xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision 95777591)
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_VIA:
714 			NL_SET_ERR_MSG(extack, "IPv4 does not support RTA_VIA attribute");
715 			err = -EINVAL;
716 			goto errout;
717 		case RTA_PRIORITY:
718 			cfg->fc_priority = nla_get_u32(attr);
719 			break;
720 		case RTA_PREFSRC:
721 			cfg->fc_prefsrc = nla_get_be32(attr);
722 			break;
723 		case RTA_METRICS:
724 			cfg->fc_mx = nla_data(attr);
725 			cfg->fc_mx_len = nla_len(attr);
726 			break;
727 		case RTA_MULTIPATH:
728 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
729 							     nla_len(attr),
730 							     extack);
731 			if (err < 0)
732 				goto errout;
733 			cfg->fc_mp = nla_data(attr);
734 			cfg->fc_mp_len = nla_len(attr);
735 			break;
736 		case RTA_FLOW:
737 			cfg->fc_flow = nla_get_u32(attr);
738 			break;
739 		case RTA_TABLE:
740 			cfg->fc_table = nla_get_u32(attr);
741 			break;
742 		case RTA_ENCAP:
743 			cfg->fc_encap = attr;
744 			break;
745 		case RTA_ENCAP_TYPE:
746 			cfg->fc_encap_type = nla_get_u16(attr);
747 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
748 							extack);
749 			if (err < 0)
750 				goto errout;
751 			break;
752 		}
753 	}
754 
755 	return 0;
756 errout:
757 	return err;
758 }
759 
760 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
761 			     struct netlink_ext_ack *extack)
762 {
763 	struct net *net = sock_net(skb->sk);
764 	struct fib_config cfg;
765 	struct fib_table *tb;
766 	int err;
767 
768 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
769 	if (err < 0)
770 		goto errout;
771 
772 	tb = fib_get_table(net, cfg.fc_table);
773 	if (!tb) {
774 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
775 		err = -ESRCH;
776 		goto errout;
777 	}
778 
779 	err = fib_table_delete(net, tb, &cfg, extack);
780 errout:
781 	return err;
782 }
783 
784 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
785 			     struct netlink_ext_ack *extack)
786 {
787 	struct net *net = sock_net(skb->sk);
788 	struct fib_config cfg;
789 	struct fib_table *tb;
790 	int err;
791 
792 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
793 	if (err < 0)
794 		goto errout;
795 
796 	tb = fib_new_table(net, cfg.fc_table);
797 	if (!tb) {
798 		err = -ENOBUFS;
799 		goto errout;
800 	}
801 
802 	err = fib_table_insert(net, tb, &cfg, extack);
803 	if (!err && cfg.fc_type == RTN_LOCAL)
804 		net->ipv4.fib_has_custom_local_routes = true;
805 errout:
806 	return err;
807 }
808 
809 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
810 			  struct fib_dump_filter *filter,
811 			  struct netlink_callback *cb)
812 {
813 	struct netlink_ext_ack *extack = cb->extack;
814 	struct nlattr *tb[RTA_MAX + 1];
815 	struct rtmsg *rtm;
816 	int err, i;
817 
818 	ASSERT_RTNL();
819 
820 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
821 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
822 		return -EINVAL;
823 	}
824 
825 	rtm = nlmsg_data(nlh);
826 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
827 	    rtm->rtm_scope) {
828 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
829 		return -EINVAL;
830 	}
831 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
832 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
833 		return -EINVAL;
834 	}
835 
836 	filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
837 	filter->flags    = rtm->rtm_flags;
838 	filter->protocol = rtm->rtm_protocol;
839 	filter->rt_type  = rtm->rtm_type;
840 	filter->table_id = rtm->rtm_table;
841 
842 	err = nlmsg_parse_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
843 				 rtm_ipv4_policy, extack);
844 	if (err < 0)
845 		return err;
846 
847 	for (i = 0; i <= RTA_MAX; ++i) {
848 		int ifindex;
849 
850 		if (!tb[i])
851 			continue;
852 
853 		switch (i) {
854 		case RTA_TABLE:
855 			filter->table_id = nla_get_u32(tb[i]);
856 			break;
857 		case RTA_OIF:
858 			ifindex = nla_get_u32(tb[i]);
859 			filter->dev = __dev_get_by_index(net, ifindex);
860 			if (!filter->dev)
861 				return -ENODEV;
862 			break;
863 		default:
864 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
865 			return -EINVAL;
866 		}
867 	}
868 
869 	if (filter->flags || filter->protocol || filter->rt_type ||
