xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision 7d545e77)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: FIB frontend.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/bitops.h>
15 #include <linux/capability.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/errno.h>
23 #include <linux/in.h>
24 #include <linux/inet.h>
25 #include <linux/inetdevice.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_addr.h>
28 #include <linux/if_arp.h>
29 #include <linux/skbuff.h>
30 #include <linux/cache.h>
31 #include <linux/init.h>
32 #include <linux/list.h>
33 #include <linux/slab.h>
34 
35 #include <net/ip.h>
36 #include <net/protocol.h>
37 #include <net/route.h>
38 #include <net/tcp.h>
39 #include <net/sock.h>
40 #include <net/arp.h>
41 #include <net/ip_fib.h>
42 #include <net/nexthop.h>
43 #include <net/rtnetlink.h>
44 #include <net/xfrm.h>
45 #include <net/l3mdev.h>
46 #include <net/lwtunnel.h>
47 #include <trace/events/fib.h>
48 
49 #ifndef CONFIG_IP_MULTIPLE_TABLES
50 
51 static int __net_init fib4_rules_init(struct net *net)
52 {
53 	struct fib_table *local_table, *main_table;
54 
55 	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
56 	if (!main_table)
57 		return -ENOMEM;
58 
59 	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
60 	if (!local_table)
61 		goto fail;
62 
63 	hlist_add_head_rcu(&local_table->tb_hlist,
64 				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
65 	hlist_add_head_rcu(&main_table->tb_hlist,
66 				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
67 	return 0;
68 
69 fail:
70 	fib_free_table(main_table);
71 	return -ENOMEM;
72 }
73 
74 static bool fib4_has_custom_rules(struct net *net)
75 {
76 	return false;
77 }
78 #else
79 
80 struct fib_table *fib_new_table(struct net *net, u32 id)
81 {
82 	struct fib_table *tb, *alias = NULL;
83 	unsigned int h;
84 
85 	if (id == 0)
86 		id = RT_TABLE_MAIN;
87 	tb = fib_get_table(net, id);
88 	if (tb)
89 		return tb;
90 
91 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
92 		alias = fib_new_table(net, RT_TABLE_MAIN);
93 
94 	tb = fib_trie_table(id, alias);
95 	if (!tb)
96 		return NULL;
97 
98 	switch (id) {
99 	case RT_TABLE_MAIN:
100 		rcu_assign_pointer(net->ipv4.fib_main, tb);
101 		break;
102 	case RT_TABLE_DEFAULT:
103 		rcu_assign_pointer(net->ipv4.fib_default, tb);
104 		break;
105 	default:
106 		break;
107 	}
108 
109 	h = id & (FIB_TABLE_HASHSZ - 1);
110 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
111 	return tb;
112 }
113 EXPORT_SYMBOL_GPL(fib_new_table);
114 
115 /* caller must hold either rtnl or rcu read lock */
116 struct fib_table *fib_get_table(struct net *net, u32 id)
117 {
118 	struct fib_table *tb;
119 	struct hlist_head *head;
120 	unsigned int h;
121 
122 	if (id == 0)
123 		id = RT_TABLE_MAIN;
124 	h = id & (FIB_TABLE_HASHSZ - 1);
125 
126 	head = &net->ipv4.fib_table_hash[h];
127 	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
128 		if (tb->tb_id == id)
129 			return tb;
130 	}
131 	return NULL;
132 }
133 
134 static bool fib4_has_custom_rules(struct net *net)
135 {
136 	return net->ipv4.fib_has_custom_rules;
137 }
138 #endif /* CONFIG_IP_MULTIPLE_TABLES */
139 
140 static void fib_replace_table(struct net *net, struct fib_table *old,
141 			      struct fib_table *new)
142 {
143 #ifdef CONFIG_IP_MULTIPLE_TABLES
144 	switch (new->tb_id) {
145 	case RT_TABLE_MAIN:
146 		rcu_assign_pointer(net->ipv4.fib_main, new);
147 		break;
148 	case RT_TABLE_DEFAULT:
149 		rcu_assign_pointer(net->ipv4.fib_default, new);
150 		break;
151 	default:
152 		break;
153 	}
154 
155 #endif
156 	/* replace the old table in the hlist */
157 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
158 }
159 
160 int fib_unmerge(struct net *net)
161 {
162 	struct fib_table *old, *new, *main_table;
163 
164 	/* attempt to fetch local table if it has been allocated */
165 	old = fib_get_table(net, RT_TABLE_LOCAL);
166 	if (!old)
167 		return 0;
168 
169 	new = fib_trie_unmerge(old);
170 	if (!new)
171 		return -ENOMEM;
172 
173 	/* table is already unmerged */
174 	if (new == old)
175 		return 0;
176 
177 	/* replace merged table with clean table */
178 	fib_replace_table(net, old, new);
179 	fib_free_table(old);
180 
181 	/* attempt to fetch main table if it has been allocated */
182 	main_table = fib_get_table(net, RT_TABLE_MAIN);
183 	if (!main_table)
184 		return 0;
185 
186 	/* flush local entries from main table */
187 	fib_table_flush_external(main_table);
188 
189 	return 0;
190 }
191 
192 void fib_flush(struct net *net)
193 {
194 	int flushed = 0;
195 	unsigned int h;
196 
197 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
198 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
199 		struct hlist_node *tmp;
200 		struct fib_table *tb;
201 
202 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
203 			flushed += fib_table_flush(net, tb, false);
204 	}
205 
206 	if (flushed)
207 		rt_cache_flush(net);
208 }
209 
210 /*
211  * Find address type as if only "dev" was present in the system. If
212  * on_dev is NULL then all interfaces are taken into consideration.
