xref: /linux/net/ipv4/fib_frontend.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
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_hash_table(RT_TABLE_LOCAL);
55 	if (local_table == NULL)
56 		return -ENOMEM;
57 
58 	main_table  = fib_hash_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_hash_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 void fib_select_default(struct net *net,
118 			const struct flowi *flp, struct fib_result *res)
119 {
120 	struct fib_table *tb;
121 	int table = RT_TABLE_MAIN;
122 #ifdef CONFIG_IP_MULTIPLE_TABLES
123 	if (res->r == NULL || res->r->action != FR_ACT_TO_TBL)
124 		return;
125 	table = res->r->table;
126 #endif
127 	tb = fib_get_table(net, table);
128 	if (FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
129 		fib_table_select_default(tb, flp, res);
130 }
131 
132 static void fib_flush(struct net *net)
133 {
134 	int flushed = 0;
135 	struct fib_table *tb;
136 	struct hlist_node *node;
137 	struct hlist_head *head;
138 	unsigned int h;
139 
140 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
141 		head = &net->ipv4.fib_table_hash[h];
142 		hlist_for_each_entry(tb, node, head, tb_hlist)
143 			flushed += fib_table_flush(tb);
144 	}
145 
146 	if (flushed)
147 		rt_cache_flush(net, -1);
148 }
149 
150 /*
151  *	Find the first device with a given source address.
152  */
153 
154 struct net_device * ip_dev_find(struct net *net, __be32 addr)
155 {
156 	struct flowi fl = { .nl_u = { .ip4_u = { .daddr = addr } } };
157 	struct fib_result res;
158 	struct net_device *dev = NULL;
159 	struct fib_table *local_table;
160 
161 #ifdef CONFIG_IP_MULTIPLE_TABLES
162 	res.r = NULL;
163 #endif
164 
165 	local_table = fib_get_table(net, RT_TABLE_LOCAL);
166 	if (!local_table || fib_table_lookup(local_table, &fl, &res))
167 		return NULL;
168 	if (res.type != RTN_LOCAL)
169 		goto out;
170 	dev = FIB_RES_DEV(res);
171 
172 	if (dev)
173 		dev_hold(dev);
174 out:
175 	fib_res_put(&res);
176 	return dev;
177 }
178 EXPORT_SYMBOL(ip_dev_find);
179 
180 /*
181  * Find address type as if only "dev" was present in the system. If
182  * on_dev is NULL then all interfaces are taken into consideration.
183  */
184 static inline unsigned __inet_dev_addr_type(struct net *net,
185 					    const struct net_device *dev,
186 					    __be32 addr)
187 {
188 	struct flowi		fl = { .nl_u = { .ip4_u = { .daddr = addr } } };
189 	struct fib_result	res;
190 	unsigned ret = RTN_BROADCAST;
191 	struct fib_table *local_table;
192 
193 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
194 		return RTN_BROADCAST;
195 	if (ipv4_is_multicast(addr))
196 		return RTN_MULTICAST;
197 
198 #ifdef CONFIG_IP_MULTIPLE_TABLES
199 	res.r = NULL;
200 #endif
201 
202 	local_table = fib_get_table(net, RT_TABLE_LOCAL);
203 	if (local_table) {
204 		ret = RTN_UNICAST;
205 		if (!fib_table_lookup(local_table, &fl, &res)) {
206 			if (!dev || dev == res.fi->fib_dev)
207 				ret = res.type;
208 			fib_res_put(&res);
209 		}
210 	}
211 	return ret;
212 }
213 
214 unsigned int inet_addr_type(struct net *net, __be32 addr)
215 {
216 	return __inet_dev_addr_type(net, NULL, addr);
217 }
218 EXPORT_SYMBOL(inet_addr_type);
219 
220 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
221 				__be32 addr)
222 {
223        return __inet_dev_addr_type(net, dev, addr);
224 }
225 EXPORT_SYMBOL(inet_dev_addr_type);
226 
227 /* Given (packet source, input interface) and optional (dst, oif, tos):
228    - (main) check, that source is valid i.e. not broadcast or our local
229      address.
