xref: /linux/net/ipv4/fib_frontend.c (revision c5aec4c76af1a2d89ee2f2d4d5463b2ad2d85de5)
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 <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 
49 #ifndef CONFIG_IP_MULTIPLE_TABLES
50 
51 static int __net_init fib4_rules_init(struct net *net)
52 {
53 	struct fib_table *local_table, *main_table;
54 
55 	local_table = fib_trie_table(RT_TABLE_LOCAL);
56 	if (local_table == NULL)
57 		return -ENOMEM;
58 
59 	main_table  = fib_trie_table(RT_TABLE_MAIN);
60 	if (main_table == NULL)
61 		goto fail;
62 
63 	hlist_add_head_rcu(&local_table->tb_hlist,
64 				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
65 	hlist_add_head_rcu(&main_table->tb_hlist,
66 				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
67 	return 0;
68 
69 fail:
70 	kfree(local_table);
71 	return -ENOMEM;
72 }
73 #else
74 
75 struct fib_table *fib_new_table(struct net *net, u32 id)
76 {
77 	struct fib_table *tb;
78 	unsigned int h;
79 
80 	if (id == 0)
81 		id = RT_TABLE_MAIN;
82 	tb = fib_get_table(net, id);
83 	if (tb)
84 		return tb;
85 
86 	tb = fib_trie_table(id);
87 	if (!tb)
88 		return NULL;
89 
90 	switch (id) {
91 	case RT_TABLE_LOCAL:
92 		net->ipv4.fib_local = tb;
93 		break;
94 
95 	case RT_TABLE_MAIN:
96 		net->ipv4.fib_main = tb;
97 		break;
98 
99 	case RT_TABLE_DEFAULT:
100 		net->ipv4.fib_default = tb;
101 		break;
102 
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 
112 struct fib_table *fib_get_table(struct net *net, u32 id)
113 {
114 	struct fib_table *tb;
115 	struct hlist_head *head;
116 	unsigned int h;
117 
118 	if (id == 0)
119 		id = RT_TABLE_MAIN;
120 	h = id & (FIB_TABLE_HASHSZ - 1);
121 
122 	rcu_read_lock();
123 	head = &net->ipv4.fib_table_hash[h];
124 	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
125 		if (tb->tb_id == id) {
126 			rcu_read_unlock();
127 			return tb;
128 		}
129 	}
130 	rcu_read_unlock();
131 	return NULL;
132 }
133 #endif /* CONFIG_IP_MULTIPLE_TABLES */
134 
135 static void fib_flush(struct net *net)
136 {
137 	int flushed = 0;
138 	struct fib_table *tb;
139 	struct hlist_head *head;
140 	unsigned int h;
141 
142 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
143 		head = &net->ipv4.fib_table_hash[h];
144 		hlist_for_each_entry(tb, head, tb_hlist)
145 			flushed += fib_table_flush(tb);
146 	}
147 
148 	if (flushed)
149 		rt_cache_flush(net);
150 }
151 
152 /*
153  * Find address type as if only "dev" was present in the system. If
154  * on_dev is NULL then all interfaces are taken into consideration.
