xref: /linux/net/ipv4/fib_frontend.c (revision 95e9fd10f06cb5642028b6b851e32b8c8afb4571)
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_node *node;
116 	struct hlist_head *head;
117 	unsigned int h;
118 
119 	if (id == 0)
120 		id = RT_TABLE_MAIN;
121 	h = id & (FIB_TABLE_HASHSZ - 1);
122 
123 	rcu_read_lock();
124 	head = &net->ipv4.fib_table_hash[h];
125 	hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
126 		if (tb->tb_id == id) {
127 			rcu_read_unlock();
128 			return tb;
129 		}
130 	}
131 	rcu_read_unlock();
132 	return NULL;
133 }
134 #endif /* CONFIG_IP_MULTIPLE_TABLES */
135 
136 static void fib_flush(struct net *net)
137 {
138 	int flushed = 0;
139 	struct fib_table *tb;
140 	struct hlist_node *node;
141 	struct hlist_head *head;
142 	unsigned int h;
143 
144 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
145 		head = &net->ipv4.fib_table_hash[h];
146 		hlist_for_each_entry(tb, node, head, tb_hlist)
147 			flushed += fib_table_flush(tb);
148 	}
149 
150 	if (flushed)
151 		rt_cache_flush(net, -1);
152 }
153 
154 /*
155  * Find address type as if only "dev" was present in the system. If
156  * on_dev is NULL then all interfaces are taken into consideration.
157  */
158 static inline unsigned int __inet_dev_addr_type(struct net *net,
159 						const struct net_device *dev,
160 						__be32 addr)
161 {
162 	struct flowi4		fl4 = { .daddr = addr };
163 	struct fib_result	res;
164 	unsigned int ret = RTN_BROADCAST;
165 	struct fib_table *local_table;
166 
167 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
168 		return RTN_BROADCAST;
169 	if (ipv4_is_multicast(addr))
170 		return RTN_MULTICAST;
171 
172 	local_table = fib_get_table(net, RT_TABLE_LOCAL);
173 	if (local_table) {
174 		ret = RTN_UNICAST;
175 		rcu_read_lock();
176 		if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) {
177 			if (!dev || dev == res.fi->fib_dev)
178 				ret = res.type;
179 		}
180 		rcu_read_unlock();
181 	}
182 	return ret;
183 }
184 
185 unsigned int inet_addr_type(struct net *net, __be32 addr)
186 {
187 	return __inet_dev_addr_type(net, NULL, addr);
188 }
189 EXPORT_SYMBOL(inet_addr_type);
190 
191 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
192 				__be32 addr)
193 {
194 	return __inet_dev_addr_type(net, dev, addr);
195 }
196 EXPORT_SYMBOL(inet_dev_addr_type);
197 
198 __be32 fib_compute_spec_dst(struct sk_buff *skb)
199 {
200 	struct net_device *dev = skb->dev;
201 	struct in_device *in_dev;
202 	struct fib_result res;
203 	struct rtable *rt;
204 	struct flowi4 fl4;
205 	struct net *net;
206 	int scope;
207 
208 	rt = skb_rtable(skb);
209 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
210 	    RTCF_LOCAL)
211 		return ip_hdr(skb)->daddr;
212 
213 	in_dev = __in_dev_get_rcu(dev);
214 	BUG_ON(!in_dev);
215 
216 	net = dev_net(dev);
217 
218 	scope = RT_SCOPE_UNIVERSE;
219 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
220 		fl4.flowi4_oif = 0;
221 		fl4.flowi4_iif = net->loopback_dev->ifindex;
222 		fl4.daddr = ip_hdr(skb)->saddr;
223 		fl4.saddr = 0;
224 		fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
225 		fl4.flowi4_scope = scope;
226 		fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
227 		if (!fib_lookup(net, &fl4, &res))
228 			return FIB_RES_PREFSRC(net, res);
229 	} else {
230 		scope = RT_SCOPE_LINK;
231 	}
232 
233 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
234 }
235 
236 /* Given (packet source, input interface) and optional (dst, oif, tos):
237  * - (main) check, that source is valid i.e. not broadcast or our local
238  *   address.
239  * - figure out what "logical" interface this packet arrived
240  *   and calculate "specific destination" address.
241  * - check, that packet arrived from expected physical interface.