230    - figure out what "logical" interface this packet arrived
231      and calculate "specific destination" address.
232    - check, that packet arrived from expected physical interface.
233  */
234 
235 int fib_validate_source(__be32 src, __be32 dst, u8 tos, int oif,
236 			struct net_device *dev, __be32 *spec_dst,
237 			u32 *itag, u32 mark)
238 {
239 	struct in_device *in_dev;
240 	struct flowi fl = { .nl_u = { .ip4_u =
241 				      { .daddr = src,
242 					.saddr = dst,
243 					.tos = tos } },
244 			    .mark = mark,
245 			    .iif = oif };
246 
247 	struct fib_result res;
248 	int no_addr, rpf, accept_local;
249 	int ret;
250 	struct net *net;
251 
252 	no_addr = rpf = accept_local = 0;
253 	rcu_read_lock();
254 	in_dev = __in_dev_get_rcu(dev);
255 	if (in_dev) {
256 		no_addr = in_dev->ifa_list == NULL;
257 		rpf = IN_DEV_RPFILTER(in_dev);
258 		accept_local = IN_DEV_ACCEPT_LOCAL(in_dev);
259 		if (mark && !IN_DEV_SRC_VMARK(in_dev))
260 			fl.mark = 0;
261 	}
262 	rcu_read_unlock();
263 
264 	if (in_dev == NULL)
265 		goto e_inval;
266 
267 	net = dev_net(dev);
268 	if (fib_lookup(net, &fl, &res))
269 		goto last_resort;
270 	if (res.type != RTN_UNICAST) {
271 		if (res.type != RTN_LOCAL || !accept_local)
272 			goto e_inval_res;
273 	}
274 	*spec_dst = FIB_RES_PREFSRC(res);
275 	fib_combine_itag(itag, &res);
276 #ifdef CONFIG_IP_ROUTE_MULTIPATH
277 	if (FIB_RES_DEV(res) == dev || res.fi->fib_nhs > 1)
278 #else
279 	if (FIB_RES_DEV(res) == dev)
280 #endif
281 	{
282 		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
283 		fib_res_put(&res);
284 		return ret;
285 	}
286 	fib_res_put(&res);
287 	if (no_addr)
288 		goto last_resort;
289 	if (rpf == 1)
290 		goto e_rpf;
291 	fl.oif = dev->ifindex;
292 
293 	ret = 0;
294 	if (fib_lookup(net, &fl, &res) == 0) {
295 		if (res.type == RTN_UNICAST) {
296 			*spec_dst = FIB_RES_PREFSRC(res);
297 			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
298 		}
299 		fib_res_put(&res);
300 	}
301 	return ret;
302 
303 last_resort:
304 	if (rpf)
305 		goto e_rpf;
306 	*spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
307 	*itag = 0;
308 	return 0;
309 
310 e_inval_res:
311 	fib_res_put(&res);
312 e_inval:
313 	return -EINVAL;
314 e_rpf:
315 	return -EXDEV;
316 }
317 
318 static inline __be32 sk_extract_addr(struct sockaddr *addr)
319 {
320 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
321 }
322 
323 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
324 {
325 	struct nlattr *nla;
326 
327 	nla = (struct nlattr *) ((char *) mx + len);
328 	nla->nla_type = type;
329 	nla->nla_len = nla_attr_size(4);
330 	*(u32 *) nla_data(nla) = value;
331 
332 	return len + nla_total_size(4);
333 }
334 
335 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
336 				 struct fib_config *cfg)
337 {
338 	__be32 addr;
339 	int plen;
340 
341 	memset(cfg, 0, sizeof(*cfg));
342 	cfg->fc_nlinfo.nl_net = net;
343 
344 	if (rt->rt_dst.sa_family != AF_INET)
345 		return -EAFNOSUPPORT;
346 
347 	/*
348 	 * Check mask for validity:
349 	 * a) it must be contiguous.
