xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision 2fa49589)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		IPv4 Forwarding Information Base: FIB frontend.
7  *
8  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9  *
10  *		This program is free software; you can redistribute it and/or
11  *		modify it under the terms of the GNU General Public License
12  *		as published by the Free Software Foundation; either version
13  *		2 of the License, or (at your option) any later version.
14  */
15 
16 #include <linux/module.h>
17 #include <linux/uaccess.h>
18 #include <linux/bitops.h>
19 #include <linux/capability.h>
20 #include <linux/types.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
27 #include <linux/in.h>
28 #include <linux/inet.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_addr.h>
32 #include <linux/if_arp.h>
33 #include <linux/skbuff.h>
34 #include <linux/cache.h>
35 #include <linux/init.h>
36 #include <linux/list.h>
37 #include <linux/slab.h>
38 
39 #include <net/ip.h>
40 #include <net/protocol.h>
41 #include <net/route.h>
42 #include <net/tcp.h>
43 #include <net/sock.h>
44 #include <net/arp.h>
45 #include <net/ip_fib.h>
46 #include <net/rtnetlink.h>
47 #include <net/xfrm.h>
48 #include <net/l3mdev.h>
49 #include <net/lwtunnel.h>
50 #include <trace/events/fib.h>
51 
52 #ifndef CONFIG_IP_MULTIPLE_TABLES
53 
54 static int __net_init fib4_rules_init(struct net *net)
55 {
56 	struct fib_table *local_table, *main_table;
57 
58 	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
59 	if (!main_table)
60 		return -ENOMEM;
61 
62 	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
63 	if (!local_table)
64 		goto fail;
65 
66 	hlist_add_head_rcu(&local_table->tb_hlist,
67 				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
68 	hlist_add_head_rcu(&main_table->tb_hlist,
69 				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
70 	return 0;
71 
72 fail:
73 	fib_free_table(main_table);
74 	return -ENOMEM;
75 }
76 
77 static bool fib4_has_custom_rules(struct net *net)
78 {
79 	return false;
80 }
81 #else
82 
83 struct fib_table *fib_new_table(struct net *net, u32 id)
84 {
85 	struct fib_table *tb, *alias = NULL;
86 	unsigned int h;
87 
88 	if (id == 0)
89 		id = RT_TABLE_MAIN;
90 	tb = fib_get_table(net, id);
91 	if (tb)
92 		return tb;
93 
94 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
95 		alias = fib_new_table(net, RT_TABLE_MAIN);
96 
97 	tb = fib_trie_table(id, alias);
98 	if (!tb)
99 		return NULL;
100 
101 	switch (id) {
102 	case RT_TABLE_MAIN:
103 		rcu_assign_pointer(net->ipv4.fib_main, tb);
104 		break;
105 	case RT_TABLE_DEFAULT:
106 		rcu_assign_pointer(net->ipv4.fib_default, tb);
107 		break;
108 	default:
109 		break;
110 	}
111 
112 	h = id & (FIB_TABLE_HASHSZ - 1);
113 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
114 	return tb;
115 }
116 EXPORT_SYMBOL_GPL(fib_new_table);
117 
118 /* caller must hold either rtnl or rcu read lock */
119 struct fib_table *fib_get_table(struct net *net, u32 id)
120 {
121 	struct fib_table *tb;
122 	struct hlist_head *head;
123 	unsigned int h;
124 
125 	if (id == 0)
126 		id = RT_TABLE_MAIN;
127 	h = id & (FIB_TABLE_HASHSZ - 1);
128 
129 	head = &net->ipv4.fib_table_hash[h];
130 	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
131 		if (tb->tb_id == id)
132 			return tb;
133 	}
134 	return NULL;
135 }
136 
137 static bool fib4_has_custom_rules(struct net *net)
138 {
139 	return net->ipv4.fib_has_custom_rules;
140 }
141 #endif /* CONFIG_IP_MULTIPLE_TABLES */
142 
143 static void fib_replace_table(struct net *net, struct fib_table *old,
144 			      struct fib_table *new)
145 {
146 #ifdef CONFIG_IP_MULTIPLE_TABLES
147 	switch (new->tb_id) {
148 	case RT_TABLE_MAIN:
149 		rcu_assign_pointer(net->ipv4.fib_main, new);
150 		break;
151 	case RT_TABLE_DEFAULT:
152 		rcu_assign_pointer(net->ipv4.fib_default, new);
153 		break;
154 	default:
155 		break;
156 	}
157 
158 #endif
159 	/* replace the old table in the hlist */
160 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
161 }
162 
163 int fib_unmerge(struct net *net)
164 {
165 	struct fib_table *old, *new, *main_table;
166 
167 	/* attempt to fetch local table if it has been allocated */
168 	old = fib_get_table(net, RT_TABLE_LOCAL);
169 	if (!old)
170 		return 0;
171 
172 	new = fib_trie_unmerge(old);
173 	if (!new)
174 		return -ENOMEM;
175 
176 	/* table is already unmerged */
177 	if (new == old)
178 		return 0;
179 
180 	/* replace merged table with clean table */
181 	fib_replace_table(net, old, new);
182 	fib_free_table(old);
183 
184 	/* attempt to fetch main table if it has been allocated */
185 	main_table = fib_get_table(net, RT_TABLE_MAIN);
186 	if (!main_table)
187 		return 0;
188 
189 	/* flush local entries from main table */
190 	fib_table_flush_external(main_table);
191 
192 	return 0;
193 }
194 
195 static void fib_flush(struct net *net)
196 {
197 	int flushed = 0;
198 	unsigned int h;
199 
200 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
201 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
202 		struct hlist_node *tmp;
203 		struct fib_table *tb;
204 
205 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
206 			flushed += fib_table_flush(net, tb, false);
207 	}
208 
209 	if (flushed)
210 		rt_cache_flush(net);
211 }
212 
213 /*
214  * Find address type as if only "dev" was present in the system. If
215  * on_dev is NULL then all interfaces are taken into consideration.
