xref: /linux/net/ipv4/fib_frontend.c (revision 0a9f4a20c6153d187c8ee58133357ac671372f5f)
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_oif = 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_iif = l3mdev_master_ifindex_rcu(dev);
357 	if (!fl4.flowi4_iif)
358 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
359 	fl4.daddr = src;
360 	fl4.saddr = dst;
361 	fl4.flowi4_tos = tos;
362 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
363 	fl4.flowi4_tun_key.tun_id = 0;
364 	fl4.flowi4_flags = 0;
365 	fl4.flowi4_uid = sock_net_uid(net, NULL);
366 	fl4.flowi4_multipath_hash = 0;
367 
368 	no_addr = idev->ifa_list == NULL;
369 
370 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
371 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
372 		fl4.flowi4_proto = 0;
373 		fl4.fl4_sport = 0;
374 		fl4.fl4_dport = 0;
375 	} else {
376 		swap(fl4.fl4_sport, fl4.fl4_dport);
377 	}
378 
379 	if (fib_lookup(net, &fl4, &res, 0))
380 		goto last_resort;
381 	if (res.type != RTN_UNICAST &&
382 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
383 		goto e_inval;
384 	fib_combine_itag(itag, &res);
385 
386 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
387 	/* This is not common, loopback packets retain skb_dst so normally they
388 	 * would not even hit this slow path.
389 	 */
390 	dev_match = dev_match || (res.type == RTN_LOCAL &&
391 				  dev == net->loopback_dev);
392 	if (dev_match) {
393 		ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
394 		return ret;
395 	}
396 	if (no_addr)
397 		goto last_resort;
398 	if (rpf == 1)
399 		goto e_rpf;
400 	fl4.flowi4_oif = dev->ifindex;
401 
402 	ret = 0;
403 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
404 		if (res.type == RTN_UNICAST)
405 			ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
406 	}
407 	return ret;
408 
409 last_resort:
410 	if (rpf)
411 		goto e_rpf;
412 	*itag = 0;
413 	return 0;
414 
415 e_inval:
416 	return -EINVAL;
417 e_rpf:
418 	return -EXDEV;
419 }
420 
421 /* Ignore rp_filter for packets protected by IPsec. */
422 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
423 			u8 tos, int oif, struct net_device *dev,
424 			struct in_device *idev, u32 *itag)
425 {
426 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
427 	struct net *net = dev_net(dev);
428 
429 	if (!r && !fib_num_tclassid_users(net) &&
430 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
431 		if (IN_DEV_ACCEPT_LOCAL(idev))
432 			goto ok;
433 		/* with custom local routes in place, checking local addresses
434 		 * only will be too optimistic, with custom rules, checking
435 		 * local addresses only can be too strict, e.g. due to vrf
436 		 */
437 		if (net->ipv4.fib_has_custom_local_routes ||
438 		    fib4_has_custom_rules(net))
439 			goto full_check;
440 		if (inet_lookup_ifaddr_rcu(net, src))
441 			return -EINVAL;
442 
443 ok:
444 		*itag = 0;
445 		return 0;
446 	}
447 
448 full_check:
449 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
450 }
451 
452 static inline __be32 sk_extract_addr(struct sockaddr *addr)
453 {
454 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
455 }
456 
457 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
458 {
459 	struct nlattr *nla;
460 
461 	nla = (struct nlattr *) ((char *) mx + len);
462 	nla->nla_type = type;
463 	nla->nla_len = nla_attr_size(4);
464 	*(u32 *) nla_data(nla) = value;
465 
466 	return len + nla_total_size(4);
467 }
468 
469 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
470 				 struct fib_config *cfg)
471 {
472 	__be32 addr;
473 	int plen;
474 
475 	memset(cfg, 0, sizeof(*cfg));
476 	cfg->fc_nlinfo.nl_net = net;
477 
478 	if (rt->rt_dst.sa_family != AF_INET)
479 		return -EAFNOSUPPORT;
480 
481 	/*
482 	 * Check mask for validity:
483 	 * a) it must be contiguous.
484 	 * b) destination must have all host bits clear.
485 	 * c) if application forgot to set correct family (AF_INET),
486 	 *    reject request unless it is absolutely clear i.e.
487 	 *    both family and mask are zero.