350 	 * b) destination must have all host bits clear.
351 	 * c) if application forgot to set correct family (AF_INET),
352 	 *    reject request unless it is absolutely clear i.e.
353 	 *    both family and mask are zero.
354 	 */
355 	plen = 32;
356 	addr = sk_extract_addr(&rt->rt_dst);
357 	if (!(rt->rt_flags & RTF_HOST)) {
358 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
359 
360 		if (rt->rt_genmask.sa_family != AF_INET) {
361 			if (mask || rt->rt_genmask.sa_family)
362 				return -EAFNOSUPPORT;
363 		}
364 
365 		if (bad_mask(mask, addr))
366 			return -EINVAL;
367 
368 		plen = inet_mask_len(mask);
369 	}
370 
371 	cfg->fc_dst_len = plen;
372 	cfg->fc_dst = addr;
373 
374 	if (cmd != SIOCDELRT) {
375 		cfg->fc_nlflags = NLM_F_CREATE;
376 		cfg->fc_protocol = RTPROT_BOOT;
377 	}
378 
379 	if (rt->rt_metric)
380 		cfg->fc_priority = rt->rt_metric - 1;
381 
382 	if (rt->rt_flags & RTF_REJECT) {
383 		cfg->fc_scope = RT_SCOPE_HOST;
384 		cfg->fc_type = RTN_UNREACHABLE;
385 		return 0;
386 	}
387 
388 	cfg->fc_scope = RT_SCOPE_NOWHERE;
389 	cfg->fc_type = RTN_UNICAST;
390 
391 	if (rt->rt_dev) {
392 		char *colon;
393 		struct net_device *dev;
394 		char devname[IFNAMSIZ];
395 
396 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
397 			return -EFAULT;
398 
399 		devname[IFNAMSIZ-1] = 0;
400 		colon = strchr(devname, ':');
401 		if (colon)
402 			*colon = 0;
403 		dev = __dev_get_by_name(net, devname);
404 		if (!dev)
405 			return -ENODEV;
406 		cfg->fc_oif = dev->ifindex;
407 		if (colon) {
408 			struct in_ifaddr *ifa;
409 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
410 			if (!in_dev)
411 				return -ENODEV;
412 			*colon = ':';
413 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
414 				if (strcmp(ifa->ifa_label, devname) == 0)
415 					break;
416 			if (ifa == NULL)
417 				return -ENODEV;
418 			cfg->fc_prefsrc = ifa->ifa_local;
419 		}
420 	}
421 
422 	addr = sk_extract_addr(&rt->rt_gateway);
423 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
424 		cfg->fc_gw = addr;
425 		if (rt->rt_flags & RTF_GATEWAY &&
426 		    inet_addr_type(net, addr) == RTN_UNICAST)
427 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
428 	}
429 
430 	if (cmd == SIOCDELRT)
431 		return 0;
432 
433 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
434 		return -EINVAL;
435 
436 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
437 		cfg->fc_scope = RT_SCOPE_LINK;
438 
439 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
440 		struct nlattr *mx;
441 		int len = 0;
442 
443 		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
444 		if (mx == NULL)
445 			return -ENOMEM;
446 
447 		if (rt->rt_flags & RTF_MTU)
448 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
449 
450 		if (rt->rt_flags & RTF_WINDOW)
451 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
452 
453 		if (rt->rt_flags & RTF_IRTT)
454 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
455 
456 		cfg->fc_mx = mx;
457 		cfg->fc_mx_len = len;
458 	}
459 
460 	return 0;
461 }
462 
463 /*
464  *	Handle IP routing ioctl calls. These are used to manipulate the routing tables
465  */
466 
467 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
468 {
469 	struct fib_config cfg;
470 	struct rtentry rt;
