 
87 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
88 		alias = fib_new_table(net, RT_TABLE_MAIN);
89 
90 	tb = fib_trie_table(id, alias);
91 	if (!tb)
92 		return NULL;
93 
94 	switch (id) {
95 	case RT_TABLE_MAIN:
96 		rcu_assign_pointer(net->ipv4.fib_main, tb);
97 		break;
98 	case RT_TABLE_DEFAULT:
99 		rcu_assign_pointer(net->ipv4.fib_default, tb);
100 		break;
101 	default:
102 		break;
103 	}
104 
105 	h = id & (FIB_TABLE_HASHSZ - 1);
106 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
107 	return tb;
108 }
109 EXPORT_SYMBOL_GPL(fib_new_table);
110 
111 /* caller must hold either rtnl or rcu read lock */
112 struct fib_table *fib_get_table(struct net *net, u32 id)
113 {
114 	struct fib_table *tb;
115 	struct hlist_head *head;
116 	unsigned int h;
117 
118 	if (id == 0)
119 		id = RT_TABLE_MAIN;
120 	h = id & (FIB_TABLE_HASHSZ - 1);
121 
122 	head = &net->ipv4.fib_table_hash[h];
123 	hlist_for_each_entry_rcu(tb, head, tb_hlist,
124 				 lockdep_rtnl_is_held()) {
125 		if (tb->tb_id == id)
126 			return tb;
127 	}
128 	return NULL;
129 }
130 #endif /* CONFIG_IP_MULTIPLE_TABLES */
131 
132 static void fib_replace_table(struct net *net, struct fib_table *old,
133 			      struct fib_table *new)
134 {
135 #ifdef CONFIG_IP_MULTIPLE_TABLES
136 	switch (new->tb_id) {
137 	case RT_TABLE_MAIN:
138 		rcu_assign_pointer(net->ipv4.fib_main, new);
139 		break;
140 	case RT_TABLE_DEFAULT:
141 		rcu_assign_pointer(net->ipv4.fib_default, new);
142 		break;
143 	default:
144 		break;
145 	}
146 
147 #endif
148 	/* replace the old table in the hlist */
149 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
150 }
151 
152 int fib_unmerge(struct net *net)
153 {
154 	struct fib_table *old, *new, *main_table;
155 
156 	/* attempt to fetch local table if it has been allocated */
157 	old = fib_get_table(net, RT_TABLE_LOCAL);
158 	if (!old)
159 		return 0;
160 
161 	new = fib_trie_unmerge(old);
162 	if (!new)
163 		return -ENOMEM;
164 
165 	/* table is already unmerged */
166 	if (new == old)
167 		return 0;
168 
169 	/* replace merged table with clean table */
170 	fib_replace_table(net, old, new);
171 	fib_free_table(old);
172 
173 	/* attempt to fetch main table if it has been allocated */
174 	main_table = fib_get_table(net, RT_TABLE_MAIN);
175 	if (!main_table)
176 		return 0;
177 
178 	/* flush local entries from main table */
179 	fib_table_flush_external(main_table);
180 
181 	return 0;
182 }
183 
184 void fib_flush(struct net *net)
185 {
186 	int flushed = 0;
187 	unsigned int h;
188 
189 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
190 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
191 		struct hlist_node *tmp;
192 		struct fib_table *tb;
193 
194 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
195 			flushed += fib_table_flush(net, tb, false);
196 	}
197 
198 	if (flushed)
199 		rt_cache_flush(net);
200 }
201 
202 /*
203  * Find address type as if only "dev" was present in the system. If
204  * on_dev is NULL then all interfaces are taken into consideration.
205  */
206 static inline unsigned int __inet_dev_addr_type(struct net *net,
207 						const struct net_device *dev,
208 						__be32 addr, u32 tb_id)
209 {
210 	struct flowi4		fl4 = { .daddr = addr };
211 	struct fib_result	res;
212 	unsigned int ret = RTN_BROADCAST;
213 	struct fib_table *table;
214 
215 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
216 		return RTN_BROADCAST;
217 	if (ipv4_is_multicast(addr))
218 		return RTN_MULTICAST;
219 
220 	rcu_read_lock();
221 
222 	table = fib_get_table(net, tb_id);
223 	if (table) {
224 		ret = RTN_UNICAST;
225 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
226 			struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
227 
228 			if (!dev || dev == nhc->nhc_dev)
229 				ret = res.type;
230 		}
231 	}
232 
233 	rcu_read_unlock();
234 	return ret;
235 }
236 
237 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
238 {
239 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
240 }
241 EXPORT_SYMBOL(inet_addr_type_table);
242 
243 unsigned int inet_addr_type(struct net *net, __be32 addr)
244 {
245 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
246 }
247 EXPORT_SYMBOL(inet_addr_type);
248 
249 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
250 				__be32 addr)
251 {
252 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
253 
254 	return __inet_dev_addr_type(net, dev, addr, rt_table);
255 }
256 EXPORT_SYMBOL(inet_dev_addr_type);
257 
258 /* inet_addr_type with dev == NULL but using the table from a dev
259  * if one is associated
260  */
261 unsigned int inet_addr_type_dev_table(struct net *net,
262 				      const struct net_device *dev,
263 				      __be32 addr)
264 {
265 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
266 
267 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
268 }
269 EXPORT_SYMBOL(inet_addr_type_dev_table);
270 
271 __be32 fib_compute_spec_dst(struct sk_buff *skb)
272 {
273 	struct net_device *dev = skb->dev;
274 	struct in_device *in_dev;
275 	struct fib_result res;
276 	struct rtable *rt;
277 	struct net *net;
278 	int scope;
279 
280 	rt = skb_rtable(skb);
281 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
282 	    RTCF_LOCAL)
283 		return ip_hdr(skb)->daddr;
284 
285 	in_dev = __in_dev_get_rcu(dev);
286 
287 	net = dev_net(dev);
288 
289 	scope = RT_SCOPE_UNIVERSE;
290 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
291 		bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
292 		struct flowi4 fl4 = {
293 			.flowi4_iif = LOOPBACK_IFINDEX,
294 			.flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev),
295 			.daddr = ip_hdr(skb)->saddr,
296 			.flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
297 			.flowi4_scope = scope,
298 			.flowi4_mark = vmark ? skb->mark : 0,
299 		};
300 		if (!fib_lookup(net, &fl4, &res, 0))
301 			return fib_result_prefsrc(net, &res);
302 	} else {
303 		scope = RT_SCOPE_LINK;
304 	}
305 
306 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
307 }
308 
309 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
310 {
311 	bool dev_match = false;
312 #ifdef CONFIG_IP_ROUTE_MULTIPATH
313 	if (unlikely(fi->nh)) {
314 		dev_match = nexthop_uses_dev(fi->nh, dev);
315 	} else {
316 		int ret;
317 
318 		for (ret = 0; ret < fib_info_num_path(fi); ret++) {
319 			const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
320 
321 			if (nhc_l3mdev_matches_dev(nhc, dev)) {
322 				dev_match = true;
323 				break;
324 			}
325 		}
326 	}
327 #else
328 	if (fib_info_nhc(fi, 0)->nhc_dev == dev)
329 		dev_match = true;
330 #endif
331 
332 	return dev_match;
333 }
334 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
335 
336 /* Given (packet source, input interface) and optional (dst, oif, tos):
337  * - (main) check, that source is valid i.e. not broadcast or our local
338  *   address.
