xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision 8dde5715)
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/rtnetlink.h>
43 #include <net/xfrm.h>
44 #include <net/l3mdev.h>
45 #include <net/lwtunnel.h>
46 #include <trace/events/fib.h>
47 
48 #ifndef CONFIG_IP_MULTIPLE_TABLES
49 
50 static int __net_init fib4_rules_init(struct net *net)
51 {
52 	struct fib_table *local_table, *main_table;
53 
54 	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
55 	if (!main_table)
56 		return -ENOMEM;
57 
58 	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
59 	if (!local_table)
60 		goto fail;
61 
62 	hlist_add_head_rcu(&local_table->tb_hlist,
63 				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
64 	hlist_add_head_rcu(&main_table->tb_hlist,
65 				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
66 	return 0;
67 
68 fail:
69 	fib_free_table(main_table);
70 	return -ENOMEM;
71 }
72 
73 static bool fib4_has_custom_rules(struct net *net)
74 {
75 	return false;
76 }
77 #else
78 
79 struct fib_table *fib_new_table(struct net *net, u32 id)
80 {
81 	struct fib_table *tb, *alias = NULL;
82 	unsigned int h;
83 
84 	if (id == 0)
85 		id = RT_TABLE_MAIN;
86 	tb = fib_get_table(net, id);
87 	if (tb)
88 		return tb;
89 
90 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
91 		alias = fib_new_table(net, RT_TABLE_MAIN);
92 
93 	tb = fib_trie_table(id, alias);
94 	if (!tb)
95 		return NULL;
96 
97 	switch (id) {
98 	case RT_TABLE_MAIN:
99 		rcu_assign_pointer(net->ipv4.fib_main, tb);
100 		break;
101 	case RT_TABLE_DEFAULT:
102 		rcu_assign_pointer(net->ipv4.fib_default, tb);
103 		break;
104 	default:
105 		break;
106 	}
107 
108 	h = id & (FIB_TABLE_HASHSZ - 1);
109 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
110 	return tb;
111 }
112 EXPORT_SYMBOL_GPL(fib_new_table);
113 
114 /* caller must hold either rtnl or rcu read lock */
115 struct fib_table *fib_get_table(struct net *net, u32 id)
116 {
117 	struct fib_table *tb;
118 	struct hlist_head *head;
119 	unsigned int h;
120 
121 	if (id == 0)
122 		id = RT_TABLE_MAIN;
123 	h = id & (FIB_TABLE_HASHSZ - 1);
124 
125 	head = &net->ipv4.fib_table_hash[h];
126 	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
127 		if (tb->tb_id == id)
128 			return tb;
129 	}
130 	return NULL;
131 }
132 
133 static bool fib4_has_custom_rules(struct net *net)
134 {
135 	return net->ipv4.fib_has_custom_rules;
136 }
137 #endif /* CONFIG_IP_MULTIPLE_TABLES */
138 
139 static void fib_replace_table(struct net *net, struct fib_table *old,
140 			      struct fib_table *new)
141 {
142 #ifdef CONFIG_IP_MULTIPLE_TABLES
143 	switch (new->tb_id) {
144 	case RT_TABLE_MAIN:
145 		rcu_assign_pointer(net->ipv4.fib_main, new);
146 		break;
147 	case RT_TABLE_DEFAULT:
148 		rcu_assign_pointer(net->ipv4.fib_default, new);
149 		break;
150 	default:
151 		break;
152 	}
153 
154 #endif
155 	/* replace the old table in the hlist */
156 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
157 }
158 
159 int fib_unmerge(struct net *net)
160 {
161 	struct fib_table *old, *new, *main_table;
162 
163 	/* attempt to fetch local table if it has been allocated */
164 	old = fib_get_table(net, RT_TABLE_LOCAL);
165 	if (!old)
166 		return 0;
167 
168 	new = fib_trie_unmerge(old);
169 	if (!new)
170 		return -ENOMEM;
171 
172 	/* table is already unmerged */
173 	if (new == old)
174 		return 0;
175 
176 	/* replace merged table with clean table */
177 	fib_replace_table(net, old, new);
178 	fib_free_table(old);
179 
180 	/* attempt to fetch main table if it has been allocated */
181 	main_table = fib_get_table(net, RT_TABLE_MAIN);
182 	if (!main_table)
183 		return 0;
184 
185 	/* flush local entries from main table */
186 	fib_table_flush_external(main_table);
187 
188 	return 0;
189 }
190 
191 static void fib_flush(struct net *net)
192 {
193 	int flushed = 0;
194 	unsigned int h;
195 
196 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
197 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
198 		struct hlist_node *tmp;
199 		struct fib_table *tb;
200 
201 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
202 			flushed += fib_table_flush(net, tb, false);
203 	}
204 
205 	if (flushed)
206 		rt_cache_flush(net);
207 }
208 
209 /*
210  * Find address type as if only "dev" was present in the system. If
211  * on_dev is NULL then all interfaces are taken into consideration.