471 	int err;
472 
473 	switch (cmd) {
474 	case SIOCADDRT:		/* Add a route */
475 	case SIOCDELRT:		/* Delete a route */
476 		if (!capable(CAP_NET_ADMIN))
477 			return -EPERM;
478 
479 		if (copy_from_user(&rt, arg, sizeof(rt)))
480 			return -EFAULT;
481 
482 		rtnl_lock();
483 		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
484 		if (err == 0) {
485 			struct fib_table *tb;
486 
487 			if (cmd == SIOCDELRT) {
488 				tb = fib_get_table(net, cfg.fc_table);
489 				if (tb)
490 					err = fib_table_delete(tb, &cfg);
491 				else
492 					err = -ESRCH;
493 			} else {
494 				tb = fib_new_table(net, cfg.fc_table);
495 				if (tb)
496 					err = fib_table_insert(tb, &cfg);
497 				else
498 					err = -ENOBUFS;
499 			}
500 
501 			/* allocated by rtentry_to_fib_config() */
502 			kfree(cfg.fc_mx);
503 		}
504 		rtnl_unlock();
505 		return err;
506 	}
507 	return -EINVAL;
508 }
509 
510 const struct nla_policy rtm_ipv4_policy[RTA_MAX+1] = {
511 	[RTA_DST]		= { .type = NLA_U32 },
512 	[RTA_SRC]		= { .type = NLA_U32 },
513 	[RTA_IIF]		= { .type = NLA_U32 },
514 	[RTA_OIF]		= { .type = NLA_U32 },
515 	[RTA_GATEWAY]		= { .type = NLA_U32 },
516 	[RTA_PRIORITY]		= { .type = NLA_U32 },
517 	[RTA_PREFSRC]		= { .type = NLA_U32 },
518 	[RTA_METRICS]		= { .type = NLA_NESTED },
519 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
520 	[RTA_FLOW]		= { .type = NLA_U32 },
521 };
522 
523 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
524 			    struct nlmsghdr *nlh, struct fib_config *cfg)
525 {
526 	struct nlattr *attr;
527 	int err, remaining;
528 	struct rtmsg *rtm;
529 
530 	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
531 	if (err < 0)
532 		goto errout;
533 
534 	memset(cfg, 0, sizeof(*cfg));
535 
536 	rtm = nlmsg_data(nlh);
537 	cfg->fc_dst_len = rtm->rtm_dst_len;
538 	cfg->fc_tos = rtm->rtm_tos;
539 	cfg->fc_table = rtm->rtm_table;
540 	cfg->fc_protocol = rtm->rtm_protocol;
541 	cfg->fc_scope = rtm->rtm_scope;
542 	cfg->fc_type = rtm->rtm_type;
543 	cfg->fc_flags = rtm->rtm_flags;
544 	cfg->fc_nlflags = nlh->nlmsg_flags;
545 
546 	cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
547 	cfg->fc_nlinfo.nlh = nlh;
548 	cfg->fc_nlinfo.nl_net = net;
549 
550 	if (cfg->fc_type > RTN_MAX) {
551 		err = -EINVAL;
552 		goto errout;
553 	}
554 
555 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
556 		switch (nla_type(attr)) {
557 		case RTA_DST:
558 			cfg->fc_dst = nla_get_be32(attr);
559 			break;
560 		case RTA_OIF:
561 			cfg->fc_oif = nla_get_u32(attr);
562 			break;
563 		case RTA_GATEWAY:
564 			cfg->fc_gw = nla_get_be32(attr);
565 			break;
566 		case RTA_PRIORITY:
567 			cfg->fc_priority = nla_get_u32(attr);
568 			break;
569 		case RTA_PREFSRC:
570 			cfg->fc_prefsrc = nla_get_be32(attr);
571 			break;
572 		case RTA_METRICS:
573 			cfg->fc_mx = nla_data(attr);
574 			cfg->fc_mx_len = nla_len(attr);
575 			break;
576 		case RTA_MULTIPATH:
577 			cfg->fc_mp = nla_data(attr);
578 			cfg->fc_mp_len = nla_len(attr);
579 			break;
580 		case RTA_FLOW:
581 			cfg->fc_flow = nla_get_u32(attr);
582 			break;
583 		case RTA_TABLE:
584 			cfg->fc_table = nla_get_u32(attr);
585 			break;
586 		}
587 	}
588 
589 	return 0;
590 errout:
591 	return err;
592 }
593 
