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#include <net/l3mdev.h>
49#include <trace/events/fib.h>
50
51#ifndef CONFIG_IP_MULTIPLE_TABLES
52
53static int __net_init fib4_rules_init(struct net *net)
54{
55	struct fib_table *local_table, *main_table;
56
57	main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
58	if (!main_table)
59		return -ENOMEM;
60
61	local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
62	if (!local_table)
63		goto fail;
64
65	hlist_add_head_rcu(&local_table->tb_hlist,
66				&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
67	hlist_add_head_rcu(&main_table->tb_hlist,
68				&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
69	return 0;
70
71fail:
72	fib_free_table(main_table);
73	return -ENOMEM;
74}
75#else
76
77struct fib_table *fib_new_table(struct net *net, u32 id)
78{
79	struct fib_table *tb, *alias = NULL;
80	unsigned int h;
81
82	if (id == 0)
83		id = RT_TABLE_MAIN;
84	tb = fib_get_table(net, id);
85	if (tb)
86		return tb;
87
88	if (id == RT_TABLE_LOCAL)
89		alias = fib_new_table(net, RT_TABLE_MAIN);
90
91	tb = fib_trie_table(id, alias);
92	if (!tb)
93		return NULL;
94
95	switch (id) {
96	case RT_TABLE_LOCAL:
97		rcu_assign_pointer(net->ipv4.fib_local, tb);
98		break;
99	case RT_TABLE_MAIN:
100		rcu_assign_pointer(net->ipv4.fib_main, tb);
101		break;
102	case RT_TABLE_DEFAULT:
103		rcu_assign_pointer(net->ipv4.fib_default, tb);
104		break;
105	default:
106		break;
107	}
108
109	h = id & (FIB_TABLE_HASHSZ - 1);
110	hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
111	return tb;
112}
113
114/* caller must hold either rtnl or rcu read lock */
115struct fib_table *fib_get_table(struct net *net, u32 id)
116{
117	struct fib_table *tb;
118	struct hlist_head *head;
119	unsigned int h;
120
121	if (id == 0)
122		id = RT_TABLE_MAIN;
123	h = id & (FIB_TABLE_HASHSZ - 1);
124
125	head = &net->ipv4.fib_table_hash[h];
126	hlist_for_each_entry_rcu(tb, head, tb_hlist) {
127		if (tb->tb_id == id)
128			return tb;
129	}
130	return NULL;
131}
132#endif /* CONFIG_IP_MULTIPLE_TABLES */
133
134static void fib_replace_table(struct net *net, struct fib_table *old,
135			      struct fib_table *new)
136{
137#ifdef CONFIG_IP_MULTIPLE_TABLES
138	switch (new->tb_id) {
139	case RT_TABLE_LOCAL:
140		rcu_assign_pointer(net->ipv4.fib_local, new);
141		break;
142	case RT_TABLE_MAIN:
143		rcu_assign_pointer(net->ipv4.fib_main, new);
144		break;
145	case RT_TABLE_DEFAULT:
146		rcu_assign_pointer(net->ipv4.fib_default, new);
147		break;
148	default:
149		break;
150	}
151
152#endif
153	/* replace the old table in the hlist */
154	hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
155}
156
157int fib_unmerge(struct net *net)
158{
159	struct fib_table *old, *new;
160
161	/* attempt to fetch local table if it has been allocated */
162	old = fib_get_table(net, RT_TABLE_LOCAL);
163	if (!old)
164		return 0;
165
166	new = fib_trie_unmerge(old);
167	if (!new)
168		return -ENOMEM;
169
170	/* replace merged table with clean table */
171	if (new != old) {
172		fib_replace_table(net, old, new);
173		fib_free_table(old);
174	}
175
176	return 0;
177}
178
179static void fib_flush(struct net *net)
180{
181	int flushed = 0;
182	unsigned int h;
183
184	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
185		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
186		struct hlist_node *tmp;
187		struct fib_table *tb;
188
189		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
190			flushed += fib_table_flush(tb);
191	}
192
193	if (flushed)
194		rt_cache_flush(net);
195}
196
197void fib_flush_external(struct net *net)
198{
199	struct fib_table *tb;
200	struct hlist_head *head;
201	unsigned int h;
202
203	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
204		head = &net->ipv4.fib_table_hash[h];
205		hlist_for_each_entry(tb, head, tb_hlist)
206			fib_table_flush_external(tb);
207	}
208}
209
210/*
211 * Find address type as if only "dev" was present in the system. If
212 * on_dev is NULL then all interfaces are taken into consideration.