212  */
213 static inline unsigned int __inet_dev_addr_type(struct net *net,
214 						const struct net_device *dev,
215 						__be32 addr, u32 tb_id)
216 {
217 	struct flowi4		fl4 = { .daddr = addr };
218 	struct fib_result	res;
219 	unsigned int ret = RTN_BROADCAST;
220 	struct fib_table *table;
221 
222 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
223 		return RTN_BROADCAST;
224 	if (ipv4_is_multicast(addr))
225 		return RTN_MULTICAST;
226 
227 	rcu_read_lock();
228 
229 	table = fib_get_table(net, tb_id);
230 	if (table) {
231 		ret = RTN_UNICAST;
232 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
233 			if (!dev || dev == res.fi->fib_dev)
234 				ret = res.type;
235 		}
236 	}
237 
238 	rcu_read_unlock();
239 	return ret;
240 }
241 
242 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
243 {
244 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
245 }
246 EXPORT_SYMBOL(inet_addr_type_table);
247 
248 unsigned int inet_addr_type(struct net *net, __be32 addr)
249 {
250 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
251 }
252 EXPORT_SYMBOL(inet_addr_type);
253 
254 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
255 				__be32 addr)
256 {
257 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
258 
259 	return __inet_dev_addr_type(net, dev, addr, rt_table);
260 }
261 EXPORT_SYMBOL(inet_dev_addr_type);
262 
263 /* inet_addr_type with dev == NULL but using the table from a dev
264  * if one is associated
265  */
266 unsigned int inet_addr_type_dev_table(struct net *net,
267 				      const struct net_device *dev,
268 				      __be32 addr)
269 {
270 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
271 
272 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
273 }
274 EXPORT_SYMBOL(inet_addr_type_dev_table);
275 
276 __be32 fib_compute_spec_dst(struct sk_buff *skb)
277 {
278 	struct net_device *dev = skb->dev;
279 	struct in_device *in_dev;
280 	struct fib_result res;
281 	struct rtable *rt;
282 	struct net *net;
283 	int scope;
284 
285 	rt = skb_rtable(skb);
286 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
287 	    RTCF_LOCAL)
288 		return ip_hdr(skb)->daddr;
289 
290 	in_dev = __in_dev_get_rcu(dev);
291 
292 	net = dev_net(dev);
293 
294 	scope = RT_SCOPE_UNIVERSE;
295 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
296 		bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
297 		struct flowi4 fl4 = {
298 			.flowi4_iif = LOOPBACK_IFINDEX,
299 			.flowi4_oif = l3mdev_master_ifindex_rcu(dev),
300 			.daddr = ip_hdr(skb)->saddr,
301 			.flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
302 			.flowi4_scope = scope,
303 			.flowi4_mark = vmark ? skb->mark : 0,
304 		};
305 		if (!fib_lookup(net, &fl4, &res, 0))
306 			return fib_result_prefsrc(net, &res);
307 	} else {
308 		scope = RT_SCOPE_LINK;
309 	}
310 
311 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
312 }
313 
314 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
315 {
316 	bool dev_match = false;
317 #ifdef CONFIG_IP_ROUTE_MULTIPATH
318 	int ret;
319 
320 	for (ret = 0; ret < fi->fib_nhs; ret++) {
321 		struct fib_nh *nh = &fi->fib_nh[ret];
322 
323 		if (nh->fib_nh_dev == dev) {
324 			dev_match = true;
325 			break;
326 		} else if (l3mdev_master_ifindex_rcu(nh->fib_nh_dev) == dev->ifindex) {
327 			dev_match = true;
328 			break;
329 		}
330 	}
331 #else
332 	if (fi->fib_nh[0].fib_nh_dev == dev)
333 		dev_match = true;
334 #endif
335 
336 	return dev_match;
337 }
338 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
339 
340 /* Given (packet source, input interface) and optional (dst, oif, tos):
341  * - (main) check, that source is valid i.e. not broadcast or our local
342  *   address.
343  * - figure out what "logical" interface this packet arrived
344  *   and calculate "specific destination" address.
345  * - check, that packet arrived from expected physical interface.
346  * called with rcu_read_lock()
347  */
348 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
349 				 u8 tos, int oif, struct net_device *dev,
350 				 int rpf, struct in_device *idev, u32 *itag)
351 {
352 	struct net *net = dev_net(dev);
353 	struct flow_keys flkeys;
354 	int ret, no_addr;
355 	struct fib_result res;
356 	struct flowi4 fl4;
357 	bool dev_match;
358 
359 	fl4.flowi4_oif = 0;
360 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
361 	if (!fl4.flowi4_iif)
362 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
363 	fl4.daddr = src;
364 	fl4.saddr = dst;
365 	fl4.flowi4_tos = tos;
366 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
367 	fl4.flowi4_tun_key.tun_id = 0;
368 	fl4.flowi4_flags = 0;
369 	fl4.flowi4_uid = sock_net_uid(net, NULL);
370 
371 	no_addr = idev->ifa_list == NULL;
372 
373 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
374 	if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
375 		fl4.flowi4_proto = 0;
376 		fl4.fl4_sport = 0;
377 		fl4.fl4_dport = 0;
378 	}
379 
380 	if (fib_lookup(net, &fl4, &res, 0))
381 		goto last_resort;
382 	if (res.type != RTN_UNICAST &&
383 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
384 		goto e_inval;
385 	fib_combine_itag(itag, &res);
386 
387 	dev_match = fib_info_nh_uses_dev(res.fi, dev);
388 	if (dev_match) {
389 		ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
390 		return ret;
391 	}
392 	if (no_addr)
393 		goto last_resort;
394 	if (rpf == 1)
395 		goto e_rpf;
396 	fl4.flowi4_oif = dev->ifindex;
397 
398 	ret = 0;
399 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
400 		if (res.type == RTN_UNICAST)
401 			ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
402 	}
403 	return ret;
404 
405 last_resort:
406 	if (rpf)
407 		goto e_rpf;
408 	*itag = 0;
409 	return 0;
410 
411 e_inval:
412 	return -EINVAL;
413 e_rpf:
414 	return -EXDEV;
415 }
416 
417 /* Ignore rp_filter for packets protected by IPsec. */
418 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
419 			u8 tos, int oif, struct net_device *dev,
420 			struct in_device *idev, u32 *itag)
421 {
422 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
423 	struct net *net = dev_net(dev);
424 
425 	if (!r && !fib_num_tclassid_users(net) &&
426 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
427 		if (IN_DEV_ACCEPT_LOCAL(idev))
428 			goto ok;
429 		/* with custom local routes in place, checking local addresses
430 		 * only will be too optimistic, with custom rules, checking
431 		 * local addresses only can be too strict, e.g. due to vrf
432 		 */
433 		if (net->ipv4.fib_has_custom_local_routes ||
434 		    fib4_has_custom_rules(net))
435 			goto full_check;
436 		if (inet_lookup_ifaddr_rcu(net, src))
437 			return -EINVAL;
438 
439 ok:
440 		*itag = 0;
441 		return 0;
442 	}
443 
444 full_check:
445 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
446 }
447 
448 static inline __be32 sk_extract_addr(struct sockaddr *addr)
449 {
450 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
451 }
452 
453 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
454 {
455 	struct nlattr *nla;
456 
457 	nla = (struct nlattr *) ((char *) mx + len);
458 	nla->nla_type = type;
459 	nla->nla_len = nla_attr_size(4);
460 	*(u32 *) nla_data(nla) = value;
461 
462 	return len + nla_total_size(4);
463 }
464 
465 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
466 				 struct fib_config *cfg)
467 {
468 	__be32 addr;
469 	int plen;
470 
471 	memset(cfg, 0, sizeof(*cfg));
472 	cfg->fc_nlinfo.nl_net = net;
473 
474 	if (rt->rt_dst.sa_family != AF_INET)
475 		return -EAFNOSUPPORT;
476 
477 	/*
478 	 * Check mask for validity:
479 	 * a) it must be contiguous.