488 	 */
489 	plen = 32;
490 	addr = sk_extract_addr(&rt->rt_dst);
491 	if (!(rt->rt_flags & RTF_HOST)) {
492 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
493 
494 		if (rt->rt_genmask.sa_family != AF_INET) {
495 			if (mask || rt->rt_genmask.sa_family)
496 				return -EAFNOSUPPORT;
497 		}
498 
499 		if (bad_mask(mask, addr))
500 			return -EINVAL;
501 
502 		plen = inet_mask_len(mask);
503 	}
504 
505 	cfg->fc_dst_len = plen;
506 	cfg->fc_dst = addr;
507 
508 	if (cmd != SIOCDELRT) {
509 		cfg->fc_nlflags = NLM_F_CREATE;
510 		cfg->fc_protocol = RTPROT_BOOT;
511 	}
512 
513 	if (rt->rt_metric)
514 		cfg->fc_priority = rt->rt_metric - 1;
515 
516 	if (rt->rt_flags & RTF_REJECT) {
517 		cfg->fc_scope = RT_SCOPE_HOST;
518 		cfg->fc_type = RTN_UNREACHABLE;
519 		return 0;
520 	}
521 
522 	cfg->fc_scope = RT_SCOPE_NOWHERE;
523 	cfg->fc_type = RTN_UNICAST;
524 
525 	if (rt->rt_dev) {
526 		char *colon;
527 		struct net_device *dev;
528 		char devname[IFNAMSIZ];
529 
530 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
531 			return -EFAULT;
532 
533 		devname[IFNAMSIZ-1] = 0;
534 		colon = strchr(devname, ':');
535 		if (colon)
536 			*colon = 0;
537 		dev = __dev_get_by_name(net, devname);
538 		if (!dev)
539 			return -ENODEV;
540 		cfg->fc_oif = dev->ifindex;
541 		cfg->fc_table = l3mdev_fib_table(dev);
542 		if (colon) {
543 			const struct in_ifaddr *ifa;
544 			struct in_device *in_dev;
545 
546 			in_dev = __in_dev_get_rtnl(dev);
547 			if (!in_dev)
548 				return -ENODEV;
549 
550 			*colon = ':';
551 
552 			rcu_read_lock();
553 			in_dev_for_each_ifa_rcu(ifa, in_dev) {
554 				if (strcmp(ifa->ifa_label, devname) == 0)
555 					break;
556 			}
557 			rcu_read_unlock();
558 
559 			if (!ifa)
560 				return -ENODEV;
561 			cfg->fc_prefsrc = ifa->ifa_local;
562 		}
563 	}
564 
565 	addr = sk_extract_addr(&rt->rt_gateway);
566 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
567 		unsigned int addr_type;
568 
569 		cfg->fc_gw4 = addr;
570 		cfg->fc_gw_family = AF_INET;
571 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
572 		if (rt->rt_flags & RTF_GATEWAY &&
573 		    addr_type == RTN_UNICAST)
574 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
575 	}
576 
577 	if (cmd == SIOCDELRT)
578 		return 0;
579 
580 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
581 		return -EINVAL;
582 
583 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
584 		cfg->fc_scope = RT_SCOPE_LINK;
585 
586 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
587 		struct nlattr *mx;
588 		int len = 0;
589 
590 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
591 		if (!mx)
592 			return -ENOMEM;
593 
594 		if (rt->rt_flags & RTF_MTU)
595 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
596 
597 		if (rt->rt_flags & RTF_WINDOW)
598 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
599 
600 		if (rt->rt_flags & RTF_IRTT)
601 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
602 
603 		cfg->fc_mx = mx;
604 		cfg->fc_mx_len = len;
605 	}
606 
607 	return 0;
608 }
609 
610 /*
611  * Handle IP routing ioctl calls.
612  * These are used to manipulate the routing tables
613  */
614 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
615 {
616 	struct fib_config cfg;
617 	int err;
618 
619 	switch (cmd) {
620 	case SIOCADDRT:		/* Add a route */
621 	case SIOCDELRT:		/* Delete a route */
622 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
623 			return -EPERM;
624 
625 		rtnl_lock();
626 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
627 		if (err == 0) {
628 			struct fib_table *tb;
629 
630 			if (cmd == SIOCDELRT) {
631 				tb = fib_get_table(net, cfg.fc_table);
632 				if (tb)
633 					err = fib_table_delete(net, tb, &cfg,
634 							       NULL);
635 				else
636 					err = -ESRCH;
637 			} else {
638 				tb = fib_new_table(net, cfg.fc_table);
639 				if (tb)
640 					err = fib_table_insert(net, tb,
641 							       &cfg, NULL);
642 				else
643 					err = -ENOBUFS;
644 			}
645 
646 			/* allocated by rtentry_to_fib_config() */
647 			kfree(cfg.fc_mx);
648 		}
649 		rtnl_unlock();
650 		return err;
651 	}
652 	return -EINVAL;
653 }
654 
655 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
656 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
657 	[RTA_DST]		= { .type = NLA_U32 },
658 	[RTA_SRC]		= { .type = NLA_U32 },
659 	[RTA_IIF]		= { .type = NLA_U32 },
660 	[RTA_OIF]		= { .type = NLA_U32 },
661 	[RTA_GATEWAY]		= { .type = NLA_U32 },
662 	[RTA_PRIORITY]		= { .type = NLA_U32 },
663 	[RTA_PREFSRC]		= { .type = NLA_U32 },
664 	[RTA_METRICS]		= { .type = NLA_NESTED },
665 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
666 	[RTA_FLOW]		= { .type = NLA_U32 },
667 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
668 	[RTA_ENCAP]		= { .type = NLA_NESTED },
669 	[RTA_UID]		= { .type = NLA_U32 },
670 	[RTA_MARK]		= { .type = NLA_U32 },
671 	[RTA_TABLE]		= { .type = NLA_U32 },
672 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
673 	[RTA_SPORT]		= { .type = NLA_U16 },
674 	[RTA_DPORT]		= { .type = NLA_U16 },