213  */
214 static inline unsigned int __inet_dev_addr_type(struct net *net,
215 						const struct net_device *dev,
216 						__be32 addr, u32 tb_id)
217 {
218 	struct flowi4		fl4 = { .daddr = addr };
219 	struct fib_result	res;
220 	unsigned int ret = RTN_BROADCAST;
221 	struct fib_table *table;
222 
223 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
224 		return RTN_BROADCAST;
225 	if (ipv4_is_multicast(addr))
226 		return RTN_MULTICAST;
227 
228 	rcu_read_lock();
229 
230 	table = fib_get_table(net, tb_id);
231 	if (table) {
232 		ret = RTN_UNICAST;
233 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
234 			struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
235 
236 			if (!dev || dev == nhc->nhc_dev)
237 				ret = res.type;
238 		}
239 	}
240 
241 	rcu_read_unlock();
242 	return ret;
243 }
244 
245 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
246 {
247 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
248 }
249 EXPORT_SYMBOL(inet_addr_type_table);
250 
251 unsigned int inet_addr_type(struct net *net, __be32 addr)
252 {
253 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
254 }
255 EXPORT_SYMBOL(inet_addr_type);
256 
257 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
258 				__be32 addr)
259 {
260 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
261 
262 	return __inet_dev_addr_type(net, dev, addr, rt_table);
263 }
264 EXPORT_SYMBOL(inet_dev_addr_type);
265 
266 /* inet_addr_type with dev == NULL but using the table from a dev
267  * if one is associated
268  */
269 unsigned int inet_addr_type_dev_table(struct net *net,
270 				      const struct net_device *dev,
271 				      __be32 addr)
272 {
273 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
274 
275 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
276 }
277 EXPORT_SYMBOL(inet_addr_type_dev_table);
278 
279 __be32 fib_compute_spec_dst(struct sk_buff *skb)
280 {
281 	struct net_device *dev = skb->dev;
282 	struct in_device *in_dev;
283 	struct fib_result res;
284 	struct rtable *rt;
285 	struct net *net;
286 	int scope;
287 
288 	rt = skb_rtable(skb);
289 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
290 	    RTCF_LOCAL)
291 		return ip_hdr(skb)->daddr;
292 
293 	in_dev = __in_dev_get_rcu(dev);
294 
295 	net = dev_net(dev);
296 
297 	scope = RT_SCOPE_UNIVERSE;
298 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
299 		bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
300 		struct flowi4 fl4 = {
301 			.flowi4_iif = LOOPBACK_IFINDEX,
302 			.flowi4_oif = l3mdev_master_ifindex_rcu(dev),
303 			.daddr = ip_hdr(skb)->saddr,
304 			.flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
305 			.flowi4_scope = scope,
306 			.flowi4_mark = vmark ? skb->mark : 0,
307 		};
308 		if (!fib_lookup(net, &fl4, &res, 0))
309 			return fib_result_prefsrc(net, &res);
310 	} else {
311 		scope = RT_SCOPE_LINK;
312 	}
313 
314 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
315 }
316 
317 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
318 {
319 	bool dev_match = false;
320 #ifdef CONFIG_IP_ROUTE_MULTIPATH
321 	int ret;
322 
323 	for (ret = 0; ret < fib_info_num_path(fi); ret++) {
324 		const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
325 
326 		if (nhc->nhc_dev == dev) {
327 			dev_match = true;
328 			break;
329 		} else if (l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex) {
330 			dev_match = true;
331 			break;
332 		}
333 	}
334 #else
335 	if (fib_info_nhc(fi, 0)->nhc_dev == dev)
336 		dev_match = true;
337 #endif
338 
339 	return dev_match;
340 }
341 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
342 
343 /* Given (packet source, input interface) and optional (dst, oif, tos):
344  * - (main) check, that source is valid i.e. not broadcast or our local
345  *   address.
346  * - figure out what "logical" interface this packet arrived
347  *   and calculate "specific destination" address.
348  * - check, that packet arrived from expected physical interface.
349  * called with rcu_read_lock()
350  */
351 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
352 				 u8 tos, int oif, struct net_device *dev,
353 				 int rpf, struct in_device *idev, u32 *itag)
354 {
355 	struct net *net = dev_net(dev);
356 	struct flow_keys flkeys;
357 	int ret, no_addr;
358 	struct fib_result res;
359 	struct flowi4 fl4;
360 	bool dev_match;
361 
362 	fl4.flowi4_oif = 0;
363 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
364 	if (!fl4.flowi4_iif)
365 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
366 	fl4.daddr = src;
367 	fl4.saddr = dst;
368 	fl4.flowi4_tos = tos;
369 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
370 	fl4.flowi4_tun_key.tun_id = 0;
371 	fl4.flowi4_flags = 0;
372 	fl4.flowi4_uid = sock_net_uid(net, NULL);
373 
374 	no_addr = idev->ifa_list == NULL;
375 
376 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
377 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
378 		fl4.flowi4_proto = 0;
379 		fl4.fl4_sport = 0;
380 		fl4.fl4_dport = 0;
381 	}
382 
383 	if (fib_lookup(net, &fl4, &res, 0))
384 		goto last_resort;
385 	if (res.type != RTN_UNICAST &&
386 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
387 		goto e_inval;
388 	fib_combine_itag(itag, &res);
389 
390 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
391 	/* This is not common, loopback packets retain skb_dst so normally they
392 	 * would not even hit this slow path.