480 	 * b) destination must have all host bits clear.
481 	 * c) if application forgot to set correct family (AF_INET),
482 	 *    reject request unless it is absolutely clear i.e.
483 	 *    both family and mask are zero.
484 	 */
485 	plen = 32;
486 	addr = sk_extract_addr(&rt->rt_dst);
487 	if (!(rt->rt_flags & RTF_HOST)) {
488 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
489 
490 		if (rt->rt_genmask.sa_family != AF_INET) {
491 			if (mask || rt->rt_genmask.sa_family)
492 				return -EAFNOSUPPORT;
493 		}
494 
495 		if (bad_mask(mask, addr))
496 			return -EINVAL;
497 
498 		plen = inet_mask_len(mask);
499 	}
500 
501 	cfg->fc_dst_len = plen;
502 	cfg->fc_dst = addr;
503 
504 	if (cmd != SIOCDELRT) {
505 		cfg->fc_nlflags = NLM_F_CREATE;
506 		cfg->fc_protocol = RTPROT_BOOT;
507 	}
508 
509 	if (rt->rt_metric)
510 		cfg->fc_priority = rt->rt_metric - 1;
511 
512 	if (rt->rt_flags & RTF_REJECT) {
513 		cfg->fc_scope = RT_SCOPE_HOST;
514 		cfg->fc_type = RTN_UNREACHABLE;
515 		return 0;
516 	}
517 
518 	cfg->fc_scope = RT_SCOPE_NOWHERE;
519 	cfg->fc_type = RTN_UNICAST;
520 
521 	if (rt->rt_dev) {
522 		char *colon;
523 		struct net_device *dev;
524 		char devname[IFNAMSIZ];
525 
526 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
527 			return -EFAULT;
528 
529 		devname[IFNAMSIZ-1] = 0;
530 		colon = strchr(devname, ':');
531 		if (colon)
532 			*colon = 0;
533 		dev = __dev_get_by_name(net, devname);
534 		if (!dev)
535 			return -ENODEV;
536 		cfg->fc_oif = dev->ifindex;
537 		cfg->fc_table = l3mdev_fib_table(dev);
538 		if (colon) {
539 			struct in_ifaddr *ifa;
540 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
541 			if (!in_dev)
542 				return -ENODEV;
543 			*colon = ':';
544 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
545 				if (strcmp(ifa->ifa_label, devname) == 0)
546 					break;
547 			if (!ifa)
548 				return -ENODEV;
549 			cfg->fc_prefsrc = ifa->ifa_local;
550 		}
551 	}
552 
553 	addr = sk_extract_addr(&rt->rt_gateway);
554 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
555 		unsigned int addr_type;
556 
557 		cfg->fc_gw4 = addr;
558 		cfg->fc_gw_family = AF_INET;
559 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
560 		if (rt->rt_flags & RTF_GATEWAY &&
561 		    addr_type == RTN_UNICAST)
562 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
563 	}
564 
565 	if (cmd == SIOCDELRT)
566 		return 0;
567 
568 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
569 		return -EINVAL;
570 
571 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
572 		cfg->fc_scope = RT_SCOPE_LINK;
573 
574 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
575 		struct nlattr *mx;
576 		int len = 0;
577 
578 		mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
579 		if (!mx)
580 			return -ENOMEM;
581 
582 		if (rt->rt_flags & RTF_MTU)
583 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
584 
585 		if (rt->rt_flags & RTF_WINDOW)
586 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
587 
588 		if (rt->rt_flags & RTF_IRTT)
589 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
590 
591 		cfg->fc_mx = mx;
592 		cfg->fc_mx_len = len;
593 	}
594 
595 	return 0;
596 }
597 
598 /*
599  * Handle IP routing ioctl calls.
