xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision 565d76cb)
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 <asm/uaccess.h>
18 #include <asm/system.h>
19 #include <linux/bitops.h>
20 #include <linux/capability.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/string.h>
25 #include <linux/socket.h>
26 #include <linux/sockios.h>
27 #include <linux/errno.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/inetdevice.h>
31 #include <linux/netdevice.h>
32 #include <linux/if_addr.h>
33 #include <linux/if_arp.h>
34 #include <linux/skbuff.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 
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 	local_table = fib_trie_table(RT_TABLE_LOCAL);
55 	if (local_table == NULL)
56 		return -ENOMEM;
57 
58 	main_table  = fib_trie_table(RT_TABLE_MAIN);
59 	if (main_table == NULL)
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 	kfree(local_table);
70 	return -ENOMEM;
71 }
72 #else
73 
74 struct fib_table *fib_new_table(struct net *net, u32 id)
75 {
76 	struct fib_table *tb;
77 	unsigned int h;
78 
79 	if (id == 0)
80 		id = RT_TABLE_MAIN;
81 	tb = fib_get_table(net, id);
82 	if (tb)
83 		return tb;
84 
85 	tb = fib_trie_table(id);
86 	if (!tb)
87 		return NULL;
88 	h = id & (FIB_TABLE_HASHSZ - 1);
89 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
90 	return tb;
91 }
92 
93 struct fib_table *fib_get_table(struct net *net, u32 id)
94 {
95 	struct fib_table *tb;
96 	struct hlist_node *node;
97 	struct hlist_head *head;
98 	unsigned int h;
99 
100 	if (id == 0)
101 		id = RT_TABLE_MAIN;
102 	h = id & (FIB_TABLE_HASHSZ - 1);
103 
104 	rcu_read_lock();
105 	head = &net->ipv4.fib_table_hash[h];
106 	hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
107 		if (tb->tb_id == id) {
108 			rcu_read_unlock();
109 			return tb;
110 		}
111 	}
112 	rcu_read_unlock();
113 	return NULL;
114 }
115 #endif /* CONFIG_IP_MULTIPLE_TABLES */
116 
117 static void fib_flush(struct net *net)
118 {
119 	int flushed = 0;
120 	struct fib_table *tb;
121 	struct hlist_node *node;
122 	struct hlist_head *head;
123 	unsigned int h;
124 
125 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
126 		head = &net->ipv4.fib_table_hash[h];
127 		hlist_for_each_entry(tb, node, head, tb_hlist)
128 			flushed += fib_table_flush(tb);
129 	}
130 
131 	if (flushed)
132 		rt_cache_flush(net, -1);
133 }
134 
135 /*
136  * Find address type as if only "dev" was present in the system. If
137  * on_dev is NULL then all interfaces are taken into consideration.
138  */
139 static inline unsigned __inet_dev_addr_type(struct net *net,
140 					    const struct net_device *dev,
141 					    __be32 addr)
142 {
143 	struct flowi4		fl4 = { .daddr = addr };
144 	struct fib_result	res;
145 	unsigned ret = RTN_BROADCAST;
146 	struct fib_table *local_table;
147 
148 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
149 		return RTN_BROADCAST;
150 	if (ipv4_is_multicast(addr))
151 		return RTN_MULTICAST;
152 
153 #ifdef CONFIG_IP_MULTIPLE_TABLES
154 	res.r = NULL;
155 #endif
156 
157 	local_table = fib_get_table(net, RT_TABLE_LOCAL);
158 	if (local_table) {
159 		ret = RTN_UNICAST;
160 		rcu_read_lock();
161 		if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) {
162 			if (!dev || dev == res.fi->fib_dev)
163 				ret = res.type;
164 		}
165 		rcu_read_unlock();
166 	}
167 	return ret;
168 }
169 
170 unsigned int inet_addr_type(struct net *net, __be32 addr)
171 {
172 	return __inet_dev_addr_type(net, NULL, addr);
173 }
174 EXPORT_SYMBOL(inet_addr_type);
175 
176 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
177 				__be32 addr)
178 {
179 	return __inet_dev_addr_type(net, dev, addr);
180 }
181 EXPORT_SYMBOL(inet_dev_addr_type);
182 
183 /* Given (packet source, input interface) and optional (dst, oif, tos):
184  * - (main) check, that source is valid i.e. not broadcast or our local
185  *   address.