339  * - figure out what "logical" interface this packet arrived
340  *   and calculate "specific destination" address.
341  * - check, that packet arrived from expected physical interface.
342  * called with rcu_read_lock()
343  */
344 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
345 				 u8 tos, int oif, struct net_device *dev,
346 				 int rpf, struct in_device *idev, u32 *itag)
347 {
348 	struct net *net = dev_net(dev);
349 	struct flow_keys flkeys;
350 	int ret, no_addr;
351 	struct fib_result res;
352 	struct flowi4 fl4;
353 	bool dev_match;
354 
355 	fl4.flowi4_oif = 0;
356 	fl4.flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev);
357 	fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
358 	fl4.daddr = src;
359 	fl4.saddr = dst;
360 	fl4.flowi4_tos = tos;
361 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
362 	fl4.flowi4_tun_key.tun_id = 0;
363 	fl4.flowi4_flags = 0;
364 	fl4.flowi4_uid = sock_net_uid(net, NULL);
365 	fl4.flowi4_multipath_hash = 0;
366 
367 	no_addr = idev->ifa_list == NULL;
368 
369 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
370 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
371 		fl4.flowi4_proto = 0;
372 		fl4.fl4_sport = 0;
373 		fl4.fl4_dport = 0;
374 	} else {
375 		swap(fl4.fl4_sport, fl4.fl4_dport);
376 	}
377 
378 	if (fib_lookup(net, &fl4, &res, 0))
379 		goto last_resort;
380 	if (res.type != RTN_UNICAST &&
381 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
382 		goto e_inval;
383 	fib_combine_itag(itag, &res);
384 
385 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
386 	/* This is not common, loopback packets retain skb_dst so normally they
387 	 * would not even hit this slow path.
388 	 */
389 	dev_match = dev_match || (res.type == RTN_LOCAL &&
390 				  dev == net->loopback_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 		/* Within the same container, it is regarded as a martian source,
440 		 * and the same host but different containers are not.
441 		 */
442 		if (inet_lookup_ifaddr_rcu(net, src))
443 			return -EINVAL;
444 
445 ok:
446 		*itag = 0;
447 		return 0;
448 	}
449 
450 full_check:
451 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
452 }
453 
454 static inline __be32 sk_extract_addr(struct sockaddr *addr)
455 {
456 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
457 }
458 
459 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
460 {
461 	struct nlattr *nla;
462 
463 	nla = (struct nlattr *) ((char *) mx + len);
464 	nla->nla_type = type;
465 	nla->nla_len = nla_attr_size(4);
466 	*(u32 *) nla_data(nla) = value;
467 
468 	return len + nla_total_size(4);
469 }
470 
471 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
472 				 struct fib_config *cfg)
473 {
474 	__be32 addr;
475 	int plen;
476 
477 	memset(cfg, 0, sizeof(*cfg));
478 	cfg->fc_nlinfo.nl_net = net;
479 
480 	if (rt->rt_dst.sa_family != AF_INET)
481 		return -EAFNOSUPPORT;
482 
483 	/*
484 	 * Check mask for validity:
485 	 * a) it must be contiguous.
486 	 * b) destination must have all host bits clear.
487 	 * c) if application forgot to set correct family (AF_INET),
488 	 *    reject request unless it is absolutely clear i.e.
489 	 *    both family and mask are zero.
