fib_frontend.c revision 5c04c819a20af40adb7d282199f4e34e14fa05c5
1/*
2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
3 *		operating system.  INET is implemented using the  BSD Socket
4 *		interface as the means of communication with the user level.
5 *
6 *		IPv4 Forwarding Information Base: FIB frontend.
7 *
8 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9 *
10 *		This program is free software; you can redistribute it and/or
11 *		modify it under the terms of the GNU General Public License
12 *		as published by the Free Software Foundation; either version
13 *		2 of the License, or (at your option) any later version.
14 */
15
16#include <linux/module.h>
17#include <asm/uaccess.h>
18#include <asm/system.h>
19#include <linux/bitops.h>
20#include <linux/capability.h>
21#include <linux/types.h>
22#include <linux/kernel.h>
23#include <linux/mm.h>
24#include <linux/string.h>
25#include <linux/socket.h>
26#include <linux/sockios.h>
27#include <linux/errno.h>
28#include <linux/in.h>
29#include <linux/inet.h>
30#include <linux/inetdevice.h>
31#include <linux/netdevice.h>
32#include <linux/if_addr.h>
33#include <linux/if_arp.h>
34#include <linux/skbuff.h>
35#include <linux/init.h>
36#include <linux/list.h>
37#include <linux/slab.h>
38
39#include <net/ip.h>
40#include <net/protocol.h>
41#include <net/route.h>
42#include <net/tcp.h>
43#include <net/sock.h>
44#include <net/arp.h>
45#include <net/ip_fib.h>
46#include <net/rtnetlink.h>
47
48#ifndef CONFIG_IP_MULTIPLE_TABLES
49
50static int __net_init fib4_rules_init(struct net *net)
51{
52	struct fib_table *local_table, *main_table;
53
54	local_table = fib_trie_table(RT_TABLE_LOCAL);
55	if (local_table == NULL)
56		return -ENOMEM;
57
58	main_table  = fib_trie_table(RT_TABLE_MAIN);
59	if (main_table == NULL)
60		goto fail;
61
62	hlist_add_head_rcu(&local_table->tb_hlist,
63				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
64	hlist_add_head_rcu(&main_table->tb_hlist,
65				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
66	return 0;
67
68fail:
69	kfree(local_table);
70	return -ENOMEM;
71}
72#else
73
74struct fib_table *fib_new_table(struct net *net, u32 id)
75{
76	struct fib_table *tb;
77	unsigned int h;
78
79	if (id == 0)
80		id = RT_TABLE_MAIN;
81	tb = fib_get_table(net, id);
82	if (tb)
83		return tb;
84
85	tb = fib_trie_table(id);
86	if (!tb)
87		return NULL;
88	h = id & (FIB_TABLE_HASHSZ - 1);
89	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
90	return tb;
91}
92
93struct fib_table *fib_get_table(struct net *net, u32 id)
94{
95	struct fib_table *tb;
96	struct hlist_node *node;
97	struct hlist_head *head;
98	unsigned int h;
99
100	if (id == 0)
101		id = RT_TABLE_MAIN;
102	h = id & (FIB_TABLE_HASHSZ - 1);
103
104	rcu_read_lock();
105	head = &net->ipv4.fib_table_hash[h];
106	hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
107		if (tb->tb_id == id) {
108			rcu_read_unlock();
109			return tb;
110		}
111	}
112	rcu_read_unlock();
113	return NULL;
114}
115#endif /* CONFIG_IP_MULTIPLE_TABLES */
116
117static void fib_flush(struct net *net)
118{
119	int flushed = 0;
120	struct fib_table *tb;
121	struct hlist_node *node;
122	struct hlist_head *head;
123	unsigned int h;
124
125	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
126		head = &net->ipv4.fib_table_hash[h];
127		hlist_for_each_entry(tb, node, head, tb_hlist)
128			flushed += fib_table_flush(tb);
129	}
130
131	if (flushed)
132		rt_cache_flush(net, -1);
133}
134
135/*
136 * Find address type as if only "dev" was present in the system. If
137 * on_dev is NULL then all interfaces are taken into consideration.