675 	[RTA_NH_ID]		= { .type = NLA_U32 },
676 };
677 
678 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
679 		    struct netlink_ext_ack *extack)
680 {
681 	struct rtvia *via;
682 	int alen;
683 
684 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
685 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
686 		return -EINVAL;
687 	}
688 
689 	via = nla_data(nla);
690 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
691 
692 	switch (via->rtvia_family) {
693 	case AF_INET:
694 		if (alen != sizeof(__be32)) {
695 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
696 			return -EINVAL;
697 		}
698 		cfg->fc_gw_family = AF_INET;
699 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
700 		break;
701 	case AF_INET6:
702 #if IS_ENABLED(CONFIG_IPV6)
703 		if (alen != sizeof(struct in6_addr)) {
704 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
705 			return -EINVAL;
706 		}
707 		cfg->fc_gw_family = AF_INET6;
708 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
709 #else
710 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
711 		return -EINVAL;
712 #endif
713 		break;
714 	default:
715 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
716 		return -EINVAL;
717 	}
718 
719 	return 0;
720 }
721 
722 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
723 			     struct nlmsghdr *nlh, struct fib_config *cfg,
724 			     struct netlink_ext_ack *extack)
725 {
726 	bool has_gw = false, has_via = false;
727 	struct nlattr *attr;
728 	int err, remaining;
729 	struct rtmsg *rtm;
730 
731 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
732 					rtm_ipv4_policy, extack);
733 	if (err < 0)
734 		goto errout;
735 
736 	memset(cfg, 0, sizeof(*cfg));
737 
738 	rtm = nlmsg_data(nlh);
739 
740 	if (!inet_validate_dscp(rtm->rtm_tos)) {
741 		NL_SET_ERR_MSG(extack,
742 			       "Invalid dsfield (tos): ECN bits must be 0");
743 		err = -EINVAL;
744 		goto errout;
745 	}
746 	cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
747 
748 	cfg->fc_dst_len = rtm->rtm_dst_len;
749 	cfg->fc_table = rtm->rtm_table;
750 	cfg->fc_protocol = rtm->rtm_protocol;
751 	cfg->fc_scope = rtm->rtm_scope;
752 	cfg->fc_type = rtm->rtm_type;
753 	cfg->fc_flags = rtm->rtm_flags;
754 	cfg->fc_nlflags = nlh->nlmsg_flags;
755 
756 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
757 	cfg->fc_nlinfo.nlh = nlh;
758 	cfg->fc_nlinfo.nl_net = net;
759 
760 	if (cfg->fc_type > RTN_MAX) {
761 		NL_SET_ERR_MSG(extack, "Invalid route type");
762 		err = -EINVAL;
763 		goto errout;
764 	}
765 
766 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
767 		switch (nla_type(attr)) {
768 		case RTA_DST:
769 			cfg->fc_dst = nla_get_be32(attr);
770 			break;
771 		case RTA_OIF:
772 			cfg->fc_oif = nla_get_u32(attr);
773 			break;
774 		case RTA_GATEWAY:
775 			has_gw = true;
776 			cfg->fc_gw4 = nla_get_be32(attr);
777 			if (cfg->fc_gw4)
778 				cfg->fc_gw_family = AF_INET;
779 			break;
780 		case RTA_VIA:
781 			has_via = true;
782 			err = fib_gw_from_via(cfg, attr, extack);
783 			if (err)
784 				goto errout;
785 			break;
786 		case RTA_PRIORITY:
787 			cfg->fc_priority = nla_get_u32(attr);
788 			break;
789 		case RTA_PREFSRC:
790 			cfg->fc_prefsrc = nla_get_be32(attr);
791 			break;
792 		case RTA_METRICS:
793 			cfg->fc_mx = nla_data(attr);
794 			cfg->fc_mx_len = nla_len(attr);
795 			break;
796 		case RTA_MULTIPATH:
797 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
798 							     nla_len(attr),
799 							     extack);
800 			if (err < 0)
801 				goto errout;
802 			cfg->fc_mp = nla_data(attr);
803 			cfg->fc_mp_len = nla_len(attr);
804 			break;
805 		case RTA_FLOW:
806 			cfg->fc_flow = nla_get_u32(attr);
807 			break;
808 		case RTA_TABLE:
809 			cfg->fc_table = nla_get_u32(attr);
810 			break;
811 		case RTA_ENCAP:
812 			cfg->fc_encap = attr;
813 			break;
814 		case RTA_ENCAP_TYPE:
815 			cfg->fc_encap_type = nla_get_u16(attr);
816 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
817 							extack);
818 			if (err < 0)
819 				goto errout;
820 			break;
821 		case RTA_NH_ID:
822 			cfg->fc_nh_id = nla_get_u32(attr);
823 			break;
824 		}
825 	}
826 
827 	if (cfg->fc_nh_id) {
828 		if (cfg->fc_oif || cfg->fc_gw_family ||
829 		    cfg->fc_encap || cfg->fc_mp) {
830 			NL_SET_ERR_MSG(extack,
831 				       "Nexthop specification and nexthop id are mutually exclusive");
832 			return -EINVAL;
833 		}
834 	}
835 
836 	if (has_gw && has_via) {
837 		NL_SET_ERR_MSG(extack,
838 			       "Nexthop configuration can not contain both GATEWAY and VIA");
839 		return -EINVAL;
840 	}
841 
842 	return 0;
843 errout:
844 	return err;
845 }
846 
847 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
848 			     struct netlink_ext_ack *extack)
849 {
850 	struct net *net = sock_net(skb->sk);
851 	struct fib_config cfg;
852 	struct fib_table *tb;
853 	int err;
854 
855 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
856 	if (err < 0)
857 		goto errout;
858 
859 	if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