213 */
214static inline unsigned int __inet_dev_addr_type(struct net *net,
215						const struct net_device *dev,
216						__be32 addr, u32 tb_id)
217{
218	struct flowi4		fl4 = { .daddr = addr };
219	struct fib_result	res;
220	unsigned int ret = RTN_BROADCAST;
221	struct fib_table *table;
222
223	if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
224		return RTN_BROADCAST;
225	if (ipv4_is_multicast(addr))
226		return RTN_MULTICAST;
227
228	rcu_read_lock();
229
230	table = fib_get_table(net, tb_id);
231	if (table) {
232		ret = RTN_UNICAST;
233		if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
234			if (!dev || dev == res.fi->fib_dev)
235				ret = res.type;
236		}
237	}
238
239	rcu_read_unlock();
240	return ret;
241}
242
243unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
244{
245	return __inet_dev_addr_type(net, NULL, addr, tb_id);
246}
247EXPORT_SYMBOL(inet_addr_type_table);
248
249unsigned int inet_addr_type(struct net *net, __be32 addr)
250{
251	return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
252}
253EXPORT_SYMBOL(inet_addr_type);
254
255unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
256				__be32 addr)
257{
258	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
259
260	return __inet_dev_addr_type(net, dev, addr, rt_table);
261}
262EXPORT_SYMBOL(inet_dev_addr_type);
263
264/* inet_addr_type with dev == NULL but using the table from a dev
265 * if one is associated
266 */
267unsigned int inet_addr_type_dev_table(struct net *net,
268				      const struct net_device *dev,
269				      __be32 addr)
270{
271	u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
272
273	return __inet_dev_addr_type(net, NULL, addr, rt_table);
274}
275EXPORT_SYMBOL(inet_addr_type_dev_table);
276
277__be32 fib_compute_spec_dst(struct sk_buff *skb)
278{
279	struct net_device *dev = skb->dev;
280	struct in_device *in_dev;
281	struct fib_result res;
282	struct rtable *rt;
283	struct net *net;
284	int scope;
285
286	rt = skb_rtable(skb);
287	if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
288	    RTCF_LOCAL)
289		return ip_hdr(skb)->daddr;
290
291	in_dev = __in_dev_get_rcu(dev);
292	BUG_ON(!in_dev);
293
294	net = dev_net(dev);
295
296	scope = RT_SCOPE_UNIVERSE;
297	if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
298		struct flowi4 fl4 = {
299			.flowi4_iif = LOOPBACK_IFINDEX,
300			.daddr = ip_hdr(skb)->saddr,
301			.flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
302			.flowi4_scope = scope,
303			.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0,
304		};
305		if (!fib_lookup(net, &fl4, &res, 0))
306			return FIB_RES_PREFSRC(net, res);
307	} else {
308		scope = RT_SCOPE_LINK;
309	}
310
311	return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
312}
313
314/* Given (packet source, input interface) and optional (dst, oif, tos):
315 * - (main) check, that source is valid i.e. not broadcast or our local
316 *   address.
317 * - figure out what "logical" interface this packet arrived
318 *   and calculate "specific destination" address.
319 * - check, that packet arrived from expected physical interface.
320 * called with rcu_read_lock()
321 */
322static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
323				 u8 tos, int oif, struct net_device *dev,
324				 int rpf, struct in_device *idev, u32 *itag)
325{
326	int ret, no_addr;
327	struct fib_result res;
328	struct flowi4 fl4;
329	struct net *net;
330	bool dev_match;
331
332	fl4.flowi4_oif = 0;
333	fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
334	if (!fl4.flowi4_iif)
335		fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
336	fl4.daddr = src;
337	fl4.saddr = dst;
338	fl4.flowi4_tos = tos;
339	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
340	fl4.flowi4_tun_key.tun_id = 0;
341	fl4.flowi4_flags = 0;
342
343	no_addr = idev->ifa_list == NULL;
344
345	fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
346
347	trace_fib_validate_source(dev, &fl4);
348
349	net = dev_net(dev);
350	if (fib_lookup(net, &fl4, &res, 0))
351		goto last_resort;
352	if (res.type != RTN_UNICAST &&
353	    (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
354		goto e_inval;
355	if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
356	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
357		goto last_resort;
358	fib_combine_itag(itag, &res);
359	dev_match = false;
360
361#ifdef CONFIG_IP_ROUTE_MULTIPATH
362	for (ret = 0; ret < res.fi->fib_nhs; ret++) {
363		struct fib_nh *nh = &res.fi->fib_nh[ret];
364
365		if (nh->nh_dev == dev) {
366			dev_match = true;
367			break;
368		} else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
369			dev_match = true;
370			break;
371		}
372	}
373#else
374	if (FIB_RES_DEV(res) == dev)
375		dev_match = true;
376#endif
377	if (dev_match) {
378		ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
379		return ret;