186  * - figure out what "logical" interface this packet arrived
187  *   and calculate "specific destination" address.
188  * - check, that packet arrived from expected physical interface.
189  * called with rcu_read_lock()
190  */
191 int fib_validate_source(__be32 src, __be32 dst, u8 tos, int oif,
192 			struct net_device *dev, __be32 *spec_dst,
193 			u32 *itag, u32 mark)
194 {
195 	struct in_device *in_dev;
196 	struct flowi4 fl4;
197 	struct fib_result res;
198 	int no_addr, rpf, accept_local;
199 	bool dev_match;
200 	int ret;
201 	struct net *net;
202 
203 	fl4.flowi4_oif = 0;
204 	fl4.flowi4_iif = oif;
205 	fl4.flowi4_mark = mark;
206 	fl4.daddr = src;
207 	fl4.saddr = dst;
208 	fl4.flowi4_tos = tos;
209 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
210 
211 	no_addr = rpf = accept_local = 0;
212 	in_dev = __in_dev_get_rcu(dev);
213 	if (in_dev) {
214 		no_addr = in_dev->ifa_list == NULL;
215 		rpf = IN_DEV_RPFILTER(in_dev);
216 		accept_local = IN_DEV_ACCEPT_LOCAL(in_dev);
217 		if (mark && !IN_DEV_SRC_VMARK(in_dev))
218 			fl4.flowi4_mark = 0;
219 	}
220 
221 	if (in_dev == NULL)
222 		goto e_inval;
223 
224 	net = dev_net(dev);
225 	if (fib_lookup(net, &fl4, &res))
226 		goto last_resort;
227 	if (res.type != RTN_UNICAST) {
228 		if (res.type != RTN_LOCAL || !accept_local)
229 			goto e_inval;
230 	}
231 	*spec_dst = FIB_RES_PREFSRC(res);
232 	fib_combine_itag(itag, &res);
233 	dev_match = false;
234 
235 #ifdef CONFIG_IP_ROUTE_MULTIPATH
236 	for (ret = 0; ret < res.fi->fib_nhs; ret++) {
237 		struct fib_nh *nh = &res.fi->fib_nh[ret];
238 
239 		if (nh->nh_dev == dev) {
240 			dev_match = true;
241 			break;
242 		}
243 	}
244 #else
245 	if (FIB_RES_DEV(res) == dev)
246 		dev_match = true;
247 #endif
248 	if (dev_match) {
249 		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
250 		return ret;
251 	}
252 	if (no_addr)
253 		goto last_resort;
254 	if (rpf == 1)
255 		goto e_rpf;
256 	fl4.flowi4_oif = dev->ifindex;
257 
258 	ret = 0;
259 	if (fib_lookup(net, &fl4, &res) == 0) {
260 		if (res.type == RTN_UNICAST) {
261 			*spec_dst = FIB_RES_PREFSRC(res);
262 			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
263 		}
264 	}
265 	return ret;
266 
267 last_resort:
268 	if (rpf)
269 		goto e_rpf;
270 	*spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
271 	*itag = 0;
272 	return 0;
273 
274 e_inval:
275 	return -EINVAL;
276 e_rpf:
277 	return -EXDEV;
278 }
279 
280 static inline __be32 sk_extract_addr(struct sockaddr *addr)
281 {
282 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
283 }
284 
285 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
286 {
287 	struct nlattr *nla;
288 
289 	nla = (struct nlattr *) ((char *) mx + len);
290 	nla->nla_type = type;
291 	nla->nla_len = nla_attr_size(4);
292 	*(u32 *) nla_data(nla) = value;
293 
294 	return len + nla_total_size(4);
295 }
296 
297 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
298 				 struct fib_config *cfg)
299 {
300 	__be32 addr;
301 	int plen;
302 
303 	memset(cfg, 0, sizeof(*cfg));
304 	cfg->fc_nlinfo.nl_net = net;
305 
306 	if (rt->rt_dst.sa_family != AF_INET)
307 		return -EAFNOSUPPORT;
308 
309 	/*
310 	 * Check mask for validity:
311 	 * a) it must be contiguous.
