xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision bbde9fc1824aab58bc78c084163007dd6c03fe5b)
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 	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
56 	if (!main_table)
57 		return -ENOMEM;
58 
59 	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
60 	if (!local_table)
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 	fib_free_table(main_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, *alias = NULL;
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 	if (id == RT_TABLE_LOCAL)
87 		alias = fib_new_table(net, RT_TABLE_MAIN);
88 
89 	tb = fib_trie_table(id, alias);
90 	if (!tb)
91 		return NULL;
92 
93 	switch (id) {
94 	case RT_TABLE_LOCAL:
95 		rcu_assign_pointer(net->ipv4.fib_local, tb);
96 		break;
97 	case RT_TABLE_MAIN:
98 		rcu_assign_pointer(net->ipv4.fib_main, tb);
99 		break;
100 	case RT_TABLE_DEFAULT:
101 		rcu_assign_pointer(net->ipv4.fib_default, tb);
102 		break;
103 	default:
104 		break;
105 	}
106 
107 	h = id & (FIB_TABLE_HASHSZ - 1);
108 	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
109 	return tb;
110 }
111 
112 /* caller must hold either rtnl or rcu read lock */
113 struct fib_table *fib_get_table(struct net *net, u32 id)
114 {
115 	struct fib_table *tb;
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 	head = &net->ipv4.fib_table_hash[h];
124 	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
125 		if (tb->tb_id == id)
126 			return tb;
127 	}
128 	return NULL;
129 }
130 #endif /* CONFIG_IP_MULTIPLE_TABLES */
131 
132 static void fib_replace_table(struct net *net, struct fib_table *old,
133 			      struct fib_table *new)
134 {
135 #ifdef CONFIG_IP_MULTIPLE_TABLES
136 	switch (new->tb_id) {
137 	case RT_TABLE_LOCAL:
138 		rcu_assign_pointer(net->ipv4.fib_local, new);
139 		break;
140 	case RT_TABLE_MAIN:
141 		rcu_assign_pointer(net->ipv4.fib_main, new);
142 		break;
143 	case RT_TABLE_DEFAULT:
144 		rcu_assign_pointer(net->ipv4.fib_default, new);
145 		break;
146 	default:
147 		break;
148 	}
149 
150 #endif
151 	/* replace the old table in the hlist */
152 	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
153 }
154 
155 int fib_unmerge(struct net *net)
156 {
157 	struct fib_table *old, *new;
158 
159 	/* attempt to fetch local table if it has been allocated */
160 	old = fib_get_table(net, RT_TABLE_LOCAL);
161 	if (!old)
162 		return 0;
163 
164 	new = fib_trie_unmerge(old);
165 	if (!new)
166 		return -ENOMEM;
167 
168 	/* replace merged table with clean table */
169 	if (new != old) {
170 		fib_replace_table(net, old, new);
171 		fib_free_table(old);
172 	}
173 
174 	return 0;
175 }
176 
177 static void fib_flush(struct net *net)
178 {
179 	int flushed = 0;
180 	unsigned int h;
181 
182 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
183 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
184 		struct hlist_node *tmp;
185 		struct fib_table *tb;
186 
187 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
188 			flushed += fib_table_flush(tb);
189 	}
190 
191 	if (flushed)
192 		rt_cache_flush(net);
193 }
194 
195 void fib_flush_external(struct net *net)
196 {
197 	struct fib_table *tb;
198 	struct hlist_head *head;
199 	unsigned int h;
200 
201 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
202 		head = &net->ipv4.fib_table_hash[h];
203 		hlist_for_each_entry(tb, head, tb_hlist)
204 			fib_table_flush_external(tb);
205 	}
206 }
207 
208 /*
209  * Find address type as if only "dev" was present in the system. If
210  * on_dev is NULL then all interfaces are taken into consideration.