870 	    filter->table_id || filter->dev) {
871 		filter->filter_set = 1;
872 		cb->answer_flags = NLM_F_DUMP_FILTERED;
873 	}
874 
875 	return 0;
876 }
877 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
878 
879 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
880 {
881 	const struct nlmsghdr *nlh = cb->nlh;
882 	struct net *net = sock_net(skb->sk);
883 	struct fib_dump_filter filter = {};
884 	unsigned int h, s_h;
885 	unsigned int e = 0, s_e;
886 	struct fib_table *tb;
887 	struct hlist_head *head;
888 	int dumped = 0, err;
889 
890 	if (cb->strict_check) {
891 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
892 		if (err < 0)
893 			return err;
894 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
895 		struct rtmsg *rtm = nlmsg_data(nlh);
896 
897 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
898 	}
899 
900 	/* fib entries are never clones and ipv4 does not use prefix flag */
901 	if (filter.flags & (RTM_F_PREFIX | RTM_F_CLONED))
902 		return skb->len;
903 
904 	if (filter.table_id) {
905 		tb = fib_get_table(net, filter.table_id);
906 		if (!tb) {
907 			if (filter.dump_all_families)
908 				return skb->len;
909 
910 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
911 			return -ENOENT;
912 		}
913 
914 		err = fib_table_dump(tb, skb, cb, &filter);
915 		return skb->len ? : err;
916 	}
917 
918 	s_h = cb->args[0];
919 	s_e = cb->args[1];
920 
921 	rcu_read_lock();
922 
923 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
924 		e = 0;
925 		head = &net->ipv4.fib_table_hash[h];
926 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
927 			if (e < s_e)
928 				goto next;
929 			if (dumped)
930 				memset(&cb->args[2], 0, sizeof(cb->args) -
931 						 2 * sizeof(cb->args[0]));
932 			err = fib_table_dump(tb, skb, cb, &filter);
933 			if (err < 0) {
934 				if (likely(skb->len))
935 					goto out;
936 
937 				goto out_err;
938 			}
939 			dumped = 1;
940 next:
941 			e++;
942 		}
943 	}
944 out:
945 	err = skb->len;
946 out_err:
947 	rcu_read_unlock();
948 
949 	cb->args[1] = e;
950 	cb->args[0] = h;
951 
952 	return err;
953 }
954 
955 /* Prepare and feed intra-kernel routing request.
956  * Really, it should be netlink message, but :-( netlink
957  * can be not configured, so that we feed it directly
958  * to fib engine. It is legal, because all events occur
959  * only when netlink is already locked.
960  */
961 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
962 		      struct in_ifaddr *ifa, u32 rt_priority)
963 {
964 	struct net *net = dev_net(ifa->ifa_dev->dev);
965 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
966 	struct fib_table *tb;
967 	struct fib_config cfg = {
968 		.fc_protocol = RTPROT_KERNEL,
969 		.fc_type = type,
970 		.fc_dst = dst,
971 		.fc_dst_len = dst_len,
972 		.fc_priority = rt_priority,
973 		.fc_prefsrc = ifa->ifa_local,
974 		.fc_oif = ifa->ifa_dev->dev->ifindex,
975 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
976 		.fc_nlinfo = {
977 			.nl_net = net,
978 		},
979 	};
980 
981 	if (!tb_id)
982 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
983 
984 	tb = fib_new_table(net, tb_id);
985 	if (!tb)
986 		return;
987 
988 	cfg.fc_table = tb->tb_id;
989 
990 	if (type != RTN_LOCAL)
991 		cfg.fc_scope = RT_SCOPE_LINK;
992 	else
993 		cfg.fc_scope = RT_SCOPE_HOST;
994 
995 	if (cmd == RTM_NEWROUTE)
996 		fib_table_insert(net, tb, &cfg, NULL);
997 	else
998 		fib_table_delete(net, tb, &cfg, NULL);
999 }
1000 
1001 void fib_add_ifaddr(struct in_ifaddr *ifa)
1002 {
1003 	struct in_device *in_dev = ifa->ifa_dev;
1004 	struct net_device *dev = in_dev->dev;
1005 	struct in_ifaddr *prim = ifa;
1006 	__be32 mask = ifa->ifa_mask;
1007 	__be32 addr = ifa->ifa_local;
1008 	__be32 prefix = ifa->ifa_address & mask;
1009 
1010 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1011 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1012 		if (!prim) {
1013 			pr_warn("%s: bug: prim == NULL\n", __func__);
1014 			return;
1015 		}
1016 	}
1017 
1018 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1019 
1020 	if (!(dev->flags & IFF_UP))
1021 		return;
1022 
1023 	/* Add broadcast address, if it is explicitly assigned. */
