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