312 	 * b) destination must have all host bits clear.
313 	 * c) if application forgot to set correct family (AF_INET),
314 	 *    reject request unless it is absolutely clear i.e.
315 	 *    both family and mask are zero.
316 	 */
317 	plen = 32;
318 	addr = sk_extract_addr(&rt->rt_dst);
319 	if (!(rt->rt_flags & RTF_HOST)) {
320 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
321 
322 		if (rt->rt_genmask.sa_family != AF_INET) {
323 			if (mask || rt->rt_genmask.sa_family)
324 				return -EAFNOSUPPORT;
325 		}
326 
327 		if (bad_mask(mask, addr))
328 			return -EINVAL;
329 
330 		plen = inet_mask_len(mask);
331 	}
332 
333 	cfg->fc_dst_len = plen;
334 	cfg->fc_dst = addr;
335 
336 	if (cmd != SIOCDELRT) {
337 		cfg->fc_nlflags = NLM_F_CREATE;
338 		cfg->fc_protocol = RTPROT_BOOT;
339 	}
340 
341 	if (rt->rt_metric)
342 		cfg->fc_priority = rt->rt_metric - 1;
343 
344 	if (rt->rt_flags & RTF_REJECT) {
345 		cfg->fc_scope = RT_SCOPE_HOST;
346 		cfg->fc_type = RTN_UNREACHABLE;
347 		return 0;
348 	}
349 
350 	cfg->fc_scope = RT_SCOPE_NOWHERE;
351 	cfg->fc_type = RTN_UNICAST;
352 
353 	if (rt->rt_dev) {
354 		char *colon;
355 		struct net_device *dev;
356 		char devname[IFNAMSIZ];
357 
358 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
359 			return -EFAULT;
360 
361 		devname[IFNAMSIZ-1] = 0;
362 		colon = strchr(devname, ':');
363 		if (colon)
364 			*colon = 0;
365 		dev = __dev_get_by_name(net, devname);
366 		if (!dev)
367 			return -ENODEV;
368 		cfg->fc_oif = dev->ifindex;
369 		if (colon) {
370 			struct in_ifaddr *ifa;
371 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
372 			if (!in_dev)
373 				return -ENODEV;
374 			*colon = ':';
375 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
376 				if (strcmp(ifa->ifa_label, devname) == 0)
377 					break;
378 			if (ifa == NULL)
379 				return -ENODEV;
380 			cfg->fc_prefsrc = ifa->ifa_local;
381 		}
382 	}
383 
384 	addr = sk_extract_addr(&rt->rt_gateway);
385 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
386 		cfg->fc_gw = addr;
387 		if (rt->rt_flags & RTF_GATEWAY &&
388 		    inet_addr_type(net, addr) == RTN_UNICAST)
389 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
390 	}
391 
392 	if (cmd == SIOCDELRT)
393 		return 0;
394 
395 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
396 		return -EINVAL;
397 
398 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
399 		cfg->fc_scope = RT_SCOPE_LINK;
400 
401 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
402 		struct nlattr *mx;
403 		int len = 0;
404 
405 		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
406 		if (mx == NULL)
407 			return -ENOMEM;
408 
409 		if (rt->rt_flags & RTF_MTU)
410 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
411 
412 		if (rt->rt_flags & RTF_WINDOW)
413 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
414 
415 		if (rt->rt_flags & RTF_IRTT)
416 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
417 
418 		cfg->fc_mx = mx;
419 		cfg->fc_mx_len = len;
420 	}
421 
422 	return 0;
423 }
424 
425 /*
426  * Handle IP routing ioctl calls.