1024 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1025 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1026 			  prim, 0);
1027 
1028 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1029 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1030 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1031 			fib_magic(RTM_NEWROUTE,
1032 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1033 				  prefix, ifa->ifa_prefixlen, prim,
1034 				  ifa->ifa_rt_priority);
1035 
1036 		/* Add network specific broadcasts, when it takes a sense */
1037 		if (ifa->ifa_prefixlen < 31) {
1038 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1039 				  prim, 0);
1040 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1041 				  32, prim, 0);
1042 		}
1043 	}
1044 }
1045 
1046 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1047 {
1048 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1049 	struct in_device *in_dev = ifa->ifa_dev;
1050 	struct net_device *dev = in_dev->dev;
1051 
1052 	if (!(dev->flags & IFF_UP) ||
1053 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1054 	    ipv4_is_zeronet(prefix) ||
1055 	    prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1056 		return;
1057 
1058 	/* add the new */
1059 	fib_magic(RTM_NEWROUTE,
1060 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1061 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1062 
1063 	/* delete the old */
1064 	fib_magic(RTM_DELROUTE,
1065 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1066 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1067 }
1068 
1069 /* Delete primary or secondary address.
1070  * Optionally, on secondary address promotion consider the addresses
1071  * from subnet iprim as deleted, even if they are in device list.
1072  * In this case the secondary ifa can be in device list.
1073  */
1074 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1075 {
1076 	struct in_device *in_dev = ifa->ifa_dev;
1077 	struct net_device *dev = in_dev->dev;
1078 	struct in_ifaddr *ifa1;
1079 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1080 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1081 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1082 #define LOCAL_OK	1
1083 #define BRD_OK		2
1084 #define BRD0_OK		4
1085 #define BRD1_OK		8
1086 	unsigned int ok = 0;
1087 	int subnet = 0;		/* Primary network */
1088 	int gone = 1;		/* Address is missing */
1089 	int same_prefsrc = 0;	/* Another primary with same IP */
1090 
1091 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1092 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1093 		if (!prim) {
1094 			/* if the device has been deleted, we don't perform
1095 			 * address promotion
1096 			 */
1097 			if (!in_dev->dead)
1098 				pr_warn("%s: bug: prim == NULL\n", __func__);
1099 			return;
1100 		}
1101 		if (iprim && iprim != prim) {
1102 			pr_warn("%s: bug: iprim != prim\n", __func__);
1103 			return;
1104 		}
1105 	} else if (!ipv4_is_zeronet(any) &&
1106 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1107 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1108 			fib_magic(RTM_DELROUTE,
1109 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1110 				  any, ifa->ifa_prefixlen, prim, 0);
1111 		subnet = 1;
1112 	}
1113 
1114 	if (in_dev->dead)
1115 		goto no_promotions;
1116 
1117 	/* Deletion is more complicated than add.
1118 	 * We should take care of not to delete too much :-)
1119 	 *
1120 	 * Scan address list to be sure that addresses are really gone.
1121 	 */
1122 
1123 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
1124 		if (ifa1 == ifa) {
1125 			/* promotion, keep the IP */
1126 			gone = 0;
1127 			continue;
1128 		}
1129 		/* Ignore IFAs from our subnet */
1130 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1131 		    inet_ifa_match(ifa1->ifa_address, iprim))
1132 			continue;
1133 
1134 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1135 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1136 			/* Another address from our subnet? */
1137 			if (ifa1->ifa_mask == prim->ifa_mask &&
1138 			    inet_ifa_match(ifa1->ifa_address, prim))
1139 				prim1 = prim;
1140 			else {
1141 				/* We reached the secondaries, so
1142 				 * same_prefsrc should be determined.
