xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision 28efb0046512e8a13ed9f9bdf0d68d10bbfbe9cf)
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 #else
77 
78 struct fib_table *fib_new_table(struct net *net, u32 id)
79 {
80 	struct fib_table *tb, *alias = NULL;
81 	unsigned int h;
82 
83 	if (id == 0)
84 		id = RT_TABLE_MAIN;
85 	tb = fib_get_table(net, id);
86 	if (tb)
87 		return tb;
88 
89 	if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
90 		alias = fib_new_table(net, RT_TABLE_MAIN);
91 
92 	tb = fib_trie_table(id, alias);
93 	if (!tb)
94 		return NULL;
95 
96 	switch (id) {
97 	case RT_TABLE_MAIN:
98 		rcu_assign_pointer(net->ipv4.fib_main, tb);
99 		break;
100 	case RT_TABLE_DEFAULT:
101 		rcu_assign_pointer(net->ipv4.fib_default, tb);
102 		break;
103 	default:
104 		break;
105 	}
106 
107 	h = id & (FIB_TABLE_HASHSZ - 1);
108 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
109 	return tb;
110 }
111 EXPORT_SYMBOL_GPL(fib_new_table);
112 
113 /* caller must hold either rtnl or rcu read lock */
114 struct fib_table *fib_get_table(struct net *net, u32 id)
115 {
116 	struct fib_table *tb;
117 	struct hlist_head *head;
118 	unsigned int h;
119 
120 	if (id == 0)
121 		id = RT_TABLE_MAIN;
122 	h = id & (FIB_TABLE_HASHSZ - 1);
123 
124 	head = &net->ipv4.fib_table_hash[h];
125 	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
126 		if (tb->tb_id == id)
127 			return tb;
128 	}
129 	return NULL;
130 }
131 #endif /* CONFIG_IP_MULTIPLE_TABLES */
132 
133 static void fib_replace_table(struct net *net, struct fib_table *old,
134 			      struct fib_table *new)
135 {
136 #ifdef CONFIG_IP_MULTIPLE_TABLES
137 	switch (new->tb_id) {
138 	case RT_TABLE_MAIN:
139 		rcu_assign_pointer(net->ipv4.fib_main, new);
140 		break;
141 	case RT_TABLE_DEFAULT:
142 		rcu_assign_pointer(net->ipv4.fib_default, new);
143 		break;
144 	default:
145 		break;
146 	}
147 
148 #endif
149 	/* replace the old table in the hlist */
150 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
151 }
152 
153 int fib_unmerge(struct net *net)
154 {
155 	struct fib_table *old, *new, *main_table;
156 
157 	/* attempt to fetch local table if it has been allocated */
158 	old = fib_get_table(net, RT_TABLE_LOCAL);
159 	if (!old)
160 		return 0;
161 
162 	new = fib_trie_unmerge(old);
163 	if (!new)
164 		return -ENOMEM;
165 
166 	/* table is already unmerged */
167 	if (new == old)
168 		return 0;
169 
170 	/* replace merged table with clean table */
171 	fib_replace_table(net, old, new);
172 	fib_free_table(old);
173 
174 	/* attempt to fetch main table if it has been allocated */
175 	main_table = fib_get_table(net, RT_TABLE_MAIN);
176 	if (!main_table)
177 		return 0;
178 
179 	/* flush local entries from main table */
180 	fib_table_flush_external(main_table);
181 
182 	return 0;
183 }
184 
185 static void fib_flush(struct net *net)
186 {
187 	int flushed = 0;
188 	unsigned int h;
189 
190 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
191 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
192 		struct hlist_node *tmp;
193 		struct fib_table *tb;
194 
195 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
196 			flushed += fib_table_flush(net, tb);
197 	}
198 
199 	if (flushed)
200 		rt_cache_flush(net);
201 }
202 
203 /*
204  * Find address type as if only "dev" was present in the system. If
205  * on_dev is NULL then all interfaces are taken into consideration.
