xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision c94b731d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: FIB frontend.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/bitops.h>
15 #include <linux/capability.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/errno.h>
23 #include <linux/in.h>
24 #include <linux/inet.h>
25 #include <linux/inetdevice.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_addr.h>
28 #include <linux/if_arp.h>
29 #include <linux/skbuff.h>
30 #include <linux/cache.h>
31 #include <linux/init.h>
32 #include <linux/list.h>
33 #include <linux/slab.h>
34 
35 #include <net/ip.h>
36 #include <net/protocol.h>
37 #include <net/route.h>
38 #include <net/tcp.h>
39 #include <net/sock.h>
40 #include <net/arp.h>
41 #include <net/ip_fib.h>
42 #include <net/nexthop.h>
43 #include <net/rtnetlink.h>
44 #include <net/xfrm.h>
45 #include <net/l3mdev.h>
46 #include <net/lwtunnel.h>
47 #include <trace/events/fib.h>
48 
49 #ifndef CONFIG_IP_MULTIPLE_TABLES
50 
51 static int __net_init fib4_rules_init(struct net *net)
52 {
53 	struct fib_table *local_table, *main_table;
54 
55 	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
56 	if (!main_table)
57 		return -ENOMEM;
58 
59 	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
60 	if (!local_table)
61 		goto fail;
62 
63 	hlist_add_head_rcu(&local_table->tb_hlist,
64 				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
65 	hlist_add_head_rcu(&main_table->tb_hlist,
66 				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
67 	return 0;
68 
69 fail:
70 	fib_free_table(main_table);
71 	return -ENOMEM;
72 }
73 #else
74 
75 struct fib_table *fib_new_table(struct net *net, u32 id)
76 {
77 	struct fib_table *tb, *alias = NULL;
78 	unsigned int h;
79 
80 	if (id == 0)
81 		id = RT_TABLE_MAIN;
82 	tb = fib_get_table(net, id);
83 	if (tb)
84 		return tb;
85 
86 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
87 		alias = fib_new_table(net, RT_TABLE_MAIN);
88 
89 	tb = fib_trie_table(id, alias);
90 	if (!tb)
91 		return NULL;
92 
93 	switch (id) {
94 	case RT_TABLE_MAIN:
95 		rcu_assign_pointer(net->ipv4.fib_main, tb);
96 		break;
97 	case RT_TABLE_DEFAULT:
98 		rcu_assign_pointer(net->ipv4.fib_default, tb);
99 		break;
100 	default:
101 		break;
102 	}
103 
104 	h = id & (FIB_TABLE_HASHSZ - 1);
105 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
106 	return tb;
107 }
108 EXPORT_SYMBOL_GPL(fib_new_table);
109 
110 /* caller must hold either rtnl or rcu read lock */
111 struct fib_table *fib_get_table(struct net *net, u32 id)
112 {
113 	struct fib_table *tb;
114 	struct hlist_head *head;
115 	unsigned int h;
116 
117 	if (id == 0)
118 		id = RT_TABLE_MAIN;
119 	h = id & (FIB_TABLE_HASHSZ - 1);
120 
121 	head = &net->ipv4.fib_table_hash[h];
122 	hlist_for_each_entry_rcu(tb, head, tb_hlist,
123 				 lockdep_rtnl_is_held()) {
124 		if (tb->tb_id == id)
125 			return tb;
126 	}
127 	return NULL;
128 }
129 #endif /* CONFIG_IP_MULTIPLE_TABLES */
130 
131 static void fib_replace_table(struct net *net, struct fib_table *old,
132 			      struct fib_table *new)
133 {
134 #ifdef CONFIG_IP_MULTIPLE_TABLES
135 	switch (new->tb_id) {
136 	case RT_TABLE_MAIN:
137 		rcu_assign_pointer(net->ipv4.fib_main, new);
138 		break;
139 	case RT_TABLE_DEFAULT:
140 		rcu_assign_pointer(net->ipv4.fib_default, new);
141 		break;
142 	default:
143 		break;
144 	}
145 
146 #endif
147 	/* replace the old table in the hlist */
148 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
149 }
150 
151 int fib_unmerge(struct net *net)
152 {
153 	struct fib_table *old, *new, *main_table;
154 
155 	/* attempt to fetch local table if it has been allocated */
156 	old = fib_get_table(net, RT_TABLE_LOCAL);
157 	if (!old)
158 		return 0;
159 
160 	new = fib_trie_unmerge(old);
161 	if (!new)
162 		return -ENOMEM;
163 
164 	/* table is already unmerged */
165 	if (new == old)
166 		return 0;
167 
168 	/* replace merged table with clean table */
169 	fib_replace_table(net, old, new);
170 	fib_free_table(old);
171 
172 	/* attempt to fetch main table if it has been allocated */
173 	main_table = fib_get_table(net, RT_TABLE_MAIN);
174 	if (!main_table)
175 		return 0;
176 
177 	/* flush local entries from main table */
178 	fib_table_flush_external(main_table);
179 
180 	return 0;
181 }
182 
183 void fib_flush(struct net *net)
184 {
185 	int flushed = 0;
186 	unsigned int h;
187 
188 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
189 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
190 		struct hlist_node *tmp;
191 		struct fib_table *tb;
192 
193 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
194 			flushed += fib_table_flush(net, tb, false);
195 	}
196 
197 	if (flushed)
198 		rt_cache_flush(net);
199 }
200 
201 /*
202  * Find address type as if only "dev" was present in the system. If
203  * on_dev is NULL then all interfaces are taken into consideration.
