xref: /linux/net/ipv4/ipmr.c (revision c6bd5bcc4983f1a2d2f87a3769bf309482ee8c04)
1 /*
2  *	IP multicast routing support for mrouted 3.6/3.8
3  *
4  *		(c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
5  *	  Linux Consultancy and Custom Driver Development
6  *
7  *	This program is free software; you can redistribute it and/or
8  *	modify it under the terms of the GNU General Public License
9  *	as published by the Free Software Foundation; either version
10  *	2 of the License, or (at your option) any later version.
11  *
12  *	Fixes:
13  *	Michael Chastain	:	Incorrect size of copying.
14  *	Alan Cox		:	Added the cache manager code
15  *	Alan Cox		:	Fixed the clone/copy bug and device race.
16  *	Mike McLagan		:	Routing by source
17  *	Malcolm Beattie		:	Buffer handling fixes.
18  *	Alexey Kuznetsov	:	Double buffer free and other fixes.
19  *	SVR Anand		:	Fixed several multicast bugs and problems.
20  *	Alexey Kuznetsov	:	Status, optimisations and more.
21  *	Brad Parker		:	Better behaviour on mrouted upcall
22  *					overflow.
23  *      Carlos Picoto           :       PIMv1 Support
24  *	Pavlin Ivanov Radoslavov:	PIMv2 Registers must checksum only PIM header
25  *					Relax this requirement to work with older peers.
26  *
27  */
28 
29 #include <asm/uaccess.h>
30 #include <linux/types.h>
31 #include <linux/capability.h>
32 #include <linux/errno.h>
33 #include <linux/timer.h>
34 #include <linux/mm.h>
35 #include <linux/kernel.h>
36 #include <linux/fcntl.h>
37 #include <linux/stat.h>
38 #include <linux/socket.h>
39 #include <linux/in.h>
40 #include <linux/inet.h>
41 #include <linux/netdevice.h>
42 #include <linux/inetdevice.h>
43 #include <linux/igmp.h>
44 #include <linux/proc_fs.h>
45 #include <linux/seq_file.h>
46 #include <linux/mroute.h>
47 #include <linux/init.h>
48 #include <linux/if_ether.h>
49 #include <linux/slab.h>
50 #include <net/net_namespace.h>
51 #include <net/ip.h>
52 #include <net/protocol.h>
53 #include <linux/skbuff.h>
54 #include <net/route.h>
55 #include <net/sock.h>
56 #include <net/icmp.h>
57 #include <net/udp.h>
58 #include <net/raw.h>
59 #include <linux/notifier.h>
60 #include <linux/if_arp.h>
61 #include <linux/netfilter_ipv4.h>
62 #include <linux/compat.h>
63 #include <linux/export.h>
64 #include <net/ipip.h>
65 #include <net/checksum.h>
66 #include <net/netlink.h>
67 #include <net/fib_rules.h>
68 
69 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
70 #define CONFIG_IP_PIMSM	1
71 #endif
72 
73 struct mr_table {
74 	struct list_head	list;
75 #ifdef CONFIG_NET_NS
76 	struct net		*net;
77 #endif
78 	u32			id;
79 	struct sock __rcu	*mroute_sk;
80 	struct timer_list	ipmr_expire_timer;
81 	struct list_head	mfc_unres_queue;
82 	struct list_head	mfc_cache_array[MFC_LINES];
83 	struct vif_device	vif_table[MAXVIFS];
84 	int			maxvif;
85 	atomic_t		cache_resolve_queue_len;
86 	int			mroute_do_assert;
87 	int			mroute_do_pim;
88 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
89 	int			mroute_reg_vif_num;
90 #endif
91 };
92 
93 struct ipmr_rule {
94 	struct fib_rule		common;
95 };
96 
97 struct ipmr_result {
98 	struct mr_table		*mrt;
99 };
100 
101 /* Big lock, protecting vif table, mrt cache and mroute socket state.
102  * Note that the changes are semaphored via rtnl_lock.