1143 				 */
1144 				if (!same_prefsrc)
1145 					continue;
1146 				/* Search new prim1 if ifa1 is not
1147 				 * using the current prim1
1148 				 */
1149 				if (!prim1 ||
1150 				    ifa1->ifa_mask != prim1->ifa_mask ||
1151 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1152 					prim1 = inet_ifa_byprefix(in_dev,
1153 							ifa1->ifa_address,
1154 							ifa1->ifa_mask);
1155 				if (!prim1)
1156 					continue;
1157 				if (prim1->ifa_local != prim->ifa_local)
1158 					continue;
1159 			}
1160 		} else {
1161 			if (prim->ifa_local != ifa1->ifa_local)
1162 				continue;
1163 			prim1 = ifa1;
1164 			if (prim != prim1)
1165 				same_prefsrc = 1;
1166 		}
1167 		if (ifa->ifa_local == ifa1->ifa_local)
1168 			ok |= LOCAL_OK;
1169 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1170 			ok |= BRD_OK;
1171 		if (brd == ifa1->ifa_broadcast)
1172 			ok |= BRD1_OK;
1173 		if (any == ifa1->ifa_broadcast)
1174 			ok |= BRD0_OK;
1175 		/* primary has network specific broadcasts */
1176 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1177 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1178 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1179 
1180 			if (!ipv4_is_zeronet(any1)) {
1181 				if (ifa->ifa_broadcast == brd1 ||
1182 				    ifa->ifa_broadcast == any1)
1183 					ok |= BRD_OK;
1184 				if (brd == brd1 || brd == any1)
1185 					ok |= BRD1_OK;
1186 				if (any == brd1 || any == any1)
1187 					ok |= BRD0_OK;
1188 			}
1189 		}
1190 	}
1191 
1192 no_promotions:
1193 	if (!(ok & BRD_OK))
1194 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1195 			  prim, 0);
1196 	if (subnet && ifa->ifa_prefixlen < 31) {
1197 		if (!(ok & BRD1_OK))
1198 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1199 				  prim, 0);
1200 		if (!(ok & BRD0_OK))
1201 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1202 				  prim, 0);
1203 	}
1204 	if (!(ok & LOCAL_OK)) {
1205 		unsigned int addr_type;
1206 
1207 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1208 
1209 		/* Check, that this local address finally disappeared. */
1210 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1211 						     ifa->ifa_local);
1212 		if (gone && addr_type != RTN_LOCAL) {
1213 			/* And the last, but not the least thing.
1214 			 * We must flush stray FIB entries.
1215 			 *
1216 			 * First of all, we scan fib_info list searching
1217 			 * for stray nexthop entries, then ignite fib_flush.
1218 			 */
1219 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1220 				fib_flush(dev_net(dev));
1221 		}
1222 	}
1223 #undef LOCAL_OK
1224 #undef BRD_OK
1225 #undef BRD0_OK
1226 #undef BRD1_OK
1227 }
1228 
1229 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1230 {
1231 
1232 	struct fib_result       res;
1233 	struct flowi4           fl4 = {
1234 		.flowi4_mark = frn->fl_mark,
1235 		.daddr = frn->fl_addr,
1236 		.flowi4_tos = frn->fl_tos,
1237 		.flowi4_scope = frn->fl_scope,
1238 	};
1239 	struct fib_table *tb;
1240 
1241 	rcu_read_lock();
1242 
1243 	tb = fib_get_table(net, frn->tb_id_in);
1244 
1245 	frn->err = -ENOENT;
1246 	if (tb) {
1247 		local_bh_disable();
1248 
1249 		frn->tb_id = tb->tb_id;
1250 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1251 
1252 		if (!frn->err) {
1253 			frn->prefixlen = res.prefixlen;
1254 			frn->nh_sel = res.nh_sel;
1255 			frn->type = res.type;
1256 			frn->scope = res.scope;
1257 		}
1258 		local_bh_enable();
1259 	}
1260 
1261 	rcu_read_unlock();
1262 }
1263 
1264 static void nl_fib_input(struct sk_buff *skb)
1265 {
1266 	struct net *net;
1267 	struct fib_result_nl *frn;
1268 	struct nlmsghdr *nlh;
1269 	u32 portid;
1270 
1271 	net = sock_net(skb->sk);