216  */
217 static inline unsigned int __inet_dev_addr_type(struct net *net,
218 						const struct net_device *dev,
219 						__be32 addr, u32 tb_id)
220 {
221 	struct flowi4		fl4 = { .daddr = addr };
222 	struct fib_result	res;
223 	unsigned int ret = RTN_BROADCAST;
224 	struct fib_table *table;
225 
226 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
227 		return RTN_BROADCAST;
228 	if (ipv4_is_multicast(addr))
229 		return RTN_MULTICAST;
230 
231 	rcu_read_lock();
232 
233 	table = fib_get_table(net, tb_id);
234 	if (table) {
235 		ret = RTN_UNICAST;
236 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
237 			if (!dev || dev == res.fi->fib_dev)
238 				ret = res.type;
239 		}
240 	}
241 
242 	rcu_read_unlock();
243 	return ret;
244 }
245 
246 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
247 {
248 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
249 }
250 EXPORT_SYMBOL(inet_addr_type_table);
251 
252 unsigned int inet_addr_type(struct net *net, __be32 addr)
253 {
254 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
255 }
256 EXPORT_SYMBOL(inet_addr_type);
257 
258 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
259 				__be32 addr)
260 {
261 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
262 
263 	return __inet_dev_addr_type(net, dev, addr, rt_table);
264 }
265 EXPORT_SYMBOL(inet_dev_addr_type);
266 
267 /* inet_addr_type with dev == NULL but using the table from a dev
268  * if one is associated
269  */
270 unsigned int inet_addr_type_dev_table(struct net *net,
271 				      const struct net_device *dev,
272 				      __be32 addr)
273 {
274 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
275 
276 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
277 }
278 EXPORT_SYMBOL(inet_addr_type_dev_table);
279 
280 __be32 fib_compute_spec_dst(struct sk_buff *skb)
281 {
282 	struct net_device *dev = skb->dev;
283 	struct in_device *in_dev;
284 	struct fib_result res;
285 	struct rtable *rt;
286 	struct net *net;
287 	int scope;
288 
289 	rt = skb_rtable(skb);
290 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
291 	    RTCF_LOCAL)
292 		return ip_hdr(skb)->daddr;
293 
294 	in_dev = __in_dev_get_rcu(dev);
295 
296 	net = dev_net(dev);
297 
298 	scope = RT_SCOPE_UNIVERSE;
299 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
300 		bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
301 		struct flowi4 fl4 = {
302 			.flowi4_iif = LOOPBACK_IFINDEX,
303 			.flowi4_oif = l3mdev_master_ifindex_rcu(dev),
304 			.daddr = ip_hdr(skb)->saddr,
305 			.flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
306 			.flowi4_scope = scope,
307 			.flowi4_mark = vmark ? skb->mark : 0,
308 		};
309 		if (!fib_lookup(net, &fl4, &res, 0))
310 			return FIB_RES_PREFSRC(net, res);
311 	} else {
312 		scope = RT_SCOPE_LINK;
313 	}
314 
315 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
316 }
317 
318 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
319 {
320 	bool dev_match = false;
321 #ifdef CONFIG_IP_ROUTE_MULTIPATH
322 	int ret;
323 
324 	for (ret = 0; ret < fi->fib_nhs; ret++) {
325 		struct fib_nh *nh = &fi->fib_nh[ret];
326 
327 		if (nh->nh_dev == dev) {
328 			dev_match = true;
329 			break;
330 		} else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
331 			dev_match = true;
332 			break;
333 		}
334 	}
335 #else
336 	if (fi->fib_nh[0].nh_dev == dev)
337 		dev_match = true;
338 #endif
339 
340 	return dev_match;
341 }
342 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
343 
344 /* Given (packet source, input interface) and optional (dst, oif, tos):
345  * - (main) check, that source is valid i.e. not broadcast or our local
346  *   address.
347  * - figure out what "logical" interface this packet arrived
348  *   and calculate "specific destination" address.
349  * - check, that packet arrived from expected physical interface.