138 */
139static inline unsigned __inet_dev_addr_type(struct net *net,
140					    const struct net_device *dev,
141					    __be32 addr)
142{
143	struct flowi4		fl4 = { .daddr = addr };
144	struct fib_result	res;
145	unsigned ret = RTN_BROADCAST;
146	struct fib_table *local_table;
147
148	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
149		return RTN_BROADCAST;
150	if (ipv4_is_multicast(addr))
151		return RTN_MULTICAST;
152
153#ifdef CONFIG_IP_MULTIPLE_TABLES
154	res.r = NULL;
155#endif
156
157	local_table = fib_get_table(net, RT_TABLE_LOCAL);
158	if (local_table) {
159		ret = RTN_UNICAST;
160		rcu_read_lock();
161		if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) {
162			if (!dev || dev == res.fi->fib_dev)
163				ret = res.type;
164		}
165		rcu_read_unlock();
166	}
167	return ret;
168}
169
170unsigned int inet_addr_type(struct net *net, __be32 addr)
171{
172	return __inet_dev_addr_type(net, NULL, addr);
173}
174EXPORT_SYMBOL(inet_addr_type);
175
176unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
177				__be32 addr)
178{
179	return __inet_dev_addr_type(net, dev, addr);
180}
181EXPORT_SYMBOL(inet_dev_addr_type);
182
183/* Given (packet source, input interface) and optional (dst, oif, tos):
184 * - (main) check, that source is valid i.e. not broadcast or our local
185 *   address.
186 * - figure out what "logical" interface this packet arrived
187 *   and calculate "specific destination" address.
188 * - check, that packet arrived from expected physical interface.
189 * called with rcu_read_lock()
190 */
191int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, u8 tos,
192			int oif, struct net_device *dev, __be32 *spec_dst,
193			u32 *itag)
194{
195	struct in_device *in_dev;
196	struct flowi4 fl4;
197	struct fib_result res;
198	int no_addr, rpf, accept_local;
199	bool dev_match;
200	int ret;
201	struct net *net;
202
203	fl4.flowi4_oif = 0;
204	fl4.flowi4_iif = oif;
205	fl4.daddr = src;
206	fl4.saddr = dst;
207	fl4.flowi4_tos = tos;
208	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
209
210	no_addr = rpf = accept_local = 0;
211	in_dev = __in_dev_get_rcu(dev);
212	if (in_dev) {
213		no_addr = in_dev->ifa_list == NULL;
214		rpf = IN_DEV_RPFILTER(in_dev);
215		accept_local = IN_DEV_ACCEPT_LOCAL(in_dev);
216		fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
217	}
218
219	if (in_dev == NULL)
220		goto e_inval;
221
222	net = dev_net(dev);
223	if (fib_lookup(net, &fl4, &res))
224		goto last_resort;
225	if (res.type != RTN_UNICAST) {
226		if (res.type != RTN_LOCAL || !accept_local)
227			goto e_inval;
228	}
229	*spec_dst = FIB_RES_PREFSRC(net, res);
230	fib_combine_itag(itag, &res);
231	dev_match = false;
232
233#ifdef CONFIG_IP_ROUTE_MULTIPATH
234	for (ret = 0; ret < res.fi->fib_nhs; ret++) {
235		struct fib_nh *nh = &res.fi->fib_nh[ret];
236
237		if (nh->nh_dev == dev) {
238			dev_match = true;
239			break;
240		}
241	}
242#else
243	if (FIB_RES_DEV(res) == dev)
244		dev_match = true;
245#endif
246	if (dev_match) {
247		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
248		return ret;
249	}
250	if (no_addr)
251		goto last_resort;
252	if (rpf == 1)
253		goto e_rpf;
254	fl4.flowi4_oif = dev->ifindex;
255
256	ret = 0;
257	if (fib_lookup(net, &fl4, &res) == 0) {
258		if (res.type == RTN_UNICAST) {
259			*spec_dst = FIB_RES_PREFSRC(net, res);
260			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
261		}
262	}
263	return ret;
264
265last_resort:
266	if (rpf)
267		goto e_rpf;
268	*spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
269	*itag = 0;
270	return 0;
271
272e_inval:
273	return -EINVAL;
274e_rpf:
275	return -EXDEV;
276}
277
278static inline __be32 sk_extract_addr(struct sockaddr *addr)
279{
280	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
281}
282
283static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
284{
285	struct nlattr *nla;
286
287	nla = (struct nlattr *) ((char *) mx + len);
288	nla->nla_type = type;
289	nla->nla_len = nla_attr_size(4);
290	*(u32 *) nla_data(nla) = value;
291
292	return len + nla_total_size(4);
293}
294
295static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
296				 struct fib_config *cfg)
297{
298	__be32 addr;
299	int plen;
300
301	memset(cfg, 0, sizeof(*cfg));
302	cfg->fc_nlinfo.nl_net = net;
303
304	if (rt->rt_dst.sa_family != AF_INET)
305		return -EAFNOSUPPORT;
306
307	/*
308	 * Check mask for validity:
309	 * a) it must be contiguous.