600  * These are used to manipulate the routing tables
601  */
602 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
603 {
604 	struct fib_config cfg;
605 	int err;
606 
607 	switch (cmd) {
608 	case SIOCADDRT:		/* Add a route */
609 	case SIOCDELRT:		/* Delete a route */
610 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
611 			return -EPERM;
612 
613 		rtnl_lock();
614 		err = rtentry_to_fib_config(net, cmd, rt, &cfg);
615 		if (err == 0) {
616 			struct fib_table *tb;
617 
618 			if (cmd == SIOCDELRT) {
619 				tb = fib_get_table(net, cfg.fc_table);
620 				if (tb)
621 					err = fib_table_delete(net, tb, &cfg,
622 							       NULL);
623 				else
624 					err = -ESRCH;
625 			} else {
626 				tb = fib_new_table(net, cfg.fc_table);
627 				if (tb)
628 					err = fib_table_insert(net, tb,
629 							       &cfg, NULL);
630 				else
631 					err = -ENOBUFS;
632 			}
633 
634 			/* allocated by rtentry_to_fib_config() */
635 			kfree(cfg.fc_mx);
636 		}
637 		rtnl_unlock();
638 		return err;
639 	}
640 	return -EINVAL;
641 }
642 
643 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
644 	[RTA_DST]		= { .type = NLA_U32 },
645 	[RTA_SRC]		= { .type = NLA_U32 },
646 	[RTA_IIF]		= { .type = NLA_U32 },
647 	[RTA_OIF]		= { .type = NLA_U32 },
648 	[RTA_GATEWAY]		= { .type = NLA_U32 },
649 	[RTA_PRIORITY]		= { .type = NLA_U32 },
650 	[RTA_PREFSRC]		= { .type = NLA_U32 },
651 	[RTA_METRICS]		= { .type = NLA_NESTED },
652 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
653 	[RTA_FLOW]		= { .type = NLA_U32 },
654 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
655 	[RTA_ENCAP]		= { .type = NLA_NESTED },
656 	[RTA_UID]		= { .type = NLA_U32 },
657 	[RTA_MARK]		= { .type = NLA_U32 },
658 	[RTA_TABLE]		= { .type = NLA_U32 },
659 	[RTA_IP_PROTO]		= { .type = NLA_U8 },
660 	[RTA_SPORT]		= { .type = NLA_U16 },
661 	[RTA_DPORT]		= { .type = NLA_U16 },
662 };
663 
664 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
665 		    struct netlink_ext_ack *extack)
666 {
667 	struct rtvia *via;
668 	int alen;
669 
670 	if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
671 		NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
672 		return -EINVAL;
673 	}
674 
675 	via = nla_data(nla);
676 	alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
677 
678 	switch (via->rtvia_family) {
679 	case AF_INET:
680 		if (alen != sizeof(__be32)) {
681 			NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
682 			return -EINVAL;
683 		}
684 		cfg->fc_gw_family = AF_INET;
685 		cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
686 		break;
687 	case AF_INET6:
688 #ifdef CONFIG_IPV6
689 		if (alen != sizeof(struct in6_addr)) {
690 			NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
691 			return -EINVAL;
692 		}
693 		cfg->fc_gw_family = AF_INET6;
694 		cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
695 #else
696 		NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
697 		return -EINVAL;
698 #endif
699 		break;
700 	default:
701 		NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
702 		return -EINVAL;
703 	}
704 
705 	return 0;
706 }
707 
708 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
709 			     struct nlmsghdr *nlh, struct fib_config *cfg,
710 			     struct netlink_ext_ack *extack)
711 {
712 	bool has_gw = false, has_via = false;
713 	struct nlattr *attr;
714 	int err, remaining;
715 	struct rtmsg *rtm;
716 
717 	err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
718 					rtm_ipv4_policy, extack);
719 	if (err < 0)
720 		goto errout;
721 
722 	memset(cfg, 0, sizeof(*cfg));
723 
724 	rtm = nlmsg_data(nlh);
725 	cfg->fc_dst_len = rtm->rtm_dst_len;
726 	cfg->fc_tos = rtm->rtm_tos;
727 	cfg->fc_table = rtm->rtm_table;
728 	cfg->fc_protocol = rtm->rtm_protocol;
729 	cfg->fc_scope = rtm->rtm_scope;
730 	cfg->fc_type = rtm->rtm_type;
731 	cfg->fc_flags = rtm->rtm_flags;
732 	cfg->fc_nlflags = nlh->nlmsg_flags;
733 
734 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
735 	cfg->fc_nlinfo.nlh = nlh;
736 	cfg->fc_nlinfo.nl_net = net;
737 
738 	if (cfg->fc_type > RTN_MAX) {
739 		NL_SET_ERR_MSG(extack, "Invalid route type");
740 		err = -EINVAL;
741 		goto errout;
742 	}
743 
744 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
745 		switch (nla_type(attr)) {
746 		case RTA_DST:
747 			cfg->fc_dst = nla_get_be32(attr);
748 			break;
749 		case RTA_OIF:
750 			cfg->fc_oif = nla_get_u32(attr);
751 			break;
752 		case RTA_GATEWAY:
753 			has_gw = true;
754 			cfg->fc_gw4 = nla_get_be32(attr);
755 			if (cfg->fc_gw4)
756 				cfg->fc_gw_family = AF_INET;
757 			break;
758 		case RTA_VIA:
759 			has_via = true;
760 			err = fib_gw_from_via(cfg, attr, extack);
761 			if (err)
762 				goto errout;
763 			break;
764 		case RTA_PRIORITY:
765 			cfg->fc_priority = nla_get_u32(attr);
766 			break;
767 		case RTA_PREFSRC:
768 			cfg->fc_prefsrc = nla_get_be32(attr);
769 			break;
770 		case RTA_METRICS:
771 			cfg->fc_mx = nla_data(attr);
772 			cfg->fc_mx_len = nla_len(attr);
773 			break;
774 		case RTA_MULTIPATH:
775 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
776 							     nla_len(attr),
777 							     extack);
778 			if (err < 0)
779 				goto errout;
780 			cfg->fc_mp = nla_data(attr);
781 			cfg->fc_mp_len = nla_len(attr);
782 			break;
783 		case RTA_FLOW:
784 			cfg->fc_flow = nla_get_u32(attr);
785 			break;
786 		case RTA_TABLE:
787 			cfg->fc_table = nla_get_u32(attr);
788 			break;
789 		case RTA_ENCAP:
790 			cfg->fc_encap = attr;
791 			break;
792 		case RTA_ENCAP_TYPE:
793 			cfg->fc_encap_type = nla_get_u16(attr);
794 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
795 							extack);