206  */
207 static inline unsigned int __inet_dev_addr_type(struct net *net,
208 						const struct net_device *dev,
209 						__be32 addr, u32 tb_id)
210 {
211 	struct flowi4		fl4 = { .daddr = addr };
212 	struct fib_result	res;
213 	unsigned int ret = RTN_BROADCAST;
214 	struct fib_table *table;
215 
216 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
217 		return RTN_BROADCAST;
218 	if (ipv4_is_multicast(addr))
219 		return RTN_MULTICAST;
220 
221 	rcu_read_lock();
222 
223 	table = fib_get_table(net, tb_id);
224 	if (table) {
225 		ret = RTN_UNICAST;
226 		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
227 			if (!dev || dev == res.fi->fib_dev)
228 				ret = res.type;
229 		}
230 	}
231 
232 	rcu_read_unlock();
233 	return ret;
234 }
235 
236 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
237 {
238 	return __inet_dev_addr_type(net, NULL, addr, tb_id);
239 }
240 EXPORT_SYMBOL(inet_addr_type_table);
241 
242 unsigned int inet_addr_type(struct net *net, __be32 addr)
243 {
244 	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
245 }
246 EXPORT_SYMBOL(inet_addr_type);
247 
248 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
249 				__be32 addr)
250 {
251 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
252 
253 	return __inet_dev_addr_type(net, dev, addr, rt_table);
254 }
255 EXPORT_SYMBOL(inet_dev_addr_type);
256 
257 /* inet_addr_type with dev == NULL but using the table from a dev
258  * if one is associated
259  */
260 unsigned int inet_addr_type_dev_table(struct net *net,
261 				      const struct net_device *dev,
262 				      __be32 addr)
263 {
264 	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
265 
266 	return __inet_dev_addr_type(net, NULL, addr, rt_table);
267 }
268 EXPORT_SYMBOL(inet_addr_type_dev_table);
269 
270 __be32 fib_compute_spec_dst(struct sk_buff *skb)
271 {
272 	struct net_device *dev = skb->dev;
273 	struct in_device *in_dev;
274 	struct fib_result res;
275 	struct rtable *rt;
276 	struct net *net;
277 	int scope;
278 
279 	rt = skb_rtable(skb);
280 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
281 	    RTCF_LOCAL)
282 		return ip_hdr(skb)->daddr;
283 
284 	in_dev = __in_dev_get_rcu(dev);
285 	BUG_ON(!in_dev);
286 
287 	net = dev_net(dev);
288 
289 	scope = RT_SCOPE_UNIVERSE;
290 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
291 		struct flowi4 fl4 = {
292 			.flowi4_iif = LOOPBACK_IFINDEX,
293 			.daddr = ip_hdr(skb)->saddr,
294 			.flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
295 			.flowi4_scope = scope,
296 			.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0,
297 		};
298 		if (!fib_lookup(net, &fl4, &res, 0))
299 			return FIB_RES_PREFSRC(net, res);
300 	} else {
301 		scope = RT_SCOPE_LINK;
302 	}
303 
304 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
305 }
306 
307 /* Given (packet source, input interface) and optional (dst, oif, tos):
308  * - (main) check, that source is valid i.e. not broadcast or our local
309  *   address.
310  * - figure out what "logical" interface this packet arrived
311  *   and calculate "specific destination" address.
312  * - check, that packet arrived from expected physical interface.
313  * called with rcu_read_lock()
314  */
315 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
316 				 u8 tos, int oif, struct net_device *dev,
317 				 int rpf, struct in_device *idev, u32 *itag)
318 {
319 	int ret, no_addr;
320 	struct fib_result res;
321 	struct flowi4 fl4;
322 	struct net *net = dev_net(dev);
323 	bool dev_match;
324 
325 	fl4.flowi4_oif = 0;
326 	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
327 	if (!fl4.flowi4_iif)
328 		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
329 	fl4.daddr = src;
330 	fl4.saddr = dst;
331 	fl4.flowi4_tos = tos;
332 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
333 	fl4.flowi4_tun_key.tun_id = 0;
334 	fl4.flowi4_flags = 0;
335 	fl4.flowi4_uid = sock_net_uid(net, NULL);
336 
337 	no_addr = idev->ifa_list == NULL;
338 
339 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
340 
341 	trace_fib_validate_source(dev, &fl4);
342 
343 	if (fib_lookup(net, &fl4, &res, 0))
344 		goto last_resort;
345 	if (res.type != RTN_UNICAST &&
346 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
347 		goto e_inval;
348 	fib_combine_itag(itag, &res);
349 	dev_match = false;
350 
351 #ifdef CONFIG_IP_ROUTE_MULTIPATH
352 	for (ret = 0; ret < res.fi->fib_nhs; ret++) {
353 		struct fib_nh *nh = &res.fi->fib_nh[ret];
354 
355 		if (nh->nh_dev == dev) {
356 			dev_match = true;
357 			break;
358 		} else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
359 			dev_match = true;
360 			break;
361 		}
362 	}
363 #else
364 	if (FIB_RES_DEV(res) == dev)
365 		dev_match = true;
366 #endif
367 	if (dev_match) {
368 		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
369 		return ret;
370 	}
371 	if (no_addr)
372 		goto last_resort;
373 	if (rpf == 1)
374 		goto e_rpf;
375 	fl4.flowi4_oif = dev->ifindex;
376 
377 	ret = 0;
378 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
379 		if (res.type == RTN_UNICAST)
380 			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
381 	}
382 	return ret;
383 
384 last_resort:
385 	if (rpf)
386 		goto e_rpf;
387 	*itag = 0;
388 	return 0;
389 
390 e_inval:
391 	return -EINVAL;
392 e_rpf:
393 	return -EXDEV;
394 }
395 
396 /* Ignore rp_filter for packets protected by IPsec. */
397 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
398 			u8 tos, int oif, struct net_device *dev,
399 			struct in_device *idev, u32 *itag)
400 {
401 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
402 	struct net *net = dev_net(dev);
403 
404 	if (!r && !fib_num_tclassid_users(net) &&
405 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
406 		if (IN_DEV_ACCEPT_LOCAL(idev))
407 			goto ok;
408 		/* if no local routes are added from user space we can check
409 		 * for local addresses looking-up the ifaddr table
410 		 */
411 		if (net->ipv4.fib_has_custom_local_routes)
412 			goto full_check;
413 		if (inet_lookup_ifaddr_rcu(net, src))
414 			return -EINVAL;
415 
416 ok:
417 		*itag = 0;
418 		return 0;
419 	}
420 
421 full_check:
422 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
423 }
424 
425 static inline __be32 sk_extract_addr(struct sockaddr *addr)
426 {
427 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
428 }
429 
430 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
431 {
432 	struct nlattr *nla;
433 
434 	nla = (struct nlattr *) ((char *) mx + len);
435 	nla->nla_type = type;
436 	nla->nla_len = nla_attr_size(4);
437 	*(u32 *) nla_data(nla) = value;
438 
439 	return len + nla_total_size(4);
440 }
441 
442 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
443 				 struct fib_config *cfg)
444 {
445 	__be32 addr;
446 	int plen;
447 
448 	memset(cfg, 0, sizeof(*cfg));
449 	cfg->fc_nlinfo.nl_net = net;
450 
451 	if (rt->rt_dst.sa_family != AF_INET)
452 		return -EAFNOSUPPORT;
453 
454 	/*
455 	 * Check mask for validity:
456 	 * a) it must be contiguous.