310	 * b) destination must have all host bits clear.
311	 * c) if application forgot to set correct family (AF_INET),
312	 *    reject request unless it is absolutely clear i.e.
313	 *    both family and mask are zero.
314	 */
315	plen = 32;
316	addr = sk_extract_addr(&rt->rt_dst);
317	if (!(rt->rt_flags & RTF_HOST)) {
318		__be32 mask = sk_extract_addr(&rt->rt_genmask);
319
320		if (rt->rt_genmask.sa_family != AF_INET) {
321			if (mask || rt->rt_genmask.sa_family)
322				return -EAFNOSUPPORT;
323		}
324
325		if (bad_mask(mask, addr))
326			return -EINVAL;
327
328		plen = inet_mask_len(mask);
329	}
330
331	cfg->fc_dst_len = plen;
332	cfg->fc_dst = addr;
333
334	if (cmd != SIOCDELRT) {
335		cfg->fc_nlflags = NLM_F_CREATE;
336		cfg->fc_protocol = RTPROT_BOOT;
337	}
338
339	if (rt->rt_metric)
340		cfg->fc_priority = rt->rt_metric - 1;
341
342	if (rt->rt_flags & RTF_REJECT) {
343		cfg->fc_scope = RT_SCOPE_HOST;
344		cfg->fc_type = RTN_UNREACHABLE;
345		return 0;
346	}
347
348	cfg->fc_scope = RT_SCOPE_NOWHERE;
349	cfg->fc_type = RTN_UNICAST;
350
351	if (rt->rt_dev) {
352		char *colon;
353		struct net_device *dev;
354		char devname[IFNAMSIZ];
355
356		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
357			return -EFAULT;
358
359		devname[IFNAMSIZ-1] = 0;
360		colon = strchr(devname, ':');
361		if (colon)
362			*colon = 0;
363		dev = __dev_get_by_name(net, devname);
364		if (!dev)
365			return -ENODEV;
366		cfg->fc_oif = dev->ifindex;
367		if (colon) {
368			struct in_ifaddr *ifa;
369			struct in_device *in_dev = __in_dev_get_rtnl(dev);
370			if (!in_dev)
371				return -ENODEV;
372			*colon = ':';
373			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
374				if (strcmp(ifa->ifa_label, devname) == 0)
375					break;
376			if (ifa == NULL)
377				return -ENODEV;
378			cfg->fc_prefsrc = ifa->ifa_local;
379		}
380	}
381
382	addr = sk_extract_addr(&rt->rt_gateway);
383	if (rt->rt_gateway.sa_family == AF_INET && addr) {
384		cfg->fc_gw = addr;
385		if (rt->rt_flags & RTF_GATEWAY &&
386		    inet_addr_type(net, addr) == RTN_UNICAST)
387			cfg->fc_scope = RT_SCOPE_UNIVERSE;
388	}
389
390	if (cmd == SIOCDELRT)
391		return 0;
392
393	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
394		return -EINVAL;
395
396	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
397		cfg->fc_scope = RT_SCOPE_LINK;
398
399	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
400		struct nlattr *mx;
401		int len = 0;
402
403		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
404		if (mx == NULL)
405			return -ENOMEM;
406
407		if (rt->rt_flags & RTF_MTU)
408			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
409
410		if (rt->rt_flags & RTF_WINDOW)
411			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
412
413		if (rt->rt_flags & RTF_IRTT)
414			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
415
416		cfg->fc_mx = mx;
417		cfg->fc_mx_len = len;
418	}
419
420	return 0;
421}
422
423/*
424 * Handle IP routing ioctl calls.