393 	 */
394 	dev_match = dev_match || (res.type == RTN_LOCAL &&
395 				  dev == net->loopback_dev);
396 	if (dev_match) {
397 		ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
398 		return ret;
399 	}
400 	if (no_addr)
401 		goto last_resort;
402 	if (rpf == 1)
403 		goto e_rpf;
404 	fl4.flowi4_oif = dev->ifindex;
405 
406 	ret = 0;
407 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
408 		if (res.type == RTN_UNICAST)
409 			ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
410 	}
411 	return ret;
412 
413 last_resort:
414 	if (rpf)
415 		goto e_rpf;
416 	*itag = 0;
417 	return 0;
418 
419 e_inval:
420 	return -EINVAL;
421 e_rpf:
422 	return -EXDEV;
423 }
424 
425 /* Ignore rp_filter for packets protected by IPsec. */
426 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
427 			u8 tos, int oif, struct net_device *dev,
428 			struct in_device *idev, u32 *itag)
429 {
430 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
431 	struct net *net = dev_net(dev);
432 
433 	if (!r && !fib_num_tclassid_users(net) &&
434 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
435 		if (IN_DEV_ACCEPT_LOCAL(idev))
436 			goto ok;
437 		/* with custom local routes in place, checking local addresses
438 		 * only will be too optimistic, with custom rules, checking
439 		 * local addresses only can be too strict, e.g. due to vrf
440 		 */
441 		if (net->ipv4.fib_has_custom_local_routes ||
442 		    fib4_has_custom_rules(net))
443 			goto full_check;
444 		if (inet_lookup_ifaddr_rcu(net, src))
445 			return -EINVAL;
446 
447 ok:
448 		*itag = 0;
449 		return 0;
450 	}
451 
452 full_check:
453 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
454 }
455 
456 static inline __be32 sk_extract_addr(struct sockaddr *addr)
457 {
458 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
459 }
460 
461 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
462 {
463 	struct nlattr *nla;
464 
465 	nla = (struct nlattr *) ((char *) mx + len);
466 	nla->nla_type = type;
467 	nla->nla_len = nla_attr_size(4);
468 	*(u32 *) nla_data(nla) = value;
469 
470 	return len + nla_total_size(4);
471 }
472 
473 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
474 				 struct fib_config *cfg)
475 {
476 	__be32 addr;
477 	int plen;
478 
479 	memset(cfg, 0, sizeof(*cfg));
480 	cfg->fc_nlinfo.nl_net = net;
481 
482 	if (rt->rt_dst.sa_family != AF_INET)
483 		return -EAFNOSUPPORT;
484 
485 	/*
486 	 * Check mask for validity:
487 	 * a) it must be contiguous.
488 	 * b) destination must have all host bits clear.
489 	 * c) if application forgot to set correct family (AF_INET),
490 	 *    reject request unless it is absolutely clear i.e.
491 	 *    both family and mask are zero.
492 	 */
493 	plen = 32;
494 	addr = sk_extract_addr(&rt->rt_dst);
495 	if (!(rt->rt_flags & RTF_HOST)) {
496 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
497 
498 		if (rt->rt_genmask.sa_family != AF_INET) {
499 			if (mask || rt->rt_genmask.sa_family)
500 				return -EAFNOSUPPORT;
501 		}
502 
503 		if (bad_mask(mask, addr))
504 			return -EINVAL;
505 
506 		plen = inet_mask_len(mask);
507 	}
508 
509 	cfg->fc_dst_len = plen;
510 	cfg->fc_dst = addr;
511 
512 	if (cmd != SIOCDELRT) {
513 		cfg->fc_nlflags = NLM_F_CREATE;
514 		cfg->fc_protocol = RTPROT_BOOT;
515 	}
516 
517 	if (rt->rt_metric)
518 		cfg->fc_priority = rt->rt_metric - 1;
519 
520 	if (rt->rt_flags & RTF_REJECT) {
521 		cfg->fc_scope = RT_SCOPE_HOST;
522 		cfg->fc_type = RTN_UNREACHABLE;
523 		return 0;
524 	}
525 
526 	cfg->fc_scope = RT_SCOPE_NOWHERE;
527 	cfg->fc_type = RTN_UNICAST;
528 
529 	if (rt->rt_dev) {
530 		char *colon;
531 		struct net_device *dev;
532 		char devname[IFNAMSIZ];
533 
534 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
535 			return -EFAULT;
536 
537 		devname[IFNAMSIZ-1] = 0;
538 		colon = strchr(devname, ':');
539 		if (colon)
540 			*colon = 0;
541 		dev = __dev_get_by_name(net, devname);
542 		if (!dev)
543 			return -ENODEV;
544 		cfg->fc_oif = dev->ifindex;
545 		cfg->fc_table = l3mdev_fib_table(dev);
546 		if (colon) {
547 			const struct in_ifaddr *ifa;
548 			struct in_device *in_dev;
549 
550 			in_dev = __in_dev_get_rtnl(dev);
551 			if (!in_dev)
552 				return -ENODEV;
553 
554 			*colon = ':';
555 
556 			rcu_read_lock();
557 			in_dev_for_each_ifa_rcu(ifa, in_dev) {
558 				if (strcmp(ifa->ifa_label, devname) == 0)
559 					break;
560 			}
561 			rcu_read_unlock();
562 
563 			if (!ifa)
564 				return -ENODEV;
565 			cfg->fc_prefsrc = ifa->ifa_local;
566 		}
567 	}
568 
569 	addr = sk_extract_addr(&rt->rt_gateway);
570 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
571 		unsigned int addr_type;
572 
573 		cfg->fc_gw4 = addr;
574 		cfg->fc_gw_family = AF_INET;
575 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
576 		if (rt->rt_flags & RTF_GATEWAY &&
577 		    addr_type == RTN_UNICAST)
578 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
579 	}
580 
581 	if (cmd == SIOCDELRT)
582 		return 0;
583 
584 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
585 		return -EINVAL;
586 
587 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
588 		cfg->fc_scope = RT_SCOPE_LINK;
589 
590 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
591 		struct nlattr *mx;
592 		int len = 0;
593 
594 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
595 		if (!mx)
596 			return -ENOMEM;
597 
598 		if (rt->rt_flags & RTF_MTU)
599 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
600 
601 		if (rt->rt_flags & RTF_WINDOW)
602 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
603 
604 		if (rt->rt_flags & RTF_IRTT)
605 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
606 
607 		cfg->fc_mx = mx;
608 		cfg->fc_mx_len = len;
609 	}
610 
611 	return 0;
612 }
613 
614 /*
615  * Handle IP routing ioctl calls.