155  */
156 static inline unsigned int __inet_dev_addr_type(struct net *net,
157 						const struct net_device *dev,
158 						__be32 addr)
159 {
160 	struct flowi4		fl4 = { .daddr = addr };
161 	struct fib_result	res;
162 	unsigned int ret = RTN_BROADCAST;
163 	struct fib_table *local_table;
164 
165 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
166 		return RTN_BROADCAST;
167 	if (ipv4_is_multicast(addr))
168 		return RTN_MULTICAST;
169 
170 	local_table = fib_get_table(net, RT_TABLE_LOCAL);
171 	if (local_table) {
172 		ret = RTN_UNICAST;
173 		rcu_read_lock();
174 		if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) {
175 			if (!dev || dev == res.fi->fib_dev)
176 				ret = res.type;
177 		}
178 		rcu_read_unlock();
179 	}
180 	return ret;
181 }
182 
183 unsigned int inet_addr_type(struct net *net, __be32 addr)
184 {
185 	return __inet_dev_addr_type(net, NULL, addr);
186 }
187 EXPORT_SYMBOL(inet_addr_type);
188 
189 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
190 				__be32 addr)
191 {
192 	return __inet_dev_addr_type(net, dev, addr);
193 }
194 EXPORT_SYMBOL(inet_dev_addr_type);
195 
196 __be32 fib_compute_spec_dst(struct sk_buff *skb)
197 {
198 	struct net_device *dev = skb->dev;
199 	struct in_device *in_dev;
200 	struct fib_result res;
201 	struct rtable *rt;
202 	struct flowi4 fl4;
203 	struct net *net;
204 	int scope;
205 
206 	rt = skb_rtable(skb);
207 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
208 	    RTCF_LOCAL)
209 		return ip_hdr(skb)->daddr;
210 
211 	in_dev = __in_dev_get_rcu(dev);
212 	BUG_ON(!in_dev);
213 
214 	net = dev_net(dev);
215 
216 	scope = RT_SCOPE_UNIVERSE;
217 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
218 		fl4.flowi4_oif = 0;
219 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
220 		fl4.daddr = ip_hdr(skb)->saddr;
221 		fl4.saddr = 0;
222 		fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
223 		fl4.flowi4_scope = scope;
224 		fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
225 		if (!fib_lookup(net, &fl4, &res))
226 			return FIB_RES_PREFSRC(net, res);
227 	} else {
228 		scope = RT_SCOPE_LINK;
229 	}
230 
231 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
232 }
233 
234 /* Given (packet source, input interface) and optional (dst, oif, tos):
235  * - (main) check, that source is valid i.e. not broadcast or our local
236  *   address.
237  * - figure out what "logical" interface this packet arrived
238  *   and calculate "specific destination" address.
239  * - check, that packet arrived from expected physical interface.
240  * called with rcu_read_lock()
241  */
242 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
243 				 u8 tos, int oif, struct net_device *dev,
244 				 int rpf, struct in_device *idev, u32 *itag)
245 {
246 	int ret, no_addr, accept_local;
247 	struct fib_result res;
248 	struct flowi4 fl4;
249 	struct net *net;
250 	bool dev_match;
251 
252 	fl4.flowi4_oif = 0;
253 	fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
254 	fl4.daddr = src;
255 	fl4.saddr = dst;
256 	fl4.flowi4_tos = tos;
257 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
258 
259 	no_addr = idev->ifa_list == NULL;
260 
261 	accept_local = IN_DEV_ACCEPT_LOCAL(idev);
262 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
263 
264 	net = dev_net(dev);
265 	if (fib_lookup(net, &fl4, &res))
266 		goto last_resort;
267 	if (res.type != RTN_UNICAST) {
268 		if (res.type != RTN_LOCAL || !accept_local)
269 			goto e_inval;
270 	}
271 	fib_combine_itag(itag, &res);
272 	dev_match = false;
273 
274 #ifdef CONFIG_IP_ROUTE_MULTIPATH
275 	for (ret = 0; ret < res.fi->fib_nhs; ret++) {
276 		struct fib_nh *nh = &res.fi->fib_nh[ret];
277 
278 		if (nh->nh_dev == dev) {
279 			dev_match = true;
280 			break;
281 		}
282 	}
283 #else
284 	if (FIB_RES_DEV(res) == dev)
285 		dev_match = true;
286 #endif
287 	if (dev_match) {
288 		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
289 		return ret;
290 	}
291 	if (no_addr)
292 		goto last_resort;
293 	if (rpf == 1)
294 		goto e_rpf;
295 	fl4.flowi4_oif = dev->ifindex;
296 
297 	ret = 0;
298 	if (fib_lookup(net, &fl4, &res) == 0) {
299 		if (res.type == RTN_UNICAST)
300 			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
301 	}
302 	return ret;
303 
304 last_resort:
305 	if (rpf)
306 		goto e_rpf;
307 	*itag = 0;
308 	return 0;
309 
310 e_inval:
311 	return -EINVAL;
312 e_rpf:
313 	return -EXDEV;
314 }
315 
316 /* Ignore rp_filter for packets protected by IPsec. */
317 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
318 			u8 tos, int oif, struct net_device *dev,
319 			struct in_device *idev, u32 *itag)
320 {
321 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
322 
323 	if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
324 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
325 		*itag = 0;
326 		return 0;
327 	}
328 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
329 }
330 
331 static inline __be32 sk_extract_addr(struct sockaddr *addr)
332 {
333 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
334 }
335 
336 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
337 {
338 	struct nlattr *nla;
339 
340 	nla = (struct nlattr *) ((char *) mx + len);
341 	nla->nla_type = type;
342 	nla->nla_len = nla_attr_size(4);
343 	*(u32 *) nla_data(nla) = value;
344 
345 	return len + nla_total_size(4);
346 }
347 
348 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
349 				 struct fib_config *cfg)
350 {
351 	__be32 addr;
352 	int plen;
353 
354 	memset(cfg, 0, sizeof(*cfg));
355 	cfg->fc_nlinfo.nl_net = net;
356 
357 	if (rt->rt_dst.sa_family != AF_INET)
358 		return -EAFNOSUPPORT;
359 
360 	/*
361 	 * Check mask for validity:
362 	 * a) it must be contiguous.