242  * called with rcu_read_lock()
243  */
244 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
245 				 u8 tos, int oif, struct net_device *dev,
246 				 int rpf, struct in_device *idev, u32 *itag)
247 {
248 	int ret, no_addr, accept_local;
249 	struct fib_result res;
250 	struct flowi4 fl4;
251 	struct net *net;
252 	bool dev_match;
253 
254 	fl4.flowi4_oif = 0;
255 	fl4.flowi4_iif = oif;
256 	fl4.daddr = src;
257 	fl4.saddr = dst;
258 	fl4.flowi4_tos = tos;
259 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
260 
261 	no_addr = idev->ifa_list == NULL;
262 
263 	accept_local = IN_DEV_ACCEPT_LOCAL(idev);
264 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
265 
266 	net = dev_net(dev);
267 	if (fib_lookup(net, &fl4, &res))
268 		goto last_resort;
269 	if (res.type != RTN_UNICAST) {
270 		if (res.type != RTN_LOCAL || !accept_local)
271 			goto e_inval;
272 	}
273 	fib_combine_itag(itag, &res);
274 	dev_match = false;
275 
276 #ifdef CONFIG_IP_ROUTE_MULTIPATH
277 	for (ret = 0; ret < res.fi->fib_nhs; ret++) {
278 		struct fib_nh *nh = &res.fi->fib_nh[ret];
279 
280 		if (nh->nh_dev == dev) {
281 			dev_match = true;
282 			break;
283 		}
284 	}
285 #else
286 	if (FIB_RES_DEV(res) == dev)
287 		dev_match = true;
288 #endif
289 	if (dev_match) {
290 		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
291 		return ret;
292 	}
293 	if (no_addr)
294 		goto last_resort;
295 	if (rpf == 1)
296 		goto e_rpf;
297 	fl4.flowi4_oif = dev->ifindex;
298 
299 	ret = 0;
300 	if (fib_lookup(net, &fl4, &res) == 0) {
301 		if (res.type == RTN_UNICAST)
302 			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
303 	}
304 	return ret;
305 
306 last_resort:
307 	if (rpf)
308 		goto e_rpf;
309 	*itag = 0;
310 	return 0;
311 
312 e_inval:
313 	return -EINVAL;
314 e_rpf:
315 	return -EXDEV;
316 }
317 
318 /* Ignore rp_filter for packets protected by IPsec. */
319 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
320 			u8 tos, int oif, struct net_device *dev,
321 			struct in_device *idev, u32 *itag)
322 {
323 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
324 
325 	if (!r && !fib_num_tclassid_users(dev_net(dev))) {
326 		*itag = 0;
327 		return 0;
328 	}
329 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
330 }
331 
332 static inline __be32 sk_extract_addr(struct sockaddr *addr)
333 {
334 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
335 }
336 
337 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
338 {
339 	struct nlattr *nla;
340 
341 	nla = (struct nlattr *) ((char *) mx + len);
342 	nla->nla_type = type;
343 	nla->nla_len = nla_attr_size(4);
344 	*(u32 *) nla_data(nla) = value;
345 
346 	return len + nla_total_size(4);
347 }
348 
349 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
350 				 struct fib_config *cfg)
351 {
352 	__be32 addr;
353 	int plen;
354 
355 	memset(cfg, 0, sizeof(*cfg));
356 	cfg->fc_nlinfo.nl_net = net;
357 
358 	if (rt->rt_dst.sa_family != AF_INET)
359 		return -EAFNOSUPPORT;
360 
361 	/*
362 	 * Check mask for validity:
363 	 * a) it must be contiguous.
364 	 * b) destination must have all host bits clear.
365 	 * c) if application forgot to set correct family (AF_INET),
366 	 *    reject request unless it is absolutely clear i.e.
367 	 *    both family and mask are zero.