211  */
212 static inline unsigned int __inet_dev_addr_type(struct net *net,
213 						const struct net_device *dev,
214 						__be32 addr)
215 {
216 	struct flowi4		fl4 = { .daddr = addr };
217 	struct fib_result	res;
218 	unsigned int ret = RTN_BROADCAST;
219 	struct fib_table *local_table;
220 
221 	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
222 		return RTN_BROADCAST;
223 	if (ipv4_is_multicast(addr))
224 		return RTN_MULTICAST;
225 
226 	rcu_read_lock();
227 
228 	local_table = fib_get_table(net, RT_TABLE_LOCAL);
229 	if (local_table) {
230 		ret = RTN_UNICAST;
231 		if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) {
232 			if (!dev || dev == res.fi->fib_dev)
233 				ret = res.type;
234 		}
235 	}
236 
237 	rcu_read_unlock();
238 	return ret;
239 }
240 
241 unsigned int inet_addr_type(struct net *net, __be32 addr)
242 {
243 	return __inet_dev_addr_type(net, NULL, addr);
244 }
245 EXPORT_SYMBOL(inet_addr_type);
246 
247 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
248 				__be32 addr)
249 {
250 	return __inet_dev_addr_type(net, dev, addr);
251 }
252 EXPORT_SYMBOL(inet_dev_addr_type);
253 
254 __be32 fib_compute_spec_dst(struct sk_buff *skb)
255 {
256 	struct net_device *dev = skb->dev;
257 	struct in_device *in_dev;
258 	struct fib_result res;
259 	struct rtable *rt;
260 	struct flowi4 fl4;
261 	struct net *net;
262 	int scope;
263 
264 	rt = skb_rtable(skb);
265 	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
266 	    RTCF_LOCAL)
267 		return ip_hdr(skb)->daddr;
268 
269 	in_dev = __in_dev_get_rcu(dev);
270 	BUG_ON(!in_dev);
271 
272 	net = dev_net(dev);
273 
274 	scope = RT_SCOPE_UNIVERSE;
275 	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
276 		fl4.flowi4_oif = 0;
277 		fl4.flowi4_iif = LOOPBACK_IFINDEX;
278 		fl4.daddr = ip_hdr(skb)->saddr;
279 		fl4.saddr = 0;
280 		fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
281 		fl4.flowi4_scope = scope;
282 		fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
283 		fl4.flowi4_tun_key.tun_id = 0;
284 		if (!fib_lookup(net, &fl4, &res, 0))
285 			return FIB_RES_PREFSRC(net, res);
286 	} else {
287 		scope = RT_SCOPE_LINK;
288 	}
289 
290 	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
291 }
292 
293 /* Given (packet source, input interface) and optional (dst, oif, tos):
294  * - (main) check, that source is valid i.e. not broadcast or our local
295  *   address.
296  * - figure out what "logical" interface this packet arrived
297  *   and calculate "specific destination" address.
298  * - check, that packet arrived from expected physical interface.
299  * called with rcu_read_lock()
300  */
301 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
302 				 u8 tos, int oif, struct net_device *dev,
303 				 int rpf, struct in_device *idev, u32 *itag)
304 {
305 	int ret, no_addr;
306 	struct fib_result res;
307 	struct flowi4 fl4;
308 	struct net *net;
309 	bool dev_match;
310 
311 	fl4.flowi4_oif = 0;
312 	fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
313 	fl4.daddr = src;
314 	fl4.saddr = dst;
315 	fl4.flowi4_tos = tos;
316 	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
317 	fl4.flowi4_tun_key.tun_id = 0;
318 
319 	no_addr = idev->ifa_list == NULL;
320 
321 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
322 
323 	net = dev_net(dev);
324 	if (fib_lookup(net, &fl4, &res, 0))
325 		goto last_resort;
326 	if (res.type != RTN_UNICAST &&
327 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
328 		goto e_inval;
329 	if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
330 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
331 		goto last_resort;
332 	fib_combine_itag(itag, &res);
333 	dev_match = false;
334 
335 #ifdef CONFIG_IP_ROUTE_MULTIPATH
336 	for (ret = 0; ret < res.fi->fib_nhs; ret++) {
337 		struct fib_nh *nh = &res.fi->fib_nh[ret];
338 
339 		if (nh->nh_dev == dev) {
340 			dev_match = true;
341 			break;
342 		}
343 	}
344 #else
345 	if (FIB_RES_DEV(res) == dev)
346 		dev_match = true;
347 #endif
348 	if (dev_match) {
349 		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
350 		return ret;
351 	}
352 	if (no_addr)
353 		goto last_resort;
354 	if (rpf == 1)
355 		goto e_rpf;
356 	fl4.flowi4_oif = dev->ifindex;
357 
358 	ret = 0;
359 	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
360 		if (res.type == RTN_UNICAST)
361 			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
362 	}
363 	return ret;
364 
365 last_resort:
366 	if (rpf)
367 		goto e_rpf;
368 	*itag = 0;
369 	return 0;
370 
371 e_inval:
372 	return -EINVAL;
373 e_rpf:
374 	return -EXDEV;
375 }
376 
377 /* Ignore rp_filter for packets protected by IPsec. */
378 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
379 			u8 tos, int oif, struct net_device *dev,
380 			struct in_device *idev, u32 *itag)
381 {
382 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
383 
384 	if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
385 	    IN_DEV_ACCEPT_LOCAL(idev) &&
386 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
387 		*itag = 0;
388 		return 0;
389 	}
390 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
391 }
392 
393 static inline __be32 sk_extract_addr(struct sockaddr *addr)
394 {
395 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
396 }
397 
398 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
399 {
400 	struct nlattr *nla;
401 
402 	nla = (struct nlattr *) ((char *) mx + len);
403 	nla->nla_type = type;
404 	nla->nla_len = nla_attr_size(4);
405 	*(u32 *) nla_data(nla) = value;
406 
407 	return len + nla_total_size(4);
408 }
409 
410 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
411 				 struct fib_config *cfg)
412 {
413 	__be32 addr;
414 	int plen;
415 
416 	memset(cfg, 0, sizeof(*cfg));
417 	cfg->fc_nlinfo.nl_net = net;
418 
419 	if (rt->rt_dst.sa_family != AF_INET)
420 		return -EAFNOSUPPORT;
421 
422 	/*
423 	 * Check mask for validity:
424 	 * a) it must be contiguous.