594 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
595 {
596 	struct net *net = sock_net(skb->sk);
597 	struct fib_config cfg;
598 	struct fib_table *tb;
599 	int err;
600 
601 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
602 	if (err < 0)
603 		goto errout;
604 
605 	tb = fib_get_table(net, cfg.fc_table);
606 	if (tb == NULL) {
607 		err = -ESRCH;
608 		goto errout;
609 	}
610 
611 	err = fib_table_delete(tb, &cfg);
612 errout:
613 	return err;
614 }
615 
616 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
617 {
618 	struct net *net = sock_net(skb->sk);
619 	struct fib_config cfg;
620 	struct fib_table *tb;
621 	int err;
622 
623 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
624 	if (err < 0)
625 		goto errout;
626 
627 	tb = fib_new_table(net, cfg.fc_table);
628 	if (tb == NULL) {
629 		err = -ENOBUFS;
630 		goto errout;
631 	}
632 
633 	err = fib_table_insert(tb, &cfg);
634 errout:
635 	return err;
636 }
637 
638 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
639 {
640 	struct net *net = sock_net(skb->sk);
641 	unsigned int h, s_h;
642 	unsigned int e = 0, s_e;
643 	struct fib_table *tb;
644 	struct hlist_node *node;
645 	struct hlist_head *head;
646 	int dumped = 0;
647 
648 	if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
649 	    ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
650 		return ip_rt_dump(skb, cb);
651 
652 	s_h = cb->args[0];
653 	s_e = cb->args[1];
654 
655 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
656 		e = 0;
657 		head = &net->ipv4.fib_table_hash[h];
658 		hlist_for_each_entry(tb, node, head, tb_hlist) {
659 			if (e < s_e)
660 				goto next;
661 			if (dumped)
662 				memset(&cb->args[2], 0, sizeof(cb->args) -
663 						 2 * sizeof(cb->args[0]));
664 			if (fib_table_dump(tb, skb, cb) < 0)
665 				goto out;
666 			dumped = 1;
667 next:
668 			e++;
669 		}
670 	}
671 out:
672 	cb->args[1] = e;
673 	cb->args[0] = h;
674 
675 	return skb->len;
676 }
677 
678 /* Prepare and feed intra-kernel routing request.
679    Really, it should be netlink message, but :-( netlink
680    can be not configured, so that we feed it directly
681    to fib engine. It is legal, because all events occur
682    only when netlink is already locked.
683  */
684 
685 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
686 {
687 	struct net *net = dev_net(ifa->ifa_dev->dev);
688 	struct fib_table *tb;
689 	struct fib_config cfg = {
690 		.fc_protocol = RTPROT_KERNEL,
691 		.fc_type = type,
692 		.fc_dst = dst,
693 		.fc_dst_len = dst_len,
694 		.fc_prefsrc = ifa->ifa_local,
695 		.fc_oif = ifa->ifa_dev->dev->ifindex,
696 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
697 		.fc_nlinfo = {
698 			.nl_net = net,
699 		},
700 	};
701 
702 	if (type == RTN_UNICAST)
703 		tb = fib_new_table(net, RT_TABLE_MAIN);
704 	else
705 		tb = fib_new_table(net, RT_TABLE_LOCAL);
706 
707 	if (tb == NULL)
708 		return;
709 
710 	cfg.fc_table = tb->tb_id;
711 
712 	if (type != RTN_LOCAL)
713 		cfg.fc_scope = RT_SCOPE_LINK;
714 	else
715 		cfg.fc_scope = RT_SCOPE_HOST;
716 
717 	if (cmd == RTM_NEWROUTE)
718 		fib_table_insert(tb, &cfg);
719 	else
720 		fib_table_delete(tb, &cfg);
721 }
722 
723 void fib_add_ifaddr(struct in_ifaddr *ifa)
724 {
725 	struct in_device *in_dev = ifa->ifa_dev;