350  * called with rcu_read_lock()
351  */
352 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
353 				 u8 tos, int oif, struct net_device *dev,
354 				 int rpf, struct in_device *idev, u32 *itag)
355 {
356 	struct net *net = dev_net(dev);
357 	struct flow_keys flkeys;
358 	int ret, no_addr;
359 	struct fib_result res;
360 	struct flowi4 fl4;
361 	bool dev_match;
362 
363 	fl4.flowi4_oif = 0;
364 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
365 	if (!fl4.flowi4_iif)
366 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
367 	fl4.daddr = src;
368 	fl4.saddr = dst;
369 	fl4.flowi4_tos = tos;
370 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
371 	fl4.flowi4_tun_key.tun_id = 0;
372 	fl4.flowi4_flags = 0;
373 	fl4.flowi4_uid = sock_net_uid(net, NULL);
374 
375 	no_addr = idev->ifa_list == NULL;
376 
377 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
378 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
379 		fl4.flowi4_proto = 0;
380 		fl4.fl4_sport = 0;
381 		fl4.fl4_dport = 0;
382 	}
383 
384 	if (fib_lookup(net, &fl4, &res, 0))
385 		goto last_resort;
386 	if (res.type != RTN_UNICAST &&
387 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
388 		goto e_inval;
389 	fib_combine_itag(itag, &res);
390 
391 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
392 	if (dev_match) {
393 		ret = FIB_RES_NH(res).nh_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_NH(res).nh_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 			struct in_ifaddr *ifa;
544 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
545 			if (!in_dev)
546 				return -ENODEV;
547 			*colon = ':';
548 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
549 				if (strcmp(ifa->ifa_label, devname) == 0)
550 					break;
551 			if (!ifa)
552 				return -ENODEV;
553 			cfg->fc_prefsrc = ifa->ifa_local;
554 		}
555 	}
556 
557 	addr = sk_extract_addr(&rt->rt_gateway);
558 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
559 		unsigned int addr_type;
560 
561 		cfg->fc_gw = addr;
562 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
563 		if (rt->rt_flags & RTF_GATEWAY &&
564 		    addr_type == RTN_UNICAST)
565 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
566 	}
567 
568 	if (cmd == SIOCDELRT)
569 		return 0;
570 
571 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
572 		return -EINVAL;
573 
574 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
575 		cfg->fc_scope = RT_SCOPE_LINK;
576 
577 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
578 		struct nlattr *mx;
579 		int len = 0;
580 
581 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
582 		if (!mx)
583 			return -ENOMEM;
584 
585 		if (rt->rt_flags & RTF_MTU)
586 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
587 
588 		if (rt->rt_flags & RTF_WINDOW)
589 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
590 
591 		if (rt->rt_flags & RTF_IRTT)
592 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
593 
594 		cfg->fc_mx = mx;
595 		cfg->fc_mx_len = len;
596 	}
597 
598 	return 0;
599 }
600 
601 /*
602  * Handle IP routing ioctl calls.
603  * These are used to manipulate the routing tables
604  */
605 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
606 {
607 	struct fib_config cfg;
608 	int err;
609 
610 	switch (cmd) {
611 	case SIOCADDRT:		/* Add a route */
612 	case SIOCDELRT:		/* Delete a route */
613 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
614 			return -EPERM;
615 
616 		rtnl_lock();
617 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
618 		if (err == 0) {
619 			struct fib_table *tb;
620 
621 			if (cmd == SIOCDELRT) {
622 				tb = fib_get_table(net, cfg.fc_table);
623 				if (tb)
624 					err = fib_table_delete(net, tb, &cfg,
625 							       NULL);
626 				else
627 					err = -ESRCH;
628 			} else {
629 				tb = fib_new_table(net, cfg.fc_table);
630 				if (tb)
631 					err = fib_table_insert(net, tb,
632 							       &cfg, NULL);
633 				else
634 					err = -ENOBUFS;
635 			}
636 
637 			/* allocated by rtentry_to_fib_config() */
638 			kfree(cfg.fc_mx);
639 		}
640 		rtnl_unlock();
641 		return err;
642 	}
643 	return -EINVAL;
644 }
645 
646 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
647 	[RTA_DST]		= { .type = NLA_U32 },
648 	[RTA_SRC]		= { .type = NLA_U32 },
649 	[RTA_IIF]		= { .type = NLA_U32 },
650 	[RTA_OIF]		= { .type = NLA_U32 },
651 	[RTA_GATEWAY]		= { .type = NLA_U32 },
652 	[RTA_PRIORITY]		= { .type = NLA_U32 },
653 	[RTA_PREFSRC]		= { .type = NLA_U32 },
654 	[RTA_METRICS]		= { .type = NLA_NESTED },
655 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
656 	[RTA_FLOW]		= { .type = NLA_U32 },
657 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
658 	[RTA_ENCAP]		= { .type = NLA_NESTED },
659 	[RTA_UID]		= { .type = NLA_U32 },
660 	[RTA_MARK]		= { .type = NLA_U32 },
661 	[RTA_TABLE]		= { .type = NLA_U32 },
662 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
663 	[RTA_SPORT]		= { .type = NLA_U16 },
664 	[RTA_DPORT]		= { .type = NLA_U16 },
665 };
666 
667 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
668 			     struct nlmsghdr *nlh, struct fib_config *cfg,
669 			     struct netlink_ext_ack *extack)
670 {
671 	struct nlattr *attr;
672 	int err, remaining;
673 	struct rtmsg *rtm;
674 
675 	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
676 			     extack);
677 	if (err < 0)
678 		goto errout;
679 