427  * These are used to manipulate the routing tables
428  */
429 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
430 {
431 	struct fib_config cfg;
432 	struct rtentry rt;
433 	int err;
434 
435 	switch (cmd) {
436 	case SIOCADDRT:		/* Add a route */
437 	case SIOCDELRT:		/* Delete a route */
438 		if (!capable(CAP_NET_ADMIN))
439 			return -EPERM;
440 
441 		if (copy_from_user(&rt, arg, sizeof(rt)))
442 			return -EFAULT;
443 
444 		rtnl_lock();
445 		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
446 		if (err == 0) {
447 			struct fib_table *tb;
448 
449 			if (cmd == SIOCDELRT) {
450 				tb = fib_get_table(net, cfg.fc_table);
451 				if (tb)
452 					err = fib_table_delete(tb, &cfg);
453 				else
454 					err = -ESRCH;
455 			} else {
456 				tb = fib_new_table(net, cfg.fc_table);
457 				if (tb)
458 					err = fib_table_insert(tb, &cfg);
459 				else
460 					err = -ENOBUFS;
461 			}
462 
463 			/* allocated by rtentry_to_fib_config() */
464 			kfree(cfg.fc_mx);
465 		}
466 		rtnl_unlock();
467 		return err;
468 	}
469 	return -EINVAL;
470 }
471 
472 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
473 	[RTA_DST]		= { .type = NLA_U32 },
474 	[RTA_SRC]		= { .type = NLA_U32 },
475 	[RTA_IIF]		= { .type = NLA_U32 },
476 	[RTA_OIF]		= { .type = NLA_U32 },
477 	[RTA_GATEWAY]		= { .type = NLA_U32 },
478 	[RTA_PRIORITY]		= { .type = NLA_U32 },
479 	[RTA_PREFSRC]		= { .type = NLA_U32 },
480 	[RTA_METRICS]		= { .type = NLA_NESTED },
481 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
482 	[RTA_FLOW]		= { .type = NLA_U32 },
483 };
484 
485 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
486 			     struct nlmsghdr *nlh, struct fib_config *cfg)
487 {
488 	struct nlattr *attr;
489 	int err, remaining;
490 	struct rtmsg *rtm;
491 
492 	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
493 	if (err < 0)
494 		goto errout;
495 
496 	memset(cfg, 0, sizeof(*cfg));
497 
498 	rtm = nlmsg_data(nlh);
499 	cfg->fc_dst_len = rtm->rtm_dst_len;
500 	cfg->fc_tos = rtm->rtm_tos;
501 	cfg->fc_table = rtm->rtm_table;
502 	cfg->fc_protocol = rtm->rtm_protocol;
503 	cfg->fc_scope = rtm->rtm_scope;
504 	cfg->fc_type = rtm->rtm_type;
505 	cfg->fc_flags = rtm->rtm_flags;
506 	cfg->fc_nlflags = nlh->nlmsg_flags;
507 
508 	cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
509 	cfg->fc_nlinfo.nlh = nlh;
510 	cfg->fc_nlinfo.nl_net = net;
511 
512 	if (cfg->fc_type > RTN_MAX) {
513 		err = -EINVAL;
514 		goto errout;
515 	}
516 
517 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
518 		switch (nla_type(attr)) {
519 		case RTA_DST:
520 			cfg->fc_dst = nla_get_be32(attr);
521 			break;
522 		case RTA_OIF:
523 			cfg->fc_oif = nla_get_u32(attr);
524 			break;
525 		case RTA_GATEWAY:
526 			cfg->fc_gw = nla_get_be32(attr);
527 			break;
528 		case RTA_PRIORITY:
529 			cfg->fc_priority = nla_get_u32(attr);
530 			break;
531 		case RTA_PREFSRC:
532 			cfg->fc_prefsrc = nla_get_be32(attr);
533 			break;
534 		case RTA_METRICS:
535 			cfg->fc_mx = nla_data(attr);
536 			cfg->fc_mx_len = nla_len(attr);
537 			break;
538 		case RTA_MULTIPATH:
539 			cfg->fc_mp = nla_data(attr);
540 			cfg->fc_mp_len = nla_len(attr);
541 			break;
542 		case RTA_FLOW:
543 			cfg->fc_flow = nla_get_u32(attr);
544 			break;
545 		case RTA_TABLE:
546 			cfg->fc_table = nla_get_u32(attr);
547 			break;
548 		}
549 	}
550 
551 	return 0;
552 errout:
553 	return err;
554 }
555 
556 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
557 {
558 	struct net *net = sock_net(skb->sk);
559 	struct fib_config cfg;
560 	struct fib_table *tb;
561 	int err;
562 
563 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
564 	if (err < 0)
565 		goto errout;
566 
567 	tb = fib_get_table(net, cfg.fc_table);
568 	if (tb == NULL) {
569 		err = -ESRCH;
570 		goto errout;
571 	}
572 
573 	err = fib_table_delete(tb, &cfg);
574 errout:
575 	return err;
576 }
577 
578 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
579 {
580 	struct net *net = sock_net(skb->sk);
581 	struct fib_config cfg;
582 	struct fib_table *tb;
583 	int err;
584 
585 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
586 	if (err < 0)
587 		goto errout;
588 
589 	tb = fib_new_table(net, cfg.fc_table);
590 	if (tb == NULL) {
591 		err = -ENOBUFS;
592 		goto errout;
593 	}
594 
595 	err = fib_table_insert(tb, &cfg);
596 errout:
597 	return err;
598 }
599 
600 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
601 {
602 	struct net *net = sock_net(skb->sk);
603 	unsigned int h, s_h;
604 	unsigned int e = 0, s_e;
605 	struct fib_table *tb;
606 	struct hlist_node *node;
607 	struct hlist_head *head;
608 	int dumped = 0;
609 
610 	if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
611 	    ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
612 		return ip_rt_dump(skb, cb);
613 
614 	s_h = cb->args[0];
615 	s_e = cb->args[1];
616 
617 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
618 		e = 0;
619 		head = &net->ipv4.fib_table_hash[h];
620 		hlist_for_each_entry(tb, node, head, tb_hlist) {
621 			if (e < s_e)
622 				goto next;
623 			if (dumped)
624 				memset(&cb->args[2], 0, sizeof(cb->args) -
625 						 2 * sizeof(cb->args[0]));
626 			if (fib_table_dump(tb, skb, cb) < 0)
627 				goto out;
628 			dumped = 1;
629 next:
630 			e++;
631 		}
632 	}
633 out:
634 	cb->args[1] = e;
635 	cb->args[0] = h;
636 
637 	return skb->len;
638 }
639 
640 /* Prepare and feed intra-kernel routing request.
641  * Really, it should be netlink message, but :-( netlink
642  * can be not configured, so that we feed it directly
643  * to fib engine. It is legal, because all events occur
644  * only when netlink is already locked.