425 * These are used to manipulate the routing tables
426 */
427int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
428{
429	struct fib_config cfg;
430	struct rtentry rt;
431	int err;
432
433	switch (cmd) {
434	case SIOCADDRT:		/* Add a route */
435	case SIOCDELRT:		/* Delete a route */
436		if (!capable(CAP_NET_ADMIN))
437			return -EPERM;
438
439		if (copy_from_user(&rt, arg, sizeof(rt)))
440			return -EFAULT;
441
442		rtnl_lock();
443		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
444		if (err == 0) {
445			struct fib_table *tb;
446
447			if (cmd == SIOCDELRT) {
448				tb = fib_get_table(net, cfg.fc_table);
449				if (tb)
450					err = fib_table_delete(tb, &cfg);
451				else
452					err = -ESRCH;
453			} else {
454				tb = fib_new_table(net, cfg.fc_table);
455				if (tb)
456					err = fib_table_insert(tb, &cfg);
457				else
458					err = -ENOBUFS;
459			}
460
461			/* allocated by rtentry_to_fib_config() */
462			kfree(cfg.fc_mx);
463		}
464		rtnl_unlock();
465		return err;
466	}
467	return -EINVAL;
468}
469
470const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
471	[RTA_DST]		= { .type = NLA_U32 },
472	[RTA_SRC]		= { .type = NLA_U32 },
473	[RTA_IIF]		= { .type = NLA_U32 },
474	[RTA_OIF]		= { .type = NLA_U32 },
475	[RTA_GATEWAY]		= { .type = NLA_U32 },
476	[RTA_PRIORITY]		= { .type = NLA_U32 },
477	[RTA_PREFSRC]		= { .type = NLA_U32 },
478	[RTA_METRICS]		= { .type = NLA_NESTED },
479	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
480	[RTA_FLOW]		= { .type = NLA_U32 },
481};
482
483static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
484			     struct nlmsghdr *nlh, struct fib_config *cfg)
485{
486	struct nlattr *attr;
487	int err, remaining;
488	struct rtmsg *rtm;
489
490	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
491	if (err < 0)
492		goto errout;
493
494	memset(cfg, 0, sizeof(*cfg));
495
496	rtm = nlmsg_data(nlh);
497	cfg->fc_dst_len = rtm->rtm_dst_len;
498	cfg->fc_tos = rtm->rtm_tos;
499	cfg->fc_table = rtm->rtm_table;
500	cfg->fc_protocol = rtm->rtm_protocol;
501	cfg->fc_scope = rtm->rtm_scope;
502	cfg->fc_type = rtm->rtm_type;
503	cfg->fc_flags = rtm->rtm_flags;
504	cfg->fc_nlflags = nlh->nlmsg_flags;
505
506	cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
507	cfg->fc_nlinfo.nlh = nlh;
508	cfg->fc_nlinfo.nl_net = net;
509
510	if (cfg->fc_type > RTN_MAX) {
511		err = -EINVAL;
512		goto errout;
513	}
514
515	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
516		switch (nla_type(attr)) {
517		case RTA_DST:
518			cfg->fc_dst = nla_get_be32(attr);
519			break;
520		case RTA_OIF:
521			cfg->fc_oif = nla_get_u32(attr);
522			break;
523		case RTA_GATEWAY:
524			cfg->fc_gw = nla_get_be32(attr);
525			break;
526		case RTA_PRIORITY:
527			cfg->fc_priority = nla_get_u32(attr);
528			break;
529		case RTA_PREFSRC:
530			cfg->fc_prefsrc = nla_get_be32(attr);
531			break;
532		case RTA_METRICS:
533			cfg->fc_mx = nla_data(attr);
534			cfg->fc_mx_len = nla_len(attr);
535			break;
536		case RTA_MULTIPATH:
537			cfg->fc_mp = nla_data(attr);
538			cfg->fc_mp_len = nla_len(attr);
539			break;
540		case RTA_FLOW:
541			cfg->fc_flow = nla_get_u32(attr);
542			break;
543		case RTA_TABLE:
544			cfg->fc_table = nla_get_u32(attr);
545			break;
546		}
547	}
548
549	return 0;
550errout:
551	return err;
552}
553
554static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
555{
556	struct net *net = sock_net(skb->sk);
557	struct fib_config cfg;
558	struct fib_table *tb;
559	int err;
560
561	err = rtm_to_fib_config(net, skb, nlh, &cfg);
562	if (err < 0)
563		goto errout;
564
565	tb = fib_get_table(net, cfg.fc_table);
566	if (tb == NULL) {
567		err = -ESRCH;
568		goto errout;
569	}
570
571	err = fib_table_delete(tb, &cfg);
572errout:
573	return err;
574}
575
576static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
577{
578	struct net *net = sock_net(skb->sk);