457 	 * b) destination must have all host bits clear.
458 	 * c) if application forgot to set correct family (AF_INET),
459 	 *    reject request unless it is absolutely clear i.e.
460 	 *    both family and mask are zero.
461 	 */
462 	plen = 32;
463 	addr = sk_extract_addr(&rt->rt_dst);
464 	if (!(rt->rt_flags & RTF_HOST)) {
465 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
466 
467 		if (rt->rt_genmask.sa_family != AF_INET) {
468 			if (mask || rt->rt_genmask.sa_family)
469 				return -EAFNOSUPPORT;
470 		}
471 
472 		if (bad_mask(mask, addr))
473 			return -EINVAL;
474 
475 		plen = inet_mask_len(mask);
476 	}
477 
478 	cfg->fc_dst_len = plen;
479 	cfg->fc_dst = addr;
480 
481 	if (cmd != SIOCDELRT) {
482 		cfg->fc_nlflags = NLM_F_CREATE;
483 		cfg->fc_protocol = RTPROT_BOOT;
484 	}
485 
486 	if (rt->rt_metric)
487 		cfg->fc_priority = rt->rt_metric - 1;
488 
489 	if (rt->rt_flags & RTF_REJECT) {
490 		cfg->fc_scope = RT_SCOPE_HOST;
491 		cfg->fc_type = RTN_UNREACHABLE;
492 		return 0;
493 	}
494 
495 	cfg->fc_scope = RT_SCOPE_NOWHERE;
496 	cfg->fc_type = RTN_UNICAST;
497 
498 	if (rt->rt_dev) {
499 		char *colon;
500 		struct net_device *dev;
501 		char devname[IFNAMSIZ];
502 
503 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
504 			return -EFAULT;
505 
506 		devname[IFNAMSIZ-1] = 0;
507 		colon = strchr(devname, ':');
508 		if (colon)
509 			*colon = 0;
510 		dev = __dev_get_by_name(net, devname);
511 		if (!dev)
512 			return -ENODEV;
513 		cfg->fc_oif = dev->ifindex;
514 		cfg->fc_table = l3mdev_fib_table(dev);
515 		if (colon) {
516 			struct in_ifaddr *ifa;
517 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
518 			if (!in_dev)
519 				return -ENODEV;
520 			*colon = ':';
521 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
522 				if (strcmp(ifa->ifa_label, devname) == 0)
523 					break;
524 			if (!ifa)
525 				return -ENODEV;
526 			cfg->fc_prefsrc = ifa->ifa_local;
527 		}
528 	}
529 
530 	addr = sk_extract_addr(&rt->rt_gateway);
531 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
532 		unsigned int addr_type;
533 
534 		cfg->fc_gw = addr;
535 		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
536 		if (rt->rt_flags & RTF_GATEWAY &&
537 		    addr_type == RTN_UNICAST)
538 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
539 	}
540 
541 	if (cmd == SIOCDELRT)
542 		return 0;
543 
544 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
545 		return -EINVAL;
546 
547 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
548 		cfg->fc_scope = RT_SCOPE_LINK;
549 
550 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
551 		struct nlattr *mx;
552 		int len = 0;
553 
554 		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
555 		if (!mx)
556 			return -ENOMEM;
557 
558 		if (rt->rt_flags & RTF_MTU)
559 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
560 
561 		if (rt->rt_flags & RTF_WINDOW)
562 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
563 
564 		if (rt->rt_flags & RTF_IRTT)
565 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
566 
567 		cfg->fc_mx = mx;
568 		cfg->fc_mx_len = len;
569 	}
570 
571 	return 0;
572 }
573 
574 /*
575  * Handle IP routing ioctl calls.