860 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
861 		err = -EINVAL;
862 		goto errout;
863 	}
864 
865 	tb = fib_get_table(net, cfg.fc_table);
866 	if (!tb) {
867 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
868 		err = -ESRCH;
869 		goto errout;
870 	}
871 
872 	err = fib_table_delete(net, tb, &cfg, extack);
873 errout:
874 	return err;
875 }
876 
877 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
878 			     struct netlink_ext_ack *extack)
879 {
880 	struct net *net = sock_net(skb->sk);
881 	struct fib_config cfg;
882 	struct fib_table *tb;
883 	int err;
884 
885 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
886 	if (err < 0)
887 		goto errout;
888 
889 	tb = fib_new_table(net, cfg.fc_table);
890 	if (!tb) {
891 		err = -ENOBUFS;
892 		goto errout;
893 	}
894 
895 	err = fib_table_insert(net, tb, &cfg, extack);
896 	if (!err && cfg.fc_type == RTN_LOCAL)
897 		net->ipv4.fib_has_custom_local_routes = true;
898 errout:
899 	return err;
900 }
901 
902 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
903 			  struct fib_dump_filter *filter,
904 			  struct netlink_callback *cb)
905 {
906 	struct netlink_ext_ack *extack = cb->extack;
907 	struct nlattr *tb[RTA_MAX + 1];
908 	struct rtmsg *rtm;
909 	int err, i;
910 
911 	ASSERT_RTNL();
912 
913 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
914 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
915 		return -EINVAL;
916 	}
917 
918 	rtm = nlmsg_data(nlh);
919 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
920 	    rtm->rtm_scope) {
921 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
922 		return -EINVAL;
923 	}
924 
925 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
926 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
927 		return -EINVAL;
928 	}
929 	if (rtm->rtm_flags & RTM_F_CLONED)
930 		filter->dump_routes = false;
931 	else
932 		filter->dump_exceptions = false;
933 
934 	filter->flags    = rtm->rtm_flags;
935 	filter->protocol = rtm->rtm_protocol;
936 	filter->rt_type  = rtm->rtm_type;
937 	filter->table_id = rtm->rtm_table;
938 
939 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
940 					    rtm_ipv4_policy, extack);
941 	if (err < 0)
942 		return err;
943 
944 	for (i = 0; i <= RTA_MAX; ++i) {
945 		int ifindex;
946 
947 		if (!tb[i])
948 			continue;
949 
950 		switch (i) {
951 		case RTA_TABLE:
952 			filter->table_id = nla_get_u32(tb[i]);
953 			break;
954 		case RTA_OIF:
955 			ifindex = nla_get_u32(tb[i]);
956 			filter->dev = __dev_get_by_index(net, ifindex);
957 			if (!filter->dev)
958 				return -ENODEV;
959 			break;
960 		default:
961 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
962 			return -EINVAL;
963 		}
964 	}
965 
966 	if (filter->flags || filter->protocol || filter->rt_type ||
967 	    filter->table_id || filter->dev) {
968 		filter->filter_set = 1;
969 		cb->answer_flags = NLM_F_DUMP_FILTERED;
970 	}
971 
972 	return 0;
973 }
974 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
975 
976 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
977 {
978 	struct fib_dump_filter filter = { .dump_routes = true,
979 					  .dump_exceptions = true };
980 	const struct nlmsghdr *nlh = cb->nlh;
981 	struct net *net = sock_net(skb->sk);
982 	unsigned int h, s_h;
983 	unsigned int e = 0, s_e;
984 	struct fib_table *tb;
985 	struct hlist_head *head;
986 	int dumped = 0, err;
987 
988 	if (cb->strict_check) {
989 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
990 		if (err < 0)
991 			return err;
992 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
993 		struct rtmsg *rtm = nlmsg_data(nlh);
994 
995 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
996 	}
997 
998 	/* ipv4 does not use prefix flag */
999 	if (filter.flags & RTM_F_PREFIX)
1000 		return skb->len;
1001 
1002 	if (filter.table_id) {
1003 		tb = fib_get_table(net, filter.table_id);
1004 		if (!tb) {
1005 			if (rtnl_msg_family(cb->nlh) != PF_INET)
1006 				return skb->len;
1007 
1008 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1009 			return -ENOENT;
1010 		}
1011 
1012 		rcu_read_lock();
1013 		err = fib_table_dump(tb, skb, cb, &filter);
1014 		rcu_read_unlock();
1015 		return skb->len ? : err;
1016 	}
1017 
1018 	s_h = cb->args[0];
1019 	s_e = cb->args[1];
1020 
1021 	rcu_read_lock();
1022 
1023 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1024 		e = 0;
1025 		head = &net->ipv4.fib_table_hash[h];
1026 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1027 			if (e < s_e)
1028 				goto next;
1029 			if (dumped)
1030 				memset(&cb->args[2], 0, sizeof(cb->args) -
1031 						 2 * sizeof(cb->args[0]));
1032 			err = fib_table_dump(tb, skb, cb, &filter);
1033 			if (err < 0) {
1034 				if (likely(skb->len))
1035 					goto out;
1036 
1037 				goto out_err;
1038 			}
1039 			dumped = 1;
1040 next:
1041 			e++;
1042 		}
1043 	}
1044 out:
1045 	err = skb->len;
1046 out_err:
1047 	rcu_read_unlock();
1048 
1049 	cb->args[1] = e;
1050 	cb->args[0] = h;
1051 
1052 	return err;
1053 }
1054 
1055 /* Prepare and feed intra-kernel routing request.