616  * These are used to manipulate the routing tables
617  */
618 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
619 {
620 	struct fib_config cfg;
621 	int err;
622 
623 	switch (cmd) {
624 	case SIOCADDRT:		/* Add a route */
625 	case SIOCDELRT:		/* Delete a route */
626 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
627 			return -EPERM;
628 
629 		rtnl_lock();
630 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
631 		if (err == 0) {
632 			struct fib_table *tb;
633 
634 			if (cmd == SIOCDELRT) {
635 				tb = fib_get_table(net, cfg.fc_table);
636 				if (tb)
637 					err = fib_table_delete(net, tb, &cfg,
638 							       NULL);
639 				else
640 					err = -ESRCH;
641 			} else {
642 				tb = fib_new_table(net, cfg.fc_table);
643 				if (tb)
644 					err = fib_table_insert(net, tb,
645 							       &cfg, NULL);
646 				else
647 					err = -ENOBUFS;
648 			}
649 
650 			/* allocated by rtentry_to_fib_config() */
651 			kfree(cfg.fc_mx);
652 		}
653 		rtnl_unlock();
654 		return err;
655 	}
656 	return -EINVAL;
657 }
658 
659 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
660 	[RTA_UNSPEC]		= { .strict_start_type = RTA_DPORT + 1 },
661 	[RTA_DST]		= { .type = NLA_U32 },
662 	[RTA_SRC]		= { .type = NLA_U32 },
663 	[RTA_IIF]		= { .type = NLA_U32 },
664 	[RTA_OIF]		= { .type = NLA_U32 },
665 	[RTA_GATEWAY]		= { .type = NLA_U32 },
666 	[RTA_PRIORITY]		= { .type = NLA_U32 },
667 	[RTA_PREFSRC]		= { .type = NLA_U32 },
668 	[RTA_METRICS]		= { .type = NLA_NESTED },
669 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
670 	[RTA_FLOW]		= { .type = NLA_U32 },
671 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
672 	[RTA_ENCAP]		= { .type = NLA_NESTED },
673 	[RTA_UID]		= { .type = NLA_U32 },
674 	[RTA_MARK]		= { .type = NLA_U32 },
675 	[RTA_TABLE]		= { .type = NLA_U32 },
676 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
677 	[RTA_SPORT]		= { .type = NLA_U16 },
678 	[RTA_DPORT]		= { .type = NLA_U16 },
679 	[RTA_NH_ID]		= { .type = NLA_U32 },
680 };
681 
682 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
683 		    struct netlink_ext_ack *extack)
684 {
685 	struct rtvia *via;
686 	int alen;
687 
688 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
689 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
690 		return -EINVAL;
691 	}
692 
693 	via = nla_data(nla);
694 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
695 
696 	switch (via->rtvia_family) {
697 	case AF_INET:
698 		if (alen != sizeof(__be32)) {
699 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
700 			return -EINVAL;
701 		}
702 		cfg->fc_gw_family = AF_INET;
703 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
704 		break;
705 	case AF_INET6:
706 #ifdef CONFIG_IPV6
707 		if (alen != sizeof(struct in6_addr)) {
708 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
709 			return -EINVAL;
710 		}
711 		cfg->fc_gw_family = AF_INET6;
712 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
713 #else
714 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
715 		return -EINVAL;
716 #endif
717 		break;
718 	default:
719 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
720 		return -EINVAL;
721 	}
722 
723 	return 0;
724 }
725 
726 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
727 			     struct nlmsghdr *nlh, struct fib_config *cfg,
728 			     struct netlink_ext_ack *extack)
729 {
730 	bool has_gw = false, has_via = false;
731 	struct nlattr *attr;
732 	int err, remaining;
733 	struct rtmsg *rtm;
734 
735 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
736 					rtm_ipv4_policy, extack);
737 	if (err < 0)
738 		goto errout;
739 
740 	memset(cfg, 0, sizeof(*cfg));
741 
742 	rtm = nlmsg_data(nlh);
743 	cfg->fc_dst_len = rtm->rtm_dst_len;
744 	cfg->fc_tos = rtm->rtm_tos;
745 	cfg->fc_table = rtm->rtm_table;
746 	cfg->fc_protocol = rtm->rtm_protocol;
747 	cfg->fc_scope = rtm->rtm_scope;
748 	cfg->fc_type = rtm->rtm_type;
749 	cfg->fc_flags = rtm->rtm_flags;
750 	cfg->fc_nlflags = nlh->nlmsg_flags;
751 
752 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
753 	cfg->fc_nlinfo.nlh = nlh;
754 	cfg->fc_nlinfo.nl_net = net;
755 
756 	if (cfg->fc_type > RTN_MAX) {
757 		NL_SET_ERR_MSG(extack, "Invalid route type");
758 		err = -EINVAL;
759 		goto errout;
760 	}
761 
762 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
763 		switch (nla_type(attr)) {
764 		case RTA_DST:
765 			cfg->fc_dst = nla_get_be32(attr);
766 			break;
767 		case RTA_OIF:
768 			cfg->fc_oif = nla_get_u32(attr);
769 			break;
770 		case RTA_GATEWAY:
771 			has_gw = true;
772 			cfg->fc_gw4 = nla_get_be32(attr);
773 			if (cfg->fc_gw4)
774 				cfg->fc_gw_family = AF_INET;
775 			break;
776 		case RTA_VIA:
777 			has_via = true;
778 			err = fib_gw_from_via(cfg, attr, extack);
779 			if (err)
780 				goto errout;
781 			break;
782 		case RTA_PRIORITY:
783 			cfg->fc_priority = nla_get_u32(attr);
784 			break;
785 		case RTA_PREFSRC:
786 			cfg->fc_prefsrc = nla_get_be32(attr);
787 			break;
788 		case RTA_METRICS:
789 			cfg->fc_mx = nla_data(attr);
790 			cfg->fc_mx_len = nla_len(attr);
791 			break;
792 		case RTA_MULTIPATH:
793 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
794 							     nla_len(attr),
795 							     extack);
796 			if (err < 0)
797 				goto errout;
798 			cfg->fc_mp = nla_data(attr);
799 			cfg->fc_mp_len = nla_len(attr);
800 			break;
801 		case RTA_FLOW:
802 			cfg->fc_flow = nla_get_u32(attr);
803 			break;
804 		case RTA_TABLE:
805 			cfg->fc_table = nla_get_u32(attr);
806 			break;
807 		case RTA_ENCAP:
808 			cfg->fc_encap = attr;
809 			break;
810 		case RTA_ENCAP_TYPE:
811 			cfg->fc_encap_type = nla_get_u16(attr);
812 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
813 							extack);
814 			if (err < 0)
815 				goto errout;
816 			break;
817 		case RTA_NH_ID:
818 			cfg->fc_nh_id = nla_get_u32(attr);
819 			break;
820 		}
821 	}
822 
823 	if (cfg->fc_nh_id) {
