xref: /openbmc/linux/net/ipv4/fib_frontend.c (revision 609e478b)
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;
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 	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
262 
263 	net = dev_net(dev);
264 	if (fib_lookup(net, &fl4, &res))
265 		goto last_resort;
266 	if (res.type != RTN_UNICAST &&
267 	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
268 		goto e_inval;
269 	if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
270 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
271 		goto last_resort;
272 	fib_combine_itag(itag, &res);
273 	dev_match = false;
274 
275 #ifdef CONFIG_IP_ROUTE_MULTIPATH
276 	for (ret = 0; ret < res.fi->fib_nhs; ret++) {
277 		struct fib_nh *nh = &res.fi->fib_nh[ret];
278 
279 		if (nh->nh_dev == dev) {
280 			dev_match = true;
281 			break;
282 		}
283 	}
284 #else
285 	if (FIB_RES_DEV(res) == dev)
286 		dev_match = true;
287 #endif
288 	if (dev_match) {
289 		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
290 		return ret;
291 	}
292 	if (no_addr)
293 		goto last_resort;
294 	if (rpf == 1)
295 		goto e_rpf;
296 	fl4.flowi4_oif = dev->ifindex;
297 
298 	ret = 0;
299 	if (fib_lookup(net, &fl4, &res) == 0) {
300 		if (res.type == RTN_UNICAST)
301 			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
302 	}
303 	return ret;
304 
305 last_resort:
306 	if (rpf)
307 		goto e_rpf;
308 	*itag = 0;
309 	return 0;
310 
311 e_inval:
312 	return -EINVAL;
313 e_rpf:
314 	return -EXDEV;
315 }
316 
317 /* Ignore rp_filter for packets protected by IPsec. */
318 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
319 			u8 tos, int oif, struct net_device *dev,
320 			struct in_device *idev, u32 *itag)
321 {
322 	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
323 
324 	if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
325 	    IN_DEV_ACCEPT_LOCAL(idev) &&
326 	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
327 		*itag = 0;
328 		return 0;
329 	}
330 	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
331 }
332 
333 static inline __be32 sk_extract_addr(struct sockaddr *addr)
334 {
335 	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
336 }
337 
338 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
339 {
340 	struct nlattr *nla;
341 
342 	nla = (struct nlattr *) ((char *) mx + len);
343 	nla->nla_type = type;
344 	nla->nla_len = nla_attr_size(4);
345 	*(u32 *) nla_data(nla) = value;
346 
347 	return len + nla_total_size(4);
348 }
349 
350 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
351 				 struct fib_config *cfg)
352 {
353 	__be32 addr;
354 	int plen;
355 
356 	memset(cfg, 0, sizeof(*cfg));
357 	cfg->fc_nlinfo.nl_net = net;
358 
359 	if (rt->rt_dst.sa_family != AF_INET)
360 		return -EAFNOSUPPORT;
361 
362 	/*
363 	 * Check mask for validity:
364 	 * a) it must be contiguous.
365 	 * b) destination must have all host bits clear.
366 	 * c) if application forgot to set correct family (AF_INET),
367 	 *    reject request unless it is absolutely clear i.e.
368 	 *    both family and mask are zero.