680 	memset(cfg, 0, sizeof(*cfg));
681 
682 	rtm = nlmsg_data(nlh);
683 	cfg->fc_dst_len = rtm->rtm_dst_len;
684 	cfg->fc_tos = rtm->rtm_tos;
685 	cfg->fc_table = rtm->rtm_table;
686 	cfg->fc_protocol = rtm->rtm_protocol;
687 	cfg->fc_scope = rtm->rtm_scope;
688 	cfg->fc_type = rtm->rtm_type;
689 	cfg->fc_flags = rtm->rtm_flags;
690 	cfg->fc_nlflags = nlh->nlmsg_flags;
691 
692 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
693 	cfg->fc_nlinfo.nlh = nlh;
694 	cfg->fc_nlinfo.nl_net = net;
695 
696 	if (cfg->fc_type > RTN_MAX) {
697 		NL_SET_ERR_MSG(extack, "Invalid route type");
698 		err = -EINVAL;
699 		goto errout;
700 	}
701 
702 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
703 		switch (nla_type(attr)) {
704 		case RTA_DST:
705 			cfg->fc_dst = nla_get_be32(attr);
706 			break;
707 		case RTA_OIF:
708 			cfg->fc_oif = nla_get_u32(attr);
709 			break;
710 		case RTA_GATEWAY:
711 			cfg->fc_gw = nla_get_be32(attr);
712 			break;
713 		case RTA_PRIORITY:
714 			cfg->fc_priority = nla_get_u32(attr);
715 			break;
716 		case RTA_PREFSRC:
717 			cfg->fc_prefsrc = nla_get_be32(attr);
718 			break;
719 		case RTA_METRICS:
720 			cfg->fc_mx = nla_data(attr);
721 			cfg->fc_mx_len = nla_len(attr);
722 			break;
723 		case RTA_MULTIPATH:
724 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
725 							     nla_len(attr),
726 							     extack);
727 			if (err < 0)
728 				goto errout;
729 			cfg->fc_mp = nla_data(attr);
730 			cfg->fc_mp_len = nla_len(attr);
731 			break;
732 		case RTA_FLOW:
733 			cfg->fc_flow = nla_get_u32(attr);
734 			break;
735 		case RTA_TABLE:
736 			cfg->fc_table = nla_get_u32(attr);
737 			break;
738 		case RTA_ENCAP:
739 			cfg->fc_encap = attr;
740 			break;
741 		case RTA_ENCAP_TYPE:
742 			cfg->fc_encap_type = nla_get_u16(attr);
743 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
744 							extack);
745 			if (err < 0)
746 				goto errout;
747 			break;
748 		}
749 	}
750 
751 	return 0;
752 errout:
753 	return err;
754 }
755 
756 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
757 			     struct netlink_ext_ack *extack)
758 {
759 	struct net *net = sock_net(skb->sk);
760 	struct fib_config cfg;
761 	struct fib_table *tb;
762 	int err;
763 
764 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
765 	if (err < 0)
766 		goto errout;
767 
768 	tb = fib_get_table(net, cfg.fc_table);
769 	if (!tb) {
770 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
771 		err = -ESRCH;
772 		goto errout;
773 	}
774 
775 	err = fib_table_delete(net, tb, &cfg, extack);
776 errout:
777 	return err;
778 }
779 
780 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
781 			     struct netlink_ext_ack *extack)
782 {
783 	struct net *net = sock_net(skb->sk);
784 	struct fib_config cfg;
785 	struct fib_table *tb;
786 	int err;
787 
788 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
789 	if (err < 0)
790 		goto errout;
791 
792 	tb = fib_new_table(net, cfg.fc_table);
793 	if (!tb) {
794 		err = -ENOBUFS;
795 		goto errout;
796 	}
797 
798 	err = fib_table_insert(net, tb, &cfg, extack);
799 	if (!err && cfg.fc_type == RTN_LOCAL)
800 		net->ipv4.fib_has_custom_local_routes = true;
801 errout:
802 	return err;
803 }
804 
805 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
806 			  struct fib_dump_filter *filter,
807 			  struct netlink_callback *cb)
808 {
809 	struct netlink_ext_ack *extack = cb->extack;
810 	struct nlattr *tb[RTA_MAX + 1];
811 	struct rtmsg *rtm;
812 	int err, i;
813 
814 	ASSERT_RTNL();
815 
816 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
817 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
818 		return -EINVAL;
819 	}
820 
821 	rtm = nlmsg_data(nlh);
822 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
823 	    rtm->rtm_scope) {
824 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
825 		return -EINVAL;
826 	}
827 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
828 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
829 		return -EINVAL;
830 	}
831 
832 	filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
833 	filter->flags    = rtm->rtm_flags;
834 	filter->protocol = rtm->rtm_protocol;
835 	filter->rt_type  = rtm->rtm_type;
836 	filter->table_id = rtm->rtm_table;
837 
838 	err = nlmsg_parse_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
839 				 rtm_ipv4_policy, extack);
840 	if (err < 0)
841 		return err;
842 
843 	for (i = 0; i <= RTA_MAX; ++i) {
844 		int ifindex;
845 
846 		if (!tb[i])
847 			continue;
848 
849 		switch (i) {
850 		case RTA_TABLE:
851 			filter->table_id = nla_get_u32(tb[i]);
852 			break;
853 		case RTA_OIF:
854 			ifindex = nla_get_u32(tb[i]);
855 			filter->dev = __dev_get_by_index(net, ifindex);
856 			if (!filter->dev)
857 				return -ENODEV;
858 			break;
859 		default:
860 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
861 			return -EINVAL;
862 		}
863 	}
864 
865 	if (filter->flags || filter->protocol || filter->rt_type ||
866 	    filter->table_id || filter->dev) {
867 		filter->filter_set = 1;
868 		cb->answer_flags = NLM_F_DUMP_FILTERED;
869 	}
870 
871 	return 0;
872 }
873 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
874 
875 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
876 {
877 	const struct nlmsghdr *nlh = cb->nlh;
878 	struct net *net = sock_net(skb->sk);
879 	struct fib_dump_filter filter = {};
880 	unsigned int h, s_h;