368 	 */
369 	plen = 32;
370 	addr = sk_extract_addr(&rt->rt_dst);
371 	if (!(rt->rt_flags & RTF_HOST)) {
372 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
373 
374 		if (rt->rt_genmask.sa_family != AF_INET) {
375 			if (mask || rt->rt_genmask.sa_family)
376 				return -EAFNOSUPPORT;
377 		}
378 
379 		if (bad_mask(mask, addr))
380 			return -EINVAL;
381 
382 		plen = inet_mask_len(mask);
383 	}
384 
385 	cfg->fc_dst_len = plen;
386 	cfg->fc_dst = addr;
387 
388 	if (cmd != SIOCDELRT) {
389 		cfg->fc_nlflags = NLM_F_CREATE;
390 		cfg->fc_protocol = RTPROT_BOOT;
391 	}
392 
393 	if (rt->rt_metric)
394 		cfg->fc_priority = rt->rt_metric - 1;
395 
396 	if (rt->rt_flags & RTF_REJECT) {
397 		cfg->fc_scope = RT_SCOPE_HOST;
398 		cfg->fc_type = RTN_UNREACHABLE;
399 		return 0;
400 	}
401 
402 	cfg->fc_scope = RT_SCOPE_NOWHERE;
403 	cfg->fc_type = RTN_UNICAST;
404 
405 	if (rt->rt_dev) {
406 		char *colon;
407 		struct net_device *dev;
408 		char devname[IFNAMSIZ];
409 
410 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
411 			return -EFAULT;
412 
413 		devname[IFNAMSIZ-1] = 0;
414 		colon = strchr(devname, ':');
415 		if (colon)
416 			*colon = 0;
417 		dev = __dev_get_by_name(net, devname);
418 		if (!dev)
419 			return -ENODEV;
420 		cfg->fc_oif = dev->ifindex;
421 		if (colon) {
422 			struct in_ifaddr *ifa;
423 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
424 			if (!in_dev)
425 				return -ENODEV;
426 			*colon = ':';
427 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
428 				if (strcmp(ifa->ifa_label, devname) == 0)
429 					break;
430 			if (ifa == NULL)
431 				return -ENODEV;
432 			cfg->fc_prefsrc = ifa->ifa_local;
433 		}
434 	}
435 
436 	addr = sk_extract_addr(&rt->rt_gateway);
437 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
438 		cfg->fc_gw = addr;
439 		if (rt->rt_flags & RTF_GATEWAY &&
440 		    inet_addr_type(net, addr) == RTN_UNICAST)
441 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
442 	}
443 
444 	if (cmd == SIOCDELRT)
445 		return 0;
446 
447 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
448 		return -EINVAL;
449 
450 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
451 		cfg->fc_scope = RT_SCOPE_LINK;
452 
453 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
454 		struct nlattr *mx;
455 		int len = 0;
456 
457 		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
458 		if (mx == NULL)
459 			return -ENOMEM;
460 
461 		if (rt->rt_flags & RTF_MTU)
462 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
463 
464 		if (rt->rt_flags & RTF_WINDOW)
465 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
466 
467 		if (rt->rt_flags & RTF_IRTT)
468 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
469 
470 		cfg->fc_mx = mx;
471 		cfg->fc_mx_len = len;
472 	}
473 
474 	return 0;
475 }
476 
477 /*
478  * Handle IP routing ioctl calls.
479  * These are used to manipulate the routing tables
480  */
481 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
482 {
483 	struct fib_config cfg;
484 	struct rtentry rt;
485 	int err;
486 
487 	switch (cmd) {
488 	case SIOCADDRT:		/* Add a route */
489 	case SIOCDELRT:		/* Delete a route */
490 		if (!capable(CAP_NET_ADMIN))
491 			return -EPERM;
492 
493 		if (copy_from_user(&rt, arg, sizeof(rt)))
494 			return -EFAULT;
495 
496 		rtnl_lock();
497 		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
498 		if (err == 0) {
499 			struct fib_table *tb;
500 
501 			if (cmd == SIOCDELRT) {
502 				tb = fib_get_table(net, cfg.fc_table);
503 				if (tb)
504 					err = fib_table_delete(tb, &cfg);
505 				else
506 					err = -ESRCH;
507 			} else {
508 				tb = fib_new_table(net, cfg.fc_table);
509 				if (tb)
510 					err = fib_table_insert(tb, &cfg);
511 				else
512 					err = -ENOBUFS;
513 			}
514 
515 			/* allocated by rtentry_to_fib_config() */
516 			kfree(cfg.fc_mx);
517 		}
518 		rtnl_unlock();
519 		return err;
520 	}
521 	return -EINVAL;
522 }
523 
524 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
525 	[RTA_DST]		= { .type = NLA_U32 },
526 	[RTA_SRC]		= { .type = NLA_U32 },
527 	[RTA_IIF]		= { .type = NLA_U32 },
528 	[RTA_OIF]		= { .type = NLA_U32 },