369 	 */
370 	plen = 32;
371 	addr = sk_extract_addr(&rt->rt_dst);
372 	if (!(rt->rt_flags & RTF_HOST)) {
373 		__be32 mask = sk_extract_addr(&rt->rt_genmask);
374 
375 		if (rt->rt_genmask.sa_family != AF_INET) {
376 			if (mask || rt->rt_genmask.sa_family)
377 				return -EAFNOSUPPORT;
378 		}
379 
380 		if (bad_mask(mask, addr))
381 			return -EINVAL;
382 
383 		plen = inet_mask_len(mask);
384 	}
385 
386 	cfg->fc_dst_len = plen;
387 	cfg->fc_dst = addr;
388 
389 	if (cmd != SIOCDELRT) {
390 		cfg->fc_nlflags = NLM_F_CREATE;
391 		cfg->fc_protocol = RTPROT_BOOT;
392 	}
393 
394 	if (rt->rt_metric)
395 		cfg->fc_priority = rt->rt_metric - 1;
396 
397 	if (rt->rt_flags & RTF_REJECT) {
398 		cfg->fc_scope = RT_SCOPE_HOST;
399 		cfg->fc_type = RTN_UNREACHABLE;
400 		return 0;
401 	}
402 
403 	cfg->fc_scope = RT_SCOPE_NOWHERE;
404 	cfg->fc_type = RTN_UNICAST;
405 
406 	if (rt->rt_dev) {
407 		char *colon;
408 		struct net_device *dev;
409 		char devname[IFNAMSIZ];
410 
411 		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
412 			return -EFAULT;
413 
414 		devname[IFNAMSIZ-1] = 0;
415 		colon = strchr(devname, ':');
416 		if (colon)
417 			*colon = 0;
418 		dev = __dev_get_by_name(net, devname);
419 		if (!dev)
420 			return -ENODEV;
421 		cfg->fc_oif = dev->ifindex;
422 		if (colon) {
423 			struct in_ifaddr *ifa;
424 			struct in_device *in_dev = __in_dev_get_rtnl(dev);
425 			if (!in_dev)
426 				return -ENODEV;
427 			*colon = ':';
428 			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
429 				if (strcmp(ifa->ifa_label, devname) == 0)
430 					break;
431 			if (ifa == NULL)
432 				return -ENODEV;
433 			cfg->fc_prefsrc = ifa->ifa_local;
434 		}
435 	}
436 
437 	addr = sk_extract_addr(&rt->rt_gateway);
438 	if (rt->rt_gateway.sa_family == AF_INET && addr) {
439 		cfg->fc_gw = addr;
440 		if (rt->rt_flags & RTF_GATEWAY &&
441 		    inet_addr_type(net, addr) == RTN_UNICAST)
442 			cfg->fc_scope = RT_SCOPE_UNIVERSE;
443 	}
444 
445 	if (cmd == SIOCDELRT)
446 		return 0;
447 
448 	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
449 		return -EINVAL;
450 
451 	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
452 		cfg->fc_scope = RT_SCOPE_LINK;
453 
454 	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
455 		struct nlattr *mx;
456 		int len = 0;
457 
458 		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
459 		if (mx == NULL)
460 			return -ENOMEM;
461 
462 		if (rt->rt_flags & RTF_MTU)
463 			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
464 
465 		if (rt->rt_flags & RTF_WINDOW)
466 			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
467 
468 		if (rt->rt_flags & RTF_IRTT)
469 			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
470 
471 		cfg->fc_mx = mx;
472 		cfg->fc_mx_len = len;
473 	}
474 
475 	return 0;
476 }
477 
478 /*
479  * Handle IP routing ioctl calls.
480  * These are used to manipulate the routing tables
481  */
482 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
483 {
484 	struct fib_config cfg;
485 	struct rtentry rt;
486 	int err;
487 
488 	switch (cmd) {
489 	case SIOCADDRT:		/* Add a route */
490 	case SIOCDELRT:		/* Delete a route */
491 		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
492 			return -EPERM;
493 
494 		if (copy_from_user(&rt, arg, sizeof(rt)))
495 			return -EFAULT;
496 
497 		rtnl_lock();
498 		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