824 		if (cfg->fc_oif || cfg->fc_gw_family ||
825 		    cfg->fc_encap || cfg->fc_mp) {
826 			NL_SET_ERR_MSG(extack,
827 				       "Nexthop specification and nexthop id are mutually exclusive");
828 			return -EINVAL;
829 		}
830 	}
831 
832 	if (has_gw && has_via) {
833 		NL_SET_ERR_MSG(extack,
834 			       "Nexthop configuration can not contain both GATEWAY and VIA");
835 		goto errout;
836 	}
837 
838 	return 0;
839 errout:
840 	return err;
841 }
842 
843 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
844 			     struct netlink_ext_ack *extack)
845 {
846 	struct net *net = sock_net(skb->sk);
847 	struct fib_config cfg;
848 	struct fib_table *tb;
849 	int err;
850 
851 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
852 	if (err < 0)
853 		goto errout;
854 
855 	if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
856 		NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
857 		err = -EINVAL;
858 		goto errout;
859 	}
860 
861 	tb = fib_get_table(net, cfg.fc_table);
862 	if (!tb) {
863 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
864 		err = -ESRCH;
865 		goto errout;
866 	}
867 
868 	err = fib_table_delete(net, tb, &cfg, extack);
869 errout:
870 	return err;
871 }
872 
873 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
874 			     struct netlink_ext_ack *extack)
875 {
876 	struct net *net = sock_net(skb->sk);
877 	struct fib_config cfg;
878 	struct fib_table *tb;
879 	int err;
880 
881 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
882 	if (err < 0)
883 		goto errout;
884 
885 	tb = fib_new_table(net, cfg.fc_table);
886 	if (!tb) {
887 		err = -ENOBUFS;
888 		goto errout;
889 	}
890 
891 	err = fib_table_insert(net, tb, &cfg, extack);
892 	if (!err && cfg.fc_type == RTN_LOCAL)
893 		net->ipv4.fib_has_custom_local_routes = true;
894 errout:
895 	return err;
896 }
897 
898 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
899 			  struct fib_dump_filter *filter,
900 			  struct netlink_callback *cb)
901 {
902 	struct netlink_ext_ack *extack = cb->extack;
903 	struct nlattr *tb[RTA_MAX + 1];
904 	struct rtmsg *rtm;
905 	int err, i;
906 
907 	ASSERT_RTNL();
908 
909 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
910 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
911 		return -EINVAL;
912 	}
913 
914 	rtm = nlmsg_data(nlh);
915 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
916 	    rtm->rtm_scope) {
917 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
918 		return -EINVAL;
919 	}
920 
921 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
922 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
923 		return -EINVAL;
924 	}
925 	if (rtm->rtm_flags & RTM_F_CLONED)
926 		filter->dump_routes = false;
927 	else
928 		filter->dump_exceptions = false;
929 
930 	filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
931 	filter->flags    = rtm->rtm_flags;
932 	filter->protocol = rtm->rtm_protocol;
933 	filter->rt_type  = rtm->rtm_type;
934 	filter->table_id = rtm->rtm_table;
935 
936 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
937 					    rtm_ipv4_policy, extack);
938 	if (err < 0)
939 		return err;
940 
941 	for (i = 0; i <= RTA_MAX; ++i) {
942 		int ifindex;
943 
944 		if (!tb[i])
945 			continue;
946 
947 		switch (i) {
948 		case RTA_TABLE:
949 			filter->table_id = nla_get_u32(tb[i]);
950 			break;
951 		case RTA_OIF:
952 			ifindex = nla_get_u32(tb[i]);
953 			filter->dev = __dev_get_by_index(net, ifindex);
954 			if (!filter->dev)
955 				return -ENODEV;
956 			break;
957 		default:
958 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
959 			return -EINVAL;
960 		}
961 	}
962 
963 	if (filter->flags || filter->protocol || filter->rt_type ||
964 	    filter->table_id || filter->dev) {
965 		filter->filter_set = 1;
966 		cb->answer_flags = NLM_F_DUMP_FILTERED;
967 	}
968 
969 	return 0;
970 }
971 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
972 
973 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
974 {
975 	struct fib_dump_filter filter = { .dump_routes = true,
976 					  .dump_exceptions = true };
977 	const struct nlmsghdr *nlh = cb->nlh;
978 	struct net *net = sock_net(skb->sk);
979 	unsigned int h, s_h;
980 	unsigned int e = 0, s_e;
981 	struct fib_table *tb;
982 	struct hlist_head *head;
983 	int dumped = 0, err;
984 
985 	if (cb->strict_check) {
986 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
987 		if (err < 0)
988 			return err;
989 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
990 		struct rtmsg *rtm = nlmsg_data(nlh);
991 
992 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
993 	}
994 
995 	/* ipv4 does not use prefix flag */
996 	if (filter.flags & RTM_F_PREFIX)
997 		return skb->len;
998 
999 	if (filter.table_id) {
1000 		tb = fib_get_table(net, filter.table_id);
1001 		if (!tb) {
1002 			if (filter.dump_all_families)
1003 				return skb->len;
1004 
1005 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1006 			return -ENOENT;
1007 		}
1008 
1009 		err = fib_table_dump(tb, skb, cb, &filter);
1010 		return skb->len ? : err;
1011 	}
1012 
1013 	s_h = cb->args[0];
1014 	s_e = cb->args[1];
1015 
1016 	rcu_read_lock();
1017 
1018 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1019 		e = 0;
1020 		head = &net->ipv4.fib_table_hash[h];
1021 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1022 			if (e < s_e)
1023 				goto next;
1024 			if (dumped)
1025 				memset(&cb->args[2], 0, sizeof(cb->args) -
1026 						 2 * sizeof(cb->args[0]));
1027 			err = fib_table_dump(tb, skb, cb, &filter);
1028 			if (err < 0) {
1029 				if (likely(skb->len))
1030 					goto out;
1031 
1032 				goto out_err;
1033 			}
1034 			dumped = 1;
1035 next:
1036 			e++;
1037 		}
1038 	}
1039 out:
1040 	err = skb->len;
1041 out_err:
1042 	rcu_read_unlock();
1043 
1044 	cb->args[1] = e;
1045 	cb->args[0] = h;
1046 
1047 	return err;
1048 }
1049 
1050 /* Prepare and feed intra-kernel routing request.