380	}
381	if (no_addr)
382		goto last_resort;
383	if (rpf == 1)
384		goto e_rpf;
385	fl4.flowi4_oif = dev->ifindex;
386
387	ret = 0;
388	if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
389		if (res.type == RTN_UNICAST)
390			ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
391	}
392	return ret;
393
394last_resort:
395	if (rpf)
396		goto e_rpf;
397	*itag = 0;
398	return 0;
399
400e_inval:
401	return -EINVAL;
402e_rpf:
403	return -EXDEV;
404}
405
406/* Ignore rp_filter for packets protected by IPsec. */
407int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
408			u8 tos, int oif, struct net_device *dev,
409			struct in_device *idev, u32 *itag)
410{
411	int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
412
413	if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
414	    IN_DEV_ACCEPT_LOCAL(idev) &&
415	    (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
416		*itag = 0;
417		return 0;
418	}
419	return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
420}
421
422static inline __be32 sk_extract_addr(struct sockaddr *addr)
423{
424	return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
425}
426
427static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
428{
429	struct nlattr *nla;
430
431	nla = (struct nlattr *) ((char *) mx + len);
432	nla->nla_type = type;
433	nla->nla_len = nla_attr_size(4);
434	*(u32 *) nla_data(nla) = value;
435
436	return len + nla_total_size(4);
437}
438
439static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
440				 struct fib_config *cfg)
441{
442	__be32 addr;
443	int plen;
444
445	memset(cfg, 0, sizeof(*cfg));
446	cfg->fc_nlinfo.nl_net = net;
447
448	if (rt->rt_dst.sa_family != AF_INET)
449		return -EAFNOSUPPORT;
450
451	/*
452	 * Check mask for validity:
453	 * a) it must be contiguous.
454	 * b) destination must have all host bits clear.
455	 * c) if application forgot to set correct family (AF_INET),
456	 *    reject request unless it is absolutely clear i.e.
457	 *    both family and mask are zero.
458	 */
459	plen = 32;
460	addr = sk_extract_addr(&rt->rt_dst);
461	if (!(rt->rt_flags & RTF_HOST)) {
462		__be32 mask = sk_extract_addr(&rt->rt_genmask);
463
464		if (rt->rt_genmask.sa_family != AF_INET) {
465			if (mask || rt->rt_genmask.sa_family)
466				return -EAFNOSUPPORT;
467		}
468
469		if (bad_mask(mask, addr))
470			return -EINVAL;
471
472		plen = inet_mask_len(mask);
473	}
474
475	cfg->fc_dst_len = plen;
476	cfg->fc_dst = addr;
477
478	if (cmd != SIOCDELRT) {
479		cfg->fc_nlflags = NLM_F_CREATE;
480		cfg->fc_protocol = RTPROT_BOOT;
481	}
482
483	if (rt->rt_metric)
484		cfg->fc_priority = rt->rt_metric - 1;
485
486	if (rt->rt_flags & RTF_REJECT) {
487		cfg->fc_scope = RT_SCOPE_HOST;
488		cfg->fc_type = RTN_UNREACHABLE;
489		return 0;
490	}
491
492	cfg->fc_scope = RT_SCOPE_NOWHERE;
493	cfg->fc_type = RTN_UNICAST;
494
495	if (rt->rt_dev) {
496		char *colon;
497		struct net_device *dev;
498		char devname[IFNAMSIZ];
499
500		if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
501			return -EFAULT;
502
503		devname[IFNAMSIZ-1] = 0;
504		colon = strchr(devname, ':');
505		if (colon)
506			*colon = 0;
507		dev = __dev_get_by_name(net, devname);
508		if (!dev)
509			return -ENODEV;
510		cfg->fc_oif = dev->ifindex;
511		if (colon) {
512			struct in_ifaddr *ifa;
513			struct in_device *in_dev = __in_dev_get_rtnl(dev);
514			if (!in_dev)
515				return -ENODEV;
516			*colon = ':';
517			for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
518				if (strcmp(ifa->ifa_label, devname) == 0)
519					break;
520			if (!ifa)
521				return -ENODEV;
522			cfg->fc_prefsrc = ifa->ifa_local;
523		}
524	}
525
526	addr = sk_extract_addr(&rt->rt_gateway);
527	if (rt->rt_gateway.sa_family == AF_INET && addr) {
528		unsigned int addr_type;
529
530		cfg->fc_gw = addr;
531		addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
532		if (rt->rt_flags & RTF_GATEWAY &&
533		    addr_type == RTN_UNICAST)
534			cfg->fc_scope = RT_SCOPE_UNIVERSE;
535	}
536
537	if (cmd == SIOCDELRT)
538		return 0;
539
540	if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
541		return -EINVAL;
542
543	if (cfg->fc_scope == RT_SCOPE_NOWHERE)
544		cfg->fc_scope = RT_SCOPE_LINK;
545
546	if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
547		struct nlattr *mx;
548		int len = 0;
549
550		mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
551		if (!mx)
552			return -ENOMEM;
553
554		if (rt->rt_flags & RTF_MTU)
555			len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
556
557		if (rt->rt_flags & RTF_WINDOW)
558			len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
559
560		if (rt->rt_flags & RTF_IRTT)
561			len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
562
563		cfg->fc_mx = mx;
564		cfg->fc_mx_len = len;
565	}
566
567	return 0;
568}
569
570/*
571 * Handle IP routing ioctl calls.