576  * These are used to manipulate the routing tables
577  */
578 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
579 {
580 	struct fib_config cfg;
581 	struct rtentry rt;
582 	int err;
583 
584 	switch (cmd) {
585 	case SIOCADDRT:		/* Add a route */
586 	case SIOCDELRT:		/* Delete a route */
587 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
588 			return -EPERM;
589 
590 		if (copy_from_user(&rt, arg, sizeof(rt)))
591 			return -EFAULT;
592 
593 		rtnl_lock();
594 		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
595 		if (err == 0) {
596 			struct fib_table *tb;
597 
598 			if (cmd == SIOCDELRT) {
599 				tb = fib_get_table(net, cfg.fc_table);
600 				if (tb)
601 					err = fib_table_delete(net, tb, &cfg,
602 							       NULL);
603 				else
604 					err = -ESRCH;
605 			} else {
606 				tb = fib_new_table(net, cfg.fc_table);
607 				if (tb)
608 					err = fib_table_insert(net, tb,
609 							       &cfg, NULL);
610 				else
611 					err = -ENOBUFS;
612 			}
613 
614 			/* allocated by rtentry_to_fib_config() */
615 			kfree(cfg.fc_mx);
616 		}
617 		rtnl_unlock();
618 		return err;
619 	}
620 	return -EINVAL;
621 }
622 
623 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
624 	[RTA_DST]		= { .type = NLA_U32 },
625 	[RTA_SRC]		= { .type = NLA_U32 },
626 	[RTA_IIF]		= { .type = NLA_U32 },
627 	[RTA_OIF]		= { .type = NLA_U32 },
628 	[RTA_GATEWAY]		= { .type = NLA_U32 },
629 	[RTA_PRIORITY]		= { .type = NLA_U32 },
630 	[RTA_PREFSRC]		= { .type = NLA_U32 },
631 	[RTA_METRICS]		= { .type = NLA_NESTED },
632 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
633 	[RTA_FLOW]		= { .type = NLA_U32 },
634 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
635 	[RTA_ENCAP]		= { .type = NLA_NESTED },
636 	[RTA_UID]		= { .type = NLA_U32 },
637 	[RTA_MARK]		= { .type = NLA_U32 },
638 };
639 
640 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
641 			     struct nlmsghdr *nlh, struct fib_config *cfg,
642 			     struct netlink_ext_ack *extack)
643 {
644 	struct nlattr *attr;
645 	int err, remaining;
646 	struct rtmsg *rtm;
647 
648 	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
649 			     extack);
650 	if (err < 0)
651 		goto errout;
652 
653 	memset(cfg, 0, sizeof(*cfg));
654 
655 	rtm = nlmsg_data(nlh);
656 	cfg->fc_dst_len = rtm->rtm_dst_len;
657 	cfg->fc_tos = rtm->rtm_tos;
658 	cfg->fc_table = rtm->rtm_table;
659 	cfg->fc_protocol = rtm->rtm_protocol;
660 	cfg->fc_scope = rtm->rtm_scope;
661 	cfg->fc_type = rtm->rtm_type;
662 	cfg->fc_flags = rtm->rtm_flags;
663 	cfg->fc_nlflags = nlh->nlmsg_flags;
664 
665 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
666 	cfg->fc_nlinfo.nlh = nlh;
667 	cfg->fc_nlinfo.nl_net = net;
668 
669 	if (cfg->fc_type > RTN_MAX) {
670 		NL_SET_ERR_MSG(extack, "Invalid route type");
671 		err = -EINVAL;
672 		goto errout;
673 	}
674 
675 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
676 		switch (nla_type(attr)) {
677 		case RTA_DST:
678 			cfg->fc_dst = nla_get_be32(attr);
679 			break;
680 		case RTA_OIF:
681 			cfg->fc_oif = nla_get_u32(attr);
682 			break;
683 		case RTA_GATEWAY:
684 			cfg->fc_gw = nla_get_be32(attr);
685 			break;
686 		case RTA_PRIORITY:
687 			cfg->fc_priority = nla_get_u32(attr);
688 			break;
689 		case RTA_PREFSRC:
690 			cfg->fc_prefsrc = nla_get_be32(attr);
691 			break;
692 		case RTA_METRICS:
693 			cfg->fc_mx = nla_data(attr);
694 			cfg->fc_mx_len = nla_len(attr);
695 			break;
696 		case RTA_MULTIPATH:
697 			err = lwtunnel_valid_encap_type_attr(nla_data(attr),
698 							     nla_len(attr),
699 							     extack);
700 			if (err < 0)
701 				goto errout;
702 			cfg->fc_mp = nla_data(attr);
703 			cfg->fc_mp_len = nla_len(attr);
704 			break;
705 		case RTA_FLOW:
706 			cfg->fc_flow = nla_get_u32(attr);
707 			break;
708 		case RTA_TABLE:
709 			cfg->fc_table = nla_get_u32(attr);
710 			break;
711 		case RTA_ENCAP:
712 			cfg->fc_encap = attr;
713 			break;
714 		case RTA_ENCAP_TYPE:
715 			cfg->fc_encap_type = nla_get_u16(attr);
716 			err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
717 							extack);
718 			if (err < 0)
719 				goto errout;
720 			break;
721 		}
722 	}
723 
724 	return 0;
725 errout:
726 	return err;
727 }
728 
729 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
730 			     struct netlink_ext_ack *extack)
731 {
732 	struct net *net = sock_net(skb->sk);
733 	struct fib_config cfg;
734 	struct fib_table *tb;
735 	int err;
736 
737 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
738 	if (err < 0)
739 		goto errout;
740 
741 	tb = fib_get_table(net, cfg.fc_table);
742 	if (!tb) {
743 		NL_SET_ERR_MSG(extack, "FIB table does not exist");