204  */
205 static inline unsigned int __inet_dev_addr_type(struct net *net,
206 						const struct net_device *dev,
207 						__be32 addr, u32 tb_id)
208 {
209 	struct flowi4		fl4 = { .daddr = addr };
210 	struct fib_result	res;
211 	unsigned int ret = RTN_BROADCAST;
212 	struct fib_table *table;
213 
214 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
215 		return RTN_BROADCAST;
216 	if (ipv4_is_multicast(addr))
217 		return RTN_MULTICAST;
218 
219 	rcu_read_lock();
220 
221 	table = fib_get_table(net, tb_id);
222 	if (table) {
223 		ret = RTN_UNICAST;
224 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
225 			struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
226 
227 			if (!dev || dev == nhc->nhc_dev)
228 				ret = res.type;
229 		}
230 	}
231 
232 	rcu_read_unlock();
233 	return ret;
234 }
235 
236 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
237 {
238 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
239 }
240 EXPORT_SYMBOL(inet_addr_type_table);
241 
242 unsigned int inet_addr_type(struct net *net, __be32 addr)
243 {
244 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
245 }
246 EXPORT_SYMBOL(inet_addr_type);
247 
248 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
249 				__be32 addr)
250 {
251 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
252 
253 	return __inet_dev_addr_type(net, dev, addr, rt_table);
254 }
255 EXPORT_SYMBOL(inet_dev_addr_type);
256 
257 /* inet_addr_type with dev == NULL but using the table from a dev
258  * if one is associated
259  */
260 unsigned int inet_addr_type_dev_table(struct net *net,
261 				      const struct net_device *dev,
262 				      __be32 addr)
263 {
264 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
265 
266 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
267 }
268 EXPORT_SYMBOL(inet_addr_type_dev_table);
269 
270 __be32 fib_compute_spec_dst(struct sk_buff *skb)
271 {
272 	struct net_device *dev = skb->dev;
273 	struct in_device *in_dev;
274 	struct fib_result res;
275 	struct rtable *rt;
276 	struct net *net;
277 	int scope;
278 
279 	rt = skb_rtable(skb);
280 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
281 	    RTCF_LOCAL)
282 		return ip_hdr(skb)->daddr;
283 
284 	in_dev = __in_dev_get_rcu(dev);
285 
286 	net = dev_net(dev);
287 
288 	scope = RT_SCOPE_UNIVERSE;
289 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
290 		bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
291 		struct flowi4 fl4 = {
292 			.flowi4_iif = LOOPBACK_IFINDEX,
293 			.flowi4_oif = l3mdev_master_ifindex_rcu(dev),
294 			.daddr = ip_hdr(skb)->saddr,
295 			.flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
296 			.flowi4_scope = scope,
297 			.flowi4_mark = vmark ? skb->mark : 0,
298 		};
299 		if (!fib_lookup(net, &fl4, &res, 0))
300 			return fib_result_prefsrc(net, &res);
301 	} else {
302 		scope = RT_SCOPE_LINK;
303 	}
304 
305 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
306 }
307 
308 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
309 {
310 	bool dev_match = false;
311 #ifdef CONFIG_IP_ROUTE_MULTIPATH
312 	if (unlikely(fi->nh)) {
313 		dev_match = nexthop_uses_dev(fi->nh, dev);
314 	} else {
315 		int ret;
316 
317 		for (ret = 0; ret < fib_info_num_path(fi); ret++) {
318 			const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
319 
320 			if (nhc_l3mdev_matches_dev(nhc, dev)) {
321 				dev_match = true;
322 				break;
323 			}
324 		}
325 	}
326 #else
327 	if (fib_info_nhc(fi, 0)->nhc_dev == dev)
328 		dev_match = true;
329 #endif
330 
331 	return dev_match;
332 }
333 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
334 
335 /* Given (packet source, input interface) and optional (dst, oif, tos):
336  * - (main) check, that source is valid i.e. not broadcast or our local
337  *   address.
338  * - figure out what "logical" interface this packet arrived
339  *   and calculate "specific destination" address.
340  * - check, that packet arrived from expected physical interface.
341  * called with rcu_read_lock()
342  */
343 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
344 				 u8 tos, int oif, struct net_device *dev,
345 				 int rpf, struct in_device *idev, u32 *itag)
346 {
347 	struct net *net = dev_net(dev);
348 	struct flow_keys flkeys;
349 	int ret, no_addr;
350 	struct fib_result res;
351 	struct flowi4 fl4;
352 	bool dev_match;
353 
354 	fl4.flowi4_oif = 0;
355 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
356 	if (!fl4.flowi4_iif)
357 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
358 	fl4.daddr = src;
359 	fl4.saddr = dst;
360 	fl4.flowi4_tos = tos;
361 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
362 	fl4.flowi4_tun_key.tun_id = 0;
363 	fl4.flowi4_flags = 0;
364 	fl4.flowi4_uid = sock_net_uid(net, NULL);
365 	fl4.flowi4_multipath_hash = 0;
366 
367 	no_addr = idev->ifa_list == NULL;
368 
369 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
370 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
371 		fl4.flowi4_proto = 0;
372 		fl4.fl4_sport = 0;
373 		fl4.fl4_dport = 0;
374 	} 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 (cmd == SIOCDELRT)
580 		return 0;
581 
582 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
583 		return -EINVAL;
584 
585 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
586 		cfg->fc_scope = RT_SCOPE_LINK;
587 
588 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
589 		struct nlattr *mx;
590 		int len = 0;
591 
592 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
593 		if (!mx)
594 			return -ENOMEM;
595 
596 		if (rt->rt_flags & RTF_MTU)
597 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
598 
599 		if (rt->rt_flags & RTF_WINDOW)
600 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
601 
602 		if (rt->rt_flags & RTF_IRTT)
603 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
604 
605 		cfg->fc_mx = mx;
606 		cfg->fc_mx_len = len;
607 	}
608 
609 	return 0;
610 }
611 
612 /*
613  * Handle IP routing ioctl calls.