881 	unsigned int e = 0, s_e;
882 	struct fib_table *tb;
883 	struct hlist_head *head;
884 	int dumped = 0, err;
885 
886 	if (cb->strict_check) {
887 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
888 		if (err < 0)
889 			return err;
890 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
891 		struct rtmsg *rtm = nlmsg_data(nlh);
892 
893 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
894 	}
895 
896 	/* fib entries are never clones and ipv4 does not use prefix flag */
897 	if (filter.flags & (RTM_F_PREFIX | RTM_F_CLONED))
898 		return skb->len;
899 
900 	if (filter.table_id) {
901 		tb = fib_get_table(net, filter.table_id);
902 		if (!tb) {
903 			if (filter.dump_all_families)
904 				return skb->len;
905 
906 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
907 			return -ENOENT;
908 		}
909 
910 		err = fib_table_dump(tb, skb, cb, &filter);
911 		return skb->len ? : err;
912 	}
913 
914 	s_h = cb->args[0];
915 	s_e = cb->args[1];
916 
917 	rcu_read_lock();
918 
919 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
920 		e = 0;
921 		head = &net->ipv4.fib_table_hash[h];
922 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
923 			if (e < s_e)
924 				goto next;
925 			if (dumped)
926 				memset(&cb->args[2], 0, sizeof(cb->args) -
927 						 2 * sizeof(cb->args[0]));
928 			err = fib_table_dump(tb, skb, cb, &filter);
929 			if (err < 0) {
930 				if (likely(skb->len))
931 					goto out;
932 
933 				goto out_err;
934 			}
935 			dumped = 1;
936 next:
937 			e++;
938 		}
939 	}
940 out:
941 	err = skb->len;
942 out_err:
943 	rcu_read_unlock();
944 
945 	cb->args[1] = e;
946 	cb->args[0] = h;
947 
948 	return err;
949 }
950 
951 /* Prepare and feed intra-kernel routing request.
952  * Really, it should be netlink message, but :-( netlink
953  * can be not configured, so that we feed it directly
954  * to fib engine. It is legal, because all events occur
955  * only when netlink is already locked.
956  */
957 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
958 		      struct in_ifaddr *ifa, u32 rt_priority)
959 {
960 	struct net *net = dev_net(ifa->ifa_dev->dev);
961 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
962 	struct fib_table *tb;
963 	struct fib_config cfg = {
964 		.fc_protocol = RTPROT_KERNEL,
965 		.fc_type = type,
966 		.fc_dst = dst,
967 		.fc_dst_len = dst_len,
968 		.fc_priority = rt_priority,
969 		.fc_prefsrc = ifa->ifa_local,
970 		.fc_oif = ifa->ifa_dev->dev->ifindex,
971 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
972 		.fc_nlinfo = {
973 			.nl_net = net,
974 		},
975 	};
976 
977 	if (!tb_id)
978 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
979 
980 	tb = fib_new_table(net, tb_id);
981 	if (!tb)
982 		return;
983 
984 	cfg.fc_table = tb->tb_id;
985 
986 	if (type != RTN_LOCAL)
987 		cfg.fc_scope = RT_SCOPE_LINK;
988 	else
989 		cfg.fc_scope = RT_SCOPE_HOST;
990 
991 	if (cmd == RTM_NEWROUTE)
992 		fib_table_insert(net, tb, &cfg, NULL);
993 	else
994 		fib_table_delete(net, tb, &cfg, NULL);
995 }
996 
997 void fib_add_ifaddr(struct in_ifaddr *ifa)
998 {
999 	struct in_device *in_dev = ifa->ifa_dev;
1000 	struct net_device *dev = in_dev->dev;
1001 	struct in_ifaddr *prim = ifa;
1002 	__be32 mask = ifa->ifa_mask;
1003 	__be32 addr = ifa->ifa_local;
1004 	__be32 prefix = ifa->ifa_address & mask;
1005 
1006 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1007 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1008 		if (!prim) {
1009 			pr_warn("%s: bug: prim == NULL\n", __func__);
1010 			return;
1011 		}
1012 	}
1013 
1014 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1015 
1016 	if (!(dev->flags & IFF_UP))
1017 		return;
1018 
1019 	/* Add broadcast address, if it is explicitly assigned. */
1020 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1021 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1022 			  prim, 0);
1023 
1024 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1025 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1026 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1027 			fib_magic(RTM_NEWROUTE,
1028 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1029 				  prefix, ifa->ifa_prefixlen, prim,
1030 				  ifa->ifa_rt_priority);
1031 
1032 		/* Add network specific broadcasts, when it takes a sense */
1033 		if (ifa->ifa_prefixlen < 31) {
1034 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1035 				  prim, 0);
1036 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1037 				  32, prim, 0);
1038 		}
1039 	}
1040 }
1041 
1042 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1043 {
1044 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1045 	struct in_device *in_dev = ifa->ifa_dev;
1046 	struct net_device *dev = in_dev->dev;
1047 
1048 	if (!(dev->flags & IFF_UP) ||
1049 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1050 	    ipv4_is_zeronet(prefix) ||
1051 	    prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1052 		return;
1053 
1054 	/* add the new */
1055 	fib_magic(RTM_NEWROUTE,
1056 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1057 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1058 
1059 	/* delete the old */
1060 	fib_magic(RTM_DELROUTE,
1061 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1062 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1063 }
1064 
1065 /* Delete primary or secondary address.