425 	 * b) destination must have all host bits clear.
426 	 * c) if application forgot to set correct family (AF_INET),
427 	 *    reject request unless it is absolutely clear i.e.
428 	 *    both family and mask are zero.
429 	 */
430 	plen = 32;
431 	addr = sk_extract_addr(&rt->rt_dst);
432 	if (!(rt->rt_flags & RTF_HOST)) {
433 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
434 
435 		if (rt->rt_genmask.sa_family != AF_INET) {
436 			if (mask || rt->rt_genmask.sa_family)
437 				return -EAFNOSUPPORT;
438 		}
439 
440 		if (bad_mask(mask, addr))
441 			return -EINVAL;
442 
443 		plen = inet_mask_len(mask);
444 	}
445 
446 	cfg->fc_dst_len = plen;
447 	cfg->fc_dst = addr;
448 
449 	if (cmd != SIOCDELRT) {
450 		cfg->fc_nlflags = NLM_F_CREATE;
451 		cfg->fc_protocol = RTPROT_BOOT;
452 	}
453 
454 	if (rt->rt_metric)
455 		cfg->fc_priority = rt->rt_metric - 1;
456 
457 	if (rt->rt_flags & RTF_REJECT) {
458 		cfg->fc_scope = RT_SCOPE_HOST;
459 		cfg->fc_type = RTN_UNREACHABLE;
460 		return 0;
461 	}
462 
463 	cfg->fc_scope = RT_SCOPE_NOWHERE;
464 	cfg->fc_type = RTN_UNICAST;
465 
466 	if (rt->rt_dev) {
467 		char *colon;
468 		struct net_device *dev;
469 		char devname[IFNAMSIZ];
470 
471 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
472 			return -EFAULT;
473 
474 		devname[IFNAMSIZ-1] = 0;
475 		colon = strchr(devname, ':');
476 		if (colon)
477 			*colon = 0;
478 		dev = __dev_get_by_name(net, devname);
479 		if (!dev)
480 			return -ENODEV;
481 		cfg->fc_oif = dev->ifindex;
482 		if (colon) {
483 			struct in_ifaddr *ifa;
484 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
485 			if (!in_dev)
486 				return -ENODEV;
487 			*colon = ':';
488 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
489 				if (strcmp(ifa->ifa_label, devname) == 0)
490 					break;
491 			if (!ifa)
492 				return -ENODEV;
493 			cfg->fc_prefsrc = ifa->ifa_local;
494 		}
495 	}
496 
497 	addr = sk_extract_addr(&rt->rt_gateway);
498 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
499 		cfg->fc_gw = addr;
500 		if (rt->rt_flags & RTF_GATEWAY &&
501 		    inet_addr_type(net, addr) == RTN_UNICAST)
502 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
503 	}
504 
505 	if (cmd == SIOCDELRT)
506 		return 0;
507 
508 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
509 		return -EINVAL;
510 
511 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
512 		cfg->fc_scope = RT_SCOPE_LINK;
513 
514 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
515 		struct nlattr *mx;
516 		int len = 0;
517 
518 		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
519 		if (!mx)
520 			return -ENOMEM;
521 
522 		if (rt->rt_flags & RTF_MTU)
523 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
524 
525 		if (rt->rt_flags & RTF_WINDOW)
526 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
527 
528 		if (rt->rt_flags & RTF_IRTT)
529 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
530 
531 		cfg->fc_mx = mx;
532 		cfg->fc_mx_len = len;
533 	}
534 
535 	return 0;
536 }
537 
538 /*
539  * Handle IP routing ioctl calls.