1056  * Really, it should be netlink message, but :-( netlink
1057  * can be not configured, so that we feed it directly
1058  * to fib engine. It is legal, because all events occur
1059  * only when netlink is already locked.
1060  */
1061 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1062 		      struct in_ifaddr *ifa, u32 rt_priority)
1063 {
1064 	struct net *net = dev_net(ifa->ifa_dev->dev);
1065 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1066 	struct fib_table *tb;
1067 	struct fib_config cfg = {
1068 		.fc_protocol = RTPROT_KERNEL,
1069 		.fc_type = type,
1070 		.fc_dst = dst,
1071 		.fc_dst_len = dst_len,
1072 		.fc_priority = rt_priority,
1073 		.fc_prefsrc = ifa->ifa_local,
1074 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1075 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1076 		.fc_nlinfo = {
1077 			.nl_net = net,
1078 		},
1079 	};
1080 
1081 	if (!tb_id)
1082 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1083 
1084 	tb = fib_new_table(net, tb_id);
1085 	if (!tb)
1086 		return;
1087 
1088 	cfg.fc_table = tb->tb_id;
1089 
1090 	if (type != RTN_LOCAL)
1091 		cfg.fc_scope = RT_SCOPE_LINK;
1092 	else
1093 		cfg.fc_scope = RT_SCOPE_HOST;
1094 
1095 	if (cmd == RTM_NEWROUTE)
1096 		fib_table_insert(net, tb, &cfg, NULL);
1097 	else
1098 		fib_table_delete(net, tb, &cfg, NULL);
1099 }
1100 
1101 void fib_add_ifaddr(struct in_ifaddr *ifa)
1102 {
1103 	struct in_device *in_dev = ifa->ifa_dev;
1104 	struct net_device *dev = in_dev->dev;
1105 	struct in_ifaddr *prim = ifa;
1106 	__be32 mask = ifa->ifa_mask;
1107 	__be32 addr = ifa->ifa_local;
1108 	__be32 prefix = ifa->ifa_address & mask;
1109 
1110 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1111 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1112 		if (!prim) {
1113 			pr_warn("%s: bug: prim == NULL\n", __func__);
1114 			return;
1115 		}
1116 	}
1117 
1118 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1119 
1120 	if (!(dev->flags & IFF_UP))
1121 		return;
1122 
1123 	/* Add broadcast address, if it is explicitly assigned. */
1124 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1125 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1126 			  prim, 0);
1127 
1128 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1129 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1130 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1131 			fib_magic(RTM_NEWROUTE,
1132 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1133 				  prefix, ifa->ifa_prefixlen, prim,
1134 				  ifa->ifa_rt_priority);
1135 
1136 		/* Add the network broadcast address, when it makes sense */
1137 		if (ifa->ifa_prefixlen < 31) {
1138 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1139 				  32, prim, 0);
1140 		}
1141 	}
1142 }
1143 
1144 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1145 {
1146 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1147 	struct in_device *in_dev = ifa->ifa_dev;
1148 	struct net_device *dev = in_dev->dev;
1149 
1150 	if (!(dev->flags & IFF_UP) ||
1151 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1152 	    ipv4_is_zeronet(prefix) ||
1153 	    (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1154 		return;
1155 
1156 	/* add the new */
1157 	fib_magic(RTM_NEWROUTE,
1158 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1159 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1160 
1161 	/* delete the old */
1162 	fib_magic(RTM_DELROUTE,
1163 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1164 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1165 }
1166 
1167 /* Delete primary or secondary address.
1168  * Optionally, on secondary address promotion consider the addresses
1169  * from subnet iprim as deleted, even if they are in device list.
1170  * In this case the secondary ifa can be in device list.