726 	struct net_device *dev = in_dev->dev;
727 	struct in_ifaddr *prim = ifa;
728 	__be32 mask = ifa->ifa_mask;
729 	__be32 addr = ifa->ifa_local;
730 	__be32 prefix = ifa->ifa_address&mask;
731 
732 	if (ifa->ifa_flags&IFA_F_SECONDARY) {
733 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
734 		if (prim == NULL) {
735 			printk(KERN_WARNING "fib_add_ifaddr: bug: prim == NULL\n");
736 			return;
737 		}
738 	}
739 
740 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
741 
742 	if (!(dev->flags&IFF_UP))
743 		return;
744 
745 	/* Add broadcast address, if it is explicitly assigned. */
746 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
747 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
748 
749 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags&IFA_F_SECONDARY) &&
750 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
751 		fib_magic(RTM_NEWROUTE, dev->flags&IFF_LOOPBACK ? RTN_LOCAL :
752 			  RTN_UNICAST, prefix, ifa->ifa_prefixlen, prim);
753 
754 		/* Add network specific broadcasts, when it takes a sense */
755 		if (ifa->ifa_prefixlen < 31) {
756 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
757 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix|~mask, 32, prim);
758 		}
759 	}
760 }
761 
762 static void fib_del_ifaddr(struct in_ifaddr *ifa)
763 {
764 	struct in_device *in_dev = ifa->ifa_dev;
765 	struct net_device *dev = in_dev->dev;
766 	struct in_ifaddr *ifa1;
767 	struct in_ifaddr *prim = ifa;
768 	__be32 brd = ifa->ifa_address|~ifa->ifa_mask;
769 	__be32 any = ifa->ifa_address&ifa->ifa_mask;
770 #define LOCAL_OK	1
771 #define BRD_OK		2
772 #define BRD0_OK		4
773 #define BRD1_OK		8
774 	unsigned ok = 0;
775 
776 	if (!(ifa->ifa_flags&IFA_F_SECONDARY))
777 		fib_magic(RTM_DELROUTE, dev->flags&IFF_LOOPBACK ? RTN_LOCAL :
778 			  RTN_UNICAST, any, ifa->ifa_prefixlen, prim);
779 	else {
780 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
781 		if (prim == NULL) {
782 			printk(KERN_WARNING "fib_del_ifaddr: bug: prim == NULL\n");
783 			return;
784 		}
785 	}
786 
787 	/* Deletion is more complicated than add.
788 	   We should take care of not to delete too much :-)
789 
790 	   Scan address list to be sure that addresses are really gone.
791 	 */
792 
793 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
794 		if (ifa->ifa_local == ifa1->ifa_local)
795 			ok |= LOCAL_OK;
796 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
797 			ok |= BRD_OK;
798 		if (brd == ifa1->ifa_broadcast)
799 			ok |= BRD1_OK;
800 		if (any == ifa1->ifa_broadcast)
801 			ok |= BRD0_OK;
802 	}
803 
804 	if (!(ok&BRD_OK))
805 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
806 	if (!(ok&BRD1_OK))
807 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
808 	if (!(ok&BRD0_OK))
809 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
810 	if (!(ok&LOCAL_OK)) {
811 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
812 
813 		/* Check, that this local address finally disappeared. */
814 		if (inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
815 			/* And the last, but not the least thing.
816 			   We must flush stray FIB entries.
817 
818 			   First of all, we scan fib_info list searching
819 			   for stray nexthop entries, then ignite fib_flush.