490 	 */
491 	plen = 32;
492 	addr = sk_extract_addr(&rt->rt_dst);
493 	if (!(rt->rt_flags & RTF_HOST)) {
494 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
495 
496 		if (rt->rt_genmask.sa_family != AF_INET) {
497 			if (mask || rt->rt_genmask.sa_family)
498 				return -EAFNOSUPPORT;
499 		}
500 
501 		if (bad_mask(mask, addr))
502 			return -EINVAL;
503 
504 		plen = inet_mask_len(mask);
505 	}
506 
507 	cfg->fc_dst_len = plen;
508 	cfg->fc_dst = addr;
509 
510 	if (cmd != SIOCDELRT) {
511 		cfg->fc_nlflags = NLM_F_CREATE;
512 		cfg->fc_protocol = RTPROT_BOOT;
513 	}
514 
515 	if (rt->rt_metric)
516 		cfg->fc_priority = rt->rt_metric - 1;
517 
518 	if (rt->rt_flags & RTF_REJECT) {
519 		cfg->fc_scope = RT_SCOPE_HOST;
520 		cfg->fc_type = RTN_UNREACHABLE;
521 		return 0;
522 	}
523 
524 	cfg->fc_scope = RT_SCOPE_NOWHERE;
525 	cfg->fc_type = RTN_UNICAST;
526 
527 	if (rt->rt_dev) {
528 		char *colon;
529 		struct net_device *dev;
530 		char devname[IFNAMSIZ];
531 
532 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
533 			return -EFAULT;
534 
535 		devname[IFNAMSIZ-1] = 0;
536 		colon = strchr(devname, ':');
537 		if (colon)
538 			*colon = 0;
539 		dev = __dev_get_by_name(net, devname);
540 		if (!dev)
541 			return -ENODEV;
542 		cfg->fc_oif = dev->ifindex;
543 		cfg->fc_table = l3mdev_fib_table(dev);
544 		if (colon) {
545 			const struct in_ifaddr *ifa;
546 			struct in_device *in_dev;
547 
548 			in_dev = __in_dev_get_rtnl(dev);
549 			if (!in_dev)
550 				return -ENODEV;
551 
552 			*colon = ':';
553 
554 			rcu_read_lock();
555 			in_dev_for_each_ifa_rcu(ifa, in_dev) {
556 				if (strcmp(ifa->ifa_label, devname) == 0)
557 					break;
558 			}
559 			rcu_read_unlock();
560 
561 			if (!ifa)
562 				return -ENODEV;
563 			cfg->fc_prefsrc = ifa->ifa_local;
564 		}
565 	}
566 
567 	addr = sk_extract_addr(&rt->rt_gateway);
568 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
569 		unsigned int addr_type;
570 
571 		cfg->fc_gw4 = addr;
572 		cfg->fc_gw_family = AF_INET;
573 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
574 		if (rt->rt_flags & RTF_GATEWAY &&
575 		    addr_type == RTN_UNICAST)
576 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
577 	}
578 
579 	if (!cfg->fc_table)
580 		cfg->fc_table = RT_TABLE_MAIN;
581 
582 	if (cmd == SIOCDELRT)
583 		return 0;
584 
585 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
586 		return -EINVAL;
587 
588 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
589 		cfg->fc_scope = RT_SCOPE_LINK;
590 
591 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
592 		struct nlattr *mx;
593 		int len = 0;
594 
595 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
596 		if (!mx)
597 			return -ENOMEM;
598 
599 		if (rt->rt_flags & RTF_MTU)
600 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
601 
602 		if (rt->rt_flags & RTF_WINDOW)
603 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
604 
605 		if (rt->rt_flags & RTF_IRTT)
606 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
607 
608 		cfg->fc_mx = mx;
609 		cfg->fc_mx_len = len;
610 	}
611 
612 	return 0;
613 }
614 
615 /*
616  * Handle IP routing ioctl calls.
617  * These are used to manipulate the routing tables
618  */
619 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
620 {
621 	struct fib_config cfg;
622 	int err;
623 
624 	switch (cmd) {
625 	case SIOCADDRT:		/* Add a route */
626 	case SIOCDELRT:		/* Delete a route */
627 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
628 			return -EPERM;
629 
630 		rtnl_lock();
631 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
632 		if (err == 0) {
633 			struct fib_table *tb;
634 
635 			if (cmd == SIOCDELRT) {
636 				tb = fib_get_table(net, cfg.fc_table);
637 				if (tb)
638 					err = fib_table_delete(net, tb, &cfg,
639 							       NULL);
640 				else
641 					err = -ESRCH;
642 			} else {
643 				tb = fib_new_table(net, cfg.fc_table);
644 				if (tb)
645 					err = fib_table_insert(net, tb,
646 							       &cfg, NULL);
647 				else
648 					err = -ENOBUFS;
649 			}
650 
651 			/* allocated by rtentry_to_fib_config() */
652 			kfree(cfg.fc_mx);
653 		}
654 		rtnl_unlock();
655 		return err;
656 	}
657 	return -EINVAL;
658 }
659 
660 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
661 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
662 	[RTA_DST]		= { .type = NLA_U32 },
663 	[RTA_SRC]		= { .type = NLA_U32 },
664 	[RTA_IIF]		= { .type = NLA_U32 },
665 	[RTA_OIF]		= { .type = NLA_U32 },
666 	[RTA_GATEWAY]		= { .type = NLA_U32 },
667 	[RTA_PRIORITY]		= { .type = NLA_U32 },
668 	[RTA_PREFSRC]		= { .type = NLA_U32 },
669 	[RTA_METRICS]		= { .type = NLA_NESTED },
670 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
671 	[RTA_FLOW]		= { .type = NLA_U32 },
672 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
673 	[RTA_ENCAP]		= { .type = NLA_NESTED },
674 	[RTA_UID]		= { .type = NLA_U32 },
675 	[RTA_MARK]		= { .type = NLA_U32 },
676 	[RTA_TABLE]		= { .type = NLA_U32 },
677 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
678 	[RTA_SPORT]		= { .type = NLA_U16 },
679 	[RTA_DPORT]		= { .type = NLA_U16 },
680 	[RTA_NH_ID]		= { .type = NLA_U32 },
681 };
682 