1171  */
1172 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1173 {
1174 	struct in_device *in_dev = ifa->ifa_dev;
1175 	struct net_device *dev = in_dev->dev;
1176 	struct in_ifaddr *ifa1;
1177 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1178 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1179 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1180 #define LOCAL_OK	1
1181 #define BRD_OK		2
1182 #define BRD0_OK		4
1183 #define BRD1_OK		8
1184 	unsigned int ok = 0;
1185 	int subnet = 0;		/* Primary network */
1186 	int gone = 1;		/* Address is missing */
1187 	int same_prefsrc = 0;	/* Another primary with same IP */
1188 
1189 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1190 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1191 		if (!prim) {
1192 			/* if the device has been deleted, we don't perform
1193 			 * address promotion
1194 			 */
1195 			if (!in_dev->dead)
1196 				pr_warn("%s: bug: prim == NULL\n", __func__);
1197 			return;
1198 		}
1199 		if (iprim && iprim != prim) {
1200 			pr_warn("%s: bug: iprim != prim\n", __func__);
1201 			return;
1202 		}
1203 	} else if (!ipv4_is_zeronet(any) &&
1204 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1205 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1206 			fib_magic(RTM_DELROUTE,
1207 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1208 				  any, ifa->ifa_prefixlen, prim, 0);
1209 		subnet = 1;
1210 	}
1211 
1212 	if (in_dev->dead)
1213 		goto no_promotions;
1214 
1215 	/* Deletion is more complicated than add.
1216 	 * We should take care of not to delete too much :-)
1217 	 *
1218 	 * Scan address list to be sure that addresses are really gone.
1219 	 */
1220 	rcu_read_lock();
1221 	in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1222 		if (ifa1 == ifa) {
1223 			/* promotion, keep the IP */
1224 			gone = 0;
1225 			continue;
1226 		}
1227 		/* Ignore IFAs from our subnet */
1228 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1229 		    inet_ifa_match(ifa1->ifa_address, iprim))
1230 			continue;
1231 
1232 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1233 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1234 			/* Another address from our subnet? */
1235 			if (ifa1->ifa_mask == prim->ifa_mask &&
1236 			    inet_ifa_match(ifa1->ifa_address, prim))
1237 				prim1 = prim;
1238 			else {
1239 				/* We reached the secondaries, so
1240 				 * same_prefsrc should be determined.
1241 				 */
1242 				if (!same_prefsrc)
1243 					continue;
1244 				/* Search new prim1 if ifa1 is not
1245 				 * using the current prim1
1246 				 */
1247 				if (!prim1 ||
1248 				    ifa1->ifa_mask != prim1->ifa_mask ||
1249 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1250 					prim1 = inet_ifa_byprefix(in_dev,
1251 							ifa1->ifa_address,
1252 							ifa1->ifa_mask);
1253 				if (!prim1)
1254 					continue;
1255 				if (prim1->ifa_local != prim->ifa_local)
1256 					continue;
1257 			}
1258 		} else {
1259 			if (prim->ifa_local != ifa1->ifa_local)
1260 				continue;
1261 			prim1 = ifa1;
1262 			if (prim != prim1)
1263 				same_prefsrc = 1;
1264 		}
1265 		if (ifa->ifa_local == ifa1->ifa_local)
1266 			ok |= LOCAL_OK;
1267 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1268 			ok |= BRD_OK;
1269 		if (brd == ifa1->ifa_broadcast)
1270 			ok |= BRD1_OK;
1271 		if (any == ifa1->ifa_broadcast)
1272 			ok |= BRD0_OK;
1273 		/* primary has network specific broadcasts */
1274 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1275 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1276 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1277 
1278 			if (!ipv4_is_zeronet(any1)) {
1279 				if (ifa->ifa_broadcast == brd1 ||
1280 				    ifa->ifa_broadcast == any1)
1281 					ok |= BRD_OK;
1282 				if (brd == brd1 || brd == any1)
1283 					ok |= BRD1_OK;
1284 				if (any == brd1 || any == any1)
1285 					ok |= BRD0_OK;
1286 			}
1287 		}
1288 	}
1289 	rcu_read_unlock();
1290 
1291 no_promotions:
1292 	if (!(ok & BRD_OK))
1293 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1294 			  prim, 0);
1295 	if (subnet && ifa->ifa_prefixlen < 31) {
1296 		if (!(ok & BRD1_OK))
1297 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1298 				  prim, 0);
1299 		if (!(ok & BRD0_OK))
1300 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1301 				  prim, 0);
1302 	}
1303 	if (!(ok & LOCAL_OK)) {
1304 		unsigned int addr_type;
1305 
1306 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1307 
1308 		/* Check, that this local address finally disappeared. */
1309 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1310 						     ifa->ifa_local);
1311 		if (gone && addr_type != RTN_LOCAL) {
1312 			/* And the last, but not the least thing.
1313 			 * We must flush stray FIB entries.
1314 			 *
1315 			 * First of all, we scan fib_info list searching
1316 			 * for stray nexthop entries, then ignite fib_flush.