540  * These are used to manipulate the routing tables
541  */
542 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
543 {
544 	struct fib_config cfg;
545 	struct rtentry rt;
546 	int err;
547 
548 	switch (cmd) {
549 	case SIOCADDRT:		/* Add a route */
550 	case SIOCDELRT:		/* Delete a route */
551 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
552 			return -EPERM;
553 
554 		if (copy_from_user(&rt, arg, sizeof(rt)))
555 			return -EFAULT;
556 
557 		rtnl_lock();
558 		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
559 		if (err == 0) {
560 			struct fib_table *tb;
561 
562 			if (cmd == SIOCDELRT) {
563 				tb = fib_get_table(net, cfg.fc_table);
564 				if (tb)
565 					err = fib_table_delete(tb, &cfg);
566 				else
567 					err = -ESRCH;
568 			} else {
569 				tb = fib_new_table(net, cfg.fc_table);
570 				if (tb)
571 					err = fib_table_insert(tb, &cfg);
572 				else
573 					err = -ENOBUFS;
574 			}
575 
576 			/* allocated by rtentry_to_fib_config() */
577 			kfree(cfg.fc_mx);
578 		}
579 		rtnl_unlock();
580 		return err;
581 	}
582 	return -EINVAL;
583 }
584 
585 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
586 	[RTA_DST]		= { .type = NLA_U32 },
587 	[RTA_SRC]		= { .type = NLA_U32 },
588 	[RTA_IIF]		= { .type = NLA_U32 },
589 	[RTA_OIF]		= { .type = NLA_U32 },
590 	[RTA_GATEWAY]		= { .type = NLA_U32 },
591 	[RTA_PRIORITY]		= { .type = NLA_U32 },
592 	[RTA_PREFSRC]		= { .type = NLA_U32 },
593 	[RTA_METRICS]		= { .type = NLA_NESTED },
594 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
595 	[RTA_FLOW]		= { .type = NLA_U32 },
596 	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
597 	[RTA_ENCAP]		= { .type = NLA_NESTED },
598 };
599 
600 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
601 			     struct nlmsghdr *nlh, struct fib_config *cfg)
602 {
603 	struct nlattr *attr;
604 	int err, remaining;
605 	struct rtmsg *rtm;
606 
607 	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
608 	if (err < 0)
609 		goto errout;
610 
611 	memset(cfg, 0, sizeof(*cfg));
612 
613 	rtm = nlmsg_data(nlh);
614 	cfg->fc_dst_len = rtm->rtm_dst_len;
615 	cfg->fc_tos = rtm->rtm_tos;
616 	cfg->fc_table = rtm->rtm_table;
617 	cfg->fc_protocol = rtm->rtm_protocol;
618 	cfg->fc_scope = rtm->rtm_scope;
619 	cfg->fc_type = rtm->rtm_type;
620 	cfg->fc_flags = rtm->rtm_flags;
621 	cfg->fc_nlflags = nlh->nlmsg_flags;
622 
623 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
624 	cfg->fc_nlinfo.nlh = nlh;
625 	cfg->fc_nlinfo.nl_net = net;
626 
627 	if (cfg->fc_type > RTN_MAX) {
628 		err = -EINVAL;
629 		goto errout;
630 	}
631 
632 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
633 		switch (nla_type(attr)) {
634 		case RTA_DST:
635 			cfg->fc_dst = nla_get_be32(attr);
636 			break;
637 		case RTA_OIF:
638 			cfg->fc_oif = nla_get_u32(attr);
639 			break;
640 		case RTA_GATEWAY:
641 			cfg->fc_gw = nla_get_be32(attr);
642 			break;
643 		case RTA_PRIORITY:
644 			cfg->fc_priority = nla_get_u32(attr);
645 			break;
646 		case RTA_PREFSRC:
647 			cfg->fc_prefsrc = nla_get_be32(attr);
648 			break;
649 		case RTA_METRICS:
650 			cfg->fc_mx = nla_data(attr);
651 			cfg->fc_mx_len = nla_len(attr);
652 			break;
653 		case RTA_MULTIPATH:
654 			cfg->fc_mp = nla_data(attr);
655 			cfg->fc_mp_len = nla_len(attr);
656 			break;
657 		case RTA_FLOW:
658 			cfg->fc_flow = nla_get_u32(attr);
659 			break;
660 		case RTA_TABLE:
661 			cfg->fc_table = nla_get_u32(attr);
662 			break;
663 		case RTA_ENCAP:
664 			cfg->fc_encap = attr;
665 			break;
666 		case RTA_ENCAP_TYPE:
667 			cfg->fc_encap_type = nla_get_u16(attr);
668 			break;
669 		}
670 	}
671 
672 	return 0;
673 errout:
674 	return err;
675 }
676 
677 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
678 {
679 	struct net *net = sock_net(skb->sk);
680 	struct fib_config cfg;
681 	struct fib_table *tb;
682 	int err;
683 
684 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
685 	if (err < 0)
686 		goto errout;
687 
688 	tb = fib_get_table(net, cfg.fc_table);
689 	if (!tb) {
690 		err = -ESRCH;
691 		goto errout;
692 	}
693 
694 	err = fib_table_delete(tb, &cfg);
695 errout:
696 	return err;
697 }
698 
699 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
700 {
701 	struct net *net = sock_net(skb->sk);
702 	struct fib_config cfg;
703 	struct fib_table *tb;
704 	int err;
705 
706 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
707 	if (err < 0)
708 		goto errout;
709 
710 	tb = fib_new_table(net, cfg.fc_table);
711 	if (!tb) {
712 		err = -ENOBUFS;
713 		goto errout;
714 	}
715 
716 	err = fib_table_insert(tb, &cfg);
717 errout:
718 	return err;
719 }
720 
721 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
722 {
723 	struct net *net = sock_net(skb->sk);
724 	unsigned int h, s_h;
725 	unsigned int e = 0, s_e;
726 	struct fib_table *tb;
727 	struct hlist_head *head;
728 	int dumped = 0;
729 
730 	if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
731 	    ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
732 		return skb->len;
733 
734 	s_h = cb->args[0];
735 	s_e = cb->args[1];
736 
737 	rcu_read_lock();
738 
739 	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
740 		e = 0;
741 		head = &net->ipv4.fib_table_hash[h];
742 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
743 			if (e < s_e)
744 				goto next;
745 			if (dumped)
746 				memset(&cb->args[2], 0, sizeof(cb->args) -
747 						 2 * sizeof(cb->args[0]));
748 			if (fib_table_dump(tb, skb, cb) < 0)
749 				goto out;
750 			dumped = 1;
751 next:
752 			e++;
753 		}
754 	}
755 out:
756 	rcu_read_unlock();
757 
758 	cb->args[1] = e;
759 	cb->args[0] = h;
760 
761 	return skb->len;
762 }
763 
764 /* Prepare and feed intra-kernel routing request.
765  * Really, it should be netlink message, but :-( netlink
766  * can be not configured, so that we feed it directly
767  * to fib engine. It is legal, because all events occur
768  * only when netlink is already locked.
769  */
770 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
771 {
772 	struct net *net = dev_net(ifa->ifa_dev->dev);
773 	struct fib_table *tb;
774 	struct fib_config cfg = {
775 		.fc_protocol = RTPROT_KERNEL,
776 		.fc_type = type,
777 		.fc_dst = dst,
778 		.fc_dst_len = dst_len,
779 		.fc_prefsrc = ifa->ifa_local,
780 		.fc_oif = ifa->ifa_dev->dev->ifindex,
781 		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
782 		.fc_nlinfo = {
783 			.nl_net = net,
784 		},
785 	};
786 
787 	if (type == RTN_UNICAST)
788 		tb = fib_new_table(net, RT_TABLE_MAIN);
789 	else
790 		tb = fib_new_table(net, RT_TABLE_LOCAL);
791 
792 	if (!tb)
793 		return;
794 
795 	cfg.fc_table = tb->tb_id;
796 
797 	if (type != RTN_LOCAL)
798 		cfg.fc_scope = RT_SCOPE_LINK;
799 	else
800 		cfg.fc_scope = RT_SCOPE_HOST;
801 
802 	if (cmd == RTM_NEWROUTE)
803 		fib_table_insert(tb, &cfg);
804 	else
805 		fib_table_delete(tb, &cfg);
806 }
807 
808 void fib_add_ifaddr(struct in_ifaddr *ifa)
809 {
810 	struct in_device *in_dev = ifa->ifa_dev;
811 	struct net_device *dev = in_dev->dev;
812 	struct in_ifaddr *prim = ifa;
813 	__be32 mask = ifa->ifa_mask;
814 	__be32 addr = ifa->ifa_local;
815 	__be32 prefix = ifa->ifa_address & mask;
816 
817 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
818 		prim = inet_ifa_byprefix(in_dev, prefix, mask);
819 		if (!prim) {
820 			pr_warn("%s: bug: prim == NULL\n", __func__);
821 			return;
822 		}
823 	}
824 
825 	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
826 
827 	if (!(dev->flags & IFF_UP))
828 		return;
829 
830 	/* Add broadcast address, if it is explicitly assigned. */
831 	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
832 		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
833 
834 	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
835 	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
836 		fib_magic(RTM_NEWROUTE,
837 			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
838 			  prefix, ifa->ifa_prefixlen, prim);
839 
840 		/* Add network specific broadcasts, when it takes a sense */
841 		if (ifa->ifa_prefixlen < 31) {
842 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
843 			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
844 				  32, prim);
845 		}
846 	}
847 }
848 
849 /* Delete primary or secondary address.