1066  * Optionally, on secondary address promotion consider the addresses
1067  * from subnet iprim as deleted, even if they are in device list.
1068  * In this case the secondary ifa can be in device list.
1069  */
1070 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1071 {
1072 	struct in_device *in_dev = ifa->ifa_dev;
1073 	struct net_device *dev = in_dev->dev;
1074 	struct in_ifaddr *ifa1;
1075 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1076 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1077 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1078 #define LOCAL_OK	1
1079 #define BRD_OK		2
1080 #define BRD0_OK		4
1081 #define BRD1_OK		8
1082 	unsigned int ok = 0;
1083 	int subnet = 0;		/* Primary network */
1084 	int gone = 1;		/* Address is missing */
1085 	int same_prefsrc = 0;	/* Another primary with same IP */
1086 
1087 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1088 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1089 		if (!prim) {
1090 			/* if the device has been deleted, we don't perform
1091 			 * address promotion
1092 			 */
1093 			if (!in_dev->dead)
1094 				pr_warn("%s: bug: prim == NULL\n", __func__);
1095 			return;
1096 		}
1097 		if (iprim && iprim != prim) {
1098 			pr_warn("%s: bug: iprim != prim\n", __func__);
1099 			return;
1100 		}
1101 	} else if (!ipv4_is_zeronet(any) &&
1102 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1103 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1104 			fib_magic(RTM_DELROUTE,
1105 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1106 				  any, ifa->ifa_prefixlen, prim, 0);
1107 		subnet = 1;
1108 	}
1109 
1110 	if (in_dev->dead)
1111 		goto no_promotions;
1112 
1113 	/* Deletion is more complicated than add.
1114 	 * We should take care of not to delete too much :-)
1115 	 *
1116 	 * Scan address list to be sure that addresses are really gone.
1117 	 */
1118 
1119 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
1120 		if (ifa1 == ifa) {
1121 			/* promotion, keep the IP */
1122 			gone = 0;
1123 			continue;
1124 		}
1125 		/* Ignore IFAs from our subnet */
1126 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1127 		    inet_ifa_match(ifa1->ifa_address, iprim))
1128 			continue;
1129 
1130 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1131 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1132 			/* Another address from our subnet? */
1133 			if (ifa1->ifa_mask == prim->ifa_mask &&
1134 			    inet_ifa_match(ifa1->ifa_address, prim))
1135 				prim1 = prim;
1136 			else {
1137 				/* We reached the secondaries, so
1138 				 * same_prefsrc should be determined.
1139 				 */
1140 				if (!same_prefsrc)
1141 					continue;
1142 				/* Search new prim1 if ifa1 is not
1143 				 * using the current prim1
1144 				 */
1145 				if (!prim1 ||
1146 				    ifa1->ifa_mask != prim1->ifa_mask ||
1147 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1148 					prim1 = inet_ifa_byprefix(in_dev,
1149 							ifa1->ifa_address,
1150 							ifa1->ifa_mask);
1151 				if (!prim1)
1152 					continue;
1153 				if (prim1->ifa_local != prim->ifa_local)
1154 					continue;
1155 			}
1156 		} else {
1157 			if (prim->ifa_local != ifa1->ifa_local)
1158 				continue;
1159 			prim1 = ifa1;
1160 			if (prim != prim1)
1161 				same_prefsrc = 1;
1162 		}
1163 		if (ifa->ifa_local == ifa1->ifa_local)
1164 			ok |= LOCAL_OK;
1165 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1166 			ok |= BRD_OK;
1167 		if (brd == ifa1->ifa_broadcast)
1168 			ok |= BRD1_OK;
1169 		if (any == ifa1->ifa_broadcast)
1170 			ok |= BRD0_OK;
1171 		/* primary has network specific broadcasts */
1172 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1173 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1174 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1175 
1176 			if (!ipv4_is_zeronet(any1)) {
1177 				if (ifa->ifa_broadcast == brd1 ||
1178 				    ifa->ifa_broadcast == any1)
1179 					ok |= BRD_OK;
1180 				if (brd == brd1 || brd == any1)
1181 					ok |= BRD1_OK;
1182 				if (any == brd1 || any == any1)
1183 					ok |= BRD0_OK;
1184 			}
1185 		}
1186 	}
1187 
1188 no_promotions:
1189 	if (!(ok & BRD_OK))
1190 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1191 			  prim, 0);
1192 	if (subnet && ifa->ifa_prefixlen < 31) {
1193 		if (!(ok & BRD1_OK))
1194 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1195 				  prim, 0);
1196 		if (!(ok & BRD0_OK))
1197 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1198 				  prim, 0);
1199 	}
1200 	if (!(ok & LOCAL_OK)) {
1201 		unsigned int addr_type;
1202 
1203 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1204 
1205 		/* Check, that this local address finally disappeared. */
1206 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1207 						     ifa->ifa_local);
1208 		if (gone && addr_type != RTN_LOCAL) {
1209 			/* And the last, but not the least thing.