363 	 * b) destination must have all host bits clear.
364 	 * c) if application forgot to set correct family (AF_INET),
365 	 *    reject request unless it is absolutely clear i.e.
366 	 *    both family and mask are zero.
367 	 */
368 	plen = 32;
369 	addr = sk_extract_addr(&rt->rt_dst);
370 	if (!(rt->rt_flags & RTF_HOST)) {
371 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
372 
373 		if (rt->rt_genmask.sa_family != AF_INET) {
374 			if (mask || rt->rt_genmask.sa_family)
375 				return -EAFNOSUPPORT;
376 		}
377 
378 		if (bad_mask(mask, addr))
379 			return -EINVAL;
380 
381 		plen = inet_mask_len(mask);
382 	}
383 
384 	cfg->fc_dst_len = plen;
385 	cfg->fc_dst = addr;
386 
387 	if (cmd != SIOCDELRT) {
388 		cfg->fc_nlflags = NLM_F_CREATE;
389 		cfg->fc_protocol = RTPROT_BOOT;
390 	}
391 
392 	if (rt->rt_metric)
393 		cfg->fc_priority = rt->rt_metric - 1;
394 
395 	if (rt->rt_flags & RTF_REJECT) {
396 		cfg->fc_scope = RT_SCOPE_HOST;
397 		cfg->fc_type = RTN_UNREACHABLE;
398 		return 0;
399 	}
400 
401 	cfg->fc_scope = RT_SCOPE_NOWHERE;
402 	cfg->fc_type = RTN_UNICAST;
403 
404 	if (rt->rt_dev) {
405 		char *colon;
406 		struct net_device *dev;
407 		char devname[IFNAMSIZ];
408 
409 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
410 			return -EFAULT;
411 
412 		devname[IFNAMSIZ-1] = 0;
413 		colon = strchr(devname, ':');
414 		if (colon)
415 			*colon = 0;
416 		dev = __dev_get_by_name(net, devname);
417 		if (!dev)
418 			return -ENODEV;
419 		cfg->fc_oif = dev->ifindex;
420 		if (colon) {
421 			struct in_ifaddr *ifa;
422 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
423 			if (!in_dev)
424 				return -ENODEV;
425 			*colon = ':';
426 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
427 				if (strcmp(ifa->ifa_label, devname) == 0)
428 					break;
429 			if (ifa == NULL)
430 				return -ENODEV;
431 			cfg->fc_prefsrc = ifa->ifa_local;
432 		}
433 	}
434 
435 	addr = sk_extract_addr(&rt->rt_gateway);
436 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
437 		cfg->fc_gw = addr;
438 		if (rt->rt_flags & RTF_GATEWAY &&
439 		    inet_addr_type(net, addr) == RTN_UNICAST)
440 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
441 	}
442 
443 	if (cmd == SIOCDELRT)
444 		return 0;
445 
446 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
447 		return -EINVAL;
448 
449 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
450 		cfg->fc_scope = RT_SCOPE_LINK;
451 
452 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
453 		struct nlattr *mx;
454 		int len = 0;
455 
456 		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
457 		if (mx == NULL)
458 			return -ENOMEM;
459 
460 		if (rt->rt_flags & RTF_MTU)
461 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
462 
463 		if (rt->rt_flags & RTF_WINDOW)
464 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
465 
466 		if (rt->rt_flags & RTF_IRTT)
467 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
468 
469 		cfg->fc_mx = mx;
470 		cfg->fc_mx_len = len;
471 	}
472 
473 	return 0;
474 }
475 
476 /*
477  * Handle IP routing ioctl calls.