796 			if (err < 0)
797 				goto errout;
798 			break;
799 		}
800 	}
801 
802 	if (has_gw && has_via) {
803 		NL_SET_ERR_MSG(extack,
804 			       "Nexthop configuration can not contain both GATEWAY and VIA");
805 		goto errout;
806 	}
807 
808 	return 0;
809 errout:
810 	return err;
811 }
812 
813 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
814 			     struct netlink_ext_ack *extack)
815 {
816 	struct net *net = sock_net(skb->sk);
817 	struct fib_config cfg;
818 	struct fib_table *tb;
819 	int err;
820 
821 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
822 	if (err < 0)
823 		goto errout;
824 
825 	tb = fib_get_table(net, cfg.fc_table);
826 	if (!tb) {
827 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
828 		err = -ESRCH;
829 		goto errout;
830 	}
831 
832 	err = fib_table_delete(net, tb, &cfg, extack);
833 errout:
834 	return err;
835 }
836 
837 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
838 			     struct netlink_ext_ack *extack)
839 {
840 	struct net *net = sock_net(skb->sk);
841 	struct fib_config cfg;
842 	struct fib_table *tb;
843 	int err;
844 
845 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
846 	if (err < 0)
847 		goto errout;
848 
849 	tb = fib_new_table(net, cfg.fc_table);
850 	if (!tb) {
851 		err = -ENOBUFS;
852 		goto errout;
853 	}
854 
855 	err = fib_table_insert(net, tb, &cfg, extack);
856 	if (!err && cfg.fc_type == RTN_LOCAL)
857 		net->ipv4.fib_has_custom_local_routes = true;
858 errout:
859 	return err;
860 }
861 
862 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
863 			  struct fib_dump_filter *filter,
864 			  struct netlink_callback *cb)
865 {
866 	struct netlink_ext_ack *extack = cb->extack;
867 	struct nlattr *tb[RTA_MAX + 1];
868 	struct rtmsg *rtm;
869 	int err, i;
870 
871 	ASSERT_RTNL();
872 
873 	if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
874 		NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
875 		return -EINVAL;
876 	}
877 
878 	rtm = nlmsg_data(nlh);
879 	if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
880 	    rtm->rtm_scope) {
881 		NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
882 		return -EINVAL;
883 	}
884 	if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
885 		NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
886 		return -EINVAL;
887 	}
888 
889 	filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
890 	filter->flags    = rtm->rtm_flags;
891 	filter->protocol = rtm->rtm_protocol;
892 	filter->rt_type  = rtm->rtm_type;
893 	filter->table_id = rtm->rtm_table;
894 
895 	err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
896 					    rtm_ipv4_policy, extack);
897 	if (err < 0)
898 		return err;
899 
900 	for (i = 0; i <= RTA_MAX; ++i) {
901 		int ifindex;
902 
903 		if (!tb[i])
904 			continue;
905 
906 		switch (i) {
907 		case RTA_TABLE:
908 			filter->table_id = nla_get_u32(tb[i]);
909 			break;
910 		case RTA_OIF:
911 			ifindex = nla_get_u32(tb[i]);
912 			filter->dev = __dev_get_by_index(net, ifindex);
913 			if (!filter->dev)
914 				return -ENODEV;
915 			break;
916 		default:
917 			NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
918 			return -EINVAL;
919 		}
920 	}
921 
922 	if (filter->flags || filter->protocol || filter->rt_type ||
923 	    filter->table_id || filter->dev) {
924 		filter->filter_set = 1;
925 		cb->answer_flags = NLM_F_DUMP_FILTERED;
926 	}
927 
928 	return 0;
929 }
930 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
931 
932 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
933 {
934 	const struct nlmsghdr *nlh = cb->nlh;
935 	struct net *net = sock_net(skb->sk);
936 	struct fib_dump_filter filter = {};
937 	unsigned int h, s_h;
938 	unsigned int e = 0, s_e;
939 	struct fib_table *tb;
940 	struct hlist_head *head;
941 	int dumped = 0, err;
942 
943 	if (cb->strict_check) {
944 		err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
945 		if (err < 0)
946 			return err;
947 	} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
948 		struct rtmsg *rtm = nlmsg_data(nlh);
949 
950 		filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
951 	}
952 
953 	/* fib entries are never clones and ipv4 does not use prefix flag */
954 	if (filter.flags & (RTM_F_PREFIX | RTM_F_CLONED))
955 		return skb->len;
956 
957 	if (filter.table_id) {
958 		tb = fib_get_table(net, filter.table_id);
959 		if (!tb) {
960 			if (filter.dump_all_families)
961 				return skb->len;
962 
963 			NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
964 			return -ENOENT;
965 		}
966 
967 		err = fib_table_dump(tb, skb, cb, &filter);
968 		return skb->len ? : err;
969 	}
970 
971 	s_h = cb->args[0];
972 	s_e = cb->args[1];
973 
974 	rcu_read_lock();
975 
976 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
977 		e = 0;
978 		head = &net->ipv4.fib_table_hash[h];
979 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
980 			if (e < s_e)
981 				goto next;
982 			if (dumped)
983 				memset(&cb->args[2], 0, sizeof(cb->args) -
984 						 2 * sizeof(cb->args[0]));
985 			err = fib_table_dump(tb, skb, cb, &filter);
986 			if (err < 0) {
987 				if (likely(skb->len))
988 					goto out;
989 
990 				goto out_err;
991 			}
992 			dumped = 1;
993 next:
994 			e++;
995 		}
996 	}
997 out:
998 	err = skb->len;
999 out_err:
1000 	rcu_read_unlock();
1001 
1002 	cb->args[1] = e;
1003 	cb->args[0] = h;
1004 
1005 	return err;
1006 }
1007 
1008 /* Prepare and feed intra-kernel routing request.
1009  * Really, it should be netlink message, but :-( netlink
1010  * can be not configured, so that we feed it directly
1011  * to fib engine. It is legal, because all events occur
1012  * only when netlink is already locked.