744 		err = -ESRCH;
745 		goto errout;
746 	}
747 
748 	err = fib_table_delete(net, tb, &cfg, extack);
749 errout:
750 	return err;
751 }
752 
753 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
754 			     struct netlink_ext_ack *extack)
755 {
756 	struct net *net = sock_net(skb->sk);
757 	struct fib_config cfg;
758 	struct fib_table *tb;
759 	int err;
760 
761 	err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
762 	if (err < 0)
763 		goto errout;
764 
765 	tb = fib_new_table(net, cfg.fc_table);
766 	if (!tb) {
767 		err = -ENOBUFS;
768 		goto errout;
769 	}
770 
771 	err = fib_table_insert(net, tb, &cfg, extack);
772 	if (!err && cfg.fc_type == RTN_LOCAL)
773 		net->ipv4.fib_has_custom_local_routes = true;
774 errout:
775 	return err;
776 }
777 
778 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
779 {
780 	struct net *net = sock_net(skb->sk);
781 	unsigned int h, s_h;
782 	unsigned int e = 0, s_e;
783 	struct fib_table *tb;
784 	struct hlist_head *head;
785 	int dumped = 0, err;
786 
787 	if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
788 	    ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
789 		return skb->len;
790 
791 	s_h = cb->args[0];
792 	s_e = cb->args[1];
793 
794 	rcu_read_lock();
795 
796 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
797 		e = 0;
798 		head = &net->ipv4.fib_table_hash[h];
799 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
800 			if (e < s_e)
801 				goto next;
802 			if (dumped)
803 				memset(&cb->args[2], 0, sizeof(cb->args) -
804 						 2 * sizeof(cb->args[0]));
805 			err = fib_table_dump(tb, skb, cb);
806 			if (err < 0) {
807 				if (likely(skb->len))
808 					goto out;
809 
810 				goto out_err;
811 			}
812 			dumped = 1;
813 next:
814 			e++;
815 		}
816 	}
817 out:
818 	err = skb->len;
819 out_err:
820 	rcu_read_unlock();
821 
822 	cb->args[1] = e;
823 	cb->args[0] = h;
824 
825 	return err;
826 }
827 
828 /* Prepare and feed intra-kernel routing request.
829  * Really, it should be netlink message, but :-( netlink
830  * can be not configured, so that we feed it directly
831  * to fib engine. It is legal, because all events occur
832  * only when netlink is already locked.
833  */
834 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
835 {
836 	struct net *net = dev_net(ifa->ifa_dev->dev);
837 	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
838 	struct fib_table *tb;
839 	struct fib_config cfg = {
840 		.fc_protocol = RTPROT_KERNEL,
841 		.fc_type = type,
842 		.fc_dst = dst,
843 		.fc_dst_len = dst_len,
844 		.fc_prefsrc = ifa->ifa_local,
845 		.fc_oif = ifa->ifa_dev->dev->ifindex,
846 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
847 		.fc_nlinfo = {
848 			.nl_net = net,
849 		},
850 	};
851 
852 	if (!tb_id)
853 		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
854 
855 	tb = fib_new_table(net, tb_id);
856 	if (!tb)
857 		return;
858 
859 	cfg.fc_table = tb->tb_id;
860 
861 	if (type != RTN_LOCAL)
862 		cfg.fc_scope = RT_SCOPE_LINK;
863 	else
864 		cfg.fc_scope = RT_SCOPE_HOST;
865 
866 	if (cmd == RTM_NEWROUTE)
867 		fib_table_insert(net, tb, &cfg, NULL);
868 	else
869 		fib_table_delete(net, tb, &cfg, NULL);
870 }
871 
872 void fib_add_ifaddr(struct in_ifaddr *ifa)
873 {
874 	struct in_device *in_dev = ifa->ifa_dev;
875 	struct net_device *dev = in_dev->dev;
876 	struct in_ifaddr *prim = ifa;
877 	__be32 mask = ifa->ifa_mask;
878 	__be32 addr = ifa->ifa_local;
879 	__be32 prefix = ifa->ifa_address & mask;
880 
881 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
882 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
883 		if (!prim) {
884 			pr_warn("%s: bug: prim == NULL\n", __func__);
885 			return;
886 		}
887 	}
888 
889 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
890 
891 	if (!(dev->flags & IFF_UP))
892 		return;
893 
894 	/* Add broadcast address, if it is explicitly assigned. */
895 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
896 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
897 
898 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
899 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
900 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
901 			fib_magic(RTM_NEWROUTE,
902 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
903 				  prefix, ifa->ifa_prefixlen, prim);
904 
905 		/* Add network specific broadcasts, when it takes a sense */
906 		if (ifa->ifa_prefixlen < 31) {
907 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
908 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
909 				  32, prim);
910 		}
911 	}
912 }
913 
914 /* Delete primary or secondary address.