614  * These are used to manipulate the routing tables
615  */
616 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
617 {
618 	struct fib_config cfg;
619 	int err;
620 
621 	switch (cmd) {
622 	case SIOCADDRT:		/* Add a route */
623 	case SIOCDELRT:		/* Delete a route */
624 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
625 			return -EPERM;
626 
627 		rtnl_lock();
628 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
629 		if (err == 0) {
630 			struct fib_table *tb;
631 
632 			if (cmd == SIOCDELRT) {
633 				tb = fib_get_table(net, cfg.fc_table);
634 				if (tb)
635 					err = fib_table_delete(net, tb, &cfg,
636 							       NULL);
637 				else
638 					err = -ESRCH;
639 			} else {
640 				tb = fib_new_table(net, cfg.fc_table);
641 				if (tb)
642 					err = fib_table_insert(net, tb,
643 							       &cfg, NULL);
644 				else
645 					err = -ENOBUFS;
646 			}
647 
648 			/* allocated by rtentry_to_fib_config() */
649 			kfree(cfg.fc_mx);
650 		}
651 		rtnl_unlock();
652 		return err;
653 	}
654 	return -EINVAL;
655 }
656 
657 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
658 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
659 	[RTA_DST]		= { .type = NLA_U32 },
660 	[RTA_SRC]		= { .type = NLA_U32 },
661 	[RTA_IIF]		= { .type = NLA_U32 },
662 	[RTA_OIF]		= { .type = NLA_U32 },
663 	[RTA_GATEWAY]		= { .type = NLA_U32 },
664 	[RTA_PRIORITY]		= { .type = NLA_U32 },
665 	[RTA_PREFSRC]		= { .type = NLA_U32 },
666 	[RTA_METRICS]		= { .type = NLA_NESTED },
667 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
668 	[RTA_FLOW]		= { .type = NLA_U32 },
669 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
670 	[RTA_ENCAP]		= { .type = NLA_NESTED },
671 	[RTA_UID]		= { .type = NLA_U32 },
672 	[RTA_MARK]		= { .type = NLA_U32 },
673 	[RTA_TABLE]		= { .type = NLA_U32 },
674 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
675 	[RTA_SPORT]		= { .type = NLA_U16 },
676 	[RTA_DPORT]		= { .type = NLA_U16 },
677 	[RTA_NH_ID]		= { .type = NLA_U32 },
678 };
679 
680 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
681 		    struct netlink_ext_ack *extack)
682 {
683 	struct rtvia *via;
684 	int alen;
685 
686 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
687 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
688 		return -EINVAL;
689 	}
690 
691 	via = nla_data(nla);
692 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
693 
694 	switch (via->rtvia_family) {
695 	case AF_INET:
696 		if (alen != sizeof(__be32)) {
697 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
698 			return -EINVAL;
699 		}
700 		cfg->fc_gw_family = AF_INET;
701 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
702 		break;
703 	case AF_INET6:
704 #if IS_ENABLED(CONFIG_IPV6)
705 		if (alen != sizeof(struct in6_addr)) {
706 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
707 			return -EINVAL;
708 		}
709 		cfg->fc_gw_family = AF_INET6;
710 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
711 #else
712 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
713 		return -EINVAL;
714 #endif
715 		break;
716 	default:
717 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
718 		return -EINVAL;
719 	}
720 
721 	return 0;
722 }
723 
724 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
725 			     struct nlmsghdr *nlh, struct fib_config *cfg,
726 			     struct netlink_ext_ack *extack)
727 {
728 	bool has_gw = false, has_via = false;
729 	struct nlattr *attr;
730 	int err, remaining;
731 	struct rtmsg *rtm;
732 
733 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
734 					rtm_ipv4_policy, extack);
735 	if (err < 0)
736 		goto errout;
737 
738 	memset(cfg, 0, sizeof(*cfg));
739 
740 	rtm = nlmsg_data(nlh);
741 	cfg->fc_dst_len = rtm->rtm_dst_len;
742 	cfg->fc_tos = rtm->rtm_tos;
743 	cfg->fc_table = rtm->rtm_table;
744 	cfg->fc_protocol = rtm->rtm_protocol;
745 	cfg->fc_scope = rtm->rtm_scope;
746 	cfg->fc_type = rtm->rtm_type;
747 	cfg->fc_flags = rtm->rtm_flags;
748 	cfg->fc_nlflags = nlh->nlmsg_flags;
749 
750 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
751 	cfg->fc_nlinfo.nlh = nlh;
752 	cfg->fc_nlinfo.nl_net = net;
753 
754 	if (cfg->fc_type > RTN_MAX) {
755 		NL_SET_ERR_MSG(extack, "Invalid route type");
756 		err = -EINVAL;
757 		goto errout;
758 	}
759 
760 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
761 		switch (nla_type(attr)) {
762 		case RTA_DST:
763 			cfg->fc_dst = nla_get_be32(attr);
764 			break;
765 		case RTA_OIF:
766 			cfg->fc_oif = nla_get_u32(attr);
767 			break;
768 		case RTA_GATEWAY:
769 			has_gw = true;
770 			cfg->fc_gw4 = nla_get_be32(attr);
771 			if (cfg->fc_gw4)
772 				cfg->fc_gw_family = AF_INET;
773 			break;
774 		case RTA_VIA:
775 			has_via = true;
776 			err = fib_gw_from_via(cfg, attr, extack);
777 			if (err)
778 				goto errout;
779 			break;
780 		case RTA_PRIORITY:
781 			cfg->fc_priority = nla_get_u32(attr);
782 			break;
783 		case RTA_PREFSRC:
784 			cfg->fc_prefsrc = nla_get_be32(attr);
785 			break;
786 		case RTA_METRICS:
787 			cfg->fc_mx = nla_data(attr);
788 			cfg->fc_mx_len = nla_len(attr);
789 			break;
790 		case RTA_MULTIPATH:
791 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
792 							     nla_len(attr),
793 							     extack);
794 			if (err < 0)
795 				goto errout;
796 			cfg->fc_mp = nla_data(attr);
797 			cfg->fc_mp_len = nla_len(attr);
798 			break;
799 		case RTA_FLOW:
800 			cfg->fc_flow = nla_get_u32(attr);
801 			break;
802 		case RTA_TABLE:
803 			cfg->fc_table = nla_get_u32(attr);
804 			break;
805 		case RTA_ENCAP:
806 			cfg->fc_encap = attr;
807 			break;
808 		case RTA_ENCAP_TYPE:
809 			cfg->fc_encap_type = nla_get_u16(attr);
810 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
811 							extack);
812 			if (err < 0)
813 				goto errout;
814 			break;
815 		case RTA_NH_ID:
816 			cfg->fc_nh_id = nla_get_u32(attr);
817 			break;
818 		}
819 	}
820 
821 	if (cfg->fc_nh_id) {
822 		if (cfg->fc_oif || cfg->fc_gw_family ||
823 		    cfg->fc_encap || cfg->fc_mp) {
824 			NL_SET_ERR_MSG(extack,
825 				       "Nexthop specification and nexthop id are mutually exclusive");
826 			return -EINVAL;
827 		}
828 	}
829 
830 	if (has_gw && has_via) {
831 		NL_SET_ERR_MSG(extack,
832 			       "Nexthop configuration can not contain both GATEWAY and VIA");
833 		return -EINVAL;
834 	}
835 
836 	return 0;
837 errout:
838 	return err;
839 }
840 
841 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
842 			     struct netlink_ext_ack *extack)
843 {
844 	struct net *net = sock_net(skb->sk);
845 	struct fib_config cfg;
846 	struct fib_table *tb;
847 	int err;
848 
849 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
850 	if (err < 0)
851 		goto errout;
852 
853 	if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
854 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
855 		err = -EINVAL;
856 		goto errout;
857 	}
858 
859 	tb = fib_get_table(net, cfg.fc_table);
860 	if (!tb) {
861 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
862 		err = -ESRCH;
863 		goto errout;
864 	}
865 
866 	err = fib_table_delete(net, tb, &cfg, extack);
867 errout:
868 	return err;
869 }
870 
871 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
872 			     struct netlink_ext_ack *extack)
873 {
874 	struct net *net = sock_net(skb->sk);
875 	struct fib_config cfg;
876 	struct fib_table *tb;
877 	int err;
878 
879 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
880 	if (err < 0)
881 		goto errout;
882 
883 	tb = fib_new_table(net, cfg.fc_table);
884 	if (!tb) {
885 		err = -ENOBUFS;
886 		goto errout;
887 	}
888 
889 	err = fib_table_insert(net, tb, &cfg, extack);
890 	if (!err && cfg.fc_type == RTN_LOCAL)
891 		net->ipv4.fib_has_custom_local_routes = true;
892 errout:
893 	return err;
894 }
895 
896 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
897 			  struct fib_dump_filter *filter,
898 			  struct netlink_callback *cb)
899 {
900 	struct netlink_ext_ack *extack = cb->extack;
901 	struct nlattr *tb[RTA_MAX + 1];
902 	struct rtmsg *rtm;
903 	int err, i;
904 
905 	ASSERT_RTNL();
906 
907 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
908 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
909 		return -EINVAL;
910 	}
911 
912 	rtm = nlmsg_data(nlh);
913 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
914 	    rtm->rtm_scope) {
915 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
916 		return -EINVAL;
917 	}
918 
919 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
920 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
921 		return -EINVAL;
922 	}
923 	if (rtm->rtm_flags & RTM_F_CLONED)
924 		filter->dump_routes = false;
925 	else
926 		filter->dump_exceptions = false;
927 
928 	filter->flags    = rtm->rtm_flags;
929 	filter->protocol = rtm->rtm_protocol;
930 	filter->rt_type  = rtm->rtm_type;
931 	filter->table_id = rtm->rtm_table;
932 
933 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
934 					    rtm_ipv4_policy, extack);
935 	if (err < 0)
936 		return err;
937 
938 	for (i = 0; i <= RTA_MAX; ++i) {
939 		int ifindex;
940 
941 		if (!tb[i])
942 			continue;
943 
944 		switch (i) {
945 		case RTA_TABLE:
946 			filter->table_id = nla_get_u32(tb[i]);
947 			break;
948 		case RTA_OIF:
949 			ifindex = nla_get_u32(tb[i]);
950 			filter->dev = __dev_get_by_index(net, ifindex);
951 			if (!filter->dev)
952 				return -ENODEV;
953 			break;
954 		default:
955 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
956 			return -EINVAL;
957 		}
958 	}
959 
960 	if (filter->flags || filter->protocol || filter->rt_type ||
961 	    filter->table_id || filter->dev) {
962 		filter->filter_set = 1;
963 		cb->answer_flags = NLM_F_DUMP_FILTERED;
964 	}
965 
966 	return 0;
967 }
968 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
969 
970 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
971 {
972 	struct fib_dump_filter filter = { .dump_routes = true,
973 					  .dump_exceptions = true };
974 	const struct nlmsghdr *nlh = cb->nlh;
975 	struct net *net = sock_net(skb->sk);
976 	unsigned int h, s_h;
977 	unsigned int e = 0, s_e;
978 	struct fib_table *tb;
979 	struct hlist_head *head;
980 	int dumped = 0, err;
981 
982 	if (cb->strict_check) {
983 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
984 		if (err < 0)
985 			return err;