683 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
684 		    struct netlink_ext_ack *extack)
685 {
686 	struct rtvia *via;
687 	int alen;
688 
689 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
690 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
691 		return -EINVAL;
692 	}
693 
694 	via = nla_data(nla);
695 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
696 
697 	switch (via->rtvia_family) {
698 	case AF_INET:
699 		if (alen != sizeof(__be32)) {
700 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
701 			return -EINVAL;
702 		}
703 		cfg->fc_gw_family = AF_INET;
704 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
705 		break;
706 	case AF_INET6:
707 #if IS_ENABLED(CONFIG_IPV6)
708 		if (alen != sizeof(struct in6_addr)) {
709 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
710 			return -EINVAL;
711 		}
712 		cfg->fc_gw_family = AF_INET6;
713 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
714 #else
715 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
716 		return -EINVAL;
717 #endif
718 		break;
719 	default:
720 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
721 		return -EINVAL;
722 	}
723 
724 	return 0;
725 }
726 
727 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
728 			     struct nlmsghdr *nlh, struct fib_config *cfg,
729 			     struct netlink_ext_ack *extack)
730 {
731 	bool has_gw = false, has_via = false;
732 	struct nlattr *attr;
733 	int err, remaining;
734 	struct rtmsg *rtm;
735 
736 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
737 					rtm_ipv4_policy, extack);
738 	if (err < 0)
739 		goto errout;
740 
741 	memset(cfg, 0, sizeof(*cfg));
742 
743 	rtm = nlmsg_data(nlh);
744 
745 	if (!inet_validate_dscp(rtm->rtm_tos)) {
746 		NL_SET_ERR_MSG(extack,
747 			       "Invalid dsfield (tos): ECN bits must be 0");
748 		err = -EINVAL;
749 		goto errout;
750 	}
751 	cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
752 
753 	cfg->fc_dst_len = rtm->rtm_dst_len;
754 	cfg->fc_table = rtm->rtm_table;
755 	cfg->fc_protocol = rtm->rtm_protocol;
756 	cfg->fc_scope = rtm->rtm_scope;
757 	cfg->fc_type = rtm->rtm_type;
758 	cfg->fc_flags = rtm->rtm_flags;
759 	cfg->fc_nlflags = nlh->nlmsg_flags;
760 
761 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
762 	cfg->fc_nlinfo.nlh = nlh;
763 	cfg->fc_nlinfo.nl_net = net;
764 
765 	if (cfg->fc_type > RTN_MAX) {
766 		NL_SET_ERR_MSG(extack, "Invalid route type");
767 		err = -EINVAL;
768 		goto errout;
769 	}
770 
771 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
772 		switch (nla_type(attr)) {
773 		case RTA_DST:
774 			cfg->fc_dst = nla_get_be32(attr);
775 			break;
776 		case RTA_OIF:
777 			cfg->fc_oif = nla_get_u32(attr);
778 			break;
779 		case RTA_GATEWAY:
780 			has_gw = true;
781 			cfg->fc_gw4 = nla_get_be32(attr);
782 			if (cfg->fc_gw4)
783 				cfg->fc_gw_family = AF_INET;
784 			break;
785 		case RTA_VIA:
786 			has_via = true;
787 			err = fib_gw_from_via(cfg, attr, extack);
788 			if (err)
789 				goto errout;
790 			break;
791 		case RTA_PRIORITY:
792 			cfg->fc_priority = nla_get_u32(attr);
793 			break;
794 		case RTA_PREFSRC:
795 			cfg->fc_prefsrc = nla_get_be32(attr);
796 			break;
797 		case RTA_METRICS:
798 			cfg->fc_mx = nla_data(attr);
799 			cfg->fc_mx_len = nla_len(attr);
800 			break;
801 		case RTA_MULTIPATH:
802 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
803 							     nla_len(attr),
804 							     extack);
805 			if (err < 0)
806 				goto errout;
807 			cfg->fc_mp = nla_data(attr);
808 			cfg->fc_mp_len = nla_len(attr);
809 			break;
810 		case RTA_FLOW:
811 			cfg->fc_flow = nla_get_u32(attr);
812 			break;
813 		case RTA_TABLE:
814 			cfg->fc_table = nla_get_u32(attr);
815 			break;
816 		case RTA_ENCAP:
817 			cfg->fc_encap = attr;
818 			break;
819 		case RTA_ENCAP_TYPE:
820 			cfg->fc_encap_type = nla_get_u16(attr);
821 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
822 							extack);
823 			if (err < 0)
824 				goto errout;
825 			break;
826 		case RTA_NH_ID:
827 			cfg->fc_nh_id = nla_get_u32(attr);
828 			break;
829 		}
830 	}
831 
832 	if (cfg->fc_nh_id) {
833 		if (cfg->fc_oif || cfg->fc_gw_family ||
834 		    cfg->fc_encap || cfg->fc_mp) {
835 			NL_SET_ERR_MSG(extack,
836 				       "Nexthop specification and nexthop id are mutually exclusive");
837 			return -EINVAL;
838 		}
839 	}
840 
841 	if (has_gw && has_via) {
842 		NL_SET_ERR_MSG(extack,
843 			       "Nexthop configuration can not contain both GATEWAY and VIA");
844 		return -EINVAL;
845 	}
846 
847 	if (!cfg->fc_table)
848 		cfg->fc_table = RT_TABLE_MAIN;
849 
850 	return 0;
851 errout:
852 	return err;
853 }
854 
855 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
856 			     struct netlink_ext_ack *extack)
857 {
858 	struct net *net = sock_net(skb->sk);
859 	struct fib_config cfg;
860 	struct fib_table *tb;
861 	int err;
862 
863 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
864 	if (err < 0)
865 		goto errout;
866 
867 	if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
868 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
869 		err = -EINVAL;
870 		goto errout;
871 	}
872 
873 	tb = fib_get_table(net, cfg.fc_table);