1013  */
1014 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1015 		      struct in_ifaddr *ifa, u32 rt_priority)
1016 {
1017 	struct net *net = dev_net(ifa->ifa_dev->dev);
1018 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1019 	struct fib_table *tb;
1020 	struct fib_config cfg = {
1021 		.fc_protocol = RTPROT_KERNEL,
1022 		.fc_type = type,
1023 		.fc_dst = dst,
1024 		.fc_dst_len = dst_len,
1025 		.fc_priority = rt_priority,
1026 		.fc_prefsrc = ifa->ifa_local,
1027 		.fc_oif = ifa->ifa_dev->dev->ifindex,
1028 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1029 		.fc_nlinfo = {
1030 			.nl_net = net,
1031 		},
1032 	};
1033 
1034 	if (!tb_id)
1035 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1036 
1037 	tb = fib_new_table(net, tb_id);
1038 	if (!tb)
1039 		return;
1040 
1041 	cfg.fc_table = tb->tb_id;
1042 
1043 	if (type != RTN_LOCAL)
1044 		cfg.fc_scope = RT_SCOPE_LINK;
1045 	else
1046 		cfg.fc_scope = RT_SCOPE_HOST;
1047 
1048 	if (cmd == RTM_NEWROUTE)
1049 		fib_table_insert(net, tb, &cfg, NULL);
1050 	else
1051 		fib_table_delete(net, tb, &cfg, NULL);
1052 }
1053 
1054 void fib_add_ifaddr(struct in_ifaddr *ifa)
1055 {
1056 	struct in_device *in_dev = ifa->ifa_dev;
1057 	struct net_device *dev = in_dev->dev;
1058 	struct in_ifaddr *prim = ifa;
1059 	__be32 mask = ifa->ifa_mask;
1060 	__be32 addr = ifa->ifa_local;
1061 	__be32 prefix = ifa->ifa_address & mask;
1062 
1063 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1064 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
1065 		if (!prim) {
1066 			pr_warn("%s: bug: prim == NULL\n", __func__);
1067 			return;
1068 		}
1069 	}
1070 
1071 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1072 
1073 	if (!(dev->flags & IFF_UP))
1074 		return;
1075 
1076 	/* Add broadcast address, if it is explicitly assigned. */
1077 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1078 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1079 			  prim, 0);
1080 
1081 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1082 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
1083 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1084 			fib_magic(RTM_NEWROUTE,
1085 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1086 				  prefix, ifa->ifa_prefixlen, prim,
1087 				  ifa->ifa_rt_priority);
1088 
1089 		/* Add network specific broadcasts, when it takes a sense */
1090 		if (ifa->ifa_prefixlen < 31) {
1091 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1092 				  prim, 0);
1093 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1094 				  32, prim, 0);
1095 		}
1096 	}
1097 }
1098 
1099 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1100 {
1101 	__be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1102 	struct in_device *in_dev = ifa->ifa_dev;
1103 	struct net_device *dev = in_dev->dev;
1104 
1105 	if (!(dev->flags & IFF_UP) ||
1106 	    ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1107 	    ipv4_is_zeronet(prefix) ||
1108 	    prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1109 		return;
1110 
1111 	/* add the new */
1112 	fib_magic(RTM_NEWROUTE,
1113 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1114 		  prefix, ifa->ifa_prefixlen, ifa, new_metric);
1115 
1116 	/* delete the old */
1117 	fib_magic(RTM_DELROUTE,
1118 		  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1119 		  prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1120 }
1121 
1122 /* Delete primary or secondary address.
1123  * Optionally, on secondary address promotion consider the addresses
1124  * from subnet iprim as deleted, even if they are in device list.
1125  * In this case the secondary ifa can be in device list.
1126  */
1127 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1128 {
1129 	struct in_device *in_dev = ifa->ifa_dev;
1130 	struct net_device *dev = in_dev->dev;
1131 	struct in_ifaddr *ifa1;
1132 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
1133 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1134 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
1135 #define LOCAL_OK	1
1136 #define BRD_OK		2
1137 #define BRD0_OK		4
1138 #define BRD1_OK		8
1139 	unsigned int ok = 0;
1140 	int subnet = 0;		/* Primary network */
1141 	int gone = 1;		/* Address is missing */
1142 	int same_prefsrc = 0;	/* Another primary with same IP */
1143 
1144 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
1145 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1146 		if (!prim) {
1147 			/* if the device has been deleted, we don't perform
1148 			 * address promotion
1149 			 */
1150 			if (!in_dev->dead)
1151 				pr_warn("%s: bug: prim == NULL\n", __func__);
1152 			return;
1153 		}
1154 		if (iprim && iprim != prim) {
1155 			pr_warn("%s: bug: iprim != prim\n", __func__);
1156 			return;
1157 		}
1158 	} else if (!ipv4_is_zeronet(any) &&
1159 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1160 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1161 			fib_magic(RTM_DELROUTE,
1162 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1163 				  any, ifa->ifa_prefixlen, prim, 0);
1164 		subnet = 1;
1165 	}
1166 
1167 	if (in_dev->dead)
1168 		goto no_promotions;
1169 
1170 	/* Deletion is more complicated than add.
1171 	 * We should take care of not to delete too much :-)
1172 	 *
1173 	 * Scan address list to be sure that addresses are really gone.
1174 	 */
1175 
1176 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
1177 		if (ifa1 == ifa) {
1178 			/* promotion, keep the IP */
1179 			gone = 0;
1180 			continue;
1181 		}
1182 		/* Ignore IFAs from our subnet */
1183 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1184 		    inet_ifa_match(ifa1->ifa_address, iprim))
1185 			continue;
1186 
1187 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
1188 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1189 			/* Another address from our subnet? */
1190 			if (ifa1->ifa_mask == prim->ifa_mask &&
1191 			    inet_ifa_match(ifa1->ifa_address, prim))
1192 				prim1 = prim;
1193 			else {
1194 				/* We reached the secondaries, so
1195 				 * same_prefsrc should be determined.