572 * These are used to manipulate the routing tables
573 */
574int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
575{
576	struct fib_config cfg;
577	struct rtentry rt;
578	int err;
579
580	switch (cmd) {
581	case SIOCADDRT:		/* Add a route */
582	case SIOCDELRT:		/* Delete a route */
583		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
584			return -EPERM;
585
586		if (copy_from_user(&rt, arg, sizeof(rt)))
587			return -EFAULT;
588
589		rtnl_lock();
590		err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
591		if (err == 0) {
592			struct fib_table *tb;
593
594			if (cmd == SIOCDELRT) {
595				tb = fib_get_table(net, cfg.fc_table);
596				if (tb)
597					err = fib_table_delete(tb, &cfg);
598				else
599					err = -ESRCH;
600			} else {
601				tb = fib_new_table(net, cfg.fc_table);
602				if (tb)
603					err = fib_table_insert(tb, &cfg);
604				else
605					err = -ENOBUFS;
606			}
607
608			/* allocated by rtentry_to_fib_config() */
609			kfree(cfg.fc_mx);
610		}
611		rtnl_unlock();
612		return err;
613	}
614	return -EINVAL;
615}
616
617const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
618	[RTA_DST]		= { .type = NLA_U32 },
619	[RTA_SRC]		= { .type = NLA_U32 },
620	[RTA_IIF]		= { .type = NLA_U32 },
621	[RTA_OIF]		= { .type = NLA_U32 },
622	[RTA_GATEWAY]		= { .type = NLA_U32 },
623	[RTA_PRIORITY]		= { .type = NLA_U32 },
624	[RTA_PREFSRC]		= { .type = NLA_U32 },
625	[RTA_METRICS]		= { .type = NLA_NESTED },
626	[RTA_MULTIPATH]		= { .len = sizeof(struct rtnexthop) },
627	[RTA_FLOW]		= { .type = NLA_U32 },
628	[RTA_ENCAP_TYPE]	= { .type = NLA_U16 },
629	[RTA_ENCAP]		= { .type = NLA_NESTED },
630};
631
632static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
633			     struct nlmsghdr *nlh, struct fib_config *cfg)
634{
635	struct nlattr *attr;
636	int err, remaining;
637	struct rtmsg *rtm;
638
639	err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
640	if (err < 0)
641		goto errout;
642
643	memset(cfg, 0, sizeof(*cfg));
644
645	rtm = nlmsg_data(nlh);
646	cfg->fc_dst_len = rtm->rtm_dst_len;
647	cfg->fc_tos = rtm->rtm_tos;
648	cfg->fc_table = rtm->rtm_table;
649	cfg->fc_protocol = rtm->rtm_protocol;
650	cfg->fc_scope = rtm->rtm_scope;
651	cfg->fc_type = rtm->rtm_type;
652	cfg->fc_flags = rtm->rtm_flags;
653	cfg->fc_nlflags = nlh->nlmsg_flags;
654
655	cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
656	cfg->fc_nlinfo.nlh = nlh;
657	cfg->fc_nlinfo.nl_net = net;
658
659	if (cfg->fc_type > RTN_MAX) {
660		err = -EINVAL;
661		goto errout;
662	}
663
664	nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
665		switch (nla_type(attr)) {
666		case RTA_DST:
667			cfg->fc_dst = nla_get_be32(attr);
668			break;
669		case RTA_OIF:
670			cfg->fc_oif = nla_get_u32(attr);
671			break;
672		case RTA_GATEWAY:
673			cfg->fc_gw = nla_get_be32(attr);
674			break;
675		case RTA_PRIORITY:
676			cfg->fc_priority = nla_get_u32(attr);
677			break;
678		case RTA_PREFSRC:
679			cfg->fc_prefsrc = nla_get_be32(attr);
680			break;
681		case RTA_METRICS:
682			cfg->fc_mx = nla_data(attr);
683			cfg->fc_mx_len = nla_len(attr);
684			break;
685		case RTA_MULTIPATH:
686			cfg->fc_mp = nla_data(attr);
687			cfg->fc_mp_len = nla_len(attr);
688			break;
689		case RTA_FLOW:
690			cfg->fc_flow = nla_get_u32(attr);
691			break;
692		case RTA_TABLE:
693			cfg->fc_table = nla_get_u32(attr);
694			break;
695		case RTA_ENCAP:
696			cfg->fc_encap = attr;
697			break;
698		case RTA_ENCAP_TYPE:
699			cfg->fc_encap_type = nla_get_u16(attr);
700			break;
701		}
702	}
703
704	return 0;
705errout:
706	return err;
707}
708
709static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
710{
711	struct net *net = sock_net(skb->sk);
712	struct fib_config cfg;
713	struct fib_table *tb;
714	int err;
715
716	err = rtm_to_fib_config(net, skb, nlh, &cfg);
717	if (err < 0)
718		goto errout;
719
720	tb = fib_get_table(net, cfg.fc_table);
721	if (!tb) {
722		err = -ESRCH;
723		goto errout;
724	}
725
726	err = fib_table_delete(tb, &cfg);
727errout:
728	return err;
729}
730