529 	[RTA_GATEWAY]		= { .type = NLA_U32 },
530 	[RTA_PRIORITY]		= { .type = NLA_U32 },
531 	[RTA_PREFSRC]		= { .type = NLA_U32 },
532 	[RTA_METRICS]		= { .type = NLA_NESTED },
533 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
534 	[RTA_FLOW]		= { .type = NLA_U32 },
535 };
536 
537 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
538 			     struct nlmsghdr *nlh, struct fib_config *cfg)
539 {
540 	struct nlattr *attr;
541 	int err, remaining;
542 	struct rtmsg *rtm;
543 
544 	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
545 	if (err < 0)
546 		goto errout;
547 
548 	memset(cfg, 0, sizeof(*cfg));
549 
550 	rtm = nlmsg_data(nlh);
551 	cfg->fc_dst_len = rtm->rtm_dst_len;
552 	cfg->fc_tos = rtm->rtm_tos;
553 	cfg->fc_table = rtm->rtm_table;
554 	cfg->fc_protocol = rtm->rtm_protocol;
555 	cfg->fc_scope = rtm->rtm_scope;
556 	cfg->fc_type = rtm->rtm_type;
557 	cfg->fc_flags = rtm->rtm_flags;
558 	cfg->fc_nlflags = nlh->nlmsg_flags;
559 
560 	cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
561 	cfg->fc_nlinfo.nlh = nlh;
562 	cfg->fc_nlinfo.nl_net = net;
563 
564 	if (cfg->fc_type > RTN_MAX) {
565 		err = -EINVAL;
566 		goto errout;
567 	}
568 
569 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
570 		switch (nla_type(attr)) {
571 		case RTA_DST:
572 			cfg->fc_dst = nla_get_be32(attr);
573 			break;
574 		case RTA_OIF:
575 			cfg->fc_oif = nla_get_u32(attr);
576 			break;
577 		case RTA_GATEWAY:
578 			cfg->fc_gw = nla_get_be32(attr);
579 			break;
580 		case RTA_PRIORITY:
581 			cfg->fc_priority = nla_get_u32(attr);
582 			break;
583 		case RTA_PREFSRC:
584 			cfg->fc_prefsrc = nla_get_be32(attr);
585 			break;
586 		case RTA_METRICS:
587 			cfg->fc_mx = nla_data(attr);
588 			cfg->fc_mx_len = nla_len(attr);
589 			break;
590 		case RTA_MULTIPATH:
591 			cfg->fc_mp = nla_data(attr);
592 			cfg->fc_mp_len = nla_len(attr);
593 			break;
594 		case RTA_FLOW:
595 			cfg->fc_flow = nla_get_u32(attr);
596 			break;
597 		case RTA_TABLE:
598 			cfg->fc_table = nla_get_u32(attr);
599 			break;
600 		}
601 	}
602 
603 	return 0;
604 errout:
605 	return err;
606 }
607 
608 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
609 {
610 	struct net *net = sock_net(skb->sk);
611 	struct fib_config cfg;
612 	struct fib_table *tb;
613 	int err;
614 
615 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
616 	if (err < 0)
617 		goto errout;
618 
619 	tb = fib_get_table(net, cfg.fc_table);
620 	if (tb == NULL) {
621 		err = -ESRCH;
622 		goto errout;
623 	}
624 
625 	err = fib_table_delete(tb, &cfg);
626 errout:
627 	return err;
628 }
629 
630 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
631 {
632 	struct net *net = sock_net(skb->sk);
633 	struct fib_config cfg;
634 	struct fib_table *tb;
635 	int err;
636 
637 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
638 	if (err < 0)
639 		goto errout;
640 
641 	tb = fib_new_table(net, cfg.fc_table);
642 	if (tb == NULL) {
643 		err = -ENOBUFS;
644 		goto errout;
645 	}
646 
647 	err = fib_table_insert(tb, &cfg);
648 errout:
649 	return err;
650 }
651 
652 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
653 {
654 	struct net *net = sock_net(skb->sk);
655 	unsigned int h, s_h;
656 	unsigned int e = 0, s_e;
657 	struct fib_table *tb;
658 	struct hlist_node *node;
659 	struct hlist_head *head;
660 	int dumped = 0;
661 
662 	if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
663 	    ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
664 		return ip_rt_dump(skb, cb);
665 
666 	s_h = cb->args[0];
667 	s_e = cb->args[1];
668 
669 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
670 		e = 0;
671 		head = &net->ipv4.fib_table_hash[h];
672 		hlist_for_each_entry(tb, node, head, tb_hlist) {
673 			if (e < s_e)
674 				goto next;
675 			if (dumped)
676 				memset(&cb->args[2], 0, sizeof(cb->args) -
677 						 2 * sizeof(cb->args[0]));
678 			if (fib_table_dump(tb, skb, cb) < 0)
679 				goto out;
680 			dumped = 1;
681 next:
682 			e++;
683 		}
684 	}
685 out:
686 	cb->args[1] = e;
687 	cb->args[0] = h;
688 
689 	return skb->len;
690 }
691 
692 /* Prepare and feed intra-kernel routing request.