820 			*/
821 			if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
822 				fib_flush(dev_net(dev));
823 		}
824 	}
825 #undef LOCAL_OK
826 #undef BRD_OK
827 #undef BRD0_OK
828 #undef BRD1_OK
829 }
830 
831 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb )
832 {
833 
834 	struct fib_result       res;
835 	struct flowi            fl = { .mark = frn->fl_mark,
836 				       .nl_u = { .ip4_u = { .daddr = frn->fl_addr,
837 							    .tos = frn->fl_tos,
838 							    .scope = frn->fl_scope } } };
839 
840 #ifdef CONFIG_IP_MULTIPLE_TABLES
841 	res.r = NULL;
842 #endif
843 
844 	frn->err = -ENOENT;
845 	if (tb) {
846 		local_bh_disable();
847 
848 		frn->tb_id = tb->tb_id;
849 		frn->err = fib_table_lookup(tb, &fl, &res);
850 
851 		if (!frn->err) {
852 			frn->prefixlen = res.prefixlen;
853 			frn->nh_sel = res.nh_sel;
854 			frn->type = res.type;
855 			frn->scope = res.scope;
856 			fib_res_put(&res);
857 		}
858 		local_bh_enable();
859 	}
860 }
861 
862 static void nl_fib_input(struct sk_buff *skb)
863 {
864 	struct net *net;
865 	struct fib_result_nl *frn;
866 	struct nlmsghdr *nlh;
867 	struct fib_table *tb;
868 	u32 pid;
869 
870 	net = sock_net(skb->sk);
871 	nlh = nlmsg_hdr(skb);
872 	if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
873 	    nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
874 		return;
875 
876 	skb = skb_clone(skb, GFP_KERNEL);
877 	if (skb == NULL)
878 		return;
879 	nlh = nlmsg_hdr(skb);
880 
881 	frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
882 	tb = fib_get_table(net, frn->tb_id_in);
883 
884 	nl_fib_lookup(frn, tb);
885 
886 	pid = NETLINK_CB(skb).pid;       /* pid of sending process */
887 	NETLINK_CB(skb).pid = 0;         /* from kernel */
888 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
889 	netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
890 }
891 
892 static int __net_init nl_fib_lookup_init(struct net *net)
893 {
894 	struct sock *sk;
895 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0,
896 				   nl_fib_input, NULL, THIS_MODULE);
897 	if (sk == NULL)
898 		return -EAFNOSUPPORT;
899 	net->ipv4.fibnl = sk;
900 	return 0;
901 }
902 
903 static void nl_fib_lookup_exit(struct net *net)
904 {
905 	netlink_kernel_release(net->ipv4.fibnl);
906 	net->ipv4.fibnl = NULL;
907 }
908 
909 static void fib_disable_ip(struct net_device *dev, int force, int delay)
910 {
911 	if (fib_sync_down_dev(dev, force))
912 		fib_flush(dev_net(dev));
913 	rt_cache_flush(dev_net(dev), delay);
914 	arp_ifdown(dev);
915 }
916 
917 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
918 {
919 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
920 	struct net_device *dev = ifa->ifa_dev->dev;
921 
922 	switch (event) {
923 	case NETDEV_UP:
924 		fib_add_ifaddr(ifa);
925 #ifdef CONFIG_IP_ROUTE_MULTIPATH
926 		fib_sync_up(dev);
927 #endif
928 		rt_cache_flush(dev_net(dev), -1);
929 		break;
930 	case NETDEV_DOWN:
931 		fib_del_ifaddr(ifa);
932 		if (ifa->ifa_dev->ifa_list == NULL) {
933 			/* Last address was deleted from this interface.