1317 			 */
1318 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1319 				fib_flush(dev_net(dev));
1320 		}
1321 	}
1322 #undef LOCAL_OK
1323 #undef BRD_OK
1324 #undef BRD0_OK
1325 #undef BRD1_OK
1326 }
1327 
1328 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1329 {
1330 
1331 	struct fib_result       res;
1332 	struct flowi4           fl4 = {
1333 		.flowi4_mark = frn->fl_mark,
1334 		.daddr = frn->fl_addr,
1335 		.flowi4_tos = frn->fl_tos,
1336 		.flowi4_scope = frn->fl_scope,
1337 	};
1338 	struct fib_table *tb;
1339 
1340 	rcu_read_lock();
1341 
1342 	tb = fib_get_table(net, frn->tb_id_in);
1343 
1344 	frn->err = -ENOENT;
1345 	if (tb) {
1346 		local_bh_disable();
1347 
1348 		frn->tb_id = tb->tb_id;
1349 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1350 
1351 		if (!frn->err) {
1352 			frn->prefixlen = res.prefixlen;
1353 			frn->nh_sel = res.nh_sel;
1354 			frn->type = res.type;
1355 			frn->scope = res.scope;
1356 		}
1357 		local_bh_enable();
1358 	}
1359 
1360 	rcu_read_unlock();
1361 }
1362 
1363 static void nl_fib_input(struct sk_buff *skb)
1364 {
1365 	struct net *net;
1366 	struct fib_result_nl *frn;
1367 	struct nlmsghdr *nlh;
1368 	u32 portid;
1369 
1370 	net = sock_net(skb->sk);
1371 	nlh = nlmsg_hdr(skb);
1372 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1373 	    skb->len < nlh->nlmsg_len ||
1374 	    nlmsg_len(nlh) < sizeof(*frn))
1375 		return;
1376 
1377 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1378 	if (!skb)
1379 		return;
1380 	nlh = nlmsg_hdr(skb);
1381 
1382 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1383 	nl_fib_lookup(net, frn);
1384 
1385 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1386 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1387 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1388 	nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1389 }
1390 
1391 static int __net_init nl_fib_lookup_init(struct net *net)
1392 {
1393 	struct sock *sk;
1394 	struct netlink_kernel_cfg cfg = {
1395 		.input	= nl_fib_input,
1396 	};
1397 
1398 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1399 	if (!sk)
1400 		return -EAFNOSUPPORT;
1401 	net->ipv4.fibnl = sk;
1402 	return 0;
1403 }
1404 
1405 static void nl_fib_lookup_exit(struct net *net)
1406 {
1407 	netlink_kernel_release(net->ipv4.fibnl);
1408 	net->ipv4.fibnl = NULL;
1409 }
1410 
1411 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1412 			   bool force)
1413 {
1414 	if (fib_sync_down_dev(dev, event, force))
1415 		fib_flush(dev_net(dev));
1416 	else
1417 		rt_cache_flush(dev_net(dev));
1418 	arp_ifdown(dev);
1419 }
1420 
1421 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1422 {
1423 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1424 	struct net_device *dev = ifa->ifa_dev->dev;
1425 	struct net *net = dev_net(dev);
1426 
1427 	switch (event) {
1428 	case NETDEV_UP:
1429 		fib_add_ifaddr(ifa);
1430 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1431 		fib_sync_up(dev, RTNH_F_DEAD);
1432 #endif
1433 		atomic_inc(&net->ipv4.dev_addr_genid);
1434 		rt_cache_flush(dev_net(dev));
1435 		break;
1436 	case NETDEV_DOWN:
1437 		fib_del_ifaddr(ifa, NULL);
1438 		atomic_inc(&net->ipv4.dev_addr_genid);
1439 		if (!ifa->ifa_dev->ifa_list) {
1440 			/* Last address was deleted from this interface.
1441 			 * Disable IP.
1442 			 */
1443 			fib_disable_ip(dev, event, true);
1444 		} else {
1445 			rt_cache_flush(dev_net(dev));
1446 		}
1447 		break;
1448 	}
1449 	return NOTIFY_DONE;
1450 }
1451 
1452 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1453 {
1454 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1455 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1456 	struct netdev_notifier_info_ext *info_ext = ptr;
1457 	struct in_device *in_dev;
1458 	struct net *net = dev_net(dev);
1459 	struct in_ifaddr *ifa;
1460 	unsigned int flags;
1461 
1462 	if (event == NETDEV_UNREGISTER) {
1463 		fib_disable_ip(dev, event, true);
1464 		rt_flush_dev(dev);
1465 		return NOTIFY_DONE;
1466 	}
1467 
1468 	in_dev = __in_dev_get_rtnl(dev);
1469 	if (!in_dev)
1470 		return NOTIFY_DONE;
1471 
1472 	switch (event) {
1473 	case NETDEV_UP:
1474 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1475 			fib_add_ifaddr(ifa);
1476 		}
1477 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1478 		fib_sync_up(dev, RTNH_F_DEAD);
1479 #endif
1480 		atomic_inc(&net->ipv4.dev_addr_genid);
1481 		rt_cache_flush(net);
1482 		break;
1483 	case NETDEV_DOWN:
1484 		fib_disable_ip(dev, event, false);
1485 		break;
1486 	case NETDEV_CHANGE:
1487 		flags = dev_get_flags(dev);