645  */
646 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
647 {
648 	struct net *net = dev_net(ifa->ifa_dev->dev);
649 	struct fib_table *tb;
650 	struct fib_config cfg = {
651 		.fc_protocol = RTPROT_KERNEL,
652 		.fc_type = type,
653 		.fc_dst = dst,
654 		.fc_dst_len = dst_len,
655 		.fc_prefsrc = ifa->ifa_local,
656 		.fc_oif = ifa->ifa_dev->dev->ifindex,
657 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
658 		.fc_nlinfo = {
659 			.nl_net = net,
660 		},
661 	};
662 
663 	if (type == RTN_UNICAST)
664 		tb = fib_new_table(net, RT_TABLE_MAIN);
665 	else
666 		tb = fib_new_table(net, RT_TABLE_LOCAL);
667 
668 	if (tb == NULL)
669 		return;
670 
671 	cfg.fc_table = tb->tb_id;
672 
673 	if (type != RTN_LOCAL)
674 		cfg.fc_scope = RT_SCOPE_LINK;
675 	else
676 		cfg.fc_scope = RT_SCOPE_HOST;
677 
678 	if (cmd == RTM_NEWROUTE)
679 		fib_table_insert(tb, &cfg);
680 	else
681 		fib_table_delete(tb, &cfg);
682 }
683 
684 void fib_add_ifaddr(struct in_ifaddr *ifa)
685 {
686 	struct in_device *in_dev = ifa->ifa_dev;
687 	struct net_device *dev = in_dev->dev;
688 	struct in_ifaddr *prim = ifa;
689 	__be32 mask = ifa->ifa_mask;
690 	__be32 addr = ifa->ifa_local;
691 	__be32 prefix = ifa->ifa_address & mask;
692 
693 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
694 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
695 		if (prim == NULL) {
696 			printk(KERN_WARNING "fib_add_ifaddr: bug: prim == NULL\n");
697 			return;
698 		}
699 	}
700 
701 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
702 
703 	if (!(dev->flags & IFF_UP))
704 		return;
705 
706 	/* Add broadcast address, if it is explicitly assigned. */
707 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
708 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
709 
710 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
711 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
712 		fib_magic(RTM_NEWROUTE,
713 			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
714 			  prefix, ifa->ifa_prefixlen, prim);
715 
716 		/* Add network specific broadcasts, when it takes a sense */
717 		if (ifa->ifa_prefixlen < 31) {
718 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
719 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
720 				  32, prim);
721 		}
722 	}
723 }
724 
725 static void fib_del_ifaddr(struct in_ifaddr *ifa)
726 {
727 	struct in_device *in_dev = ifa->ifa_dev;
728 	struct net_device *dev = in_dev->dev;
729 	struct in_ifaddr *ifa1;
730 	struct in_ifaddr *prim = ifa;
731 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
732 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
733 #define LOCAL_OK	1
734 #define BRD_OK		2
735 #define BRD0_OK		4
736 #define BRD1_OK		8
737 	unsigned ok = 0;
738 
739 	if (!(ifa->ifa_flags & IFA_F_SECONDARY))
740 		fib_magic(RTM_DELROUTE,
741 			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
742 			  any, ifa->ifa_prefixlen, prim);
743 	else {
744 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
745 		if (prim == NULL) {
746 			printk(KERN_WARNING "fib_del_ifaddr: bug: prim == NULL\n");
747 			return;
748 		}
749 	}
750 
751 	/* Deletion is more complicated than add.
752 	 * We should take care of not to delete too much :-)
753 	 *
754 	 * Scan address list to be sure that addresses are really gone.
755 	 */
756 
757 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
758 		if (ifa->ifa_local == ifa1->ifa_local)
759 			ok |= LOCAL_OK;
760 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
761 			ok |= BRD_OK;
762 		if (brd == ifa1->ifa_broadcast)
763 			ok |= BRD1_OK;
764 		if (any == ifa1->ifa_broadcast)
765 			ok |= BRD0_OK;
766 	}
767 
768 	if (!(ok & BRD_OK))
769 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
770 	if (!(ok & BRD1_OK))
771 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
772 	if (!(ok & BRD0_OK))
773 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
774 	if (!(ok & LOCAL_OK)) {
775 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
776 
777 		/* Check, that this local address finally disappeared. */
778 		if (inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
779 			/* And the last, but not the least thing.
780 			 * We must flush stray FIB entries.
781 			 *
782 			 * First of all, we scan fib_info list searching
783 			 * for stray nexthop entries, then ignite fib_flush.