1272 	nlh = nlmsg_hdr(skb);
1273 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1274 	    skb->len < nlh->nlmsg_len ||
1275 	    nlmsg_len(nlh) < sizeof(*frn))
1276 		return;
1277 
1278 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1279 	if (!skb)
1280 		return;
1281 	nlh = nlmsg_hdr(skb);
1282 
1283 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1284 	nl_fib_lookup(net, frn);
1285 
1286 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1287 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1288 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1289 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1290 }
1291 
1292 static int __net_init nl_fib_lookup_init(struct net *net)
1293 {
1294 	struct sock *sk;
1295 	struct netlink_kernel_cfg cfg = {
1296 		.input	= nl_fib_input,
1297 	};
1298 
1299 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1300 	if (!sk)
1301 		return -EAFNOSUPPORT;
1302 	net->ipv4.fibnl = sk;
1303 	return 0;
1304 }
1305 
1306 static void nl_fib_lookup_exit(struct net *net)
1307 {
1308 	netlink_kernel_release(net->ipv4.fibnl);
1309 	net->ipv4.fibnl = NULL;
1310 }
1311 
1312 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1313 			   bool force)
1314 {
1315 	if (fib_sync_down_dev(dev, event, force))
1316 		fib_flush(dev_net(dev));
1317 	else
1318 		rt_cache_flush(dev_net(dev));
1319 	arp_ifdown(dev);
1320 }
1321 
1322 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1323 {
1324 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1325 	struct net_device *dev = ifa->ifa_dev->dev;
1326 	struct net *net = dev_net(dev);
1327 
1328 	switch (event) {
1329 	case NETDEV_UP:
1330 		fib_add_ifaddr(ifa);
1331 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1332 		fib_sync_up(dev, RTNH_F_DEAD);
1333 #endif
1334 		atomic_inc(&net->ipv4.dev_addr_genid);
1335 		rt_cache_flush(dev_net(dev));
1336 		break;
1337 	case NETDEV_DOWN:
1338 		fib_del_ifaddr(ifa, NULL);
1339 		atomic_inc(&net->ipv4.dev_addr_genid);
1340 		if (!ifa->ifa_dev->ifa_list) {
1341 			/* Last address was deleted from this interface.
1342 			 * Disable IP.
1343 			 */
1344 			fib_disable_ip(dev, event, true);
1345 		} else {
1346 			rt_cache_flush(dev_net(dev));
1347 		}
1348 		break;
1349 	}
1350 	return NOTIFY_DONE;
1351 }
1352 
1353 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1354 {
1355 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1356 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1357 	struct netdev_notifier_info_ext *info_ext = ptr;
1358 	struct in_device *in_dev;
1359 	struct net *net = dev_net(dev);
1360 	unsigned int flags;
1361 
1362 	if (event == NETDEV_UNREGISTER) {
1363 		fib_disable_ip(dev, event, true);
1364 		rt_flush_dev(dev);
1365 		return NOTIFY_DONE;
1366 	}
1367 
1368 	in_dev = __in_dev_get_rtnl(dev);
1369 	if (!in_dev)
1370 		return NOTIFY_DONE;
1371 
1372 	switch (event) {
1373 	case NETDEV_UP:
1374 		for_ifa(in_dev) {
1375 			fib_add_ifaddr(ifa);
1376 		} endfor_ifa(in_dev);
1377 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1378 		fib_sync_up(dev, RTNH_F_DEAD);
1379 #endif
1380 		atomic_inc(&net->ipv4.dev_addr_genid);
1381 		rt_cache_flush(net);
1382 		break;
1383 	case NETDEV_DOWN:
1384 		fib_disable_ip(dev, event, false);
1385 		break;
1386 	case NETDEV_CHANGE:
1387 		flags = dev_get_flags(dev);
1388 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1389 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1390 		else
1391 			fib_sync_down_dev(dev, event, false);
1392 		rt_cache_flush(net);
1393 		break;
1394 	case NETDEV_CHANGEMTU:
1395 		fib_sync_mtu(dev, info_ext->ext.mtu);
1396 		rt_cache_flush(net);
1397 		break;
1398 	case NETDEV_CHANGEUPPER:
1399 		upper_info = ptr;
1400 		/* flush all routes if dev is linked to or unlinked from
1401 		 * an L3 master device (e.g., VRF)
1402 		 */
1403 		if (upper_info->upper_dev &&
1404 		    netif_is_l3_master(upper_info->upper_dev))
1405 			fib_disable_ip(dev, NETDEV_DOWN, true);
1406 		break;
1407 	}
1408 	return NOTIFY_DONE;
1409 }
1410 
1411 static struct notifier_block fib_inetaddr_notifier = {
1412 	.notifier_call = fib_inetaddr_event,
1413 };
1414 
1415 static struct notifier_block fib_netdev_notifier = {
1416 	.notifier_call = fib_netdev_event,
1417 };
1418 
1419 static int __net_init ip_fib_net_init(struct net *net)
1420 {
1421 	int err;
1422 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1423 
1424 	err = fib4_notifier_init(net);
1425 	if (err)
1426 		return err;
1427 
1428 	/* Avoid false sharing : Use at least a full cache line */
1429 	size = max_t(size_t, size, L1_CACHE_BYTES);
1430 
1431 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1432 	if (!net->ipv4.fib_table_hash) {
1433 		err = -ENOMEM;
1434 		goto err_table_hash_alloc;
1435 	}
1436 
1437 	err = fib4_rules_init(net);
1438 	if (err < 0)
1439 		goto err_rules_init;
1440 	return 0;
1441 
1442 err_rules_init:
1443 	kfree(net->ipv4.fib_table_hash);
1444 err_table_hash_alloc:
1445 	fib4_notifier_exit(net);
1446 	return err;
1447 }
1448 
1449 static void ip_fib_net_exit(struct net *net)
1450 {
1451 	int i;
1452 
1453 	rtnl_lock();
1454 #ifdef CONFIG_IP_MULTIPLE_TABLES
1455 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1456 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1457 #endif
1458 	/* Destroy the tables in reverse order to guarantee that the
1459 	 * local table, ID 255, is destroyed before the main table, ID
1460 	 * 254. This is necessary as the local table may contain
1461 	 * references to data contained in the main table.
1462 	 */
1463 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1464 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1465 		struct hlist_node *tmp;
1466 		struct fib_table *tb;
1467 
1468 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1469 			hlist_del(&tb->tb_hlist);
1470 			fib_table_flush(net, tb, true);
1471 			fib_free_table(tb);
1472 		}
1473 	}
1474 
1475 #ifdef CONFIG_IP_MULTIPLE_TABLES
1476 	fib4_rules_exit(net);
1477 #endif
1478 	rtnl_unlock();
1479 	kfree(net->ipv4.fib_table_hash);
1480 	fib4_notifier_exit(net);
1481 }
1482 
1483 static int __net_init fib_net_init(struct net *net)
1484 {
1485 	int error;
1486 
1487 #ifdef CONFIG_IP_ROUTE_CLASSID
1488 	net->ipv4.fib_num_tclassid_users = 0;
1489 #endif
1490 	error = ip_fib_net_init(net);
1491 	if (error < 0)
1492 		goto out;
1493 	error = nl_fib_lookup_init(net);
1494 	if (error < 0)
1495 		goto out_nlfl;
1496 	error = fib_proc_init(net);
1497 	if (error < 0)
1498 		goto out_proc;
1499 out:
1500 	return error;
1501 
1502 out_proc:
1503 	nl_fib_lookup_exit(net);
1504 out_nlfl:
1505 	ip_fib_net_exit(net);
1506 	goto out;
1507 }
1508 
1509 static void __net_exit fib_net_exit(struct net *net)
1510 {
1511 	fib_proc_exit(net);
1512 	nl_fib_lookup_exit(net);
1513 	ip_fib_net_exit(net);
1514 }
1515 
1516 static struct pernet_operations fib_net_ops = {
1517 	.init = fib_net_init,
1518 	.exit = fib_net_exit,
1519 };
1520 
1521 void __init ip_fib_init(void)
1522 {
1523 	fib_trie_init();
1524 
1525 	register_pernet_subsys(&fib_net_ops);
1526 
1527 	register_netdevice_notifier(&fib_netdev_notifier);
1528 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1529 
1530 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1531 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1532 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1533 }
1534