499 		if (err == 0) {
500 			struct fib_table *tb;
501 
502 			if (cmd == SIOCDELRT) {
503 				tb = fib_get_table(net, cfg.fc_table);
504 				if (tb)
505 					err = fib_table_delete(tb, &cfg);
506 				else
507 					err = -ESRCH;
508 			} else {
509 				tb = fib_new_table(net, cfg.fc_table);
510 				if (tb)
511 					err = fib_table_insert(tb, &cfg);
512 				else
513 					err = -ENOBUFS;
514 			}
515 
516 			/* allocated by rtentry_to_fib_config() */
517 			kfree(cfg.fc_mx);
518 		}
519 		rtnl_unlock();
520 		return err;
521 	}
522 	return -EINVAL;
523 }
524 
525 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
526 	[RTA_DST]		= { .type = NLA_U32 },
527 	[RTA_SRC]		= { .type = NLA_U32 },
528 	[RTA_IIF]		= { .type = NLA_U32 },
529 	[RTA_OIF]		= { .type = NLA_U32 },
530 	[RTA_GATEWAY]		= { .type = NLA_U32 },
531 	[RTA_PRIORITY]		= { .type = NLA_U32 },
532 	[RTA_PREFSRC]		= { .type = NLA_U32 },
533 	[RTA_METRICS]		= { .type = NLA_NESTED },
534 	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
535 	[RTA_FLOW]		= { .type = NLA_U32 },
536 };
537 
538 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
539 			     struct nlmsghdr *nlh, struct fib_config *cfg)
540 {
541 	struct nlattr *attr;
542 	int err, remaining;
543 	struct rtmsg *rtm;
544 
545 	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
546 	if (err < 0)
547 		goto errout;
548 
549 	memset(cfg, 0, sizeof(*cfg));
550 
551 	rtm = nlmsg_data(nlh);
552 	cfg->fc_dst_len = rtm->rtm_dst_len;
553 	cfg->fc_tos = rtm->rtm_tos;
554 	cfg->fc_table = rtm->rtm_table;
555 	cfg->fc_protocol = rtm->rtm_protocol;
556 	cfg->fc_scope = rtm->rtm_scope;
557 	cfg->fc_type = rtm->rtm_type;
558 	cfg->fc_flags = rtm->rtm_flags;
559 	cfg->fc_nlflags = nlh->nlmsg_flags;
560 
561 	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
562 	cfg->fc_nlinfo.nlh = nlh;
563 	cfg->fc_nlinfo.nl_net = net;
564 
565 	if (cfg->fc_type > RTN_MAX) {
566 		err = -EINVAL;
567 		goto errout;
568 	}
569 
570 	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
571 		switch (nla_type(attr)) {
572 		case RTA_DST:
573 			cfg->fc_dst = nla_get_be32(attr);
574 			break;
575 		case RTA_OIF:
576 			cfg->fc_oif = nla_get_u32(attr);
577 			break;
578 		case RTA_GATEWAY:
579 			cfg->fc_gw = nla_get_be32(attr);
580 			break;
581 		case RTA_PRIORITY:
582 			cfg->fc_priority = nla_get_u32(attr);
583 			break;
584 		case RTA_PREFSRC:
585 			cfg->fc_prefsrc = nla_get_be32(attr);
586 			break;
587 		case RTA_METRICS:
588 			cfg->fc_mx = nla_data(attr);
589 			cfg->fc_mx_len = nla_len(attr);
590 			break;
591 		case RTA_MULTIPATH:
592 			cfg->fc_mp = nla_data(attr);
593 			cfg->fc_mp_len = nla_len(attr);
594 			break;
595 		case RTA_FLOW:
596 			cfg->fc_flow = nla_get_u32(attr);
597 			break;
598 		case RTA_TABLE:
599 			cfg->fc_table = nla_get_u32(attr);
600 			break;
601 		}
602 	}
603 
604 	return 0;
605 errout:
606 	return err;
607 }
608 
609 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
610 {
611 	struct net *net = sock_net(skb->sk);
612 	struct fib_config cfg;
613 	struct fib_table *tb;
614 	int err;
615 
616 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
617 	if (err < 0)
618 		goto errout;
619 
620 	tb = fib_get_table(net, cfg.fc_table);
621 	if (tb == NULL) {
622 		err = -ESRCH;
623 		goto errout;
624 	}
625 
626 	err = fib_table_delete(tb, &cfg);
627 errout:
628 	return err;