731static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
732{
733	struct net *net = sock_net(skb->sk);
734	struct fib_config cfg;
735	struct fib_table *tb;
736	int err;
737
738	err = rtm_to_fib_config(net, skb, nlh, &cfg);
739	if (err < 0)
740		goto errout;
741
742	tb = fib_new_table(net, cfg.fc_table);
743	if (!tb) {
744		err = -ENOBUFS;
745		goto errout;
746	}
747
748	err = fib_table_insert(tb, &cfg);
749errout:
750	return err;
751}
752
753static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
754{
755	struct net *net = sock_net(skb->sk);
756	unsigned int h, s_h;
757	unsigned int e = 0, s_e;
758	struct fib_table *tb;
759	struct hlist_head *head;
760	int dumped = 0;
761
762	if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
763	    ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
764		return skb->len;
765
766	s_h = cb->args[0];
767	s_e = cb->args[1];
768
769	rcu_read_lock();
770
771	for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
772		e = 0;
773		head = &net->ipv4.fib_table_hash[h];
774		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
775			if (e < s_e)
776				goto next;
777			if (dumped)
778				memset(&cb->args[2], 0, sizeof(cb->args) -
779						 2 * sizeof(cb->args[0]));
780			if (fib_table_dump(tb, skb, cb) < 0)
781				goto out;
782			dumped = 1;
783next:
784			e++;
785		}
786	}
787out:
788	rcu_read_unlock();
789
790	cb->args[1] = e;
791	cb->args[0] = h;
792
793	return skb->len;
794}
795
796/* Prepare and feed intra-kernel routing request.
797 * Really, it should be netlink message, but :-( netlink
798 * can be not configured, so that we feed it directly
799 * to fib engine. It is legal, because all events occur
800 * only when netlink is already locked.
801 */
802static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
803{
804	struct net *net = dev_net(ifa->ifa_dev->dev);
805	u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
806	struct fib_table *tb;
807	struct fib_config cfg = {
808		.fc_protocol = RTPROT_KERNEL,
809		.fc_type = type,
810		.fc_dst = dst,
811		.fc_dst_len = dst_len,
812		.fc_prefsrc = ifa->ifa_local,
813		.fc_oif = ifa->ifa_dev->dev->ifindex,
814		.fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
815		.fc_nlinfo = {
816			.nl_net = net,
817		},
818	};
819
820	if (!tb_id)
821		tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
822
823	tb = fib_new_table(net, tb_id);
824	if (!tb)
825		return;
826
827	cfg.fc_table = tb->tb_id;
828
829	if (type != RTN_LOCAL)
830		cfg.fc_scope = RT_SCOPE_LINK;
831	else
832		cfg.fc_scope = RT_SCOPE_HOST;
833
834	if (cmd == RTM_NEWROUTE)
835		fib_table_insert(tb, &cfg);
836	else
837		fib_table_delete(tb, &cfg);
838}
839
840void fib_add_ifaddr(struct in_ifaddr *ifa)
841{
842	struct in_device *in_dev = ifa->ifa_dev;
843	struct net_device *dev = in_dev->dev;
844	struct in_ifaddr *prim = ifa;
845	__be32 mask = ifa->ifa_mask;
846	__be32 addr = ifa->ifa_local;
847	__be32 prefix = ifa->ifa_address & mask;
848
849	if (ifa->ifa_flags & IFA_F_SECONDARY) {
850		prim = inet_ifa_byprefix(in_dev, prefix, mask);
851		if (!prim) {
852			pr_warn("%s: bug: prim == NULL\n", __func__);
853			return;
854		}
855	}
856
857	fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
858
859	if (!(dev->flags & IFF_UP))
860		return;
861
862	/* Add broadcast address, if it is explicitly assigned. */
863	if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
864		fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
865
866	if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
867	    (prefix != addr || ifa->ifa_prefixlen < 32)) {
868		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
869			fib_magic(RTM_NEWROUTE,
870				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
871				  prefix, ifa->ifa_prefixlen, prim);
872
873		/* Add network specific broadcasts, when it takes a sense */
874		if (ifa->ifa_prefixlen < 31) {
875			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
876			fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
877				  32, prim);
878		}
879	}
880}
881
882/* Delete primary or secondary address.