1051  * Really, it should be netlink message, but :-( netlink
1052  * can be not configured, so that we feed it directly
1053  * to fib engine. It is legal, because all events occur
1054  * only when netlink is already locked.
1055  */
1056 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1057 		      struct in_ifaddr *ifa, u32 rt_priority)
1058 {
1059 	struct net *net = dev_net(ifa->ifa_dev->dev);
1060 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1061 	struct fib_table *tb;
1062 	struct fib_config cfg = {
1063 		.fc_protocol = RTPROT_KERNEL,
1064 		.fc_type = type,
1065 		.fc_dst = dst,
1066 		.fc_dst_len = dst_len,
1067 		.fc_priority = rt_priority,
1068 		.fc_prefsrc = ifa->ifa_local,
1069 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1070 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1071 		.fc_nlinfo = {
1072 			.nl_net = net,
1073 		},
1074 	};
1075 
1076 	if (!tb_id)
1077 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1078 
1079 	tb = fib_new_table(net, tb_id);
1080 	if (!tb)
1081 		return;
1082 
1083 	cfg.fc_table = tb->tb_id;
1084 
1085 	if (type != RTN_LOCAL)
1086 		cfg.fc_scope = RT_SCOPE_LINK;
1087 	else
1088 		cfg.fc_scope = RT_SCOPE_HOST;
1089 
1090 	if (cmd == RTM_NEWROUTE)
1091 		fib_table_insert(net, tb, &cfg, NULL);
1092 	else
1093 		fib_table_delete(net, tb, &cfg, NULL);
1094 }
1095 
1096 void fib_add_ifaddr(struct in_ifaddr *ifa)
1097 {
1098 	struct in_device *in_dev = ifa->ifa_dev;
1099 	struct net_device *dev = in_dev->dev;
1100 	struct in_ifaddr *prim = ifa;
1101 	__be32 mask = ifa->ifa_mask;
1102 	__be32 addr = ifa->ifa_local;
1103 	__be32 prefix = ifa->ifa_address & mask;
1104 
1105 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1106 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1107 		if (!prim) {
1108 			pr_warn("%s: bug: prim == NULL\n", __func__);
1109 			return;
1110 		}
1111 	}
1112 
1113 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1114 
1115 	if (!(dev->flags & IFF_UP))
1116 		return;
1117 
1118 	/* Add broadcast address, if it is explicitly assigned. */
1119 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1120 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1121 			  prim, 0);
1122 
1123 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1124 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1125 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1126 			fib_magic(RTM_NEWROUTE,
1127 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1128 				  prefix, ifa->ifa_prefixlen, prim,
1129 				  ifa->ifa_rt_priority);
1130 
1131 		/* Add network specific broadcasts, when it takes a sense */
1132 		if (ifa->ifa_prefixlen < 31) {
1133 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1134 				  prim, 0);
1135 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1136 				  32, prim, 0);
1137 		}
1138 	}
1139 }
1140 
1141 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1142 {
1143 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1144 	struct in_device *in_dev = ifa->ifa_dev;
1145 	struct net_device *dev = in_dev->dev;
1146 
1147 	if (!(dev->flags & IFF_UP) ||
1148 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1149 	    ipv4_is_zeronet(prefix) ||
1150 	    prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1151 		return;
1152 
1153 	/* add the new */
1154 	fib_magic(RTM_NEWROUTE,
1155 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1156 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1157 
1158 	/* delete the old */
1159 	fib_magic(RTM_DELROUTE,
1160 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1161 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1162 }
1163 
1164 /* Delete primary or secondary address.
1165  * Optionally, on secondary address promotion consider the addresses
1166  * from subnet iprim as deleted, even if they are in device list.
1167  * In this case the secondary ifa can be in device list.
1168  */
1169 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1170 {
1171 	struct in_device *in_dev = ifa->ifa_dev;
1172 	struct net_device *dev = in_dev->dev;
1173 	struct in_ifaddr *ifa1;
1174 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1175 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1176 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1177 #define LOCAL_OK	1
1178 #define BRD_OK		2
1179 #define BRD0_OK		4
1180 #define BRD1_OK		8
1181 	unsigned int ok = 0;
1182 	int subnet = 0;		/* Primary network */
1183 	int gone = 1;		/* Address is missing */
1184 	int same_prefsrc = 0;	/* Another primary with same IP */
1185 
1186 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1187 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1188 		if (!prim) {
1189 			/* if the device has been deleted, we don't perform
1190 			 * address promotion
1191 			 */
1192 			if (!in_dev->dead)
1193 				pr_warn("%s: bug: prim == NULL\n", __func__);
1194 			return;
1195 		}
1196 		if (iprim && iprim != prim) {
1197 			pr_warn("%s: bug: iprim != prim\n", __func__);
1198 			return;
1199 		}
1200 	} else if (!ipv4_is_zeronet(any) &&
1201 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1202 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1203 			fib_magic(RTM_DELROUTE,
1204 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1205 				  any, ifa->ifa_prefixlen, prim, 0);
1206 		subnet = 1;
1207 	}
1208 
1209 	if (in_dev->dead)
1210 		goto no_promotions;
1211 
1212 	/* Deletion is more complicated than add.
1213 	 * We should take care of not to delete too much :-)
1214 	 *
1215 	 * Scan address list to be sure that addresses are really gone.
1216 	 */
1217 	rcu_read_lock();
1218 	in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1219 		if (ifa1 == ifa) {
1220 			/* promotion, keep the IP */
1221 			gone = 0;
1222 			continue;
1223 		}
1224 		/* Ignore IFAs from our subnet */
1225 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1226 		    inet_ifa_match(ifa1->ifa_address, iprim))
1227 			continue;
1228 
1229 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1230 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1231 			/* Another address from our subnet? */
1232 			if (ifa1->ifa_mask == prim->ifa_mask &&
1233 			    inet_ifa_match(ifa1->ifa_address, prim))
1234 				prim1 = prim;
1235 			else {
1236 				/* We reached the secondaries, so
1237 				 * same_prefsrc should be determined.