579	struct fib_config cfg;
580	struct fib_table *tb;
581	int err;
582
583	err = rtm_to_fib_config(net, skb, nlh, &cfg);
584	if (err < 0)
585		goto errout;
586
587	tb = fib_new_table(net, cfg.fc_table);
588	if (tb == NULL) {
589		err = -ENOBUFS;
590		goto errout;
591	}
592
593	err = fib_table_insert(tb, &cfg);
594errout:
595	return err;
596}
597
598static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
599{
600	struct net *net = sock_net(skb->sk);
601	unsigned int h, s_h;
602	unsigned int e = 0, s_e;
603	struct fib_table *tb;
604	struct hlist_node *node;
605	struct hlist_head *head;
606	int dumped = 0;
607
608	if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
609	    ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
610		return ip_rt_dump(skb, cb);
611
612	s_h = cb->args[0];
613	s_e = cb->args[1];
614
615	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
616		e = 0;
617		head = &net->ipv4.fib_table_hash[h];
618		hlist_for_each_entry(tb, node, head, tb_hlist) {
619			if (e < s_e)
620				goto next;
621			if (dumped)
622				memset(&cb->args[2], 0, sizeof(cb->args) -
623						 2 * sizeof(cb->args[0]));
624			if (fib_table_dump(tb, skb, cb) < 0)
625				goto out;
626			dumped = 1;
627next:
628			e++;
629		}
630	}
631out:
632	cb->args[1] = e;
633	cb->args[0] = h;
634
635	return skb->len;
636}
637
638/* Prepare and feed intra-kernel routing request.
639 * Really, it should be netlink message, but :-( netlink
640 * can be not configured, so that we feed it directly
641 * to fib engine. It is legal, because all events occur
642 * only when netlink is already locked.
643 */
644static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
645{
646	struct net *net = dev_net(ifa->ifa_dev->dev);
647	struct fib_table *tb;
648	struct fib_config cfg = {
649		.fc_protocol = RTPROT_KERNEL,
650		.fc_type = type,
651		.fc_dst = dst,
652		.fc_dst_len = dst_len,
653		.fc_prefsrc = ifa->ifa_local,
654		.fc_oif = ifa->ifa_dev->dev->ifindex,
655		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
656		.fc_nlinfo = {
657			.nl_net = net,
658		},
659	};
660
661	if (type == RTN_UNICAST)
662		tb = fib_new_table(net, RT_TABLE_MAIN);
663	else
664		tb = fib_new_table(net, RT_TABLE_LOCAL);
665
666	if (tb == NULL)
667		return;
668
669	cfg.fc_table = tb->tb_id;
670
671	if (type != RTN_LOCAL)
672		cfg.fc_scope = RT_SCOPE_LINK;
673	else
674		cfg.fc_scope = RT_SCOPE_HOST;
675
676	if (cmd == RTM_NEWROUTE)
677		fib_table_insert(tb, &cfg);
678	else
679		fib_table_delete(tb, &cfg);
680}
681
682void fib_add_ifaddr(struct in_ifaddr *ifa)
683{
684	struct in_device *in_dev = ifa->ifa_dev;
685	struct net_device *dev = in_dev->dev;
686	struct in_ifaddr *prim = ifa;
687	__be32 mask = ifa->ifa_mask;
688	__be32 addr = ifa->ifa_local;
689	__be32 prefix = ifa->ifa_address & mask;
690
691	if (ifa->ifa_flags & IFA_F_SECONDARY) {
692		prim = inet_ifa_byprefix(in_dev, prefix, mask);
693		if (prim == NULL) {
694			printk(KERN_WARNING "fib_add_ifaddr: bug: prim == NULL\n");
695			return;
696		}
697	}
698
699	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
700
701	if (!(dev->flags & IFF_UP))
702		return;
703
704	/* Add broadcast address, if it is explicitly assigned. */
705	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
706		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
707
708	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
709	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
710		fib_magic(RTM_NEWROUTE,
711			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
712			  prefix, ifa->ifa_prefixlen, prim);
713
714		/* Add network specific broadcasts, when it takes a sense */
715		if (ifa->ifa_prefixlen < 31) {
716			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
717			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
718				  32, prim);
719		}
720	}
721}
722
723/* Delete primary or secondary address.