1210 			 * We must flush stray FIB entries.
1211 			 *
1212 			 * First of all, we scan fib_info list searching
1213 			 * for stray nexthop entries, then ignite fib_flush.
1214 			 */
1215 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1216 				fib_flush(dev_net(dev));
1217 		}
1218 	}
1219 #undef LOCAL_OK
1220 #undef BRD_OK
1221 #undef BRD0_OK
1222 #undef BRD1_OK
1223 }
1224 
1225 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1226 {
1227 
1228 	struct fib_result       res;
1229 	struct flowi4           fl4 = {
1230 		.flowi4_mark = frn->fl_mark,
1231 		.daddr = frn->fl_addr,
1232 		.flowi4_tos = frn->fl_tos,
1233 		.flowi4_scope = frn->fl_scope,
1234 	};
1235 	struct fib_table *tb;
1236 
1237 	rcu_read_lock();
1238 
1239 	tb = fib_get_table(net, frn->tb_id_in);
1240 
1241 	frn->err = -ENOENT;
1242 	if (tb) {
1243 		local_bh_disable();
1244 
1245 		frn->tb_id = tb->tb_id;
1246 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1247 
1248 		if (!frn->err) {
1249 			frn->prefixlen = res.prefixlen;
1250 			frn->nh_sel = res.nh_sel;
1251 			frn->type = res.type;
1252 			frn->scope = res.scope;
1253 		}
1254 		local_bh_enable();
1255 	}
1256 
1257 	rcu_read_unlock();
1258 }
1259 
1260 static void nl_fib_input(struct sk_buff *skb)
1261 {
1262 	struct net *net;
1263 	struct fib_result_nl *frn;
1264 	struct nlmsghdr *nlh;
1265 	u32 portid;
1266 
1267 	net = sock_net(skb->sk);
1268 	nlh = nlmsg_hdr(skb);
1269 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1270 	    skb->len < nlh->nlmsg_len ||
1271 	    nlmsg_len(nlh) < sizeof(*frn))
1272 		return;
1273 
1274 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1275 	if (!skb)
1276 		return;
1277 	nlh = nlmsg_hdr(skb);
1278 
1279 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1280 	nl_fib_lookup(net, frn);
1281 
1282 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1283 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1284 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1285 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1286 }
1287 
1288 static int __net_init nl_fib_lookup_init(struct net *net)
1289 {
1290 	struct sock *sk;
1291 	struct netlink_kernel_cfg cfg = {
1292 		.input	= nl_fib_input,
1293 	};
1294 
1295 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1296 	if (!sk)
1297 		return -EAFNOSUPPORT;
1298 	net->ipv4.fibnl = sk;
1299 	return 0;
1300 }
1301 
1302 static void nl_fib_lookup_exit(struct net *net)
1303 {
1304 	netlink_kernel_release(net->ipv4.fibnl);
1305 	net->ipv4.fibnl = NULL;
1306 }
1307 
1308 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1309 			   bool force)
1310 {
1311 	if (fib_sync_down_dev(dev, event, force))
1312 		fib_flush(dev_net(dev));
1313 	else
1314 		rt_cache_flush(dev_net(dev));
1315 	arp_ifdown(dev);
1316 }
1317 
1318 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1319 {
1320 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1321 	struct net_device *dev = ifa->ifa_dev->dev;
1322 	struct net *net = dev_net(dev);
1323 
1324 	switch (event) {
1325 	case NETDEV_UP:
1326 		fib_add_ifaddr(ifa);
1327 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1328 		fib_sync_up(dev, RTNH_F_DEAD);
1329 #endif
1330 		atomic_inc(&net->ipv4.dev_addr_genid);
1331 		rt_cache_flush(dev_net(dev));
1332 		break;
1333 	case NETDEV_DOWN:
1334 		fib_del_ifaddr(ifa, NULL);
1335 		atomic_inc(&net->ipv4.dev_addr_genid);
1336 		if (!ifa->ifa_dev->ifa_list) {
1337 			/* Last address was deleted from this interface.
1338 			 * Disable IP.