478  * These are used to manipulate the routing tables
479  */
480 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
481 {
482 	struct fib_config cfg;
483 	struct rtentry rt;
484 	int err;
485 
486 	switch (cmd) {
487 	case SIOCADDRT:		/* Add a route */
488 	case SIOCDELRT:		/* Delete a route */
489 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
490 			return -EPERM;
491 
492 		if (copy_from_user(&rt, arg, sizeof(rt)))
493 			return -EFAULT;
494 
495 		rtnl_lock();
496 		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
497 		if (err == 0) {
498 			struct fib_table *tb;
499 
500 			if (cmd == SIOCDELRT) {
501 				tb = fib_get_table(net, cfg.fc_table);
502 				if (tb)
503 					err = fib_table_delete(tb, &cfg);
504 				else
505 					err = -ESRCH;
506 			} else {
507 				tb = fib_new_table(net, cfg.fc_table);
508 				if (tb)
509 					err = fib_table_insert(tb, &cfg);
510 				else
511 					err = -ENOBUFS;
512 			}
513 
514 			/* allocated by rtentry_to_fib_config() */
515 			kfree(cfg.fc_mx);
516 		}
517 		rtnl_unlock();
518 		return err;
519 	}
520 	return -EINVAL;
521 }
522 
523 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
524 	[RTA_DST]		= { .type = NLA_U32 },
525 	[RTA_SRC]		= { .type = NLA_U32 },
526 	[RTA_IIF]		= { .type = NLA_U32 },
527 	[RTA_OIF]		= { .type = NLA_U32 },
528 	[RTA_GATEWAY]		= { .type = NLA_U32 },
529 	[RTA_PRIORITY]		= { .type = NLA_U32 },
530 	[RTA_PREFSRC]		= { .type = NLA_U32 },
531 	[RTA_METRICS]		= { .type = NLA_NESTED },
532 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
533 	[RTA_FLOW]		= { .type = NLA_U32 },
534 };
535 
536 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
537 			     struct nlmsghdr *nlh, struct fib_config *cfg)
538 {
539 	struct nlattr *attr;
540 	int err, remaining;
541 	struct rtmsg *rtm;
542 
543 	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
544 	if (err < 0)
545 		goto errout;
546 
547 	memset(cfg, 0, sizeof(*cfg));
548 
549 	rtm = nlmsg_data(nlh);
550 	cfg->fc_dst_len = rtm->rtm_dst_len;
551 	cfg->fc_tos = rtm->rtm_tos;
552 	cfg->fc_table = rtm->rtm_table;
553 	cfg->fc_protocol = rtm->rtm_protocol;
554 	cfg->fc_scope = rtm->rtm_scope;
555 	cfg->fc_type = rtm->rtm_type;
556 	cfg->fc_flags = rtm->rtm_flags;
557 	cfg->fc_nlflags = nlh->nlmsg_flags;
558 
559 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
560 	cfg->fc_nlinfo.nlh = nlh;
561 	cfg->fc_nlinfo.nl_net = net;
562 
563 	if (cfg->fc_type > RTN_MAX) {
564 		err = -EINVAL;
565 		goto errout;
566 	}
567 
568 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
569 		switch (nla_type(attr)) {
570 		case RTA_DST:
571 			cfg->fc_dst = nla_get_be32(attr);
572 			break;
573 		case RTA_OIF:
574 			cfg->fc_oif = nla_get_u32(attr);
575 			break;
576 		case RTA_GATEWAY:
577 			cfg->fc_gw = nla_get_be32(attr);
578 			break;
579 		case RTA_PRIORITY:
580 			cfg->fc_priority = nla_get_u32(attr);
581 			break;
582 		case RTA_PREFSRC:
583 			cfg->fc_prefsrc = nla_get_be32(attr);
584 			break;
585 		case RTA_METRICS:
586 			cfg->fc_mx = nla_data(attr);
587 			cfg->fc_mx_len = nla_len(attr);
588 			break;
589 		case RTA_MULTIPATH:
590 			cfg->fc_mp = nla_data(attr);
591 			cfg->fc_mp_len = nla_len(attr);
592 			break;
593 		case RTA_FLOW:
594 			cfg->fc_flow = nla_get_u32(attr);
595 			break;
596 		case RTA_TABLE:
597 			cfg->fc_table = nla_get_u32(attr);
598 			break;
599 		}
600 	}
601 
602 	return 0;
603 errout:
604 	return err;
605 }
606 
607 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
608 {
609 	struct net *net = sock_net(skb->sk);
610 	struct fib_config cfg;
611 	struct fib_table *tb;
612 	int err;
613 
614 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
615 	if (err < 0)
616 		goto errout;
617 
618 	tb = fib_get_table(net, cfg.fc_table);
619 	if (tb == NULL) {
620 		err = -ESRCH;
621 		goto errout;
622 	}
623 
624 	err = fib_table_delete(tb, &cfg);
625 errout:
626 	return err;
627 }
628 
629 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
630 {