693  * Really, it should be netlink message, but :-( netlink
694  * can be not configured, so that we feed it directly
695  * to fib engine. It is legal, because all events occur
696  * only when netlink is already locked.
697  */
698 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
699 {
700 	struct net *net = dev_net(ifa->ifa_dev->dev);
701 	struct fib_table *tb;
702 	struct fib_config cfg = {
703 		.fc_protocol = RTPROT_KERNEL,
704 		.fc_type = type,
705 		.fc_dst = dst,
706 		.fc_dst_len = dst_len,
707 		.fc_prefsrc = ifa->ifa_local,
708 		.fc_oif = ifa->ifa_dev->dev->ifindex,
709 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
710 		.fc_nlinfo = {
711 			.nl_net = net,
712 		},
713 	};
714 
715 	if (type == RTN_UNICAST)
716 		tb = fib_new_table(net, RT_TABLE_MAIN);
717 	else
718 		tb = fib_new_table(net, RT_TABLE_LOCAL);
719 
720 	if (tb == NULL)
721 		return;
722 
723 	cfg.fc_table = tb->tb_id;
724 
725 	if (type != RTN_LOCAL)
726 		cfg.fc_scope = RT_SCOPE_LINK;
727 	else
728 		cfg.fc_scope = RT_SCOPE_HOST;
729 
730 	if (cmd == RTM_NEWROUTE)
731 		fib_table_insert(tb, &cfg);
732 	else
733 		fib_table_delete(tb, &cfg);
734 }
735 
736 void fib_add_ifaddr(struct in_ifaddr *ifa)
737 {
738 	struct in_device *in_dev = ifa->ifa_dev;
739 	struct net_device *dev = in_dev->dev;
740 	struct in_ifaddr *prim = ifa;
741 	__be32 mask = ifa->ifa_mask;
742 	__be32 addr = ifa->ifa_local;
743 	__be32 prefix = ifa->ifa_address & mask;
744 
745 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
746 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
747 		if (prim == NULL) {
748 			pr_warn("%s: bug: prim == NULL\n", __func__);
749 			return;
750 		}
751 	}
752 
753 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
754 
755 	if (!(dev->flags & IFF_UP))
756 		return;
757 
758 	/* Add broadcast address, if it is explicitly assigned. */
759 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
760 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
761 
762 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
763 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
764 		fib_magic(RTM_NEWROUTE,
765 			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
766 			  prefix, ifa->ifa_prefixlen, prim);
767 
768 		/* Add network specific broadcasts, when it takes a sense */
769 		if (ifa->ifa_prefixlen < 31) {
770 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
771 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
772 				  32, prim);
773 		}
774 	}
775 }
776 
777 /* Delete primary or secondary address.
778  * Optionally, on secondary address promotion consider the addresses
779  * from subnet iprim as deleted, even if they are in device list.
780  * In this case the secondary ifa can be in device list.