986 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
987 		struct rtmsg *rtm = nlmsg_data(nlh);
988 
989 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
990 	}
991 
992 	/* ipv4 does not use prefix flag */
993 	if (filter.flags & RTM_F_PREFIX)
994 		return skb->len;
995 
996 	if (filter.table_id) {
997 		tb = fib_get_table(net, filter.table_id);
998 		if (!tb) {
999 			if (rtnl_msg_family(cb->nlh) != PF_INET)
1000 				return skb->len;
1001 
1002 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1003 			return -ENOENT;
1004 		}
1005 
1006 		rcu_read_lock();
1007 		err = fib_table_dump(tb, skb, cb, &filter);
1008 		rcu_read_unlock();
1009 		return skb->len ? : err;
1010 	}
1011 
1012 	s_h = cb->args[0];
1013 	s_e = cb->args[1];
1014 
1015 	rcu_read_lock();
1016 
1017 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1018 		e = 0;
1019 		head = &net->ipv4.fib_table_hash[h];
1020 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1021 			if (e < s_e)
1022 				goto next;
1023 			if (dumped)
1024 				memset(&cb->args[2], 0, sizeof(cb->args) -
1025 						 2 * sizeof(cb->args[0]));
1026 			err = fib_table_dump(tb, skb, cb, &filter);
1027 			if (err < 0) {
1028 				if (likely(skb->len))
1029 					goto out;
1030 
1031 				goto out_err;
1032 			}
1033 			dumped = 1;
1034 next:
1035 			e++;
1036 		}
1037 	}
1038 out:
1039 	err = skb->len;
1040 out_err:
1041 	rcu_read_unlock();
1042 
1043 	cb->args[1] = e;
1044 	cb->args[0] = h;
1045 
1046 	return err;
1047 }
1048 
1049 /* Prepare and feed intra-kernel routing request.
1050  * Really, it should be netlink message, but :-( netlink
1051  * can be not configured, so that we feed it directly
1052  * to fib engine. It is legal, because all events occur
1053  * only when netlink is already locked.
1054  */
1055 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1056 		      struct in_ifaddr *ifa, u32 rt_priority)
1057 {
1058 	struct net *net = dev_net(ifa->ifa_dev->dev);
1059 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1060 	struct fib_table *tb;
1061 	struct fib_config cfg = {
1062 		.fc_protocol = RTPROT_KERNEL,
1063 		.fc_type = type,
1064 		.fc_dst = dst,
1065 		.fc_dst_len = dst_len,
1066 		.fc_priority = rt_priority,
1067 		.fc_prefsrc = ifa->ifa_local,
1068 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1069 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1070 		.fc_nlinfo = {
1071 			.nl_net = net,
1072 		},
1073 	};
1074 
1075 	if (!tb_id)
1076 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1077 
1078 	tb = fib_new_table(net, tb_id);
1079 	if (!tb)
1080 		return;
1081 
1082 	cfg.fc_table = tb->tb_id;
1083 
1084 	if (type != RTN_LOCAL)
1085 		cfg.fc_scope = RT_SCOPE_LINK;
1086 	else
1087 		cfg.fc_scope = RT_SCOPE_HOST;
1088 
1089 	if (cmd == RTM_NEWROUTE)
1090 		fib_table_insert(net, tb, &cfg, NULL);
1091 	else
1092 		fib_table_delete(net, tb, &cfg, NULL);
1093 }
1094 
1095 void fib_add_ifaddr(struct in_ifaddr *ifa)
1096 {
1097 	struct in_device *in_dev = ifa->ifa_dev;
1098 	struct net_device *dev = in_dev->dev;
1099 	struct in_ifaddr *prim = ifa;
1100 	__be32 mask = ifa->ifa_mask;
1101 	__be32 addr = ifa->ifa_local;
1102 	__be32 prefix = ifa->ifa_address & mask;
1103 
1104 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1105 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1106 		if (!prim) {
1107 			pr_warn("%s: bug: prim == NULL\n", __func__);
1108 			return;
1109 		}
1110 	}
1111 
1112 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1113 
1114 	if (!(dev->flags & IFF_UP))
1115 		return;
1116 
1117 	/* Add broadcast address, if it is explicitly assigned. */
1118 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1119 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1120 			  prim, 0);
1121 
1122 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1123 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1124 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1125 			fib_magic(RTM_NEWROUTE,
1126 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1127 				  prefix, ifa->ifa_prefixlen, prim,
1128 				  ifa->ifa_rt_priority);
1129 
1130 		/* Add the network broadcast address, when it makes sense */
1131 		if (ifa->ifa_prefixlen < 31) {
1132 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1133 				  32, prim, 0);
1134 		}
1135 	}
1136 }
1137 
1138 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1139 {
1140 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1141 	struct in_device *in_dev = ifa->ifa_dev;
1142 	struct net_device *dev = in_dev->dev;
1143 
1144 	if (!(dev->flags & IFF_UP) ||
1145 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1146 	    ipv4_is_zeronet(prefix) ||
1147 	    (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1148 		return;
1149 
1150 	/* add the new */
1151 	fib_magic(RTM_NEWROUTE,
1152 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1153 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1154 
1155 	/* delete the old */
1156 	fib_magic(RTM_DELROUTE,
1157 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1158 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1159 }
1160 
1161 /* Delete primary or secondary address.