874 	if (!tb) {
875 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
876 		err = -ESRCH;
877 		goto errout;
878 	}
879 
880 	err = fib_table_delete(net, tb, &cfg, extack);
881 errout:
882 	return err;
883 }
884 
885 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
886 			     struct netlink_ext_ack *extack)
887 {
888 	struct net *net = sock_net(skb->sk);
889 	struct fib_config cfg;
890 	struct fib_table *tb;
891 	int err;
892 
893 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
894 	if (err < 0)
895 		goto errout;
896 
897 	tb = fib_new_table(net, cfg.fc_table);
898 	if (!tb) {
899 		err = -ENOBUFS;
900 		goto errout;
901 	}
902 
903 	err = fib_table_insert(net, tb, &cfg, extack);
904 	if (!err && cfg.fc_type == RTN_LOCAL)
905 		net->ipv4.fib_has_custom_local_routes = true;
906 errout:
907 	return err;
908 }
909 
910 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
911 			  struct fib_dump_filter *filter,
912 			  struct netlink_callback *cb)
913 {
914 	struct netlink_ext_ack *extack = cb->extack;
915 	struct nlattr *tb[RTA_MAX + 1];
916 	struct rtmsg *rtm;
917 	int err, i;
918 
919 	if (filter->rtnl_held)
920 		ASSERT_RTNL();
921 
922 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
923 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
924 		return -EINVAL;
925 	}
926 
927 	rtm = nlmsg_data(nlh);
928 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
929 	    rtm->rtm_scope) {
930 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
931 		return -EINVAL;
932 	}
933 
934 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
935 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
936 		return -EINVAL;
937 	}
938 	if (rtm->rtm_flags & RTM_F_CLONED)
939 		filter->dump_routes = false;
940 	else
941 		filter->dump_exceptions = false;
942 
943 	filter->flags    = rtm->rtm_flags;
944 	filter->protocol = rtm->rtm_protocol;
945 	filter->rt_type  = rtm->rtm_type;
946 	filter->table_id = rtm->rtm_table;
947 
948 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
949 					    rtm_ipv4_policy, extack);
950 	if (err < 0)
951 		return err;
952 
953 	for (i = 0; i <= RTA_MAX; ++i) {
954 		int ifindex;
955 
956 		if (!tb[i])
957 			continue;
958 
959 		switch (i) {
960 		case RTA_TABLE:
961 			filter->table_id = nla_get_u32(tb[i]);
962 			break;
963 		case RTA_OIF:
964 			ifindex = nla_get_u32(tb[i]);
965 			if (filter->rtnl_held)
966 				filter->dev = __dev_get_by_index(net, ifindex);
967 			else
968 				filter->dev = dev_get_by_index_rcu(net, ifindex);
969 			if (!filter->dev)
970 				return -ENODEV;
971 			break;
972 		default:
973 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
974 			return -EINVAL;
975 		}
976 	}
977 
978 	if (filter->flags || filter->protocol || filter->rt_type ||
979 	    filter->table_id || filter->dev) {
980 		filter->filter_set = 1;
981 		cb->answer_flags = NLM_F_DUMP_FILTERED;
982 	}
983 
984 	return 0;
985 }
986 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
987 
988 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
989 {
990 	struct fib_dump_filter filter = {
991 		.dump_routes = true,
992 		.dump_exceptions = true,
993 		.rtnl_held = false,
994 	};
995 	const struct nlmsghdr *nlh = cb->nlh;
996 	struct net *net = sock_net(skb->sk);
997 	unsigned int h, s_h;
998 	unsigned int e = 0, s_e;
999 	struct fib_table *tb;
1000 	struct hlist_head *head;
1001 	int dumped = 0, err = 0;
1002 
1003 	rcu_read_lock();
1004 	if (cb->strict_check) {
1005 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
1006 		if (err < 0)
1007 			goto unlock;
1008 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
1009 		struct rtmsg *rtm = nlmsg_data(nlh);
1010 
1011 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
1012 	}
1013 
1014 	/* ipv4 does not use prefix flag */
1015 	if (filter.flags & RTM_F_PREFIX)
1016 		goto unlock;
1017 
1018 	if (filter.table_id) {
1019 		tb = fib_get_table(net, filter.table_id);
1020 		if (!tb) {
1021 			if (rtnl_msg_family(cb->nlh) != PF_INET)
1022 				goto unlock;
1023 
1024 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1025 			err = -ENOENT;
1026 			goto unlock;
1027 		}
1028 		err = fib_table_dump(tb, skb, cb, &filter);
1029 		goto unlock;
1030 	}
1031 
1032 	s_h = cb->args[0];
1033 	s_e = cb->args[1];
1034 
1035 	err = 0;
1036 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1037 		e = 0;
1038 		head = &net->ipv4.fib_table_hash[h];
1039 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1040 			if (e < s_e)
1041 				goto next;
1042 			if (dumped)
1043 				memset(&cb->args[2], 0, sizeof(cb->args) -
1044 						 2 * sizeof(cb->args[0]));
1045 			err = fib_table_dump(tb, skb, cb, &filter);
1046 			if (err < 0)
1047 				goto out;
1048 			dumped = 1;
1049 next:
1050 			e++;
1051 		}
1052 	}
1053 out:
1054 
1055 	cb->args[1] = e;
1056 	cb->args[0] = h;
1057 
1058 unlock:
1059 	rcu_read_unlock();
1060 	return err;
1061 }
1062 
1063 /* Prepare and feed intra-kernel routing request.
1064  * Really, it should be netlink message, but :-( netlink
1065  * can be not configured, so that we feed it directly
1066  * to fib engine. It is legal, because all events occur
1067  * only when netlink is already locked.