631 	struct net *net = sock_net(skb->sk);
632 	struct fib_config cfg;
633 	struct fib_table *tb;
634 	int err;
635 
636 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
637 	if (err < 0)
638 		goto errout;
639 
640 	tb = fib_new_table(net, cfg.fc_table);
641 	if (tb == NULL) {
642 		err = -ENOBUFS;
643 		goto errout;
644 	}
645 
646 	err = fib_table_insert(tb, &cfg);
647 errout:
648 	return err;
649 }
650 
651 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
652 {
653 	struct net *net = sock_net(skb->sk);
654 	unsigned int h, s_h;
655 	unsigned int e = 0, s_e;
656 	struct fib_table *tb;
657 	struct hlist_head *head;
658 	int dumped = 0;
659 
660 	if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
661 	    ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
662 		return skb->len;
663 
664 	s_h = cb->args[0];
665 	s_e = cb->args[1];
666 
667 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
668 		e = 0;
669 		head = &net->ipv4.fib_table_hash[h];
670 		hlist_for_each_entry(tb, head, tb_hlist) {
671 			if (e < s_e)
672 				goto next;
673 			if (dumped)
674 				memset(&cb->args[2], 0, sizeof(cb->args) -
675 						 2 * sizeof(cb->args[0]));
676 			if (fib_table_dump(tb, skb, cb) < 0)
677 				goto out;
678 			dumped = 1;
679 next:
680 			e++;
681 		}
682 	}
683 out:
684 	cb->args[1] = e;
685 	cb->args[0] = h;
686 
687 	return skb->len;
688 }
689 
690 /* Prepare and feed intra-kernel routing request.
691  * Really, it should be netlink message, but :-( netlink
692  * can be not configured, so that we feed it directly
693  * to fib engine. It is legal, because all events occur
694  * only when netlink is already locked.
695  */
696 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
697 {
698 	struct net *net = dev_net(ifa->ifa_dev->dev);
699 	struct fib_table *tb;
700 	struct fib_config cfg = {
701 		.fc_protocol = RTPROT_KERNEL,
702 		.fc_type = type,
703 		.fc_dst = dst,
704 		.fc_dst_len = dst_len,
705 		.fc_prefsrc = ifa->ifa_local,
706 		.fc_oif = ifa->ifa_dev->dev->ifindex,
707 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
708 		.fc_nlinfo = {
709 			.nl_net = net,
710 		},
711 	};
712 
713 	if (type == RTN_UNICAST)
714 		tb = fib_new_table(net, RT_TABLE_MAIN);
715 	else
716 		tb = fib_new_table(net, RT_TABLE_LOCAL);
717 
718 	if (tb == NULL)
719 		return;
720 
721 	cfg.fc_table = tb->tb_id;
722 
723 	if (type != RTN_LOCAL)
724 		cfg.fc_scope = RT_SCOPE_LINK;
725 	else
726 		cfg.fc_scope = RT_SCOPE_HOST;
727 
728 	if (cmd == RTM_NEWROUTE)
729 		fib_table_insert(tb, &cfg);
730 	else
731 		fib_table_delete(tb, &cfg);
732 }
733 
734 void fib_add_ifaddr(struct in_ifaddr *ifa)
735 {
736 	struct in_device *in_dev = ifa->ifa_dev;
737 	struct net_device *dev = in_dev->dev;
738 	struct in_ifaddr *prim = ifa;
739 	__be32 mask = ifa->ifa_mask;
740 	__be32 addr = ifa->ifa_local;
741 	__be32 prefix = ifa->ifa_address & mask;
742 
743 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
744 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
745 		if (prim == NULL) {
746 			pr_warn("%s: bug: prim == NULL\n", __func__);
747 			return;
748 		}
749 	}
750 
751 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
752 
753 	if (!(dev->flags & IFF_UP))
754 		return;
755 
756 	/* Add broadcast address, if it is explicitly assigned. */
757 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
758 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
759 
760 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
761 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
762 		fib_magic(RTM_NEWROUTE,
763 			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
764 			  prefix, ifa->ifa_prefixlen, prim);
765 
766 		/* Add network specific broadcasts, when it takes a sense */
767 		if (ifa->ifa_prefixlen < 31) {
768 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
769 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
770 				  32, prim);
771 		}
772 	}
773 }
774 
775 /* Delete primary or secondary address.