1162  * Optionally, on secondary address promotion consider the addresses
1163  * from subnet iprim as deleted, even if they are in device list.
1164  * In this case the secondary ifa can be in device list.
1165  */
1166 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1167 {
1168 	struct in_device *in_dev = ifa->ifa_dev;
1169 	struct net_device *dev = in_dev->dev;
1170 	struct in_ifaddr *ifa1;
1171 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1172 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1173 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1174 #define LOCAL_OK	1
1175 #define BRD_OK		2
1176 #define BRD0_OK		4
1177 #define BRD1_OK		8
1178 	unsigned int ok = 0;
1179 	int subnet = 0;		/* Primary network */
1180 	int gone = 1;		/* Address is missing */
1181 	int same_prefsrc = 0;	/* Another primary with same IP */
1182 
1183 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1184 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1185 		if (!prim) {
1186 			/* if the device has been deleted, we don't perform
1187 			 * address promotion
1188 			 */
1189 			if (!in_dev->dead)
1190 				pr_warn("%s: bug: prim == NULL\n", __func__);
1191 			return;
1192 		}
1193 		if (iprim && iprim != prim) {
1194 			pr_warn("%s: bug: iprim != prim\n", __func__);
1195 			return;
1196 		}
1197 	} else if (!ipv4_is_zeronet(any) &&
1198 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1199 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1200 			fib_magic(RTM_DELROUTE,
1201 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1202 				  any, ifa->ifa_prefixlen, prim, 0);
1203 		subnet = 1;
1204 	}
1205 
1206 	if (in_dev->dead)
1207 		goto no_promotions;
1208 
1209 	/* Deletion is more complicated than add.
1210 	 * We should take care of not to delete too much :-)
1211 	 *
1212 	 * Scan address list to be sure that addresses are really gone.
1213 	 */
1214 	rcu_read_lock();
1215 	in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1216 		if (ifa1 == ifa) {
1217 			/* promotion, keep the IP */
1218 			gone = 0;
1219 			continue;
1220 		}
1221 		/* Ignore IFAs from our subnet */
1222 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1223 		    inet_ifa_match(ifa1->ifa_address, iprim))
1224 			continue;
1225 
1226 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1227 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1228 			/* Another address from our subnet? */
1229 			if (ifa1->ifa_mask == prim->ifa_mask &&
1230 			    inet_ifa_match(ifa1->ifa_address, prim))
1231 				prim1 = prim;
1232 			else {
1233 				/* We reached the secondaries, so
1234 				 * same_prefsrc should be determined.
1235 				 */
1236 				if (!same_prefsrc)
1237 					continue;
1238 				/* Search new prim1 if ifa1 is not
1239 				 * using the current prim1
1240 				 */
1241 				if (!prim1 ||
1242 				    ifa1->ifa_mask != prim1->ifa_mask ||
1243 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1244 					prim1 = inet_ifa_byprefix(in_dev,
1245 							ifa1->ifa_address,
1246 							ifa1->ifa_mask);
1247 				if (!prim1)
1248 					continue;
1249 				if (prim1->ifa_local != prim->ifa_local)
1250 					continue;
1251 			}
1252 		} else {
1253 			if (prim->ifa_local != ifa1->ifa_local)
1254 				continue;
1255 			prim1 = ifa1;
1256 			if (prim != prim1)
1257 				same_prefsrc = 1;
1258 		}
1259 		if (ifa->ifa_local == ifa1->ifa_local)
1260 			ok |= LOCAL_OK;
1261 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1262 			ok |= BRD_OK;
1263 		if (brd == ifa1->ifa_broadcast)
1264 			ok |= BRD1_OK;
1265 		if (any == ifa1->ifa_broadcast)
1266 			ok |= BRD0_OK;
1267 		/* primary has network specific broadcasts */
1268 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1269 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1270 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1271 
1272 			if (!ipv4_is_zeronet(any1)) {
1273 				if (ifa->ifa_broadcast == brd1 ||
1274 				    ifa->ifa_broadcast == any1)
1275 					ok |= BRD_OK;
1276 				if (brd == brd1 || brd == any1)
1277 					ok |= BRD1_OK;
1278 				if (any == brd1 || any == any1)
1279 					ok |= BRD0_OK;
1280 			}
1281 		}
1282 	}
1283 	rcu_read_unlock();
1284 
1285 no_promotions:
1286 	if (!(ok & BRD_OK))
1287 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1288 			  prim, 0);
1289 	if (subnet && ifa->ifa_prefixlen < 31) {
1290 		if (!(ok & BRD1_OK))
1291 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1292 				  prim, 0);
1293 		if (!(ok & BRD0_OK))
1294 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1295 				  prim, 0);
1296 	}
1297 	if (!(ok & LOCAL_OK)) {
1298 		unsigned int addr_type;
1299 
1300 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1301 
1302 		/* Check, that this local address finally disappeared. */
1303 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1304 						     ifa->ifa_local);
1305 		if (gone && addr_type != RTN_LOCAL) {
1306 			/* And the last, but not the least thing.