1196 				 */
1197 				if (!same_prefsrc)
1198 					continue;
1199 				/* Search new prim1 if ifa1 is not
1200 				 * using the current prim1
1201 				 */
1202 				if (!prim1 ||
1203 				    ifa1->ifa_mask != prim1->ifa_mask ||
1204 				    !inet_ifa_match(ifa1->ifa_address, prim1))
1205 					prim1 = inet_ifa_byprefix(in_dev,
1206 							ifa1->ifa_address,
1207 							ifa1->ifa_mask);
1208 				if (!prim1)
1209 					continue;
1210 				if (prim1->ifa_local != prim->ifa_local)
1211 					continue;
1212 			}
1213 		} else {
1214 			if (prim->ifa_local != ifa1->ifa_local)
1215 				continue;
1216 			prim1 = ifa1;
1217 			if (prim != prim1)
1218 				same_prefsrc = 1;
1219 		}
1220 		if (ifa->ifa_local == ifa1->ifa_local)
1221 			ok |= LOCAL_OK;
1222 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1223 			ok |= BRD_OK;
1224 		if (brd == ifa1->ifa_broadcast)
1225 			ok |= BRD1_OK;
1226 		if (any == ifa1->ifa_broadcast)
1227 			ok |= BRD0_OK;
1228 		/* primary has network specific broadcasts */
1229 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1230 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1231 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1232 
1233 			if (!ipv4_is_zeronet(any1)) {
1234 				if (ifa->ifa_broadcast == brd1 ||
1235 				    ifa->ifa_broadcast == any1)
1236 					ok |= BRD_OK;
1237 				if (brd == brd1 || brd == any1)
1238 					ok |= BRD1_OK;
1239 				if (any == brd1 || any == any1)
1240 					ok |= BRD0_OK;
1241 			}
1242 		}
1243 	}
1244 
1245 no_promotions:
1246 	if (!(ok & BRD_OK))
1247 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1248 			  prim, 0);
1249 	if (subnet && ifa->ifa_prefixlen < 31) {
1250 		if (!(ok & BRD1_OK))
1251 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1252 				  prim, 0);
1253 		if (!(ok & BRD0_OK))
1254 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1255 				  prim, 0);
1256 	}
1257 	if (!(ok & LOCAL_OK)) {
1258 		unsigned int addr_type;
1259 
1260 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1261 
1262 		/* Check, that this local address finally disappeared. */
1263 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1264 						     ifa->ifa_local);
1265 		if (gone && addr_type != RTN_LOCAL) {
1266 			/* And the last, but not the least thing.
1267 			 * We must flush stray FIB entries.
1268 			 *
1269 			 * First of all, we scan fib_info list searching
1270 			 * for stray nexthop entries, then ignite fib_flush.
1271 			 */
1272 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1273 				fib_flush(dev_net(dev));
1274 		}
1275 	}
1276 #undef LOCAL_OK
1277 #undef BRD_OK
1278 #undef BRD0_OK
1279 #undef BRD1_OK
1280 }
1281 
1282 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1283 {
1284 
1285 	struct fib_result       res;
1286 	struct flowi4           fl4 = {
1287 		.flowi4_mark = frn->fl_mark,
1288 		.daddr = frn->fl_addr,
1289 		.flowi4_tos = frn->fl_tos,
1290 		.flowi4_scope = frn->fl_scope,
1291 	};
1292 	struct fib_table *tb;
1293 
1294 	rcu_read_lock();
1295 
1296 	tb = fib_get_table(net, frn->tb_id_in);
1297 
1298 	frn->err = -ENOENT;
1299 	if (tb) {
1300 		local_bh_disable();
1301 
1302 		frn->tb_id = tb->tb_id;
1303 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1304 
1305 		if (!frn->err) {
1306 			frn->prefixlen = res.prefixlen;
1307 			frn->nh_sel = res.nh_sel;
1308 			frn->type = res.type;
1309 			frn->scope = res.scope;
1310 		}
1311 		local_bh_enable();
1312 	}
1313 
1314 	rcu_read_unlock();
1315 }
1316 
1317 static void nl_fib_input(struct sk_buff *skb)
1318 {
1319 	struct net *net;
1320 	struct fib_result_nl *frn;
1321 	struct nlmsghdr *nlh;
1322 	u32 portid;
1323 
1324 	net = sock_net(skb->sk);
1325 	nlh = nlmsg_hdr(skb);
1326 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1327 	    skb->len < nlh->nlmsg_len ||
1328 	    nlmsg_len(nlh) < sizeof(*frn))
1329 		return;
1330 
1331 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1332 	if (!skb)
1333 		return;
1334 	nlh = nlmsg_hdr(skb);
1335 
1336 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1337 	nl_fib_lookup(net, frn);
1338 
1339 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1340 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1341 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1342 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1343 }
1344 
1345 static int __net_init nl_fib_lookup_init(struct net *net)
1346 {
1347 	struct sock *sk;
1348 	struct netlink_kernel_cfg cfg = {
1349 		.input	= nl_fib_input,
1350 	};
1351 
1352 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1353 	if (!sk)
1354 		return -EAFNOSUPPORT;
1355 	net->ipv4.fibnl = sk;
1356 	return 0;
1357 }
1358 
1359 static void nl_fib_lookup_exit(struct net *net)
1360 {
1361 	netlink_kernel_release(net->ipv4.fibnl);
1362 	net->ipv4.fibnl = NULL;
1363 }
1364 
1365 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1366 			   bool force)
1367 {
1368 	if (fib_sync_down_dev(dev, event, force))
1369 		fib_flush(dev_net(dev));
1370 	else
1371 		rt_cache_flush(dev_net(dev));
1372 	arp_ifdown(dev);
1373 }
1374 
1375 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1376 {
1377 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1378 	struct net_device *dev = ifa->ifa_dev->dev;
1379 	struct net *net = dev_net(dev);
1380 
1381 	switch (event) {
1382 	case NETDEV_UP:
1383 		fib_add_ifaddr(ifa);
1384 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1385 		fib_sync_up(dev, RTNH_F_DEAD);
1386 #endif
1387 		atomic_inc(&net->ipv4.dev_addr_genid);
1388 		rt_cache_flush(dev_net(dev));
1389 		break;
1390 	case NETDEV_DOWN:
1391 		fib_del_ifaddr(ifa, NULL);
1392 		atomic_inc(&net->ipv4.dev_addr_genid);
1393 		if (!ifa->ifa_dev->ifa_list) {
1394 			/* Last address was deleted from this interface.
1395 			 * Disable IP.