1068  */
1069 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1070 		      struct in_ifaddr *ifa, u32 rt_priority)
1071 {
1072 	struct net *net = dev_net(ifa->ifa_dev->dev);
1073 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1074 	struct fib_table *tb;
1075 	struct fib_config cfg = {
1076 		.fc_protocol = RTPROT_KERNEL,
1077 		.fc_type = type,
1078 		.fc_dst = dst,
1079 		.fc_dst_len = dst_len,
1080 		.fc_priority = rt_priority,
1081 		.fc_prefsrc = ifa->ifa_local,
1082 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1083 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1084 		.fc_nlinfo = {
1085 			.nl_net = net,
1086 		},
1087 	};
1088 
1089 	if (!tb_id)
1090 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1091 
1092 	tb = fib_new_table(net, tb_id);
1093 	if (!tb)
1094 		return;
1095 
1096 	cfg.fc_table = tb->tb_id;
1097 
1098 	if (type != RTN_LOCAL)
1099 		cfg.fc_scope = RT_SCOPE_LINK;
1100 	else
1101 		cfg.fc_scope = RT_SCOPE_HOST;
1102 
1103 	if (cmd == RTM_NEWROUTE)
1104 		fib_table_insert(net, tb, &cfg, NULL);
1105 	else
1106 		fib_table_delete(net, tb, &cfg, NULL);
1107 }
1108 
1109 void fib_add_ifaddr(struct in_ifaddr *ifa)
1110 {
1111 	struct in_device *in_dev = ifa->ifa_dev;
1112 	struct net_device *dev = in_dev->dev;
1113 	struct in_ifaddr *prim = ifa;
1114 	__be32 mask = ifa->ifa_mask;
1115 	__be32 addr = ifa->ifa_local;
1116 	__be32 prefix = ifa->ifa_address & mask;
1117 
1118 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1119 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1120 		if (!prim) {
1121 			pr_warn("%s: bug: prim == NULL\n", __func__);
1122 			return;
1123 		}
1124 	}
1125 
1126 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1127 
1128 	if (!(dev->flags & IFF_UP))
1129 		return;
1130 
1131 	/* Add broadcast address, if it is explicitly assigned. */
1132 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1133 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1134 			  prim, 0);
1135 		arp_invalidate(dev, ifa->ifa_broadcast, false);
1136 	}
1137 
1138 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1139 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1140 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1141 			fib_magic(RTM_NEWROUTE,
1142 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1143 				  prefix, ifa->ifa_prefixlen, prim,
1144 				  ifa->ifa_rt_priority);
1145 
1146 		/* Add the network broadcast address, when it makes sense */
1147 		if (ifa->ifa_prefixlen < 31) {
1148 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1149 				  32, prim, 0);
1150 			arp_invalidate(dev, prefix | ~mask, false);
1151 		}
1152 	}
1153 }
1154 
1155 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1156 {
1157 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1158 	struct in_device *in_dev = ifa->ifa_dev;
1159 	struct net_device *dev = in_dev->dev;
1160 
1161 	if (!(dev->flags & IFF_UP) ||
1162 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1163 	    ipv4_is_zeronet(prefix) ||
1164 	    (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1165 		return;
1166 
1167 	/* add the new */
1168 	fib_magic(RTM_NEWROUTE,
1169 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1170 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1171 
1172 	/* delete the old */
1173 	fib_magic(RTM_DELROUTE,
1174 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1175 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1176 }
1177 
1178 /* Delete primary or secondary address.
1179  * Optionally, on secondary address promotion consider the addresses
1180  * from subnet iprim as deleted, even if they are in device list.
1181  * In this case the secondary ifa can be in device list.
1182  */
1183 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1184 {
1185 	struct in_device *in_dev = ifa->ifa_dev;
1186 	struct net_device *dev = in_dev->dev;
1187 	struct in_ifaddr *ifa1;
1188 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1189 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1190 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1191 #define LOCAL_OK	1
1192 #define BRD_OK		2
1193 #define BRD0_OK		4
1194 #define BRD1_OK		8
1195 	unsigned int ok = 0;
1196 	int subnet = 0;		/* Primary network */
1197 	int gone = 1;		/* Address is missing */
1198 	int same_prefsrc = 0;	/* Another primary with same IP */
1199 
1200 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1201 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1202 		if (!prim) {
1203 			/* if the device has been deleted, we don't perform
1204 			 * address promotion
1205 			 */
1206 			if (!in_dev->dead)
1207 				pr_warn("%s: bug: prim == NULL\n", __func__);
1208 			return;
1209 		}
1210 		if (iprim && iprim != prim) {
1211 			pr_warn("%s: bug: iprim != prim\n", __func__);
1212 			return;
1213 		}
1214 	} else if (!ipv4_is_zeronet(any) &&
1215 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1216 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1217 			fib_magic(RTM_DELROUTE,
1218 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1219 				  any, ifa->ifa_prefixlen, prim, 0);
1220 		subnet = 1;
1221 	}
1222 
1223 	if (in_dev->dead)
1224 		goto no_promotions;
1225 
1226 	/* Deletion is more complicated than add.
1227 	 * We should take care of not to delete too much :-)
1228 	 *
1229 	 * Scan address list to be sure that addresses are really gone.
1230 	 */
1231 	rcu_read_lock();
1232 	in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1233 		if (ifa1 == ifa) {
1234 			/* promotion, keep the IP */
1235 			gone = 0;
1236 			continue;
1237 		}
1238 		/* Ignore IFAs from our subnet */
1239 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1240 		    inet_ifa_match(ifa1->ifa_address, iprim))
1241 			continue;
1242 
1243 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1244 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1245 			/* Another address from our subnet? */
1246 			if (ifa1->ifa_mask == prim->ifa_mask &&
1247 			    inet_ifa_match(ifa1->ifa_address, prim))
1248 				prim1 = prim;
1249 			else {
1250 				/* We reached the secondaries, so
1251 				 * same_prefsrc should be determined.
1252 				 */
1253 				if (!same_prefsrc)
1254 					continue;
1255 				/* Search new prim1 if ifa1 is not
1256 				 * using the current prim1
1257 				 */
1258 				if (!prim1 ||
1259 				    ifa1->ifa_mask != prim1->ifa_mask ||
1260 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1261 					prim1 = inet_ifa_byprefix(in_dev,
1262 							ifa1->ifa_address,
1263 							ifa1->ifa_mask);
1264 				if (!prim1)
1265 					continue;
1266 				if (prim1->ifa_local != prim->ifa_local)
1267 					continue;
1268 			}
1269 		} else {
1270 			if (prim->ifa_local != ifa1->ifa_local)
1271 				continue;
1272 			prim1 = ifa1;
1273 			if (prim != prim1)
1274 				same_prefsrc = 1;
1275 		}
1276 		if (ifa->ifa_local == ifa1->ifa_local)
1277 			ok |= LOCAL_OK;
1278 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1279 			ok |= BRD_OK;
1280 		if (brd == ifa1->ifa_broadcast)
1281 			ok |= BRD1_OK;
1282 		if (any == ifa1->ifa_broadcast)
1283 			ok |= BRD0_OK;
1284 		/* primary has network specific broadcasts */
1285 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1286 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1287 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1288 
1289 			if (!ipv4_is_zeronet(any1)) {
1290 				if (ifa->ifa_broadcast == brd1 ||
1291 				    ifa->ifa_broadcast == any1)
1292 					ok |= BRD_OK;
1293 				if (brd == brd1 || brd == any1)
1294 					ok |= BRD1_OK;
1295 				if (any == brd1 || any == any1)
1296 					ok |= BRD0_OK;
1297 			}
1298 		}
1299 	}
1300 	rcu_read_unlock();
1301 
1302 no_promotions:
1303 	if (!(ok & BRD_OK))
1304 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1305 			  prim, 0);
1306 	if (subnet && ifa->ifa_prefixlen < 31) {
1307 		if (!(ok & BRD1_OK))
1308 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1309 				  prim, 0);
1310 		if (!(ok & BRD0_OK))
1311 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1312 				  prim, 0);
1313 	}
1314 	if (!(ok & LOCAL_OK)) {
1315 		unsigned int addr_type;
1316 
1317 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1318 
1319 		/* Check, that this local address finally disappeared. */
1320 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1321 						     ifa->ifa_local);
1322 		if (gone && addr_type != RTN_LOCAL) {
1323 			/* And the last, but not the least thing.