915  * Optionally, on secondary address promotion consider the addresses
916  * from subnet iprim as deleted, even if they are in device list.
917  * In this case the secondary ifa can be in device list.
918  */
919 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
920 {
921 	struct in_device *in_dev = ifa->ifa_dev;
922 	struct net_device *dev = in_dev->dev;
923 	struct in_ifaddr *ifa1;
924 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
925 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
926 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
927 #define LOCAL_OK	1
928 #define BRD_OK		2
929 #define BRD0_OK		4
930 #define BRD1_OK		8
931 	unsigned int ok = 0;
932 	int subnet = 0;		/* Primary network */
933 	int gone = 1;		/* Address is missing */
934 	int same_prefsrc = 0;	/* Another primary with same IP */
935 
936 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
937 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
938 		if (!prim) {
939 			/* if the device has been deleted, we don't perform
940 			 * address promotion
941 			 */
942 			if (!in_dev->dead)
943 				pr_warn("%s: bug: prim == NULL\n", __func__);
944 			return;
945 		}
946 		if (iprim && iprim != prim) {
947 			pr_warn("%s: bug: iprim != prim\n", __func__);
948 			return;
949 		}
950 	} else if (!ipv4_is_zeronet(any) &&
951 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
952 		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
953 			fib_magic(RTM_DELROUTE,
954 				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
955 				  any, ifa->ifa_prefixlen, prim);
956 		subnet = 1;
957 	}
958 
959 	if (in_dev->dead)
960 		goto no_promotions;
961 
962 	/* Deletion is more complicated than add.
963 	 * We should take care of not to delete too much :-)
964 	 *
965 	 * Scan address list to be sure that addresses are really gone.
966 	 */
967 
968 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
969 		if (ifa1 == ifa) {
970 			/* promotion, keep the IP */
971 			gone = 0;
972 			continue;
973 		}
974 		/* Ignore IFAs from our subnet */
975 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
976 		    inet_ifa_match(ifa1->ifa_address, iprim))
977 			continue;
978 
979 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
980 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
981 			/* Another address from our subnet? */
982 			if (ifa1->ifa_mask == prim->ifa_mask &&
983 			    inet_ifa_match(ifa1->ifa_address, prim))
984 				prim1 = prim;
985 			else {
986 				/* We reached the secondaries, so
987 				 * same_prefsrc should be determined.
988 				 */
989 				if (!same_prefsrc)
990 					continue;
991 				/* Search new prim1 if ifa1 is not
992 				 * using the current prim1
993 				 */
994 				if (!prim1 ||
995 				    ifa1->ifa_mask != prim1->ifa_mask ||
996 				    !inet_ifa_match(ifa1->ifa_address, prim1))
997 					prim1 = inet_ifa_byprefix(in_dev,
998 							ifa1->ifa_address,
999 							ifa1->ifa_mask);
1000 				if (!prim1)
1001 					continue;
1002 				if (prim1->ifa_local != prim->ifa_local)
1003 					continue;
1004 			}
1005 		} else {
1006 			if (prim->ifa_local != ifa1->ifa_local)
1007 				continue;
1008 			prim1 = ifa1;
1009 			if (prim != prim1)
1010 				same_prefsrc = 1;
1011 		}
1012 		if (ifa->ifa_local == ifa1->ifa_local)
1013 			ok |= LOCAL_OK;
1014 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1015 			ok |= BRD_OK;
1016 		if (brd == ifa1->ifa_broadcast)
1017 			ok |= BRD1_OK;
1018 		if (any == ifa1->ifa_broadcast)
1019 			ok |= BRD0_OK;
1020 		/* primary has network specific broadcasts */
1021 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1022 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1023 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1024 
1025 			if (!ipv4_is_zeronet(any1)) {
1026 				if (ifa->ifa_broadcast == brd1 ||
1027 				    ifa->ifa_broadcast == any1)
1028 					ok |= BRD_OK;
1029 				if (brd == brd1 || brd == any1)
1030 					ok |= BRD1_OK;
1031 				if (any == brd1 || any == any1)
1032 					ok |= BRD0_OK;
1033 			}
1034 		}
1035 	}
1036 
1037 no_promotions:
1038 	if (!(ok & BRD_OK))
1039 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1040 	if (subnet && ifa->ifa_prefixlen < 31) {
1041 		if (!(ok & BRD1_OK))
1042 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1043 		if (!(ok & BRD0_OK))
1044 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1045 	}
1046 	if (!(ok & LOCAL_OK)) {
1047 		unsigned int addr_type;
1048 
1049 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1050 
1051 		/* Check, that this local address finally disappeared. */
1052 		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1053 						     ifa->ifa_local);
1054 		if (gone && addr_type != RTN_LOCAL) {
1055 			/* And the last, but not the least thing.