781  */
782 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
783 {
784 	struct in_device *in_dev = ifa->ifa_dev;
785 	struct net_device *dev = in_dev->dev;
786 	struct in_ifaddr *ifa1;
787 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
788 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
789 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
790 #define LOCAL_OK	1
791 #define BRD_OK		2
792 #define BRD0_OK		4
793 #define BRD1_OK		8
794 	unsigned int ok = 0;
795 	int subnet = 0;		/* Primary network */
796 	int gone = 1;		/* Address is missing */
797 	int same_prefsrc = 0;	/* Another primary with same IP */
798 
799 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
800 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
801 		if (prim == NULL) {
802 			pr_warn("%s: bug: prim == NULL\n", __func__);
803 			return;
804 		}
805 		if (iprim && iprim != prim) {
806 			pr_warn("%s: bug: iprim != prim\n", __func__);
807 			return;
808 		}
809 	} else if (!ipv4_is_zeronet(any) &&
810 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
811 		fib_magic(RTM_DELROUTE,
812 			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
813 			  any, ifa->ifa_prefixlen, prim);
814 		subnet = 1;
815 	}
816 
817 	/* Deletion is more complicated than add.
818 	 * We should take care of not to delete too much :-)
819 	 *
820 	 * Scan address list to be sure that addresses are really gone.
821 	 */
822 
823 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
824 		if (ifa1 == ifa) {
825 			/* promotion, keep the IP */
826 			gone = 0;
827 			continue;
828 		}
829 		/* Ignore IFAs from our subnet */
830 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
831 		    inet_ifa_match(ifa1->ifa_address, iprim))
832 			continue;
833 
834 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
835 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
836 			/* Another address from our subnet? */
837 			if (ifa1->ifa_mask == prim->ifa_mask &&
838 			    inet_ifa_match(ifa1->ifa_address, prim))
839 				prim1 = prim;
840 			else {
841 				/* We reached the secondaries, so
842 				 * same_prefsrc should be determined.
843 				 */
844 				if (!same_prefsrc)
845 					continue;
846 				/* Search new prim1 if ifa1 is not
847 				 * using the current prim1
848 				 */
849 				if (!prim1 ||
850 				    ifa1->ifa_mask != prim1->ifa_mask ||
851 				    !inet_ifa_match(ifa1->ifa_address, prim1))
852 					prim1 = inet_ifa_byprefix(in_dev,
853 							ifa1->ifa_address,
854 							ifa1->ifa_mask);
855 				if (!prim1)
856 					continue;
857 				if (prim1->ifa_local != prim->ifa_local)
858 					continue;
859 			}
860 		} else {
861 			if (prim->ifa_local != ifa1->ifa_local)
862 				continue;
863 			prim1 = ifa1;
864 			if (prim != prim1)
865 				same_prefsrc = 1;
866 		}
867 		if (ifa->ifa_local == ifa1->ifa_local)
868 			ok |= LOCAL_OK;
869 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
870 			ok |= BRD_OK;
871 		if (brd == ifa1->ifa_broadcast)
872 			ok |= BRD1_OK;
873 		if (any == ifa1->ifa_broadcast)
874 			ok |= BRD0_OK;
875 		/* primary has network specific broadcasts */
876 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
877 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
878 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
879 
880 			if (!ipv4_is_zeronet(any1)) {
881 				if (ifa->ifa_broadcast == brd1 ||
882 				    ifa->ifa_broadcast == any1)
883 					ok |= BRD_OK;
884 				if (brd == brd1 || brd == any1)
885 					ok |= BRD1_OK;
886 				if (any == brd1 || any == any1)
887 					ok |= BRD0_OK;
888 			}
889 		}
890 	}
891 
892 	if (!(ok & BRD_OK))
893 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
894 	if (subnet && ifa->ifa_prefixlen < 31) {
895 		if (!(ok & BRD1_OK))
896 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
897 		if (!(ok & BRD0_OK))
898 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
899 	}
900 	if (!(ok & LOCAL_OK)) {
901 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
902 
903 		/* Check, that this local address finally disappeared. */
904 		if (gone &&
905 		    inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
906 			/* And the last, but not the least thing.
907 			 * We must flush stray FIB entries.
908 			 *
909 			 * First of all, we scan fib_info list searching
910 			 * for stray nexthop entries, then ignite fib_flush.