1324 			 * We must flush stray FIB entries.
1325 			 *
1326 			 * First of all, we scan fib_info list searching
1327 			 * for stray nexthop entries, then ignite fib_flush.
1328 			 */
1329 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1330 				fib_flush(dev_net(dev));
1331 		}
1332 	}
1333 #undef LOCAL_OK
1334 #undef BRD_OK
1335 #undef BRD0_OK
1336 #undef BRD1_OK
1337 }
1338 
1339 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1340 {
1341 
1342 	struct fib_result       res;
1343 	struct flowi4           fl4 = {
1344 		.flowi4_mark = frn->fl_mark,
1345 		.daddr = frn->fl_addr,
1346 		.flowi4_tos = frn->fl_tos,
1347 		.flowi4_scope = frn->fl_scope,
1348 	};
1349 	struct fib_table *tb;
1350 
1351 	rcu_read_lock();
1352 
1353 	tb = fib_get_table(net, frn->tb_id_in);
1354 
1355 	frn->err = -ENOENT;
1356 	if (tb) {
1357 		local_bh_disable();
1358 
1359 		frn->tb_id = tb->tb_id;
1360 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1361 
1362 		if (!frn->err) {
1363 			frn->prefixlen = res.prefixlen;
1364 			frn->nh_sel = res.nh_sel;
1365 			frn->type = res.type;
1366 			frn->scope = res.scope;
1367 		}
1368 		local_bh_enable();
1369 	}
1370 
1371 	rcu_read_unlock();
1372 }
1373 
1374 static void nl_fib_input(struct sk_buff *skb)
1375 {
1376 	struct net *net;
1377 	struct fib_result_nl *frn;
1378 	struct nlmsghdr *nlh;
1379 	u32 portid;
1380 
1381 	net = sock_net(skb->sk);
1382 	nlh = nlmsg_hdr(skb);
1383 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1384 	    skb->len < nlh->nlmsg_len ||
1385 	    nlmsg_len(nlh) < sizeof(*frn))
1386 		return;
1387 
1388 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1389 	if (!skb)
1390 		return;
1391 	nlh = nlmsg_hdr(skb);
1392 
1393 	frn = nlmsg_data(nlh);
1394 	nl_fib_lookup(net, frn);
1395 
1396 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1397 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1398 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1399 	nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1400 }
1401 
1402 static int __net_init nl_fib_lookup_init(struct net *net)
1403 {
1404 	struct sock *sk;
1405 	struct netlink_kernel_cfg cfg = {
1406 		.input	= nl_fib_input,
1407 	};
1408 
1409 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1410 	if (!sk)
1411 		return -EAFNOSUPPORT;
1412 	net->ipv4.fibnl = sk;
1413 	return 0;
1414 }
1415 
1416 static void nl_fib_lookup_exit(struct net *net)
1417 {
1418 	netlink_kernel_release(net->ipv4.fibnl);
1419 	net->ipv4.fibnl = NULL;
1420 }
1421 
1422 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1423 			   bool force)
1424 {
1425 	if (fib_sync_down_dev(dev, event, force))
1426 		fib_flush(dev_net(dev));
1427 	else
1428 		rt_cache_flush(dev_net(dev));
1429 	arp_ifdown(dev);
1430 }
1431 
1432 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1433 {
1434 	struct in_ifaddr *ifa = ptr;
1435 	struct net_device *dev = ifa->ifa_dev->dev;
1436 	struct net *net = dev_net(dev);
1437 
1438 	switch (event) {
1439 	case NETDEV_UP:
1440 		fib_add_ifaddr(ifa);
1441 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1442 		fib_sync_up(dev, RTNH_F_DEAD);
1443 #endif
1444 		atomic_inc(&net->ipv4.dev_addr_genid);
1445 		rt_cache_flush(dev_net(dev));
1446 		break;
1447 	case NETDEV_DOWN:
1448 		fib_del_ifaddr(ifa, NULL);
1449 		atomic_inc(&net->ipv4.dev_addr_genid);
1450 		if (!ifa->ifa_dev->ifa_list) {
1451 			/* Last address was deleted from this interface.
1452 			 * Disable IP.