724 * Optionally, on secondary address promotion consider the addresses
725 * from subnet iprim as deleted, even if they are in device list.
726 * In this case the secondary ifa can be in device list.
727 */
728void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
729{
730	struct in_device *in_dev = ifa->ifa_dev;
731	struct net_device *dev = in_dev->dev;
732	struct in_ifaddr *ifa1;
733	struct in_ifaddr *prim = ifa, *prim1 = NULL;
734	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
735	__be32 any = ifa->ifa_address & ifa->ifa_mask;
736#define LOCAL_OK	1
737#define BRD_OK		2
738#define BRD0_OK		4
739#define BRD1_OK		8
740	unsigned ok = 0;
741	int subnet = 0;		/* Primary network */
742	int gone = 1;		/* Address is missing */
743	int same_prefsrc = 0;	/* Another primary with same IP */
744
745	if (ifa->ifa_flags & IFA_F_SECONDARY) {
746		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
747		if (prim == NULL) {
748			printk(KERN_WARNING "fib_del_ifaddr: bug: prim == NULL\n");
749			return;
750		}
751		if (iprim && iprim != prim) {
752			printk(KERN_WARNING "fib_del_ifaddr: bug: iprim != prim\n");
753			return;
754		}
755	} else if (!ipv4_is_zeronet(any) &&
756		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
757		fib_magic(RTM_DELROUTE,
758			  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
759			  any, ifa->ifa_prefixlen, prim);
760		subnet = 1;
761	}
762
763	/* Deletion is more complicated than add.
764	 * We should take care of not to delete too much :-)
765	 *
766	 * Scan address list to be sure that addresses are really gone.
767	 */
768
769	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
770		if (ifa1 == ifa) {
771			/* promotion, keep the IP */
772			gone = 0;
773			continue;
774		}
775		/* Ignore IFAs from our subnet */
776		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
777		    inet_ifa_match(ifa1->ifa_address, iprim))
778			continue;
779
780		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
781		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
782			/* Another address from our subnet? */
783			if (ifa1->ifa_mask == prim->ifa_mask &&
784			    inet_ifa_match(ifa1->ifa_address, prim))
785				prim1 = prim;
786			else {
787				/* We reached the secondaries, so
788				 * same_prefsrc should be determined.
789				 */
790				if (!same_prefsrc)
791					continue;
792				/* Search new prim1 if ifa1 is not
793				 * using the current prim1
794				 */
795				if (!prim1 ||
796				    ifa1->ifa_mask != prim1->ifa_mask ||
797				    !inet_ifa_match(ifa1->ifa_address, prim1))
798					prim1 = inet_ifa_byprefix(in_dev,
799							ifa1->ifa_address,
800							ifa1->ifa_mask);
801				if (!prim1)
802					continue;
803				if (prim1->ifa_local != prim->ifa_local)
804					continue;
805			}
806		} else {
807			if (prim->ifa_local != ifa1->ifa_local)
808				continue;
809			prim1 = ifa1;
810			if (prim != prim1)
811				same_prefsrc = 1;
812		}
813		if (ifa->ifa_local == ifa1->ifa_local)
814			ok |= LOCAL_OK;
815		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
816			ok |= BRD_OK;
817		if (brd == ifa1->ifa_broadcast)
818			ok |= BRD1_OK;
819		if (any == ifa1->ifa_broadcast)
820			ok |= BRD0_OK;
821		/* primary has network specific broadcasts */
822		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
823			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
824			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
825
826			if (!ipv4_is_zeronet(any1)) {
827				if (ifa->ifa_broadcast == brd1 ||
828				    ifa->ifa_broadcast == any1)
829					ok |= BRD_OK;
830				if (brd == brd1 || brd == any1)
831					ok |= BRD1_OK;
832				if (any == brd1 || any == any1)
833					ok |= BRD0_OK;
834			}
835		}
836	}
837
838	if (!(ok & BRD_OK))
839		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
840	if (subnet && ifa->ifa_prefixlen < 31) {
841		if (!(ok & BRD1_OK))
842			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
843		if (!(ok & BRD0_OK))
844			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
845	}
846	if (!(ok & LOCAL_OK)) {
847		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
848
849		/* Check, that this local address finally disappeared. */
850		if (gone &&
851		    inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
852			/* And the last, but not the least thing.