911 			 */
912 			if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
913 				fib_flush(dev_net(dev));
914 		}
915 	}
916 #undef LOCAL_OK
917 #undef BRD_OK
918 #undef BRD0_OK
919 #undef BRD1_OK
920 }
921 
922 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
923 {
924 
925 	struct fib_result       res;
926 	struct flowi4           fl4 = {
927 		.flowi4_mark = frn->fl_mark,
928 		.daddr = frn->fl_addr,
929 		.flowi4_tos = frn->fl_tos,
930 		.flowi4_scope = frn->fl_scope,
931 	};
932 
933 	frn->err = -ENOENT;
934 	if (tb) {
935 		local_bh_disable();
936 
937 		frn->tb_id = tb->tb_id;
938 		rcu_read_lock();
939 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
940 
941 		if (!frn->err) {
942 			frn->prefixlen = res.prefixlen;
943 			frn->nh_sel = res.nh_sel;
944 			frn->type = res.type;
945 			frn->scope = res.scope;
946 		}
947 		rcu_read_unlock();
948 		local_bh_enable();
949 	}
950 }
951 
952 static void nl_fib_input(struct sk_buff *skb)
953 {
954 	struct net *net;
955 	struct fib_result_nl *frn;
956 	struct nlmsghdr *nlh;
957 	struct fib_table *tb;
958 	u32 pid;
959 
960 	net = sock_net(skb->sk);
961 	nlh = nlmsg_hdr(skb);
962 	if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
963 	    nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
964 		return;
965 
966 	skb = skb_clone(skb, GFP_KERNEL);
967 	if (skb == NULL)
968 		return;
969 	nlh = nlmsg_hdr(skb);
970 
971 	frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
972 	tb = fib_get_table(net, frn->tb_id_in);
973 
974 	nl_fib_lookup(frn, tb);
975 
976 	pid = NETLINK_CB(skb).pid;      /* pid of sending process */
977 	NETLINK_CB(skb).pid = 0;        /* from kernel */
978 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
979 	netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
980 }
981 
982 static int __net_init nl_fib_lookup_init(struct net *net)
983 {
984 	struct sock *sk;
985 	struct netlink_kernel_cfg cfg = {
986 		.input	= nl_fib_input,
987 	};
988 
989 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, THIS_MODULE, &cfg);
990 	if (sk == NULL)
991 		return -EAFNOSUPPORT;
992 	net->ipv4.fibnl = sk;
993 	return 0;
994 }
995 
996 static void nl_fib_lookup_exit(struct net *net)
997 {
998 	netlink_kernel_release(net->ipv4.fibnl);
999 	net->ipv4.fibnl = NULL;
1000 }
1001 
1002 static void fib_disable_ip(struct net_device *dev, int force, int delay)
1003 {
1004 	if (fib_sync_down_dev(dev, force))
1005 		fib_flush(dev_net(dev));
1006 	rt_cache_flush(dev_net(dev), delay);
1007 	arp_ifdown(dev);
1008 }
1009 
1010 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1011 {
1012 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1013 	struct net_device *dev = ifa->ifa_dev->dev;
1014 	struct net *net = dev_net(dev);
1015 
1016 	switch (event) {
1017 	case NETDEV_UP:
1018 		fib_add_ifaddr(ifa);
1019 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1020 		fib_sync_up(dev);
1021 #endif
1022 		atomic_inc(&net->ipv4.dev_addr_genid);
1023 		rt_cache_flush(dev_net(dev), -1);
1024 		break;
1025 	case NETDEV_DOWN:
1026 		fib_del_ifaddr(ifa, NULL);
1027 		atomic_inc(&net->ipv4.dev_addr_genid);
1028 		if (ifa->ifa_dev->ifa_list == NULL) {
1029 			/* Last address was deleted from this interface.
1030 			 * Disable IP.