1396 			 */
1397 			fib_disable_ip(dev, event, true);
1398 		} else {
1399 			rt_cache_flush(dev_net(dev));
1400 		}
1401 		break;
1402 	}
1403 	return NOTIFY_DONE;
1404 }
1405 
1406 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1407 {
1408 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1409 	struct netdev_notifier_changeupper_info *upper_info = ptr;
1410 	struct netdev_notifier_info_ext *info_ext = ptr;
1411 	struct in_device *in_dev;
1412 	struct net *net = dev_net(dev);
1413 	unsigned int flags;
1414 
1415 	if (event == NETDEV_UNREGISTER) {
1416 		fib_disable_ip(dev, event, true);
1417 		rt_flush_dev(dev);
1418 		return NOTIFY_DONE;
1419 	}
1420 
1421 	in_dev = __in_dev_get_rtnl(dev);
1422 	if (!in_dev)
1423 		return NOTIFY_DONE;
1424 
1425 	switch (event) {
1426 	case NETDEV_UP:
1427 		for_ifa(in_dev) {
1428 			fib_add_ifaddr(ifa);
1429 		} endfor_ifa(in_dev);
1430 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1431 		fib_sync_up(dev, RTNH_F_DEAD);
1432 #endif
1433 		atomic_inc(&net->ipv4.dev_addr_genid);
1434 		rt_cache_flush(net);
1435 		break;
1436 	case NETDEV_DOWN:
1437 		fib_disable_ip(dev, event, false);
1438 		break;
1439 	case NETDEV_CHANGE:
1440 		flags = dev_get_flags(dev);
1441 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1442 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1443 		else
1444 			fib_sync_down_dev(dev, event, false);
1445 		rt_cache_flush(net);
1446 		break;
1447 	case NETDEV_CHANGEMTU:
1448 		fib_sync_mtu(dev, info_ext->ext.mtu);
1449 		rt_cache_flush(net);
1450 		break;
1451 	case NETDEV_CHANGEUPPER:
1452 		upper_info = ptr;
1453 		/* flush all routes if dev is linked to or unlinked from
1454 		 * an L3 master device (e.g., VRF)
1455 		 */
1456 		if (upper_info->upper_dev &&
1457 		    netif_is_l3_master(upper_info->upper_dev))
1458 			fib_disable_ip(dev, NETDEV_DOWN, true);
1459 		break;
1460 	}
1461 	return NOTIFY_DONE;
1462 }
1463 
1464 static struct notifier_block fib_inetaddr_notifier = {
1465 	.notifier_call = fib_inetaddr_event,
1466 };
1467 
1468 static struct notifier_block fib_netdev_notifier = {
1469 	.notifier_call = fib_netdev_event,
1470 };
1471 
1472 static int __net_init ip_fib_net_init(struct net *net)
1473 {
1474 	int err;
1475 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1476 
1477 	err = fib4_notifier_init(net);
1478 	if (err)
1479 		return err;
1480 
1481 	/* Avoid false sharing : Use at least a full cache line */
1482 	size = max_t(size_t, size, L1_CACHE_BYTES);
1483 
1484 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1485 	if (!net->ipv4.fib_table_hash) {
1486 		err = -ENOMEM;
1487 		goto err_table_hash_alloc;
1488 	}
1489 
1490 	err = fib4_rules_init(net);
1491 	if (err < 0)
1492 		goto err_rules_init;
1493 	return 0;
1494 
1495 err_rules_init:
1496 	kfree(net->ipv4.fib_table_hash);
1497 err_table_hash_alloc:
1498 	fib4_notifier_exit(net);
1499 	return err;
1500 }
1501 
1502 static void ip_fib_net_exit(struct net *net)
1503 {
1504 	int i;
1505 
1506 	rtnl_lock();
1507 #ifdef CONFIG_IP_MULTIPLE_TABLES
1508 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1509 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1510 #endif
1511 	/* Destroy the tables in reverse order to guarantee that the
1512 	 * local table, ID 255, is destroyed before the main table, ID
1513 	 * 254. This is necessary as the local table may contain
1514 	 * references to data contained in the main table.
1515 	 */
1516 	for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1517 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1518 		struct hlist_node *tmp;
1519 		struct fib_table *tb;
1520 
1521 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1522 			hlist_del(&tb->tb_hlist);
1523 			fib_table_flush(net, tb, true);
1524 			fib_free_table(tb);
1525 		}
1526 	}
1527 
1528 #ifdef CONFIG_IP_MULTIPLE_TABLES
1529 	fib4_rules_exit(net);
1530 #endif
1531 	rtnl_unlock();
1532 	kfree(net->ipv4.fib_table_hash);
1533 	fib4_notifier_exit(net);
1534 }
1535 
1536 static int __net_init fib_net_init(struct net *net)
1537 {
1538 	int error;
1539 
1540 #ifdef CONFIG_IP_ROUTE_CLASSID
1541 	net->ipv4.fib_num_tclassid_users = 0;
1542 #endif
1543 	error = ip_fib_net_init(net);
1544 	if (error < 0)
1545 		goto out;
1546 	error = nl_fib_lookup_init(net);
1547 	if (error < 0)
1548 		goto out_nlfl;
1549 	error = fib_proc_init(net);
1550 	if (error < 0)
1551 		goto out_proc;
1552 out:
1553 	return error;
1554 
1555 out_proc:
1556 	nl_fib_lookup_exit(net);
1557 out_nlfl:
1558 	ip_fib_net_exit(net);
1559 	goto out;
1560 }
1561 
1562 static void __net_exit fib_net_exit(struct net *net)
1563 {
1564 	fib_proc_exit(net);
1565 	nl_fib_lookup_exit(net);
1566 	ip_fib_net_exit(net);
1567 }
1568 
1569 static struct pernet_operations fib_net_ops = {
1570 	.init = fib_net_init,
1571 	.exit = fib_net_exit,
1572 };
1573 
1574 void __init ip_fib_init(void)
1575 {
1576 	fib_trie_init();
1577 
1578 	register_pernet_subsys(&fib_net_ops);
1579 
1580 	register_netdevice_notifier(&fib_netdev_notifier);
1581 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1582 
1583 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1584 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1585 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1586 }
1587