934 			   Disable IP.
935 			 */
936 			fib_disable_ip(dev, 1, 0);
937 		} else {
938 			rt_cache_flush(dev_net(dev), -1);
939 		}
940 		break;
941 	}
942 	return NOTIFY_DONE;
943 }
944 
945 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
946 {
947 	struct net_device *dev = ptr;
948 	struct in_device *in_dev = __in_dev_get_rtnl(dev);
949 
950 	if (event == NETDEV_UNREGISTER) {
951 		fib_disable_ip(dev, 2, -1);
952 		return NOTIFY_DONE;
953 	}
954 
955 	if (!in_dev)
956 		return NOTIFY_DONE;
957 
958 	switch (event) {
959 	case NETDEV_UP:
960 		for_ifa(in_dev) {
961 			fib_add_ifaddr(ifa);
962 		} endfor_ifa(in_dev);
963 #ifdef CONFIG_IP_ROUTE_MULTIPATH
964 		fib_sync_up(dev);
965 #endif
966 		rt_cache_flush(dev_net(dev), -1);
967 		break;
968 	case NETDEV_DOWN:
969 		fib_disable_ip(dev, 0, 0);
970 		break;
971 	case NETDEV_CHANGEMTU:
972 	case NETDEV_CHANGE:
973 		rt_cache_flush(dev_net(dev), 0);
974 		break;
975 	case NETDEV_UNREGISTER_BATCH:
976 		rt_cache_flush_batch();
977 		break;
978 	}
979 	return NOTIFY_DONE;
980 }
981 
982 static struct notifier_block fib_inetaddr_notifier = {
983 	.notifier_call = fib_inetaddr_event,
984 };
985 
986 static struct notifier_block fib_netdev_notifier = {
987 	.notifier_call = fib_netdev_event,
988 };
989 
990 static int __net_init ip_fib_net_init(struct net *net)
991 {
992 	int err;
993 	unsigned int i;
994 
995 	net->ipv4.fib_table_hash = kzalloc(
996 			sizeof(struct hlist_head)*FIB_TABLE_HASHSZ, GFP_KERNEL);
997 	if (net->ipv4.fib_table_hash == NULL)
998 		return -ENOMEM;
999 
1000 	for (i = 0; i < FIB_TABLE_HASHSZ; i++)
1001 		INIT_HLIST_HEAD(&net->ipv4.fib_table_hash[i]);
1002 
1003 	err = fib4_rules_init(net);
1004 	if (err < 0)
1005 		goto fail;
1006 	return 0;
1007 
1008 fail:
1009 	kfree(net->ipv4.fib_table_hash);
1010 	return err;
1011 }
1012 
1013 static void ip_fib_net_exit(struct net *net)
1014 {
1015 	unsigned int i;
1016 
1017 #ifdef CONFIG_IP_MULTIPLE_TABLES
1018 	fib4_rules_exit(net);
1019 #endif
1020 
1021 	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1022 		struct fib_table *tb;
1023 		struct hlist_head *head;
1024 		struct hlist_node *node, *tmp;
1025 
1026 		head = &net->ipv4.fib_table_hash[i];
1027 		hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1028 			hlist_del(node);
1029 			fib_table_flush(tb);
1030 			kfree(tb);
1031 		}
1032 	}
1033 	kfree(net->ipv4.fib_table_hash);
1034 }
1035 
1036 static int __net_init fib_net_init(struct net *net)
1037 {
1038 	int error;
1039 
1040 	error = ip_fib_net_init(net);
1041 	if (error < 0)
1042 		goto out;
1043 	error = nl_fib_lookup_init(net);
1044 	if (error < 0)
1045 		goto out_nlfl;
1046 	error = fib_proc_init(net);
1047 	if (error < 0)
1048 		goto out_proc;
1049 out:
1050 	return error;
1051 
1052 out_proc:
1053 	nl_fib_lookup_exit(net);
1054 out_nlfl:
1055 	ip_fib_net_exit(net);
1056 	goto out;
1057 }
1058 
1059 static void __net_exit fib_net_exit(struct net *net)
1060 {
1061 	fib_proc_exit(net);
1062 	nl_fib_lookup_exit(net);
1063 	ip_fib_net_exit(net);
1064 }
1065 
1066 static struct pernet_operations fib_net_ops = {
1067 	.init = fib_net_init,
1068 	.exit = fib_net_exit,
1069 };
1070 
1071 void __init ip_fib_init(void)
1072 {
1073 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL);
1074 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL);
1075 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib);
1076 
1077 	register_pernet_subsys(&fib_net_ops);
1078 	register_netdevice_notifier(&fib_netdev_notifier);
1079 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1080 
1081 	fib_hash_init();
1082 }
1083