1056 			 * We must flush stray FIB entries.
1057 			 *
1058 			 * First of all, we scan fib_info list searching
1059 			 * for stray nexthop entries, then ignite fib_flush.
1060 			 */
1061 			if (fib_sync_down_addr(dev, ifa->ifa_local))
1062 				fib_flush(dev_net(dev));
1063 		}
1064 	}
1065 #undef LOCAL_OK
1066 #undef BRD_OK
1067 #undef BRD0_OK
1068 #undef BRD1_OK
1069 }
1070 
1071 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1072 {
1073 
1074 	struct fib_result       res;
1075 	struct flowi4           fl4 = {
1076 		.flowi4_mark = frn->fl_mark,
1077 		.daddr = frn->fl_addr,
1078 		.flowi4_tos = frn->fl_tos,
1079 		.flowi4_scope = frn->fl_scope,
1080 	};
1081 	struct fib_table *tb;
1082 
1083 	rcu_read_lock();
1084 
1085 	tb = fib_get_table(net, frn->tb_id_in);
1086 
1087 	frn->err = -ENOENT;
1088 	if (tb) {
1089 		local_bh_disable();
1090 
1091 		frn->tb_id = tb->tb_id;
1092 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1093 
1094 		if (!frn->err) {
1095 			frn->prefixlen = res.prefixlen;
1096 			frn->nh_sel = res.nh_sel;
1097 			frn->type = res.type;
1098 			frn->scope = res.scope;
1099 		}
1100 		local_bh_enable();
1101 	}
1102 
1103 	rcu_read_unlock();
1104 }
1105 
1106 static void nl_fib_input(struct sk_buff *skb)
1107 {
1108 	struct net *net;
1109 	struct fib_result_nl *frn;
1110 	struct nlmsghdr *nlh;
1111 	u32 portid;
1112 
1113 	net = sock_net(skb->sk);
1114 	nlh = nlmsg_hdr(skb);
1115 	if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1116 	    skb->len < nlh->nlmsg_len ||
1117 	    nlmsg_len(nlh) < sizeof(*frn))
1118 		return;
1119 
1120 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1121 	if (!skb)
1122 		return;
1123 	nlh = nlmsg_hdr(skb);
1124 
1125 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1126 	nl_fib_lookup(net, frn);
1127 
1128 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1129 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1130 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1131 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1132 }
1133 
1134 static int __net_init nl_fib_lookup_init(struct net *net)
1135 {
1136 	struct sock *sk;
1137 	struct netlink_kernel_cfg cfg = {
1138 		.input	= nl_fib_input,
1139 	};
1140 
1141 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1142 	if (!sk)
1143 		return -EAFNOSUPPORT;
1144 	net->ipv4.fibnl = sk;
1145 	return 0;
1146 }
1147 
1148 static void nl_fib_lookup_exit(struct net *net)
1149 {
1150 	netlink_kernel_release(net->ipv4.fibnl);
1151 	net->ipv4.fibnl = NULL;
1152 }
1153 
1154 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1155 			   bool force)
1156 {
1157 	if (fib_sync_down_dev(dev, event, force))
1158 		fib_flush(dev_net(dev));
1159 	else
1160 		rt_cache_flush(dev_net(dev));
1161 	arp_ifdown(dev);
1162 }
1163 
1164 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1165 {
1166 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1167 	struct net_device *dev = ifa->ifa_dev->dev;
1168 	struct net *net = dev_net(dev);
1169 
1170 	switch (event) {
1171 	case NETDEV_UP:
1172 		fib_add_ifaddr(ifa);
1173 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1174 		fib_sync_up(dev, RTNH_F_DEAD);
1175 #endif
1176 		atomic_inc(&net->ipv4.dev_addr_genid);
1177 		rt_cache_flush(dev_net(dev));
1178 		break;
1179 	case NETDEV_DOWN:
1180 		fib_del_ifaddr(ifa, NULL);
1181 		atomic_inc(&net->ipv4.dev_addr_genid);
1182 		if (!ifa->ifa_dev->ifa_list) {
1183 			/* Last address was deleted from this interface.
1184 			 * Disable IP.