776  * Optionally, on secondary address promotion consider the addresses
777  * from subnet iprim as deleted, even if they are in device list.
778  * In this case the secondary ifa can be in device list.
779  */
780 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
781 {
782 	struct in_device *in_dev = ifa->ifa_dev;
783 	struct net_device *dev = in_dev->dev;
784 	struct in_ifaddr *ifa1;
785 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
786 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
787 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
788 #define LOCAL_OK	1
789 #define BRD_OK		2
790 #define BRD0_OK		4
791 #define BRD1_OK		8
792 	unsigned int ok = 0;
793 	int subnet = 0;		/* Primary network */
794 	int gone = 1;		/* Address is missing */
795 	int same_prefsrc = 0;	/* Another primary with same IP */
796 
797 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
798 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
799 		if (prim == NULL) {
800 			pr_warn("%s: bug: prim == NULL\n", __func__);
801 			return;
802 		}
803 		if (iprim && iprim != prim) {
804 			pr_warn("%s: bug: iprim != prim\n", __func__);
805 			return;
806 		}
807 	} else if (!ipv4_is_zeronet(any) &&
808 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
809 		fib_magic(RTM_DELROUTE,
810 			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
811 			  any, ifa->ifa_prefixlen, prim);
812 		subnet = 1;
813 	}
814 
815 	/* Deletion is more complicated than add.
816 	 * We should take care of not to delete too much :-)
817 	 *
818 	 * Scan address list to be sure that addresses are really gone.
819 	 */
820 
821 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
822 		if (ifa1 == ifa) {
823 			/* promotion, keep the IP */
824 			gone = 0;
825 			continue;
826 		}
827 		/* Ignore IFAs from our subnet */
828 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
829 		    inet_ifa_match(ifa1->ifa_address, iprim))
830 			continue;
831 
832 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
833 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
834 			/* Another address from our subnet? */
835 			if (ifa1->ifa_mask == prim->ifa_mask &&
836 			    inet_ifa_match(ifa1->ifa_address, prim))
837 				prim1 = prim;
838 			else {
839 				/* We reached the secondaries, so
840 				 * same_prefsrc should be determined.
841 				 */
842 				if (!same_prefsrc)
843 					continue;
844 				/* Search new prim1 if ifa1 is not
845 				 * using the current prim1
846 				 */
847 				if (!prim1 ||
848 				    ifa1->ifa_mask != prim1->ifa_mask ||
849 				    !inet_ifa_match(ifa1->ifa_address, prim1))
850 					prim1 = inet_ifa_byprefix(in_dev,
851 							ifa1->ifa_address,
852 							ifa1->ifa_mask);
853 				if (!prim1)
854 					continue;
855 				if (prim1->ifa_local != prim->ifa_local)
856 					continue;
857 			}
858 		} else {
859 			if (prim->ifa_local != ifa1->ifa_local)
860 				continue;
861 			prim1 = ifa1;
862 			if (prim != prim1)
863 				same_prefsrc = 1;
864 		}
865 		if (ifa->ifa_local == ifa1->ifa_local)
866 			ok |= LOCAL_OK;
867 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
868 			ok |= BRD_OK;
869 		if (brd == ifa1->ifa_broadcast)
870 			ok |= BRD1_OK;
871 		if (any == ifa1->ifa_broadcast)
872 			ok |= BRD0_OK;
873 		/* primary has network specific broadcasts */
874 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
875 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
876 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
877 
878 			if (!ipv4_is_zeronet(any1)) {
879 				if (ifa->ifa_broadcast == brd1 ||
880 				    ifa->ifa_broadcast == any1)
881 					ok |= BRD_OK;
882 				if (brd == brd1 || brd == any1)
883 					ok |= BRD1_OK;
884 				if (any == brd1 || any == any1)
885 					ok |= BRD0_OK;
886 			}
887 		}
888 	}
889 
890 	if (!(ok & BRD_OK))
891 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
892 	if (subnet && ifa->ifa_prefixlen < 31) {
893 		if (!(ok & BRD1_OK))
894 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
895 		if (!(ok & BRD0_OK))
896 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
897 	}
898 	if (!(ok & LOCAL_OK)) {
899 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
900 
901 		/* Check, that this local address finally disappeared. */
902 		if (gone &&
903 		    inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
904 			/* And the last, but not the least thing.