853			 * We must flush stray FIB entries.
854			 *
855			 * First of all, we scan fib_info list searching
856			 * for stray nexthop entries, then ignite fib_flush.
857			 */
858			if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
859				fib_flush(dev_net(dev));
860		}
861	}
862#undef LOCAL_OK
863#undef BRD_OK
864#undef BRD0_OK
865#undef BRD1_OK
866}
867
868static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
869{
870
871	struct fib_result       res;
872	struct flowi4           fl4 = {
873		.flowi4_mark = frn->fl_mark,
874		.daddr = frn->fl_addr,
875		.flowi4_tos = frn->fl_tos,
876		.flowi4_scope = frn->fl_scope,
877	};
878
879#ifdef CONFIG_IP_MULTIPLE_TABLES
880	res.r = NULL;
881#endif
882
883	frn->err = -ENOENT;
884	if (tb) {
885		local_bh_disable();
886
887		frn->tb_id = tb->tb_id;
888		rcu_read_lock();
889		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
890
891		if (!frn->err) {
892			frn->prefixlen = res.prefixlen;
893			frn->nh_sel = res.nh_sel;
894			frn->type = res.type;
895			frn->scope = res.scope;
896		}
897		rcu_read_unlock();
898		local_bh_enable();
899	}
900}
901
902static void nl_fib_input(struct sk_buff *skb)
903{
904	struct net *net;
905	struct fib_result_nl *frn;
906	struct nlmsghdr *nlh;
907	struct fib_table *tb;
908	u32 pid;
909
910	net = sock_net(skb->sk);
911	nlh = nlmsg_hdr(skb);
912	if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
913	    nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
914		return;
915
916	skb = skb_clone(skb, GFP_KERNEL);
917	if (skb == NULL)
918		return;
919	nlh = nlmsg_hdr(skb);
920
921	frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
922	tb = fib_get_table(net, frn->tb_id_in);
923
924	nl_fib_lookup(frn, tb);
925
926	pid = NETLINK_CB(skb).pid;      /* pid of sending process */
927	NETLINK_CB(skb).pid = 0;        /* from kernel */
928	NETLINK_CB(skb).dst_group = 0;  /* unicast */
929	netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
930}
931
932static int __net_init nl_fib_lookup_init(struct net *net)
933{
934	struct sock *sk;
935	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0,
936				   nl_fib_input, NULL, THIS_MODULE);
937	if (sk == NULL)
938		return -EAFNOSUPPORT;
939	net->ipv4.fibnl = sk;
940	return 0;
941}
942
943static void nl_fib_lookup_exit(struct net *net)
944{
945	netlink_kernel_release(net->ipv4.fibnl);
946	net->ipv4.fibnl = NULL;
947}
948
949static void fib_disable_ip(struct net_device *dev, int force, int delay)
950{
951	if (fib_sync_down_dev(dev, force))
952		fib_flush(dev_net(dev));
953	rt_cache_flush(dev_net(dev), delay);
954	arp_ifdown(dev);
955}
956
957static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
958{
959	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
960	struct net_device *dev = ifa->ifa_dev->dev;
961	struct net *net = dev_net(dev);
962
963	switch (event) {
964	case NETDEV_UP:
965		fib_add_ifaddr(ifa);
966#ifdef CONFIG_IP_ROUTE_MULTIPATH
967		fib_sync_up(dev);
968#endif
969		atomic_inc(&net->ipv4.dev_addr_genid);
970		rt_cache_flush(dev_net(dev), -1);
971		break;
972	case NETDEV_DOWN:
973		fib_del_ifaddr(ifa, NULL);
974		atomic_inc(&net->ipv4.dev_addr_genid);
975		if (ifa->ifa_dev->ifa_list == NULL) {
976			/* Last address was deleted from this interface.
977			 * Disable IP.