883 * Optionally, on secondary address promotion consider the addresses
884 * from subnet iprim as deleted, even if they are in device list.
885 * In this case the secondary ifa can be in device list.
886 */
887void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
888{
889	struct in_device *in_dev = ifa->ifa_dev;
890	struct net_device *dev = in_dev->dev;
891	struct in_ifaddr *ifa1;
892	struct in_ifaddr *prim = ifa, *prim1 = NULL;
893	__be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
894	__be32 any = ifa->ifa_address & ifa->ifa_mask;
895#define LOCAL_OK	1
896#define BRD_OK		2
897#define BRD0_OK		4
898#define BRD1_OK		8
899	unsigned int ok = 0;
900	int subnet = 0;		/* Primary network */
901	int gone = 1;		/* Address is missing */
902	int same_prefsrc = 0;	/* Another primary with same IP */
903
904	if (ifa->ifa_flags & IFA_F_SECONDARY) {
905		prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
906		if (!prim) {
907			/* if the device has been deleted, we don't perform
908			 * address promotion
909			 */
910			if (!in_dev->dead)
911				pr_warn("%s: bug: prim == NULL\n", __func__);
912			return;
913		}
914		if (iprim && iprim != prim) {
915			pr_warn("%s: bug: iprim != prim\n", __func__);
916			return;
917		}
918	} else if (!ipv4_is_zeronet(any) &&
919		   (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
920		if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
921			fib_magic(RTM_DELROUTE,
922				  dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
923				  any, ifa->ifa_prefixlen, prim);
924		subnet = 1;
925	}
926
927	if (in_dev->dead)
928		goto no_promotions;
929
930	/* Deletion is more complicated than add.
931	 * We should take care of not to delete too much :-)
932	 *
933	 * Scan address list to be sure that addresses are really gone.
934	 */
935
936	for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
937		if (ifa1 == ifa) {
938			/* promotion, keep the IP */
939			gone = 0;
940			continue;
941		}
942		/* Ignore IFAs from our subnet */
943		if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
944		    inet_ifa_match(ifa1->ifa_address, iprim))
945			continue;
946
947		/* Ignore ifa1 if it uses different primary IP (prefsrc) */
948		if (ifa1->ifa_flags & IFA_F_SECONDARY) {
949			/* Another address from our subnet? */
950			if (ifa1->ifa_mask == prim->ifa_mask &&
951			    inet_ifa_match(ifa1->ifa_address, prim))
952				prim1 = prim;
953			else {
954				/* We reached the secondaries, so
955				 * same_prefsrc should be determined.
956				 */
957				if (!same_prefsrc)
958					continue;
959				/* Search new prim1 if ifa1 is not
960				 * using the current prim1
961				 */
962				if (!prim1 ||
963				    ifa1->ifa_mask != prim1->ifa_mask ||
964				    !inet_ifa_match(ifa1->ifa_address, prim1))
965					prim1 = inet_ifa_byprefix(in_dev,
966							ifa1->ifa_address,
967							ifa1->ifa_mask);
968				if (!prim1)
969					continue;
970				if (prim1->ifa_local != prim->ifa_local)
971					continue;
972			}
973		} else {
974			if (prim->ifa_local != ifa1->ifa_local)
975				continue;
976			prim1 = ifa1;
977			if (prim != prim1)
978				same_prefsrc = 1;
979		}
980		if (ifa->ifa_local == ifa1->ifa_local)
981			ok |= LOCAL_OK;
982		if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
983			ok |= BRD_OK;
984		if (brd == ifa1->ifa_broadcast)
985			ok |= BRD1_OK;
986		if (any == ifa1->ifa_broadcast)
987			ok |= BRD0_OK;
988		/* primary has network specific broadcasts */
989		if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
990			__be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
991			__be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
992
993			if (!ipv4_is_zeronet(any1)) {
994				if (ifa->ifa_broadcast == brd1 ||
995				    ifa->ifa_broadcast == any1)
996					ok |= BRD_OK;
997				if (brd == brd1 || brd == any1)
998					ok |= BRD1_OK;
999				if (any == brd1 || any == any1)
1000					ok |= BRD0_OK;
1001			}
1002		}
1003	}
1004
1005no_promotions:
1006	if (!(ok & BRD_OK))
1007		fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1008	if (subnet && ifa->ifa_prefixlen < 31) {
1009		if (!(ok & BRD1_OK))
1010			fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1011		if (!(ok & BRD0_OK))
1012			fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1013	}
1014	if (!(ok & LOCAL_OK)) {
1015		unsigned int addr_type;
1016
1017		fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1018
1019		/* Check, that this local address finally disappeared. */
1020		addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1021						     ifa->ifa_local);
1022		if (gone && addr_type != RTN_LOCAL) {
1023			/* And the last, but not the least thing.