1238 				 */
1239 				if (!same_prefsrc)
1240 					continue;
1241 				/* Search new prim1 if ifa1 is not
1242 				 * using the current prim1
1243 				 */
1244 				if (!prim1 ||
1245 				    ifa1->ifa_mask != prim1->ifa_mask ||
1246 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1247 					prim1 = inet_ifa_byprefix(in_dev,
1248 							ifa1->ifa_address,
1249 							ifa1->ifa_mask);
1250 				if (!prim1)
1251 					continue;
1252 				if (prim1->ifa_local != prim->ifa_local)
1253 					continue;
1254 			}
1255 		} else {
1256 			if (prim->ifa_local != ifa1->ifa_local)
1257 				continue;
1258 			prim1 = ifa1;
1259 			if (prim != prim1)
1260 				same_prefsrc = 1;
1261 		}
1262 		if (ifa->ifa_local == ifa1->ifa_local)
1263 			ok |= LOCAL_OK;
1264 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1265 			ok |= BRD_OK;
1266 		if (brd == ifa1->ifa_broadcast)
1267 			ok |= BRD1_OK;
1268 		if (any == ifa1->ifa_broadcast)
1269 			ok |= BRD0_OK;
1270 		/* primary has network specific broadcasts */
1271 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1272 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1273 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1274 
1275 			if (!ipv4_is_zeronet(any1)) {
1276 				if (ifa->ifa_broadcast == brd1 ||
1277 				    ifa->ifa_broadcast == any1)
1278 					ok |= BRD_OK;
1279 				if (brd == brd1 || brd == any1)
1280 					ok |= BRD1_OK;
1281 				if (any == brd1 || any == any1)
1282 					ok |= BRD0_OK;
1283 			}
1284 		}
1285 	}
1286 	rcu_read_unlock();
1287 
1288 no_promotions:
1289 	if (!(ok & BRD_OK))
1290 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1291 			  prim, 0);
1292 	if (subnet && ifa->ifa_prefixlen < 31) {
1293 		if (!(ok & BRD1_OK))
1294 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1295 				  prim, 0);
1296 		if (!(ok & BRD0_OK))
1297 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1298 				  prim, 0);
1299 	}
1300 	if (!(ok & LOCAL_OK)) {
1301 		unsigned int addr_type;
1302 
1303 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1304 
1305 		/* Check, that this local address finally disappeared. */
1306 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1307 						     ifa->ifa_local);
1308 		if (gone && addr_type != RTN_LOCAL) {
1309 			/* And the last, but not the least thing.
1310 			 * We must flush stray FIB entries.
1311 			 *
1312 			 * First of all, we scan fib_info list searching
1313 			 * for stray nexthop entries, then ignite fib_flush.
1314 			 */
1315 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1316 				fib_flush(dev_net(dev));
1317 		}
1318 	}
1319 #undef LOCAL_OK
1320 #undef BRD_OK
1321 #undef BRD0_OK
1322 #undef BRD1_OK
1323 }
1324 
1325 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1326 {
1327 
1328 	struct fib_result       res;
1329 	struct flowi4           fl4 = {
1330 		.flowi4_mark = frn->fl_mark,
1331 		.daddr = frn->fl_addr,
1332 		.flowi4_tos = frn->fl_tos,
1333 		.flowi4_scope = frn->fl_scope,
1334 	};
1335 	struct fib_table *tb;
1336 
1337 	rcu_read_lock();
1338 
1339 	tb = fib_get_table(net, frn->tb_id_in);
1340 
1341 	frn->err = -ENOENT;
1342 	if (tb) {
1343 		local_bh_disable();
1344 
1345 		frn->tb_id = tb->tb_id;
1346 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1347 
1348 		if (!frn->err) {
1349 			frn->prefixlen = res.prefixlen;
1350 			frn->nh_sel = res.nh_sel;
1351 			frn->type = res.type;
1352 			frn->scope = res.scope;
1353 		}
1354 		local_bh_enable();
1355 	}
1356 
1357 	rcu_read_unlock();
1358 }
1359 
1360 static void nl_fib_input(struct sk_buff *skb)
1361 {
1362 	struct net *net;
1363 	struct fib_result_nl *frn;
1364 	struct nlmsghdr *nlh;
1365 	u32 portid;
1366 
1367 	net = sock_net(skb->sk);
1368 	nlh = nlmsg_hdr(skb);
1369 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1370 	    skb->len < nlh->nlmsg_len ||
1371 	    nlmsg_len(nlh) < sizeof(*frn))
1372 		return;
1373 
1374 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1375 	if (!skb)
1376 		return;
1377 	nlh = nlmsg_hdr(skb);
1378 
1379 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1380 	nl_fib_lookup(net, frn);
1381 
1382 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1383 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1384 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1385 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1386 }
1387 
1388 static int __net_init nl_fib_lookup_init(struct net *net)
1389 {
1390 	struct sock *sk;
1391 	struct netlink_kernel_cfg cfg = {
1392 		.input	= nl_fib_input,
1393 	};
1394 
1395 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1396 	if (!sk)
1397 		return -EAFNOSUPPORT;
1398 	net->ipv4.fibnl = sk;
1399 	return 0;
1400 }
1401 
1402 static void nl_fib_lookup_exit(struct net *net)
1403 {
1404 	netlink_kernel_release(net->ipv4.fibnl);
1405 	net->ipv4.fibnl = NULL;
1406 }
1407 
1408 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1409 			   bool force)
1410 {
1411 	if (fib_sync_down_dev(dev, event, force))
1412 		fib_flush(dev_net(dev));
1413 	else
1414 		rt_cache_flush(dev_net(dev));
1415 	arp_ifdown(dev);
1416 }
1417 
1418 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1419 {
1420 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1421 	struct net_device *dev = ifa->ifa_dev->dev;
1422 	struct net *net = dev_net(dev);
1423 
1424 	switch (event) {
1425 	case NETDEV_UP:
1426 		fib_add_ifaddr(ifa);
1427 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1428 		fib_sync_up(dev, RTNH_F_DEAD);
1429 #endif
1430 		atomic_inc(&net->ipv4.dev_addr_genid);
1431 		rt_cache_flush(dev_net(dev));
1432 		break;
1433 	case NETDEV_DOWN:
1434 		fib_del_ifaddr(ifa, NULL);
1435 		atomic_inc(&net->ipv4.dev_addr_genid);
1436 		if (!ifa->ifa_dev->ifa_list) {
1437 			/* Last address was deleted from this interface.