629 }
630 
631 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
632 {
633 	struct net *net = sock_net(skb->sk);
634 	struct fib_config cfg;
635 	struct fib_table *tb;
636 	int err;
637 
638 	err = rtm_to_fib_config(net, skb, nlh, &cfg);
639 	if (err < 0)
640 		goto errout;
641 
642 	tb = fib_new_table(net, cfg.fc_table);
643 	if (tb == NULL) {
644 		err = -ENOBUFS;
645 		goto errout;
646 	}
647 
648 	err = fib_table_insert(tb, &cfg);
649 errout:
650 	return err;
651 }
652 
653 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
654 {
655 	struct net *net = sock_net(skb->sk);
656 	unsigned int h, s_h;
657 	unsigned int e = 0, s_e;
658 	struct fib_table *tb;
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 skb->len;
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, 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 		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
939 
940 		if (!frn->err) {
941 			frn->prefixlen = res.prefixlen;
942 			frn->nh_sel = res.nh_sel;
943 			frn->type = res.type;
944 			frn->scope = res.scope;
945 		}
946 		local_bh_enable();
947 	}
948 }
949 
950 static void nl_fib_input(struct sk_buff *skb)
951 {
952 	struct net *net;
953 	struct fib_result_nl *frn;
954 	struct nlmsghdr *nlh;
955 	struct fib_table *tb;
956 	u32 portid;
957 
958 	net = sock_net(skb->sk);
959 	nlh = nlmsg_hdr(skb);
960 	if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len ||
961 	    nlmsg_len(nlh) < sizeof(*frn))
962 		return;
963 
964 	skb = netlink_skb_clone(skb, GFP_KERNEL);
965 	if (skb == NULL)
966 		return;
967 	nlh = nlmsg_hdr(skb);
968 
969 	frn = (struct fib_result_nl *) nlmsg_data(nlh);
970 	tb = fib_get_table(net, frn->tb_id_in);
971 
972 	nl_fib_lookup(frn, tb);
973 
974 	portid = NETLINK_CB(skb).portid;      /* netlink portid */
975 	NETLINK_CB(skb).portid = 0;        /* from kernel */
976 	NETLINK_CB(skb).dst_group = 0;  /* unicast */
977 	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
978 }
979 
980 static int __net_init nl_fib_lookup_init(struct net *net)
981 {
982 	struct sock *sk;
983 	struct netlink_kernel_cfg cfg = {
984 		.input	= nl_fib_input,
985 	};
986 
987 	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
988 	if (sk == NULL)
989 		return -EAFNOSUPPORT;
990 	net->ipv4.fibnl = sk;
991 	return 0;
992 }
993 
994 static void nl_fib_lookup_exit(struct net *net)
995 {
996 	netlink_kernel_release(net->ipv4.fibnl);
997 	net->ipv4.fibnl = NULL;
998 }
999 
1000 static void fib_disable_ip(struct net_device *dev, int force)
1001 {
1002 	if (fib_sync_down_dev(dev, force))
1003 		fib_flush(dev_net(dev));
1004 	rt_cache_flush(dev_net(dev));
1005 	arp_ifdown(dev);
1006 }
1007 
1008 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1009 {
1010 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1011 	struct net_device *dev = ifa->ifa_dev->dev;
1012 	struct net *net = dev_net(dev);
1013 
1014 	switch (event) {
1015 	case NETDEV_UP:
1016 		fib_add_ifaddr(ifa);
1017 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1018 		fib_sync_up(dev);
1019 #endif
1020 		atomic_inc(&net->ipv4.dev_addr_genid);
1021 		rt_cache_flush(dev_net(dev));
1022 		break;
1023 	case NETDEV_DOWN:
1024 		fib_del_ifaddr(ifa, NULL);
1025 		atomic_inc(&net->ipv4.dev_addr_genid);
1026 		if (ifa->ifa_dev->ifa_list == NULL) {
1027 			/* Last address was deleted from this interface.
1028 			 * Disable IP.