1024			 * We must flush stray FIB entries.
1025			 *
1026			 * First of all, we scan fib_info list searching
1027			 * for stray nexthop entries, then ignite fib_flush.
1028			 */
1029			if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
1030				fib_flush(dev_net(dev));
1031		}
1032	}
1033#undef LOCAL_OK
1034#undef BRD_OK
1035#undef BRD0_OK
1036#undef BRD1_OK
1037}
1038
1039static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1040{
1041
1042	struct fib_result       res;
1043	struct flowi4           fl4 = {
1044		.flowi4_mark = frn->fl_mark,
1045		.daddr = frn->fl_addr,
1046		.flowi4_tos = frn->fl_tos,
1047		.flowi4_scope = frn->fl_scope,
1048	};
1049	struct fib_table *tb;
1050
1051	rcu_read_lock();
1052
1053	tb = fib_get_table(net, frn->tb_id_in);
1054
1055	frn->err = -ENOENT;
1056	if (tb) {
1057		local_bh_disable();
1058
1059		frn->tb_id = tb->tb_id;
1060		frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1061
1062		if (!frn->err) {
1063			frn->prefixlen = res.prefixlen;
1064			frn->nh_sel = res.nh_sel;
1065			frn->type = res.type;
1066			frn->scope = res.scope;
1067		}
1068		local_bh_enable();
1069	}
1070
1071	rcu_read_unlock();
1072}
1073
1074static void nl_fib_input(struct sk_buff *skb)
1075{
1076	struct net *net;
1077	struct fib_result_nl *frn;
1078	struct nlmsghdr *nlh;
1079	u32 portid;
1080
1081	net = sock_net(skb->sk);
1082	nlh = nlmsg_hdr(skb);
1083	if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len ||
1084	    nlmsg_len(nlh) < sizeof(*frn))
1085		return;
1086
1087	skb = netlink_skb_clone(skb, GFP_KERNEL);
1088	if (!skb)
1089		return;
1090	nlh = nlmsg_hdr(skb);
1091
1092	frn = (struct fib_result_nl *) nlmsg_data(nlh);
1093	nl_fib_lookup(net, frn);
1094
1095	portid = NETLINK_CB(skb).portid;      /* netlink portid */
1096	NETLINK_CB(skb).portid = 0;        /* from kernel */
1097	NETLINK_CB(skb).dst_group = 0;  /* unicast */
1098	netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1099}
1100
1101static int __net_init nl_fib_lookup_init(struct net *net)
1102{
1103	struct sock *sk;
1104	struct netlink_kernel_cfg cfg = {
1105		.input	= nl_fib_input,
1106	};
1107
1108	sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1109	if (!sk)
1110		return -EAFNOSUPPORT;
1111	net->ipv4.fibnl = sk;
1112	return 0;
1113}
1114
1115static void nl_fib_lookup_exit(struct net *net)
1116{
1117	netlink_kernel_release(net->ipv4.fibnl);
1118	net->ipv4.fibnl = NULL;
1119}
1120
1121static void fib_disable_ip(struct net_device *dev, unsigned long event,
1122			   bool force)
1123{
1124	if (fib_sync_down_dev(dev, event, force))
1125		fib_flush(dev_net(dev));
1126	rt_cache_flush(dev_net(dev));
1127	arp_ifdown(dev);
1128}
1129
1130static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1131{
1132	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1133	struct net_device *dev = ifa->ifa_dev->dev;
1134	struct net *net = dev_net(dev);
1135
1136	switch (event) {
1137	case NETDEV_UP:
1138		fib_add_ifaddr(ifa);
1139#ifdef CONFIG_IP_ROUTE_MULTIPATH
1140		fib_sync_up(dev, RTNH_F_DEAD);
1141#endif
1142		atomic_inc(&net->ipv4.dev_addr_genid);
1143		rt_cache_flush(dev_net(dev));
1144		break;
1145	case NETDEV_DOWN:
1146		fib_del_ifaddr(ifa, NULL);
1147		atomic_inc(&net->ipv4.dev_addr_genid);
1148		if (!ifa->ifa_dev->ifa_list) {
1149			/* Last address was deleted from this interface.
1150			 * Disable IP.