850  * Optionally, on secondary address promotion consider the addresses
851  * from subnet iprim as deleted, even if they are in device list.
852  * In this case the secondary ifa can be in device list.
853  */
854 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
855 {
856 	struct in_device *in_dev = ifa->ifa_dev;
857 	struct net_device *dev = in_dev->dev;
858 	struct in_ifaddr *ifa1;
859 	struct in_ifaddr *prim = ifa, *prim1 = NULL;
860 	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
861 	__be32 any = ifa->ifa_address & ifa->ifa_mask;
862 #define LOCAL_OK	1
863 #define BRD_OK		2
864 #define BRD0_OK		4
865 #define BRD1_OK		8
866 	unsigned int ok = 0;
867 	int subnet = 0;		/* Primary network */
868 	int gone = 1;		/* Address is missing */
869 	int same_prefsrc = 0;	/* Another primary with same IP */
870 
871 	if (ifa->ifa_flags & IFA_F_SECONDARY) {
872 		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
873 		if (!prim) {
874 			pr_warn("%s: bug: prim == NULL\n", __func__);
875 			return;
876 		}
877 		if (iprim && iprim != prim) {
878 			pr_warn("%s: bug: iprim != prim\n", __func__);
879 			return;
880 		}
881 	} else if (!ipv4_is_zeronet(any) &&
882 		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
883 		fib_magic(RTM_DELROUTE,
884 			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
885 			  any, ifa->ifa_prefixlen, prim);
886 		subnet = 1;
887 	}
888 
889 	/* Deletion is more complicated than add.
890 	 * We should take care of not to delete too much :-)
891 	 *
892 	 * Scan address list to be sure that addresses are really gone.
893 	 */
894 
895 	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
896 		if (ifa1 == ifa) {
897 			/* promotion, keep the IP */
898 			gone = 0;
899 			continue;
900 		}
901 		/* Ignore IFAs from our subnet */
902 		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
903 		    inet_ifa_match(ifa1->ifa_address, iprim))
904 			continue;
905 
906 		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
907 		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
908 			/* Another address from our subnet? */
909 			if (ifa1->ifa_mask == prim->ifa_mask &&
910 			    inet_ifa_match(ifa1->ifa_address, prim))
911 				prim1 = prim;
912 			else {
913 				/* We reached the secondaries, so
914 				 * same_prefsrc should be determined.
915 				 */
916 				if (!same_prefsrc)
917 					continue;
918 				/* Search new prim1 if ifa1 is not
919 				 * using the current prim1
920 				 */
921 				if (!prim1 ||
922 				    ifa1->ifa_mask != prim1->ifa_mask ||
923 				    !inet_ifa_match(ifa1->ifa_address, prim1))
924 					prim1 = inet_ifa_byprefix(in_dev,
925 							ifa1->ifa_address,
926 							ifa1->ifa_mask);
927 				if (!prim1)
928 					continue;
929 				if (prim1->ifa_local != prim->ifa_local)
930 					continue;
931 			}
932 		} else {
933 			if (prim->ifa_local != ifa1->ifa_local)
934 				continue;
935 			prim1 = ifa1;
936 			if (prim != prim1)
937 				same_prefsrc = 1;
938 		}
939 		if (ifa->ifa_local == ifa1->ifa_local)
940 			ok |= LOCAL_OK;
941 		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
942 			ok |= BRD_OK;
943 		if (brd == ifa1->ifa_broadcast)
944 			ok |= BRD1_OK;
945 		if (any == ifa1->ifa_broadcast)
946 			ok |= BRD0_OK;
947 		/* primary has network specific broadcasts */
948 		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
949 			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
950 			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
951 
952 			if (!ipv4_is_zeronet(any1)) {
953 				if (ifa->ifa_broadcast == brd1 ||
954 				    ifa->ifa_broadcast == any1)
955 					ok |= BRD_OK;
956 				if (brd == brd1 || brd == any1)
957 					ok |= BRD1_OK;
958 				if (any == brd1 || any == any1)
959 					ok |= BRD0_OK;
960 			}
961 		}
962 	}
963 
964 	if (!(ok & BRD_OK))
965 		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
966 	if (subnet && ifa->ifa_prefixlen < 31) {
967 		if (!(ok & BRD1_OK))
968 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
969 		if (!(ok & BRD0_OK))
970 			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
971 	}
972 	if (!(ok & LOCAL_OK)) {
973 		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
974 
975 		/* Check, that this local address finally disappeared. */
976 		if (gone &&
977 		    inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
978 			/* And the last, but not the least thing.