784 			 */
785 			if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
786 				fib_flush(dev_net(dev));
787 		}
788 	}
789 #undef LOCAL_OK
790 #undef BRD_OK
791 #undef BRD0_OK
792 #undef BRD1_OK
793 }
794 
795 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
796 {
797 
798 	struct fib_result       res;
799 	struct flowi4           fl4 = {
800 		.flowi4_mark = frn->fl_mark,
801 		.daddr = frn->fl_addr,
802 		.flowi4_tos = frn->fl_tos,
803 		.flowi4_scope = frn->fl_scope,
804 	};
805 
806 #ifdef CONFIG_IP_MULTIPLE_TABLES
807 	res.r = NULL;
808 #endif
809 
810 	frn->err = -ENOENT;
811 	if (tb) {
812 		local_bh_disable();
813 
814 		frn->tb_id = tb->tb_id;
815 		rcu_read_lock();
816 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
817 
818 		if (!frn->err) {
819 			frn->prefixlen = res.prefixlen;
820 			frn->nh_sel = res.nh_sel;
821 			frn->type = res.type;
822 			frn->scope = res.scope;
823 		}
824 		rcu_read_unlock();
825 		local_bh_enable();
826 	}
827 }
828 
829 static void nl_fib_input(struct sk_buff *skb)
830 {
831 	struct net *net;
832 	struct fib_result_nl *frn;
833 	struct nlmsghdr *nlh;
834 	struct fib_table *tb;
835 	u32 pid;
836 
837 	net = sock_net(skb->sk);
838 	nlh = nlmsg_hdr(skb);
839 	if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
840 	    nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
841 		return;
842 
843 	skb = skb_clone(skb, GFP_KERNEL);
844 	if (skb == NULL)
845 		return;
846 	nlh = nlmsg_hdr(skb);
847 
848 	frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
849 	tb = fib_get_table(net, frn->tb_id_in);
850 
851 	nl_fib_lookup(frn, tb);
852 
853 	pid = NETLINK_CB(skb).pid;      /* pid of sending process */
854 	NETLINK_CB(skb).pid = 0;        /* from kernel */
855 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
856 	netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
857 }
858 
859 static int __net_init nl_fib_lookup_init(struct net *net)
860 {
861 	struct sock *sk;
862 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0,
863 				   nl_fib_input, NULL, THIS_MODULE);
864 	if (sk == NULL)
865 		return -EAFNOSUPPORT;
866 	net->ipv4.fibnl = sk;
867 	return 0;
868 }
869 
870 static void nl_fib_lookup_exit(struct net *net)
871 {
872 	netlink_kernel_release(net->ipv4.fibnl);
873 	net->ipv4.fibnl = NULL;
874 }
875 
876 static void fib_disable_ip(struct net_device *dev, int force, int delay)
877 {
878 	if (fib_sync_down_dev(dev, force))
879 		fib_flush(dev_net(dev));
880 	rt_cache_flush(dev_net(dev), delay);
881 	arp_ifdown(dev);
882 }
883 
884 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
885 {
886 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
887 	struct net_device *dev = ifa->ifa_dev->dev;
888 
889 	switch (event) {
890 	case NETDEV_UP:
891 		fib_add_ifaddr(ifa);
892 #ifdef CONFIG_IP_ROUTE_MULTIPATH
893 		fib_sync_up(dev);
894 #endif
895 		fib_update_nh_saddrs(dev);
896 		rt_cache_flush(dev_net(dev), -1);
897 		break;
898 	case NETDEV_DOWN:
899 		fib_del_ifaddr(ifa);
900 		fib_update_nh_saddrs(dev);
901 		if (ifa->ifa_dev->ifa_list == NULL) {
902 			/* Last address was deleted from this interface.
903 			 * Disable IP.