978			 */
979			fib_disable_ip(dev, 1, 0);
980		} else {
981			rt_cache_flush(dev_net(dev), -1);
982		}
983		break;
984	}
985	return NOTIFY_DONE;
986}
987
988static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
989{
990	struct net_device *dev = ptr;
991	struct in_device *in_dev = __in_dev_get_rtnl(dev);
992	struct net *net = dev_net(dev);
993
994	if (event == NETDEV_UNREGISTER) {
995		fib_disable_ip(dev, 2, -1);
996		return NOTIFY_DONE;
997	}
998
999	if (!in_dev)
1000		return NOTIFY_DONE;
1001
1002	switch (event) {
1003	case NETDEV_UP:
1004		for_ifa(in_dev) {
1005			fib_add_ifaddr(ifa);
1006		} endfor_ifa(in_dev);
1007#ifdef CONFIG_IP_ROUTE_MULTIPATH
1008		fib_sync_up(dev);
1009#endif
1010		atomic_inc(&net->ipv4.dev_addr_genid);
1011		rt_cache_flush(dev_net(dev), -1);
1012		break;
1013	case NETDEV_DOWN:
1014		fib_disable_ip(dev, 0, 0);
1015		break;
1016	case NETDEV_CHANGEMTU:
1017	case NETDEV_CHANGE:
1018		rt_cache_flush(dev_net(dev), 0);
1019		break;
1020	case NETDEV_UNREGISTER_BATCH:
1021		/* The batch unregister is only called on the first
1022		 * device in the list of devices being unregistered.
1023		 * Therefore we should not pass dev_net(dev) in here.
1024		 */
1025		rt_cache_flush_batch(NULL);
1026		break;
1027	}
1028	return NOTIFY_DONE;
1029}
1030
1031static struct notifier_block fib_inetaddr_notifier = {
1032	.notifier_call = fib_inetaddr_event,
1033};
1034
1035static struct notifier_block fib_netdev_notifier = {
1036	.notifier_call = fib_netdev_event,
1037};
1038
1039static int __net_init ip_fib_net_init(struct net *net)
1040{
1041	int err;
1042	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1043
1044	/* Avoid false sharing : Use at least a full cache line */
1045	size = max_t(size_t, size, L1_CACHE_BYTES);
1046
1047	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1048	if (net->ipv4.fib_table_hash == NULL)
1049		return -ENOMEM;
1050
1051	err = fib4_rules_init(net);
1052	if (err < 0)
1053		goto fail;
1054	return 0;
1055
1056fail:
1057	kfree(net->ipv4.fib_table_hash);
1058	return err;
1059}
1060
1061static void ip_fib_net_exit(struct net *net)
1062{
1063	unsigned int i;
1064
1065#ifdef CONFIG_IP_MULTIPLE_TABLES
1066	fib4_rules_exit(net);
1067#endif
1068
1069	rtnl_lock();
1070	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1071		struct fib_table *tb;
1072		struct hlist_head *head;
1073		struct hlist_node *node, *tmp;
1074
1075		head = &net->ipv4.fib_table_hash[i];
1076		hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1077			hlist_del(node);
1078			fib_table_flush(tb);
1079			fib_free_table(tb);
1080		}
1081	}
1082	rtnl_unlock();
1083	kfree(net->ipv4.fib_table_hash);
1084}
1085
1086static int __net_init fib_net_init(struct net *net)
1087{
1088	int error;
1089
1090	error = ip_fib_net_init(net);
1091	if (error < 0)
1092		goto out;
1093	error = nl_fib_lookup_init(net);
1094	if (error < 0)
1095		goto out_nlfl;
1096	error = fib_proc_init(net);
1097	if (error < 0)
1098		goto out_proc;
1099out:
1100	return error;
1101
1102out_proc:
1103	nl_fib_lookup_exit(net);
1104out_nlfl:
1105	ip_fib_net_exit(net);
1106	goto out;
1107}
1108
1109static void __net_exit fib_net_exit(struct net *net)
1110{
1111	fib_proc_exit(net);
1112	nl_fib_lookup_exit(net);
1113	ip_fib_net_exit(net);
1114}
1115
1116static struct pernet_operations fib_net_ops = {
1117	.init = fib_net_init,
1118	.exit = fib_net_exit,
1119};
1120
1121void __init ip_fib_init(void)
1122{
1123	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL);
1124	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL);
1125	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib);
1126
1127	register_pernet_subsys(&fib_net_ops);
1128	register_netdevice_notifier(&fib_netdev_notifier);
1129	register_inetaddr_notifier(&fib_inetaddr_notifier);
1130
1131	fib_trie_init();
1132}
1133