979 			 * We must flush stray FIB entries.
980 			 *
981 			 * First of all, we scan fib_info list searching
982 			 * for stray nexthop entries, then ignite fib_flush.
983 			 */
984 			if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
985 				fib_flush(dev_net(dev));
986 		}
987 	}
988 #undef LOCAL_OK
989 #undef BRD_OK
990 #undef BRD0_OK
991 #undef BRD1_OK
992 }
993 
994 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
995 {
996 
997 	struct fib_result       res;
998 	struct flowi4           fl4 = {
999 		.flowi4_mark = frn->fl_mark,
1000 		.daddr = frn->fl_addr,
1001 		.flowi4_tos = frn->fl_tos,
1002 		.flowi4_scope = frn->fl_scope,
1003 	};
1004 	struct fib_table *tb;
1005 
1006 	rcu_read_lock();
1007 
1008 	tb = fib_get_table(net, frn->tb_id_in);
1009 
1010 	frn->err = -ENOENT;
1011 	if (tb) {
1012 		local_bh_disable();
1013 
1014 		frn->tb_id = tb->tb_id;
1015 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1016 
1017 		if (!frn->err) {
1018 			frn->prefixlen = res.prefixlen;
1019 			frn->nh_sel = res.nh_sel;
1020 			frn->type = res.type;
1021 			frn->scope = res.scope;
1022 		}
1023 		local_bh_enable();
1024 	}
1025 
1026 	rcu_read_unlock();
1027 }
1028 
1029 static void nl_fib_input(struct sk_buff *skb)
1030 {
1031 	struct net *net;
1032 	struct fib_result_nl *frn;
1033 	struct nlmsghdr *nlh;
1034 	u32 portid;
1035 
1036 	net = sock_net(skb->sk);
1037 	nlh = nlmsg_hdr(skb);
1038 	if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len ||
1039 	    nlmsg_len(nlh) < sizeof(*frn))
1040 		return;
1041 
1042 	skb = netlink_skb_clone(skb, GFP_KERNEL);
1043 	if (!skb)
1044 		return;
1045 	nlh = nlmsg_hdr(skb);
1046 
1047 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1048 	nl_fib_lookup(net, frn);
1049 
1050 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1051 	NETLINK_CB(skb).portid = 0;        /* from kernel */
1052 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1053 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1054 }
1055 
1056 static int __net_init nl_fib_lookup_init(struct net *net)
1057 {
1058 	struct sock *sk;
1059 	struct netlink_kernel_cfg cfg = {
1060 		.input	= nl_fib_input,
1061 	};
1062 
1063 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1064 	if (!sk)
1065 		return -EAFNOSUPPORT;
1066 	net->ipv4.fibnl = sk;
1067 	return 0;
1068 }
1069 
1070 static void nl_fib_lookup_exit(struct net *net)
1071 {
1072 	netlink_kernel_release(net->ipv4.fibnl);
1073 	net->ipv4.fibnl = NULL;
1074 }
1075 
1076 static void fib_disable_ip(struct net_device *dev, unsigned long event)
1077 {
1078 	if (fib_sync_down_dev(dev, event))
1079 		fib_flush(dev_net(dev));
1080 	rt_cache_flush(dev_net(dev));
1081 	arp_ifdown(dev);
1082 }
1083 
1084 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1085 {
1086 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1087 	struct net_device *dev = ifa->ifa_dev->dev;
1088 	struct net *net = dev_net(dev);
1089 
1090 	switch (event) {
1091 	case NETDEV_UP:
1092 		fib_add_ifaddr(ifa);
1093 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1094 		fib_sync_up(dev, RTNH_F_DEAD);
1095 #endif
1096 		atomic_inc(&net->ipv4.dev_addr_genid);
1097 		rt_cache_flush(dev_net(dev));
1098 		break;
1099 	case NETDEV_DOWN:
1100 		fib_del_ifaddr(ifa, NULL);
1101 		atomic_inc(&net->ipv4.dev_addr_genid);
1102 		if (!ifa->ifa_dev->ifa_list) {
1103 			/* Last address was deleted from this interface.
1104 			 * Disable IP.