905 			 * We must flush stray FIB entries.
906 			 *
907 			 * First of all, we scan fib_info list searching
908 			 * for stray nexthop entries, then ignite fib_flush.
909 			 */
910 			if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
911 				fib_flush(dev_net(dev));
912 		}
913 	}
914 #undef LOCAL_OK
915 #undef BRD_OK
916 #undef BRD0_OK
917 #undef BRD1_OK
918 }
919 
920 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
921 {
922 
923 	struct fib_result       res;
924 	struct flowi4           fl4 = {
925 		.flowi4_mark = frn->fl_mark,
926 		.daddr = frn->fl_addr,
927 		.flowi4_tos = frn->fl_tos,
928 		.flowi4_scope = frn->fl_scope,
929 	};
930 
931 	frn->err = -ENOENT;
932 	if (tb) {
933 		local_bh_disable();
934 
935 		frn->tb_id = tb->tb_id;
936 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
937 
938 		if (!frn->err) {
939 			frn->prefixlen = res.prefixlen;
940 			frn->nh_sel = res.nh_sel;
941 			frn->type = res.type;
942 			frn->scope = res.scope;
943 		}
944 		local_bh_enable();
945 	}
946 }
947 
948 static void nl_fib_input(struct sk_buff *skb)
949 {
950 	struct net *net;
951 	struct fib_result_nl *frn;
952 	struct nlmsghdr *nlh;
953 	struct fib_table *tb;
954 	u32 portid;
955 
956 	net = sock_net(skb->sk);
957 	nlh = nlmsg_hdr(skb);
958 	if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len ||
959 	    nlmsg_len(nlh) < sizeof(*frn))
960 		return;
961 
962 	skb = netlink_skb_clone(skb, GFP_KERNEL);
963 	if (skb == NULL)
964 		return;
965 	nlh = nlmsg_hdr(skb);
966 
967 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
968 	tb = fib_get_table(net, frn->tb_id_in);
969 
970 	nl_fib_lookup(frn, tb);
971 
972 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
973 	NETLINK_CB(skb).portid = 0;        /* from kernel */
974 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
975 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
976 }
977 
978 static int __net_init nl_fib_lookup_init(struct net *net)
979 {
980 	struct sock *sk;
981 	struct netlink_kernel_cfg cfg = {
982 		.input	= nl_fib_input,
983 	};
984 
985 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
986 	if (sk == NULL)
987 		return -EAFNOSUPPORT;
988 	net->ipv4.fibnl = sk;
989 	return 0;
990 }
991 
992 static void nl_fib_lookup_exit(struct net *net)
993 {
994 	netlink_kernel_release(net->ipv4.fibnl);
995 	net->ipv4.fibnl = NULL;
996 }
997 
998 static void fib_disable_ip(struct net_device *dev, int force)
999 {
1000 	if (fib_sync_down_dev(dev, force))
1001 		fib_flush(dev_net(dev));
1002 	rt_cache_flush(dev_net(dev));
1003 	arp_ifdown(dev);
1004 }
1005 
1006 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1007 {
1008 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1009 	struct net_device *dev = ifa->ifa_dev->dev;
1010 	struct net *net = dev_net(dev);
1011 
1012 	switch (event) {
1013 	case NETDEV_UP:
1014 		fib_add_ifaddr(ifa);
1015 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1016 		fib_sync_up(dev);
1017 #endif
1018 		atomic_inc(&net->ipv4.dev_addr_genid);
1019 		rt_cache_flush(dev_net(dev));
1020 		break;
1021 	case NETDEV_DOWN:
1022 		fib_del_ifaddr(ifa, NULL);
1023 		atomic_inc(&net->ipv4.dev_addr_genid);
1024 		if (ifa->ifa_dev->ifa_list == NULL) {
1025 			/* Last address was deleted from this interface.
1026 			 * Disable IP.