1185 			 */
1186 			fib_disable_ip(dev, event, true);
1187 		} else {
1188 			rt_cache_flush(dev_net(dev));
1189 		}
1190 		break;
1191 	}
1192 	return NOTIFY_DONE;
1193 }
1194 
1195 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1196 {
1197 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1198 	struct netdev_notifier_changeupper_info *info;
1199 	struct in_device *in_dev;
1200 	struct net *net = dev_net(dev);
1201 	unsigned int flags;
1202 
1203 	if (event == NETDEV_UNREGISTER) {
1204 		fib_disable_ip(dev, event, true);
1205 		rt_flush_dev(dev);
1206 		return NOTIFY_DONE;
1207 	}
1208 
1209 	in_dev = __in_dev_get_rtnl(dev);
1210 	if (!in_dev)
1211 		return NOTIFY_DONE;
1212 
1213 	switch (event) {
1214 	case NETDEV_UP:
1215 		for_ifa(in_dev) {
1216 			fib_add_ifaddr(ifa);
1217 		} endfor_ifa(in_dev);
1218 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1219 		fib_sync_up(dev, RTNH_F_DEAD);
1220 #endif
1221 		atomic_inc(&net->ipv4.dev_addr_genid);
1222 		rt_cache_flush(net);
1223 		break;
1224 	case NETDEV_DOWN:
1225 		fib_disable_ip(dev, event, false);
1226 		break;
1227 	case NETDEV_CHANGE:
1228 		flags = dev_get_flags(dev);
1229 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1230 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1231 		else
1232 			fib_sync_down_dev(dev, event, false);
1233 		/* fall through */
1234 	case NETDEV_CHANGEMTU:
1235 		rt_cache_flush(net);
1236 		break;
1237 	case NETDEV_CHANGEUPPER:
1238 		info = ptr;
1239 		/* flush all routes if dev is linked to or unlinked from
1240 		 * an L3 master device (e.g., VRF)
1241 		 */
1242 		if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1243 			fib_disable_ip(dev, NETDEV_DOWN, true);
1244 		break;
1245 	}
1246 	return NOTIFY_DONE;
1247 }
1248 
1249 static struct notifier_block fib_inetaddr_notifier = {
1250 	.notifier_call = fib_inetaddr_event,
1251 };
1252 
1253 static struct notifier_block fib_netdev_notifier = {
1254 	.notifier_call = fib_netdev_event,
1255 };
1256 
1257 static int __net_init ip_fib_net_init(struct net *net)
1258 {
1259 	int err;
1260 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1261 
1262 	err = fib4_notifier_init(net);
1263 	if (err)
1264 		return err;
1265 
1266 	/* Avoid false sharing : Use at least a full cache line */
1267 	size = max_t(size_t, size, L1_CACHE_BYTES);
1268 
1269 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1270 	if (!net->ipv4.fib_table_hash) {
1271 		err = -ENOMEM;
1272 		goto err_table_hash_alloc;
1273 	}
1274 
1275 	err = fib4_rules_init(net);
1276 	if (err < 0)
1277 		goto err_rules_init;
1278 	return 0;
1279 
1280 err_rules_init:
1281 	kfree(net->ipv4.fib_table_hash);
1282 err_table_hash_alloc:
1283 	fib4_notifier_exit(net);
1284 	return err;
1285 }
1286 
1287 static void ip_fib_net_exit(struct net *net)
1288 {
1289 	unsigned int i;
1290 
1291 	rtnl_lock();
1292 #ifdef CONFIG_IP_MULTIPLE_TABLES
1293 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1294 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1295 #endif
1296 	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1297 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1298 		struct hlist_node *tmp;
1299 		struct fib_table *tb;
1300 
1301 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1302 			hlist_del(&tb->tb_hlist);
1303 			fib_table_flush(net, tb);
1304 			fib_free_table(tb);
1305 		}
1306 	}
1307 
1308 #ifdef CONFIG_IP_MULTIPLE_TABLES
1309 	fib4_rules_exit(net);
1310 #endif
1311 	rtnl_unlock();
1312 	kfree(net->ipv4.fib_table_hash);
1313 	fib4_notifier_exit(net);
1314 }
1315 
1316 static int __net_init fib_net_init(struct net *net)
1317 {
1318 	int error;
1319 
1320 #ifdef CONFIG_IP_ROUTE_CLASSID
1321 	net->ipv4.fib_num_tclassid_users = 0;
1322 #endif
1323 	error = ip_fib_net_init(net);
1324 	if (error < 0)
1325 		goto out;
1326 	error = nl_fib_lookup_init(net);
1327 	if (error < 0)
1328 		goto out_nlfl;
1329 	error = fib_proc_init(net);
1330 	if (error < 0)
1331 		goto out_proc;
1332 out:
1333 	return error;
1334 
1335 out_proc:
1336 	nl_fib_lookup_exit(net);
1337 out_nlfl:
1338 	ip_fib_net_exit(net);
1339 	goto out;
1340 }
1341 
1342 static void __net_exit fib_net_exit(struct net *net)
1343 {
1344 	fib_proc_exit(net);
1345 	nl_fib_lookup_exit(net);
1346 	ip_fib_net_exit(net);
1347 }
1348 
1349 static struct pernet_operations fib_net_ops = {
1350 	.init = fib_net_init,
1351 	.exit = fib_net_exit,
1352 };
1353 
1354 void __init ip_fib_init(void)
1355 {
1356 	fib_trie_init();
1357 
1358 	register_pernet_subsys(&fib_net_ops);
1359 
1360 	register_netdevice_notifier(&fib_netdev_notifier);
1361 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1362 
1363 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1364 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1365 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1366 }
1367