1151			 */
1152			fib_disable_ip(dev, event, true);
1153		} else {
1154			rt_cache_flush(dev_net(dev));
1155		}
1156		break;
1157	}
1158	return NOTIFY_DONE;
1159}
1160
1161static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1162{
1163	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1164	struct netdev_notifier_changeupper_info *info;
1165	struct in_device *in_dev;
1166	struct net *net = dev_net(dev);
1167	unsigned int flags;
1168
1169	if (event == NETDEV_UNREGISTER) {
1170		fib_disable_ip(dev, event, true);
1171		rt_flush_dev(dev);
1172		return NOTIFY_DONE;
1173	}
1174
1175	in_dev = __in_dev_get_rtnl(dev);
1176	if (!in_dev)
1177		return NOTIFY_DONE;
1178
1179	switch (event) {
1180	case NETDEV_UP:
1181		for_ifa(in_dev) {
1182			fib_add_ifaddr(ifa);
1183		} endfor_ifa(in_dev);
1184#ifdef CONFIG_IP_ROUTE_MULTIPATH
1185		fib_sync_up(dev, RTNH_F_DEAD);
1186#endif
1187		atomic_inc(&net->ipv4.dev_addr_genid);
1188		rt_cache_flush(net);
1189		break;
1190	case NETDEV_DOWN:
1191		fib_disable_ip(dev, event, false);
1192		break;
1193	case NETDEV_CHANGE:
1194		flags = dev_get_flags(dev);
1195		if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1196			fib_sync_up(dev, RTNH_F_LINKDOWN);
1197		else
1198			fib_sync_down_dev(dev, event, false);
1199		/* fall through */
1200	case NETDEV_CHANGEMTU:
1201		rt_cache_flush(net);
1202		break;
1203	case NETDEV_CHANGEUPPER:
1204		info = ptr;
1205		/* flush all routes if dev is linked to or unlinked from
1206		 * an L3 master device (e.g., VRF)
1207		 */
1208		if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1209			fib_disable_ip(dev, NETDEV_DOWN, true);
1210		break;
1211	}
1212	return NOTIFY_DONE;
1213}
1214
1215static struct notifier_block fib_inetaddr_notifier = {
1216	.notifier_call = fib_inetaddr_event,
1217};
1218
1219static struct notifier_block fib_netdev_notifier = {
1220	.notifier_call = fib_netdev_event,
1221};
1222
1223static int __net_init ip_fib_net_init(struct net *net)
1224{
1225	int err;
1226	size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1227
1228	/* Avoid false sharing : Use at least a full cache line */
1229	size = max_t(size_t, size, L1_CACHE_BYTES);
1230
1231	net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1232	if (!net->ipv4.fib_table_hash)
1233		return -ENOMEM;
1234
1235	err = fib4_rules_init(net);
1236	if (err < 0)
1237		goto fail;
1238	return 0;
1239
1240fail:
1241	kfree(net->ipv4.fib_table_hash);
1242	return err;
1243}
1244
1245static void ip_fib_net_exit(struct net *net)
1246{
1247	unsigned int i;
1248
1249	rtnl_lock();
1250#ifdef CONFIG_IP_MULTIPLE_TABLES
1251	RCU_INIT_POINTER(net->ipv4.fib_local, NULL);
1252	RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1253	RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1254#endif
1255	for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1256		struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1257		struct hlist_node *tmp;
1258		struct fib_table *tb;
1259
1260		hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1261			hlist_del(&tb->tb_hlist);
1262			fib_table_flush(tb);
1263			fib_free_table(tb);
1264		}
1265	}
1266
1267#ifdef CONFIG_IP_MULTIPLE_TABLES
1268	fib4_rules_exit(net);
1269#endif
1270	rtnl_unlock();
1271	kfree(net->ipv4.fib_table_hash);
1272}
1273
1274static int __net_init fib_net_init(struct net *net)
1275{
1276	int error;
1277
1278#ifdef CONFIG_IP_ROUTE_CLASSID
1279	net->ipv4.fib_num_tclassid_users = 0;
1280#endif
1281	error = ip_fib_net_init(net);
1282	if (error < 0)
1283		goto out;
1284	error = nl_fib_lookup_init(net);
1285	if (error < 0)
1286		goto out_nlfl;
1287	error = fib_proc_init(net);
1288	if (error < 0)
1289		goto out_proc;
1290out:
1291	return error;
1292
1293out_proc:
1294	nl_fib_lookup_exit(net);
1295out_nlfl:
1296	ip_fib_net_exit(net);
1297	goto out;
1298}
1299
1300static void __net_exit fib_net_exit(struct net *net)
1301{
1302	fib_proc_exit(net);
1303	nl_fib_lookup_exit(net);
1304	ip_fib_net_exit(net);
1305}
1306
1307static struct pernet_operations fib_net_ops = {
1308	.init = fib_net_init,
1309	.exit = fib_net_exit,
1310};
1311
1312void __init ip_fib_init(void)
1313{
1314	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1315	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1316	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1317
1318	register_pernet_subsys(&fib_net_ops);
1319	register_netdevice_notifier(&fib_netdev_notifier);
1320	register_inetaddr_notifier(&fib_inetaddr_notifier);
1321
1322	fib_trie_init();
1323}
1324