1105 			 */
1106 			fib_disable_ip(dev, event);
1107 		} else {
1108 			rt_cache_flush(dev_net(dev));
1109 		}
1110 		break;
1111 	}
1112 	return NOTIFY_DONE;
1113 }
1114 
1115 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1116 {
1117 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1118 	struct in_device *in_dev;
1119 	struct net *net = dev_net(dev);
1120 	unsigned int flags;
1121 
1122 	if (event == NETDEV_UNREGISTER) {
1123 		fib_disable_ip(dev, event);
1124 		rt_flush_dev(dev);
1125 		return NOTIFY_DONE;
1126 	}
1127 
1128 	in_dev = __in_dev_get_rtnl(dev);
1129 	if (!in_dev)
1130 		return NOTIFY_DONE;
1131 
1132 	switch (event) {
1133 	case NETDEV_UP:
1134 		for_ifa(in_dev) {
1135 			fib_add_ifaddr(ifa);
1136 		} endfor_ifa(in_dev);
1137 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1138 		fib_sync_up(dev, RTNH_F_DEAD);
1139 #endif
1140 		atomic_inc(&net->ipv4.dev_addr_genid);
1141 		rt_cache_flush(net);
1142 		break;
1143 	case NETDEV_DOWN:
1144 		fib_disable_ip(dev, event);
1145 		break;
1146 	case NETDEV_CHANGE:
1147 		flags = dev_get_flags(dev);
1148 		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1149 			fib_sync_up(dev, RTNH_F_LINKDOWN);
1150 		else
1151 			fib_sync_down_dev(dev, event);
1152 		/* fall through */
1153 	case NETDEV_CHANGEMTU:
1154 		rt_cache_flush(net);
1155 		break;
1156 	}
1157 	return NOTIFY_DONE;
1158 }
1159 
1160 static struct notifier_block fib_inetaddr_notifier = {
1161 	.notifier_call = fib_inetaddr_event,
1162 };
1163 
1164 static struct notifier_block fib_netdev_notifier = {
1165 	.notifier_call = fib_netdev_event,
1166 };
1167 
1168 static int __net_init ip_fib_net_init(struct net *net)
1169 {
1170 	int err;
1171 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1172 
1173 	/* Avoid false sharing : Use at least a full cache line */
1174 	size = max_t(size_t, size, L1_CACHE_BYTES);
1175 
1176 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1177 	if (!net->ipv4.fib_table_hash)
1178 		return -ENOMEM;
1179 
1180 	err = fib4_rules_init(net);
1181 	if (err < 0)
1182 		goto fail;
1183 	return 0;
1184 
1185 fail:
1186 	kfree(net->ipv4.fib_table_hash);
1187 	return err;
1188 }
1189 
1190 static void ip_fib_net_exit(struct net *net)
1191 {
1192 	unsigned int i;
1193 
1194 	rtnl_lock();
1195 #ifdef CONFIG_IP_MULTIPLE_TABLES
1196 	RCU_INIT_POINTER(net->ipv4.fib_local, NULL);
1197 	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1198 	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1199 #endif
1200 	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1201 		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1202 		struct hlist_node *tmp;
1203 		struct fib_table *tb;
1204 
1205 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1206 			hlist_del(&tb->tb_hlist);
1207 			fib_table_flush(tb);
1208 			fib_free_table(tb);
1209 		}
1210 	}
1211 
1212 #ifdef CONFIG_IP_MULTIPLE_TABLES
1213 	fib4_rules_exit(net);
1214 #endif
1215 	rtnl_unlock();
1216 	kfree(net->ipv4.fib_table_hash);
1217 }
1218 
1219 static int __net_init fib_net_init(struct net *net)
1220 {
1221 	int error;
1222 
1223 #ifdef CONFIG_IP_ROUTE_CLASSID
1224 	net->ipv4.fib_num_tclassid_users = 0;
1225 #endif
1226 	error = ip_fib_net_init(net);
1227 	if (error < 0)
1228 		goto out;
1229 	error = nl_fib_lookup_init(net);
1230 	if (error < 0)
1231 		goto out_nlfl;
1232 	error = fib_proc_init(net);
1233 	if (error < 0)
1234 		goto out_proc;
1235 out:
1236 	return error;
1237 
1238 out_proc:
1239 	nl_fib_lookup_exit(net);
1240 out_nlfl:
1241 	ip_fib_net_exit(net);
1242 	goto out;
1243 }
1244 
1245 static void __net_exit fib_net_exit(struct net *net)
1246 {
1247 	fib_proc_exit(net);
1248 	nl_fib_lookup_exit(net);
1249 	ip_fib_net_exit(net);
1250 }
1251 
1252 static struct pernet_operations fib_net_ops = {
1253 	.init = fib_net_init,
1254 	.exit = fib_net_exit,
1255 };
1256 
1257 void __init ip_fib_init(void)
1258 {
1259 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1260 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1261 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1262 
1263 	register_pernet_subsys(&fib_net_ops);
1264 	register_netdevice_notifier(&fib_netdev_notifier);
1265 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1266 
1267 	fib_trie_init();
1268 }
1269