1438 			 * Disable IP.
1439 			 */
1440 			fib_disable_ip(dev, event, true);
1441 		} else {
1442 			rt_cache_flush(dev_net(dev));
1443 		}
1444 		break;
1445 	}
1446 	return NOTIFY_DONE;
1447 }
1448 
1449 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1450 {
1451 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1452 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1453 	struct netdev_notifier_info_ext *info_ext = ptr;
1454 	struct in_device *in_dev;
1455 	struct net *net = dev_net(dev);
1456 	struct in_ifaddr *ifa;
1457 	unsigned int flags;
1458 
1459 	if (event == NETDEV_UNREGISTER) {
1460 		fib_disable_ip(dev, event, true);
1461 		rt_flush_dev(dev);
1462 		return NOTIFY_DONE;
1463 	}
1464 
1465 	in_dev = __in_dev_get_rtnl(dev);
1466 	if (!in_dev)
1467 		return NOTIFY_DONE;
1468 
1469 	switch (event) {
1470 	case NETDEV_UP:
1471 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1472 			fib_add_ifaddr(ifa);
1473 		}
1474 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1475 		fib_sync_up(dev, RTNH_F_DEAD);
1476 #endif
1477 		atomic_inc(&net->ipv4.dev_addr_genid);
1478 		rt_cache_flush(net);
1479 		break;
1480 	case NETDEV_DOWN:
1481 		fib_disable_ip(dev, event, false);
1482 		break;
1483 	case NETDEV_CHANGE:
1484 		flags = dev_get_flags(dev);
1485 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1486 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1487 		else
1488 			fib_sync_down_dev(dev, event, false);
1489 		rt_cache_flush(net);
1490 		break;
1491 	case NETDEV_CHANGEMTU:
1492 		fib_sync_mtu(dev, info_ext->ext.mtu);
1493 		rt_cache_flush(net);
1494 		break;
1495 	case NETDEV_CHANGEUPPER:
1496 		upper_info = ptr;
1497 		/* flush all routes if dev is linked to or unlinked from
1498 		 * an L3 master device (e.g., VRF)
1499 		 */
1500 		if (upper_info->upper_dev &&
1501 		    netif_is_l3_master(upper_info->upper_dev))
1502 			fib_disable_ip(dev, NETDEV_DOWN, true);
1503 		break;
1504 	}
1505 	return NOTIFY_DONE;
1506 }
1507 
1508 static struct notifier_block fib_inetaddr_notifier = {
1509 	.notifier_call = fib_inetaddr_event,
1510 };
1511 
1512 static struct notifier_block fib_netdev_notifier = {
1513 	.notifier_call = fib_netdev_event,
1514 };
1515 
1516 static int __net_init ip_fib_net_init(struct net *net)
1517 {
1518 	int err;
1519 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1520 
1521 	err = fib4_notifier_init(net);
1522 	if (err)
1523 		return err;
1524 
1525 	/* Avoid false sharing : Use at least a full cache line */
1526 	size = max_t(size_t, size, L1_CACHE_BYTES);
1527 
1528 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1529 	if (!net->ipv4.fib_table_hash) {
1530 		err = -ENOMEM;
1531 		goto err_table_hash_alloc;
1532 	}
1533 
1534 	err = fib4_rules_init(net);
1535 	if (err < 0)
1536 		goto err_rules_init;
1537 	return 0;
1538 
1539 err_rules_init:
1540 	kfree(net->ipv4.fib_table_hash);
1541 err_table_hash_alloc:
1542 	fib4_notifier_exit(net);
1543 	return err;
1544 }
1545 
1546 static void ip_fib_net_exit(struct net *net)
1547 {
1548 	int i;
1549 
1550 	rtnl_lock();
1551 #ifdef CONFIG_IP_MULTIPLE_TABLES
1552 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1553 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1554 #endif
1555 	/* Destroy the tables in reverse order to guarantee that the
1556 	 * local table, ID 255, is destroyed before the main table, ID
1557 	 * 254. This is necessary as the local table may contain
1558 	 * references to data contained in the main table.
1559 	 */
1560 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1561 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1562 		struct hlist_node *tmp;
1563 		struct fib_table *tb;
1564 
1565 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1566 			hlist_del(&tb->tb_hlist);
1567 			fib_table_flush(net, tb, true);
1568 			fib_free_table(tb);
1569 		}
1570 	}
1571 
1572 #ifdef CONFIG_IP_MULTIPLE_TABLES
1573 	fib4_rules_exit(net);
1574 #endif
1575 	rtnl_unlock();
1576 	kfree(net->ipv4.fib_table_hash);
1577 	fib4_notifier_exit(net);
1578 }
1579 
1580 static int __net_init fib_net_init(struct net *net)
1581 {
1582 	int error;
1583 
1584 #ifdef CONFIG_IP_ROUTE_CLASSID
1585 	net->ipv4.fib_num_tclassid_users = 0;
1586 #endif
1587 	error = ip_fib_net_init(net);
1588 	if (error < 0)
1589 		goto out;
1590 	error = nl_fib_lookup_init(net);
1591 	if (error < 0)
1592 		goto out_nlfl;
1593 	error = fib_proc_init(net);
1594 	if (error < 0)
1595 		goto out_proc;
1596 out:
1597 	return error;
1598 
1599 out_proc:
1600 	nl_fib_lookup_exit(net);
1601 out_nlfl:
1602 	ip_fib_net_exit(net);
1603 	goto out;
1604 }
1605 
1606 static void __net_exit fib_net_exit(struct net *net)
1607 {
1608 	fib_proc_exit(net);
1609 	nl_fib_lookup_exit(net);
1610 	ip_fib_net_exit(net);
1611 }
1612 
1613 static struct pernet_operations fib_net_ops = {
1614 	.init = fib_net_init,
1615 	.exit = fib_net_exit,
1616 };
1617 
1618 void __init ip_fib_init(void)
1619 {
1620 	fib_trie_init();
1621 
1622 	register_pernet_subsys(&fib_net_ops);
1623 
1624 	register_netdevice_notifier(&fib_netdev_notifier);
1625 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1626 
1627 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1628 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1629 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1630 }
1631