1031 			 */
1032 			fib_disable_ip(dev, 1, 0);
1033 		} else {
1034 			rt_cache_flush(dev_net(dev), -1);
1035 		}
1036 		break;
1037 	}
1038 	return NOTIFY_DONE;
1039 }
1040 
1041 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1042 {
1043 	struct net_device *dev = ptr;
1044 	struct in_device *in_dev = __in_dev_get_rtnl(dev);
1045 	struct net *net = dev_net(dev);
1046 
1047 	if (event == NETDEV_UNREGISTER) {
1048 		fib_disable_ip(dev, 2, -1);
1049 		rt_flush_dev(dev);
1050 		return NOTIFY_DONE;
1051 	}
1052 
1053 	if (!in_dev)
1054 		return NOTIFY_DONE;
1055 
1056 	switch (event) {
1057 	case NETDEV_UP:
1058 		for_ifa(in_dev) {
1059 			fib_add_ifaddr(ifa);
1060 		} endfor_ifa(in_dev);
1061 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1062 		fib_sync_up(dev);
1063 #endif
1064 		atomic_inc(&net->ipv4.dev_addr_genid);
1065 		rt_cache_flush(dev_net(dev), -1);
1066 		break;
1067 	case NETDEV_DOWN:
1068 		fib_disable_ip(dev, 0, 0);
1069 		break;
1070 	case NETDEV_CHANGEMTU:
1071 	case NETDEV_CHANGE:
1072 		rt_cache_flush(dev_net(dev), 0);
1073 		break;
1074 	case NETDEV_UNREGISTER_BATCH:
1075 		break;
1076 	}
1077 	return NOTIFY_DONE;
1078 }
1079 
1080 static struct notifier_block fib_inetaddr_notifier = {
1081 	.notifier_call = fib_inetaddr_event,
1082 };
1083 
1084 static struct notifier_block fib_netdev_notifier = {
1085 	.notifier_call = fib_netdev_event,
1086 };
1087 
1088 static int __net_init ip_fib_net_init(struct net *net)
1089 {
1090 	int err;
1091 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1092 
1093 	/* Avoid false sharing : Use at least a full cache line */
1094 	size = max_t(size_t, size, L1_CACHE_BYTES);
1095 
1096 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1097 	if (net->ipv4.fib_table_hash == NULL)
1098 		return -ENOMEM;
1099 
1100 	err = fib4_rules_init(net);
1101 	if (err < 0)
1102 		goto fail;
1103 	return 0;
1104 
1105 fail:
1106 	kfree(net->ipv4.fib_table_hash);
1107 	return err;
1108 }
1109 
1110 static void ip_fib_net_exit(struct net *net)
1111 {
1112 	unsigned int i;
1113 
1114 #ifdef CONFIG_IP_MULTIPLE_TABLES
1115 	fib4_rules_exit(net);
1116 #endif
1117 
1118 	rtnl_lock();
1119 	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1120 		struct fib_table *tb;
1121 		struct hlist_head *head;
1122 		struct hlist_node *node, *tmp;
1123 
1124 		head = &net->ipv4.fib_table_hash[i];
1125 		hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1126 			hlist_del(node);
1127 			fib_table_flush(tb);
1128 			fib_free_table(tb);
1129 		}
1130 	}
1131 	rtnl_unlock();
1132 	kfree(net->ipv4.fib_table_hash);
1133 }
1134 
1135 static int __net_init fib_net_init(struct net *net)
1136 {
1137 	int error;
1138 
1139 #ifdef CONFIG_IP_ROUTE_CLASSID
1140 	net->ipv4.fib_num_tclassid_users = 0;
1141 #endif
1142 	error = ip_fib_net_init(net);
1143 	if (error < 0)
1144 		goto out;
1145 	error = nl_fib_lookup_init(net);
1146 	if (error < 0)
1147 		goto out_nlfl;
1148 	error = fib_proc_init(net);
1149 	if (error < 0)
1150 		goto out_proc;
1151 out:
1152 	return error;
1153 
1154 out_proc:
1155 	nl_fib_lookup_exit(net);
1156 out_nlfl:
1157 	ip_fib_net_exit(net);
1158 	goto out;
1159 }
1160 
1161 static void __net_exit fib_net_exit(struct net *net)
1162 {
1163 	fib_proc_exit(net);
1164 	nl_fib_lookup_exit(net);
1165 	ip_fib_net_exit(net);
1166 }
1167 
1168 static struct pernet_operations fib_net_ops = {
1169 	.init = fib_net_init,
1170 	.exit = fib_net_exit,
1171 };
1172 
1173 void __init ip_fib_init(void)
1174 {
1175 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1176 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1177 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1178 
1179 	register_pernet_subsys(&fib_net_ops);
1180 	register_netdevice_notifier(&fib_netdev_notifier);
1181 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1182 
1183 	fib_trie_init();
1184 }
1185