904 			 */
905 			fib_disable_ip(dev, 1, 0);
906 		} else {
907 			rt_cache_flush(dev_net(dev), -1);
908 		}
909 		break;
910 	}
911 	return NOTIFY_DONE;
912 }
913 
914 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
915 {
916 	struct net_device *dev = ptr;
917 	struct in_device *in_dev = __in_dev_get_rtnl(dev);
918 
919 	if (event == NETDEV_UNREGISTER) {
920 		fib_disable_ip(dev, 2, -1);
921 		return NOTIFY_DONE;
922 	}
923 
924 	if (!in_dev)
925 		return NOTIFY_DONE;
926 
927 	switch (event) {
928 	case NETDEV_UP:
929 		for_ifa(in_dev) {
930 			fib_add_ifaddr(ifa);
931 		} endfor_ifa(in_dev);
932 #ifdef CONFIG_IP_ROUTE_MULTIPATH
933 		fib_sync_up(dev);
934 #endif
935 		rt_cache_flush(dev_net(dev), -1);
936 		break;
937 	case NETDEV_DOWN:
938 		fib_disable_ip(dev, 0, 0);
939 		break;
940 	case NETDEV_CHANGEMTU:
941 	case NETDEV_CHANGE:
942 		rt_cache_flush(dev_net(dev), 0);
943 		break;
944 	case NETDEV_UNREGISTER_BATCH:
945 		/* The batch unregister is only called on the first
946 		 * device in the list of devices being unregistered.
947 		 * Therefore we should not pass dev_net(dev) in here.
948 		 */
949 		rt_cache_flush_batch(NULL);
950 		break;
951 	}
952 	return NOTIFY_DONE;
953 }
954 
955 static struct notifier_block fib_inetaddr_notifier = {
956 	.notifier_call = fib_inetaddr_event,
957 };
958 
959 static struct notifier_block fib_netdev_notifier = {
960 	.notifier_call = fib_netdev_event,
961 };
962 
963 static int __net_init ip_fib_net_init(struct net *net)
964 {
965 	int err;
966 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
967 
968 	/* Avoid false sharing : Use at least a full cache line */
969 	size = max_t(size_t, size, L1_CACHE_BYTES);
970 
971 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
972 	if (net->ipv4.fib_table_hash == NULL)
973 		return -ENOMEM;
974 
975 	err = fib4_rules_init(net);
976 	if (err < 0)
977 		goto fail;
978 	return 0;
979 
980 fail:
981 	kfree(net->ipv4.fib_table_hash);
982 	return err;
983 }
984 
985 static void ip_fib_net_exit(struct net *net)
986 {
987 	unsigned int i;
988 
989 #ifdef CONFIG_IP_MULTIPLE_TABLES
990 	fib4_rules_exit(net);
991 #endif
992 
993 	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
994 		struct fib_table *tb;
995 		struct hlist_head *head;
996 		struct hlist_node *node, *tmp;
997 
998 		head = &net->ipv4.fib_table_hash[i];
999 		hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1000 			hlist_del(node);
1001 			fib_table_flush(tb);
1002 			fib_free_table(tb);
1003 		}
1004 	}
1005 	kfree(net->ipv4.fib_table_hash);
1006 }
1007 
1008 static int __net_init fib_net_init(struct net *net)
1009 {
1010 	int error;
1011 
1012 	error = ip_fib_net_init(net);
1013 	if (error < 0)
1014 		goto out;
1015 	error = nl_fib_lookup_init(net);
1016 	if (error < 0)
1017 		goto out_nlfl;
1018 	error = fib_proc_init(net);
1019 	if (error < 0)
1020 		goto out_proc;
1021 out:
1022 	return error;
1023 
1024 out_proc:
1025 	nl_fib_lookup_exit(net);
1026 out_nlfl:
1027 	ip_fib_net_exit(net);
1028 	goto out;
1029 }
1030 
1031 static void __net_exit fib_net_exit(struct net *net)
1032 {
1033 	fib_proc_exit(net);
1034 	nl_fib_lookup_exit(net);
1035 	ip_fib_net_exit(net);
1036 }
1037 
1038 static struct pernet_operations fib_net_ops = {
1039 	.init = fib_net_init,
1040 	.exit = fib_net_exit,
1041 };
1042 
1043 void __init ip_fib_init(void)
1044 {
1045 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL);
1046 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL);
1047 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib);
1048 
1049 	register_pernet_subsys(&fib_net_ops);
1050 	register_netdevice_notifier(&fib_netdev_notifier);
1051 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1052 
1053 	fib_trie_init();
1054 }
1055