1307 			 * We must flush stray FIB entries.
1308 			 *
1309 			 * First of all, we scan fib_info list searching
1310 			 * for stray nexthop entries, then ignite fib_flush.
1311 			 */
1312 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1313 				fib_flush(dev_net(dev));
1314 		}
1315 	}
1316 #undef LOCAL_OK
1317 #undef BRD_OK
1318 #undef BRD0_OK
1319 #undef BRD1_OK
1320 }
1321 
1322 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1323 {
1324 
1325 	struct fib_result       res;
1326 	struct flowi4           fl4 = {
1327 		.flowi4_mark = frn->fl_mark,
1328 		.daddr = frn->fl_addr,
1329 		.flowi4_tos = frn->fl_tos,
1330 		.flowi4_scope = frn->fl_scope,
1331 	};
1332 	struct fib_table *tb;
1333 
1334 	rcu_read_lock();
1335 
1336 	tb = fib_get_table(net, frn->tb_id_in);
1337 
1338 	frn->err = -ENOENT;
1339 	if (tb) {
1340 		local_bh_disable();
1341 
1342 		frn->tb_id = tb->tb_id;
1343 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1344 
1345 		if (!frn->err) {
1346 			frn->prefixlen = res.prefixlen;
1347 			frn->nh_sel = res.nh_sel;
1348 			frn->type = res.type;
1349 			frn->scope = res.scope;
1350 		}
1351 		local_bh_enable();
1352 	}
1353 
1354 	rcu_read_unlock();
1355 }
1356 
1357 static void nl_fib_input(struct sk_buff *skb)
1358 {
1359 	struct net *net;
1360 	struct fib_result_nl *frn;
1361 	struct nlmsghdr *nlh;
1362 	u32 portid;
1363 
1364 	net = sock_net(skb->sk);
1365 	nlh = nlmsg_hdr(skb);
1366 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1367 	    skb->len < nlh->nlmsg_len ||
1368 	    nlmsg_len(nlh) < sizeof(*frn))
1369 		return;
1370 
1371 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1372 	if (!skb)
1373 		return;
1374 	nlh = nlmsg_hdr(skb);
1375 
1376 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1377 	nl_fib_lookup(net, frn);
1378 
1379 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1380 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1381 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1382 	nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1383 }
1384 
1385 static int __net_init nl_fib_lookup_init(struct net *net)
1386 {
1387 	struct sock *sk;
1388 	struct netlink_kernel_cfg cfg = {
1389 		.input	= nl_fib_input,
1390 	};
1391 
1392 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1393 	if (!sk)
1394 		return -EAFNOSUPPORT;
1395 	net->ipv4.fibnl = sk;
1396 	return 0;
1397 }
1398 
1399 static void nl_fib_lookup_exit(struct net *net)
1400 {
1401 	netlink_kernel_release(net->ipv4.fibnl);
1402 	net->ipv4.fibnl = NULL;
1403 }
1404 
1405 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1406 			   bool force)
1407 {
1408 	if (fib_sync_down_dev(dev, event, force))
1409 		fib_flush(dev_net(dev));
1410 	else
1411 		rt_cache_flush(dev_net(dev));
1412 	arp_ifdown(dev);
1413 }
1414 
1415 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1416 {
1417 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1418 	struct net_device *dev = ifa->ifa_dev->dev;
1419 	struct net *net = dev_net(dev);
1420 
1421 	switch (event) {
1422 	case NETDEV_UP:
1423 		fib_add_ifaddr(ifa);
1424 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1425 		fib_sync_up(dev, RTNH_F_DEAD);
1426 #endif
1427 		atomic_inc(&net->ipv4.dev_addr_genid);
1428 		rt_cache_flush(dev_net(dev));
1429 		break;
1430 	case NETDEV_DOWN:
1431 		fib_del_ifaddr(ifa, NULL);
1432 		atomic_inc(&net->ipv4.dev_addr_genid);
1433 		if (!ifa->ifa_dev->ifa_list) {
1434 			/* Last address was deleted from this interface.
1435 			 * Disable IP.