1453 			 */
1454 			fib_disable_ip(dev, event, true);
1455 		} else {
1456 			rt_cache_flush(dev_net(dev));
1457 		}
1458 		break;
1459 	}
1460 	return NOTIFY_DONE;
1461 }
1462 
1463 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1464 {
1465 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1466 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1467 	struct netdev_notifier_info_ext *info_ext = ptr;
1468 	struct in_device *in_dev;
1469 	struct net *net = dev_net(dev);
1470 	struct in_ifaddr *ifa;
1471 	unsigned int flags;
1472 
1473 	if (event == NETDEV_UNREGISTER) {
1474 		fib_disable_ip(dev, event, true);
1475 		rt_flush_dev(dev);
1476 		return NOTIFY_DONE;
1477 	}
1478 
1479 	in_dev = __in_dev_get_rtnl(dev);
1480 	if (!in_dev)
1481 		return NOTIFY_DONE;
1482 
1483 	switch (event) {
1484 	case NETDEV_UP:
1485 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1486 			fib_add_ifaddr(ifa);
1487 		}
1488 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1489 		fib_sync_up(dev, RTNH_F_DEAD);
1490 #endif
1491 		atomic_inc(&net->ipv4.dev_addr_genid);
1492 		rt_cache_flush(net);
1493 		break;
1494 	case NETDEV_DOWN:
1495 		fib_disable_ip(dev, event, false);
1496 		break;
1497 	case NETDEV_CHANGE:
1498 		flags = dev_get_flags(dev);
1499 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1500 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1501 		else
1502 			fib_sync_down_dev(dev, event, false);
1503 		rt_cache_flush(net);
1504 		break;
1505 	case NETDEV_CHANGEMTU:
1506 		fib_sync_mtu(dev, info_ext->ext.mtu);
1507 		rt_cache_flush(net);
1508 		break;
1509 	case NETDEV_CHANGEUPPER:
1510 		upper_info = ptr;
1511 		/* flush all routes if dev is linked to or unlinked from
1512 		 * an L3 master device (e.g., VRF)
1513 		 */
1514 		if (upper_info->upper_dev &&
1515 		    netif_is_l3_master(upper_info->upper_dev))
1516 			fib_disable_ip(dev, NETDEV_DOWN, true);
1517 		break;
1518 	}
1519 	return NOTIFY_DONE;
1520 }
1521 
1522 static struct notifier_block fib_inetaddr_notifier = {
1523 	.notifier_call = fib_inetaddr_event,
1524 };
1525 
1526 static struct notifier_block fib_netdev_notifier = {
1527 	.notifier_call = fib_netdev_event,
1528 };
1529 
1530 static int __net_init ip_fib_net_init(struct net *net)
1531 {
1532 	int err;
1533 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1534 
1535 	err = fib4_notifier_init(net);
1536 	if (err)
1537 		return err;
1538 
1539 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1540 	/* Default to 3-tuple */
1541 	net->ipv4.sysctl_fib_multipath_hash_fields =
1542 		FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1543 #endif
1544 
1545 	/* Avoid false sharing : Use at least a full cache line */
1546 	size = max_t(size_t, size, L1_CACHE_BYTES);
1547 
1548 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1549 	if (!net->ipv4.fib_table_hash) {
1550 		err = -ENOMEM;
1551 		goto err_table_hash_alloc;
1552 	}
1553 
1554 	err = fib4_rules_init(net);
1555 	if (err < 0)
1556 		goto err_rules_init;
1557 	return 0;
1558 
1559 err_rules_init:
1560 	kfree(net->ipv4.fib_table_hash);
1561 err_table_hash_alloc:
1562 	fib4_notifier_exit(net);
1563 	return err;
1564 }
1565 
1566 static void ip_fib_net_exit(struct net *net)
1567 {
1568 	int i;
1569 
1570 	ASSERT_RTNL();
1571 #ifdef CONFIG_IP_MULTIPLE_TABLES
1572 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1573 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1574 #endif
1575 	/* Destroy the tables in reverse order to guarantee that the
1576 	 * local table, ID 255, is destroyed before the main table, ID
1577 	 * 254. This is necessary as the local table may contain
1578 	 * references to data contained in the main table.
1579 	 */
1580 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1581 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1582 		struct hlist_node *tmp;
1583 		struct fib_table *tb;
1584 
1585 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1586 			hlist_del(&tb->tb_hlist);
1587 			fib_table_flush(net, tb, true);
1588 			fib_free_table(tb);
1589 		}
1590 	}
1591 
1592 #ifdef CONFIG_IP_MULTIPLE_TABLES
1593 	fib4_rules_exit(net);
1594 #endif
1595 
1596 	kfree(net->ipv4.fib_table_hash);
1597 	fib4_notifier_exit(net);
1598 }
1599 
1600 static int __net_init fib_net_init(struct net *net)
1601 {
1602 	int error;
1603 
1604 #ifdef CONFIG_IP_ROUTE_CLASSID
1605 	atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1606 #endif
1607 	error = ip_fib_net_init(net);
1608 	if (error < 0)
1609 		goto out;
1610 	error = nl_fib_lookup_init(net);
1611 	if (error < 0)
1612 		goto out_nlfl;
1613 	error = fib_proc_init(net);
1614 	if (error < 0)
1615 		goto out_proc;
1616 out:
1617 	return error;
1618 
1619 out_proc:
1620 	nl_fib_lookup_exit(net);
1621 out_nlfl:
1622 	rtnl_lock();
1623 	ip_fib_net_exit(net);
1624 	rtnl_unlock();
1625 	goto out;
1626 }
1627 
1628 static void __net_exit fib_net_exit(struct net *net)
1629 {
1630 	fib_proc_exit(net);
1631 	nl_fib_lookup_exit(net);
1632 }
1633 
1634 static void __net_exit fib_net_exit_batch(struct list_head *net_list)
1635 {
1636 	struct net *net;
1637 
1638 	rtnl_lock();
1639 	list_for_each_entry(net, net_list, exit_list)
1640 		ip_fib_net_exit(net);
1641 
1642 	rtnl_unlock();
1643 }
1644 
1645 static struct pernet_operations fib_net_ops = {
1646 	.init = fib_net_init,
1647 	.exit = fib_net_exit,
1648 	.exit_batch = fib_net_exit_batch,
1649 };
1650 
1651 void __init ip_fib_init(void)
1652 {
1653 	fib_trie_init();
1654 
1655 	register_pernet_subsys(&fib_net_ops);
1656 
1657 	register_netdevice_notifier(&fib_netdev_notifier);
1658 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1659 
1660 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1661 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1662 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib,
1663 		      RTNL_FLAG_DUMP_UNLOCKED);
1664 }
1665