1029 			 */
1030 			fib_disable_ip(dev, 1);
1031 		} else {
1032 			rt_cache_flush(dev_net(dev));
1033 		}
1034 		break;
1035 	}
1036 	return NOTIFY_DONE;
1037 }
1038 
1039 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1040 {
1041 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1042 	struct in_device *in_dev;
1043 	struct net *net = dev_net(dev);
1044 
1045 	if (event == NETDEV_UNREGISTER) {
1046 		fib_disable_ip(dev, 2);
1047 		rt_flush_dev(dev);
1048 		return NOTIFY_DONE;
1049 	}
1050 
1051 	in_dev = __in_dev_get_rtnl(dev);
1052 	if (!in_dev)
1053 		return NOTIFY_DONE;
1054 
1055 	switch (event) {
1056 	case NETDEV_UP:
1057 		for_ifa(in_dev) {
1058 			fib_add_ifaddr(ifa);
1059 		} endfor_ifa(in_dev);
1060 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1061 		fib_sync_up(dev);
1062 #endif
1063 		atomic_inc(&net->ipv4.dev_addr_genid);
1064 		rt_cache_flush(net);
1065 		break;
1066 	case NETDEV_DOWN:
1067 		fib_disable_ip(dev, 0);
1068 		break;
1069 	case NETDEV_CHANGEMTU:
1070 	case NETDEV_CHANGE:
1071 		rt_cache_flush(net);
1072 		break;
1073 	}
1074 	return NOTIFY_DONE;
1075 }
1076 
1077 static struct notifier_block fib_inetaddr_notifier = {
1078 	.notifier_call = fib_inetaddr_event,
1079 };
1080 
1081 static struct notifier_block fib_netdev_notifier = {
1082 	.notifier_call = fib_netdev_event,
1083 };
1084 
1085 static int __net_init ip_fib_net_init(struct net *net)
1086 {
1087 	int err;
1088 	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1089 
1090 	/* Avoid false sharing : Use at least a full cache line */
1091 	size = max_t(size_t, size, L1_CACHE_BYTES);
1092 
1093 	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1094 	if (net->ipv4.fib_table_hash == NULL)
1095 		return -ENOMEM;
1096 
1097 	err = fib4_rules_init(net);
1098 	if (err < 0)
1099 		goto fail;
1100 	return 0;
1101 
1102 fail:
1103 	kfree(net->ipv4.fib_table_hash);
1104 	return err;
1105 }
1106 
1107 static void ip_fib_net_exit(struct net *net)
1108 {
1109 	unsigned int i;
1110 
1111 #ifdef CONFIG_IP_MULTIPLE_TABLES
1112 	fib4_rules_exit(net);
1113 #endif
1114 
1115 	rtnl_lock();
1116 	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1117 		struct fib_table *tb;
1118 		struct hlist_head *head;
1119 		struct hlist_node *tmp;
1120 
1121 		head = &net->ipv4.fib_table_hash[i];
1122 		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1123 			hlist_del(&tb->tb_hlist);
1124 			fib_table_flush(tb);
1125 			fib_free_table(tb);
1126 		}
1127 	}
1128 	rtnl_unlock();
1129 	kfree(net->ipv4.fib_table_hash);
1130 }
1131 
1132 static int __net_init fib_net_init(struct net *net)
1133 {
1134 	int error;
1135 
1136 #ifdef CONFIG_IP_ROUTE_CLASSID
1137 	net->ipv4.fib_num_tclassid_users = 0;
1138 #endif
1139 	error = ip_fib_net_init(net);
1140 	if (error < 0)
1141 		goto out;
1142 	error = nl_fib_lookup_init(net);
1143 	if (error < 0)
1144 		goto out_nlfl;
1145 	error = fib_proc_init(net);
1146 	if (error < 0)
1147 		goto out_proc;
1148 out:
1149 	return error;
1150 
1151 out_proc:
1152 	nl_fib_lookup_exit(net);
1153 out_nlfl:
1154 	ip_fib_net_exit(net);
1155 	goto out;
1156 }
1157 
1158 static void __net_exit fib_net_exit(struct net *net)
1159 {
1160 	fib_proc_exit(net);
1161 	nl_fib_lookup_exit(net);
1162 	ip_fib_net_exit(net);
1163 }
1164 
1165 static struct pernet_operations fib_net_ops = {
1166 	.init = fib_net_init,
1167 	.exit = fib_net_exit,
1168 };
1169 
1170 void __init ip_fib_init(void)
1171 {
1172 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1173 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1174 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1175 
1176 	register_pernet_subsys(&fib_net_ops);
1177 	register_netdevice_notifier(&fib_netdev_notifier);
1178 	register_inetaddr_notifier(&fib_inetaddr_notifier);
1179 
1180 	fib_trie_init();
1181 }
1182