1339 			 */
1340 			fib_disable_ip(dev, event, true);
1341 		} else {
1342 			rt_cache_flush(dev_net(dev));
1343 		}
1344 		break;
1345 	}
1346 	return NOTIFY_DONE;
1347 }
1348 
1349 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1350 {
1351 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1352 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1353 	struct netdev_notifier_info_ext *info_ext = ptr;
1354 	struct in_device *in_dev;
1355 	struct net *net = dev_net(dev);
1356 	unsigned int flags;
1357 
1358 	if (event == NETDEV_UNREGISTER) {
1359 		fib_disable_ip(dev, event, true);
1360 		rt_flush_dev(dev);
1361 		return NOTIFY_DONE;
1362 	}
1363 
1364 	in_dev = __in_dev_get_rtnl(dev);
1365 	if (!in_dev)
1366 		return NOTIFY_DONE;
1367 
1368 	switch (event) {
1369 	case NETDEV_UP:
1370 		for_ifa(in_dev) {
1371 			fib_add_ifaddr(ifa);
1372 		} endfor_ifa(in_dev);
1373 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1374 		fib_sync_up(dev, RTNH_F_DEAD);
1375 #endif
1376 		atomic_inc(&net->ipv4.dev_addr_genid);
1377 		rt_cache_flush(net);
1378 		break;
1379 	case NETDEV_DOWN:
1380 		fib_disable_ip(dev, event, false);
1381 		break;
1382 	case NETDEV_CHANGE:
1383 		flags = dev_get_flags(dev);
1384 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1385 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1386 		else
1387 			fib_sync_down_dev(dev, event, false);
1388 		rt_cache_flush(net);
1389 		break;
1390 	case NETDEV_CHANGEMTU:
1391 		fib_sync_mtu(dev, info_ext->ext.mtu);
1392 		rt_cache_flush(net);
1393 		break;
1394 	case NETDEV_CHANGEUPPER:
1395 		upper_info = ptr;
1396 		/* flush all routes if dev is linked to or unlinked from
1397 		 * an L3 master device (e.g., VRF)
1398 		 */
1399 		if (upper_info->upper_dev &&
1400 		    netif_is_l3_master(upper_info->upper_dev))
1401 			fib_disable_ip(dev, NETDEV_DOWN, true);
1402 		break;
1403 	}
1404 	return NOTIFY_DONE;
1405 }
1406 
1407 static struct notifier_block fib_inetaddr_notifier = {
1408 	.notifier_call = fib_inetaddr_event,
1409 };
1410 
1411 static struct notifier_block fib_netdev_notifier = {
1412 	.notifier_call = fib_netdev_event,
1413 };
1414 
1415 static int __net_init ip_fib_net_init(struct net *net)
1416 {
1417 	int err;
1418 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1419 
1420 	err = fib4_notifier_init(net);
1421 	if (err)
1422 		return err;
1423 
1424 	/* Avoid false sharing : Use at least a full cache line */
1425 	size = max_t(size_t, size, L1_CACHE_BYTES);
1426 
1427 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1428 	if (!net->ipv4.fib_table_hash) {
1429 		err = -ENOMEM;
1430 		goto err_table_hash_alloc;
1431 	}
1432 
1433 	err = fib4_rules_init(net);
1434 	if (err < 0)
1435 		goto err_rules_init;
1436 	return 0;
1437 
1438 err_rules_init:
1439 	kfree(net->ipv4.fib_table_hash);
1440 err_table_hash_alloc:
1441 	fib4_notifier_exit(net);
1442 	return err;
1443 }
1444 
1445 static void ip_fib_net_exit(struct net *net)
1446 {
1447 	int i;
1448 
1449 	rtnl_lock();
1450 #ifdef CONFIG_IP_MULTIPLE_TABLES
1451 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1452 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1453 #endif
1454 	/* Destroy the tables in reverse order to guarantee that the
1455 	 * local table, ID 255, is destroyed before the main table, ID
1456 	 * 254. This is necessary as the local table may contain
1457 	 * references to data contained in the main table.
1458 	 */
1459 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1460 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1461 		struct hlist_node *tmp;
1462 		struct fib_table *tb;
1463 
1464 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1465 			hlist_del(&tb->tb_hlist);
1466 			fib_table_flush(net, tb, true);
1467 			fib_free_table(tb);
1468 		}
1469 	}
1470 
1471 #ifdef CONFIG_IP_MULTIPLE_TABLES
1472 	fib4_rules_exit(net);
1473 #endif
1474 	rtnl_unlock();
1475 	kfree(net->ipv4.fib_table_hash);
1476 	fib4_notifier_exit(net);
1477 }
1478 
1479 static int __net_init fib_net_init(struct net *net)
1480 {
1481 	int error;
1482 
1483 #ifdef CONFIG_IP_ROUTE_CLASSID
1484 	net->ipv4.fib_num_tclassid_users = 0;
1485 #endif
1486 	error = ip_fib_net_init(net);
1487 	if (error < 0)
1488 		goto out;
1489 	error = nl_fib_lookup_init(net);
1490 	if (error < 0)
1491 		goto out_nlfl;
1492 	error = fib_proc_init(net);
1493 	if (error < 0)
1494 		goto out_proc;
1495 out:
1496 	return error;
1497 
1498 out_proc:
1499 	nl_fib_lookup_exit(net);
1500 out_nlfl:
1501 	ip_fib_net_exit(net);
1502 	goto out;
1503 }
1504 
1505 static void __net_exit fib_net_exit(struct net *net)
1506 {
1507 	fib_proc_exit(net);
1508 	nl_fib_lookup_exit(net);
1509 	ip_fib_net_exit(net);
1510 }
1511 
1512 static struct pernet_operations fib_net_ops = {
1513 	.init = fib_net_init,
1514 	.exit = fib_net_exit,
1515 };
1516 
1517 void __init ip_fib_init(void)
1518 {
1519 	fib_trie_init();
1520 
1521 	register_pernet_subsys(&fib_net_ops);
1522 
1523 	register_netdevice_notifier(&fib_netdev_notifier);
1524 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1525 
1526 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1527 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1528 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1529 }
1530