1488 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1489 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1490 		else
1491 			fib_sync_down_dev(dev, event, false);
1492 		rt_cache_flush(net);
1493 		break;
1494 	case NETDEV_CHANGEMTU:
1495 		fib_sync_mtu(dev, info_ext->ext.mtu);
1496 		rt_cache_flush(net);
1497 		break;
1498 	case NETDEV_CHANGEUPPER:
1499 		upper_info = ptr;
1500 		/* flush all routes if dev is linked to or unlinked from
1501 		 * an L3 master device (e.g., VRF)
1502 		 */
1503 		if (upper_info->upper_dev &&
1504 		    netif_is_l3_master(upper_info->upper_dev))
1505 			fib_disable_ip(dev, NETDEV_DOWN, true);
1506 		break;
1507 	}
1508 	return NOTIFY_DONE;
1509 }
1510 
1511 static struct notifier_block fib_inetaddr_notifier = {
1512 	.notifier_call = fib_inetaddr_event,
1513 };
1514 
1515 static struct notifier_block fib_netdev_notifier = {
1516 	.notifier_call = fib_netdev_event,
1517 };
1518 
1519 static int __net_init ip_fib_net_init(struct net *net)
1520 {
1521 	int err;
1522 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1523 
1524 	err = fib4_notifier_init(net);
1525 	if (err)
1526 		return err;
1527 
1528 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1529 	/* Default to 3-tuple */
1530 	net->ipv4.sysctl_fib_multipath_hash_fields =
1531 		FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1532 #endif
1533 
1534 	/* Avoid false sharing : Use at least a full cache line */
1535 	size = max_t(size_t, size, L1_CACHE_BYTES);
1536 
1537 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1538 	if (!net->ipv4.fib_table_hash) {
1539 		err = -ENOMEM;
1540 		goto err_table_hash_alloc;
1541 	}
1542 
1543 	err = fib4_rules_init(net);
1544 	if (err < 0)
1545 		goto err_rules_init;
1546 	return 0;
1547 
1548 err_rules_init:
1549 	kfree(net->ipv4.fib_table_hash);
1550 err_table_hash_alloc:
1551 	fib4_notifier_exit(net);
1552 	return err;
1553 }
1554 
1555 static void ip_fib_net_exit(struct net *net)
1556 {
1557 	int i;
1558 
1559 	ASSERT_RTNL();
1560 #ifdef CONFIG_IP_MULTIPLE_TABLES
1561 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1562 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1563 #endif
1564 	/* Destroy the tables in reverse order to guarantee that the
1565 	 * local table, ID 255, is destroyed before the main table, ID
1566 	 * 254. This is necessary as the local table may contain
1567 	 * references to data contained in the main table.
1568 	 */
1569 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1570 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1571 		struct hlist_node *tmp;
1572 		struct fib_table *tb;
1573 
1574 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1575 			hlist_del(&tb->tb_hlist);
1576 			fib_table_flush(net, tb, true);
1577 			fib_free_table(tb);
1578 		}
1579 	}
1580 
1581 #ifdef CONFIG_IP_MULTIPLE_TABLES
1582 	fib4_rules_exit(net);
1583 #endif
1584 
1585 	kfree(net->ipv4.fib_table_hash);
1586 	fib4_notifier_exit(net);
1587 }
1588 
1589 static int __net_init fib_net_init(struct net *net)
1590 {
1591 	int error;
1592 
1593 #ifdef CONFIG_IP_ROUTE_CLASSID
1594 	atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1595 #endif
1596 	error = ip_fib_net_init(net);
1597 	if (error < 0)
1598 		goto out;
1599 	error = nl_fib_lookup_init(net);
1600 	if (error < 0)
1601 		goto out_nlfl;
1602 	error = fib_proc_init(net);
1603 	if (error < 0)
1604 		goto out_proc;
1605 out:
1606 	return error;
1607 
1608 out_proc:
1609 	nl_fib_lookup_exit(net);
1610 out_nlfl:
1611 	rtnl_lock();
1612 	ip_fib_net_exit(net);
1613 	rtnl_unlock();
1614 	goto out;
1615 }
1616 
1617 static void __net_exit fib_net_exit(struct net *net)
1618 {
1619 	fib_proc_exit(net);
1620 	nl_fib_lookup_exit(net);
1621 }
1622 
1623 static void __net_exit fib_net_exit_batch(struct list_head *net_list)
1624 {
1625 	struct net *net;
1626 
1627 	rtnl_lock();
1628 	list_for_each_entry(net, net_list, exit_list)
1629 		ip_fib_net_exit(net);
1630 
1631 	rtnl_unlock();
1632 }
1633 
1634 static struct pernet_operations fib_net_ops = {
1635 	.init = fib_net_init,
1636 	.exit = fib_net_exit,
1637 	.exit_batch = fib_net_exit_batch,
1638 };
1639 
1640 void __init ip_fib_init(void)
1641 {
1642 	fib_trie_init();
1643 
1644 	register_pernet_subsys(&fib_net_ops);
1645 
1646 	register_netdevice_notifier(&fib_netdev_notifier);
1647 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1648 
1649 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1650 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1651 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1652 }
1653