1436 			 */
1437 			fib_disable_ip(dev, event, true);
1438 		} else {
1439 			rt_cache_flush(dev_net(dev));
1440 		}
1441 		break;
1442 	}
1443 	return NOTIFY_DONE;
1444 }
1445 
1446 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1447 {
1448 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1449 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1450 	struct netdev_notifier_info_ext *info_ext = ptr;
1451 	struct in_device *in_dev;
1452 	struct net *net = dev_net(dev);
1453 	struct in_ifaddr *ifa;
1454 	unsigned int flags;
1455 
1456 	if (event == NETDEV_UNREGISTER) {
1457 		fib_disable_ip(dev, event, true);
1458 		rt_flush_dev(dev);
1459 		return NOTIFY_DONE;
1460 	}
1461 
1462 	in_dev = __in_dev_get_rtnl(dev);
1463 	if (!in_dev)
1464 		return NOTIFY_DONE;
1465 
1466 	switch (event) {
1467 	case NETDEV_UP:
1468 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1469 			fib_add_ifaddr(ifa);
1470 		}
1471 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1472 		fib_sync_up(dev, RTNH_F_DEAD);
1473 #endif
1474 		atomic_inc(&net->ipv4.dev_addr_genid);
1475 		rt_cache_flush(net);
1476 		break;
1477 	case NETDEV_DOWN:
1478 		fib_disable_ip(dev, event, false);
1479 		break;
1480 	case NETDEV_CHANGE:
1481 		flags = dev_get_flags(dev);
1482 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1483 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1484 		else
1485 			fib_sync_down_dev(dev, event, false);
1486 		rt_cache_flush(net);
1487 		break;
1488 	case NETDEV_CHANGEMTU:
1489 		fib_sync_mtu(dev, info_ext->ext.mtu);
1490 		rt_cache_flush(net);
1491 		break;
1492 	case NETDEV_CHANGEUPPER:
1493 		upper_info = ptr;
1494 		/* flush all routes if dev is linked to or unlinked from
1495 		 * an L3 master device (e.g., VRF)
1496 		 */
1497 		if (upper_info->upper_dev &&
1498 		    netif_is_l3_master(upper_info->upper_dev))
1499 			fib_disable_ip(dev, NETDEV_DOWN, true);
1500 		break;
1501 	}
1502 	return NOTIFY_DONE;
1503 }
1504 
1505 static struct notifier_block fib_inetaddr_notifier = {
1506 	.notifier_call = fib_inetaddr_event,
1507 };
1508 
1509 static struct notifier_block fib_netdev_notifier = {
1510 	.notifier_call = fib_netdev_event,
1511 };
1512 
1513 static int __net_init ip_fib_net_init(struct net *net)
1514 {
1515 	int err;
1516 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1517 
1518 	err = fib4_notifier_init(net);
1519 	if (err)
1520 		return err;
1521 
1522 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1523 	/* Default to 3-tuple */
1524 	net->ipv4.sysctl_fib_multipath_hash_fields =
1525 		FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1526 #endif
1527 
1528 	/* Avoid false sharing : Use at least a full cache line */
1529 	size = max_t(size_t, size, L1_CACHE_BYTES);
1530 
1531 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1532 	if (!net->ipv4.fib_table_hash) {
1533 		err = -ENOMEM;
1534 		goto err_table_hash_alloc;
1535 	}
1536 
1537 	err = fib4_rules_init(net);
1538 	if (err < 0)
1539 		goto err_rules_init;
1540 	return 0;
1541 
1542 err_rules_init:
1543 	kfree(net->ipv4.fib_table_hash);
1544 err_table_hash_alloc:
1545 	fib4_notifier_exit(net);
1546 	return err;
1547 }
1548 
1549 static void ip_fib_net_exit(struct net *net)
1550 {
1551 	int i;
1552 
1553 	rtnl_lock();
1554 #ifdef CONFIG_IP_MULTIPLE_TABLES
1555 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1556 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1557 #endif
1558 	/* Destroy the tables in reverse order to guarantee that the
1559 	 * local table, ID 255, is destroyed before the main table, ID
1560 	 * 254. This is necessary as the local table may contain
1561 	 * references to data contained in the main table.
1562 	 */
1563 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1564 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1565 		struct hlist_node *tmp;
1566 		struct fib_table *tb;
1567 
1568 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1569 			hlist_del(&tb->tb_hlist);
1570 			fib_table_flush(net, tb, true);
1571 			fib_free_table(tb);
1572 		}
1573 	}
1574 
1575 #ifdef CONFIG_IP_MULTIPLE_TABLES
1576 	fib4_rules_exit(net);
1577 #endif
1578 	rtnl_unlock();
1579 	kfree(net->ipv4.fib_table_hash);
1580 	fib4_notifier_exit(net);
1581 }
1582 
1583 static int __net_init fib_net_init(struct net *net)
1584 {
1585 	int error;
1586 
1587 #ifdef CONFIG_IP_ROUTE_CLASSID
1588 	atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1589 #endif
1590 	error = ip_fib_net_init(net);
1591 	if (error < 0)
1592 		goto out;
1593 	error = nl_fib_lookup_init(net);
1594 	if (error < 0)
1595 		goto out_nlfl;
1596 	error = fib_proc_init(net);
1597 	if (error < 0)
1598 		goto out_proc;
1599 out:
1600 	return error;
1601 
1602 out_proc:
1603 	nl_fib_lookup_exit(net);
1604 out_nlfl:
1605 	ip_fib_net_exit(net);
1606 	goto out;
1607 }
1608 
1609 static void __net_exit fib_net_exit(struct net *net)
1610 {
1611 	fib_proc_exit(net);
1612 	nl_fib_lookup_exit(net);
1613 	ip_fib_net_exit(net);
1614 }
1615 
1616 static struct pernet_operations fib_net_ops = {
1617 	.init = fib_net_init,
1618 	.exit = fib_net_exit,
1619 };
1620 
1621 void __init ip_fib_init(void)
1622 {
1623 	fib_trie_init();
1624 
1625 	register_pernet_subsys(&fib_net_ops);
1626 
1627 	register_netdevice_notifier(&fib_netdev_notifier);
1628 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1629 
1630 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1631 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1632 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1633 }
1634