1027 			 */
1028 			fib_disable_ip(dev, 1);
1029 		} else {
1030 			rt_cache_flush(dev_net(dev));
1031 		}
1032 		break;
1033 	}
1034 	return NOTIFY_DONE;
1035 }
1036 
1037 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1038 {
1039 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1040 	struct in_device *in_dev;
1041 	struct net *net = dev_net(dev);
1042 
1043 	if (event == NETDEV_UNREGISTER) {
1044 		fib_disable_ip(dev, 2);
1045 		rt_flush_dev(dev);
1046 		return NOTIFY_DONE;
1047 	}
1048 
1049 	in_dev = __in_dev_get_rtnl(dev);
1050 	if (!in_dev)
1051 		return NOTIFY_DONE;
1052 
1053 	switch (event) {
1054 	case NETDEV_UP:
1055 		for_ifa(in_dev) {
1056 			fib_add_ifaddr(ifa);
1057 		} endfor_ifa(in_dev);
1058 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1059 		fib_sync_up(dev);
1060 #endif
1061 		atomic_inc(&net->ipv4.dev_addr_genid);
1062 		rt_cache_flush(net);
1063 		break;
1064 	case NETDEV_DOWN:
1065 		fib_disable_ip(dev, 0);
1066 		break;
1067 	case NETDEV_CHANGEMTU:
1068 	case NETDEV_CHANGE:
1069 		rt_cache_flush(net);
1070 		break;
1071 	}
1072 	return NOTIFY_DONE;
1073 }
1074 
1075 static struct notifier_block fib_inetaddr_notifier = {
1076 	.notifier_call = fib_inetaddr_event,
1077 };
1078 
1079 static struct notifier_block fib_netdev_notifier = {
1080 	.notifier_call = fib_netdev_event,
1081 };
1082 
1083 static int __net_init ip_fib_net_init(struct net *net)
1084 {
1085 	int err;
1086 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1087 
1088 	/* Avoid false sharing : Use at least a full cache line */
1089 	size = max_t(size_t, size, L1_CACHE_BYTES);
1090 
1091 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1092 	if (net->ipv4.fib_table_hash == NULL)
1093 		return -ENOMEM;
1094 
1095 	err = fib4_rules_init(net);
1096 	if (err < 0)
1097 		goto fail;
1098 	return 0;
1099 
1100 fail:
1101 	kfree(net->ipv4.fib_table_hash);
1102 	return err;
1103 }
1104 
1105 static void ip_fib_net_exit(struct net *net)
1106 {
1107 	unsigned int i;
1108 
1109 #ifdef CONFIG_IP_MULTIPLE_TABLES
1110 	fib4_rules_exit(net);
1111 #endif
1112 
1113 	rtnl_lock();
1114 	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1115 		struct fib_table *tb;
1116 		struct hlist_head *head;
1117 		struct hlist_node *tmp;
1118 
1119 		head = &net->ipv4.fib_table_hash[i];
1120 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1121 			hlist_del(&tb->tb_hlist);
1122 			fib_table_flush(tb);
1123 			fib_free_table(tb);
1124 		}
1125 	}
1126 	rtnl_unlock();
1127 	kfree(net->ipv4.fib_table_hash);
1128 }
1129 
1130 static int __net_init fib_net_init(struct net *net)
1131 {
1132 	int error;
1133 
1134 #ifdef CONFIG_IP_ROUTE_CLASSID
1135 	net->ipv4.fib_num_tclassid_users = 0;
1136 #endif
1137 	error = ip_fib_net_init(net);
1138 	if (error < 0)
1139 		goto out;
1140 	error = nl_fib_lookup_init(net);
1141 	if (error < 0)
1142 		goto out_nlfl;
1143 	error = fib_proc_init(net);
1144 	if (error < 0)
1145 		goto out_proc;
1146 out:
1147 	return error;
1148 
1149 out_proc:
1150 	nl_fib_lookup_exit(net);
1151 out_nlfl:
1152 	ip_fib_net_exit(net);
1153 	goto out;
1154 }
1155 
1156 static void __net_exit fib_net_exit(struct net *net)
1157 {
1158 	fib_proc_exit(net);
1159 	nl_fib_lookup_exit(net);
1160 	ip_fib_net_exit(net);
1161 }
1162 
1163 static struct pernet_operations fib_net_ops = {
1164 	.init = fib_net_init,
1165 	.exit = fib_net_exit,
1166 };
1167 
1168 void __init ip_fib_init(void)
1169 {
1170 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1171 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1172 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1173 
1174 	register_pernet_subsys(&fib_net_ops);
1175 	register_netdevice_notifier(&fib_netdev_notifier);
1176 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1177 
1178 	fib_trie_init();
1179 }
1180