103  */
104 
105 static DEFINE_RWLOCK(mrt_lock);
106 
107 /*
108  *	Multicast router control variables
109  */
110 
111 #define VIF_EXISTS(_mrt, _idx) ((_mrt)->vif_table[_idx].dev != NULL)
112 
113 /* Special spinlock for queue of unresolved entries */
114 static DEFINE_SPINLOCK(mfc_unres_lock);
115 
116 /* We return to original Alan's scheme. Hash table of resolved
117  * entries is changed only in process context and protected
118  * with weak lock mrt_lock. Queue of unresolved entries is protected
119  * with strong spinlock mfc_unres_lock.
120  *
121  * In this case data path is free of exclusive locks at all.
122  */
123 
124 static struct kmem_cache *mrt_cachep __read_mostly;
125 
126 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
127 static void ipmr_free_table(struct mr_table *mrt);
128 
129 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
130 			 struct sk_buff *skb, struct mfc_cache *cache,
131 			 int local);
132 static int ipmr_cache_report(struct mr_table *mrt,
133 			     struct sk_buff *pkt, vifi_t vifi, int assert);
134 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
135 			      struct mfc_cache *c, struct rtmsg *rtm);
136 static void mroute_clean_tables(struct mr_table *mrt);
137 static void ipmr_expire_process(unsigned long arg);
138 
139 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
140 #define ipmr_for_each_table(mrt, net) \
141 	list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
142 
143 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
144 {
145 	struct mr_table *mrt;
146 
147 	ipmr_for_each_table(mrt, net) {
148 		if (mrt->id == id)
149 			return mrt;
150 	}
151 	return NULL;
152 }
153 
154 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
155 			   struct mr_table **mrt)
156 {
157 	struct ipmr_result res;
158 	struct fib_lookup_arg arg = { .result = &res, };
159 	int err;
160 
161 	err = fib_rules_lookup(net->ipv4.mr_rules_ops,
162 			       flowi4_to_flowi(flp4), 0, &arg);
163 	if (err < 0)
164 		return err;
165 	*mrt = res.mrt;
166 	return 0;
167 }
168 
169 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
170 			    int flags, struct fib_lookup_arg *arg)
171 {
172 	struct ipmr_result *res = arg->result;
173 	struct mr_table *mrt;
174 
175 	switch (rule->action) {
176 	case FR_ACT_TO_TBL:
177 		break;
178 	case FR_ACT_UNREACHABLE:
179 		return -ENETUNREACH;
180 	case FR_ACT_PROHIBIT:
181 		return -EACCES;
182 	case FR_ACT_BLACKHOLE:
183 	default:
184 		return -EINVAL;
185 	}
186 
187 	mrt = ipmr_get_table(rule->fr_net, rule->table);
188 	if (mrt == NULL)
189 		return -EAGAIN;
190 	res->mrt = mrt;
191 	return 0;
192 }
193 
194 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
195 {
196 	return 1;
197 }
198 
199 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
200 	FRA_GENERIC_POLICY,
201 };
202 
203 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
204 			       struct fib_rule_hdr *frh, struct nlattr **tb)
205 {
206 	return 0;
207 }
208 
209 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
210 			     struct nlattr **tb)
211 {
212 	return 1;
213 }
214 
215 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
216 			  struct fib_rule_hdr *frh)
217 {
218 	frh->dst_len = 0;
219 	frh->src_len = 0;
220 	frh->tos     = 0;
221 	return 0;
222 }
223 
224 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
225 	.family		= RTNL_FAMILY_IPMR,
226 	.rule_size	= sizeof(struct ipmr_rule),
227 	.addr_size	= sizeof(u32),
228 	.action		= ipmr_rule_action,
229 	.match		= ipmr_rule_match,
230 	.configure	= ipmr_rule_configure,
231 	.compare	= ipmr_rule_compare,
232 	.default_pref	= fib_default_rule_pref,
233 	.fill		= ipmr_rule_fill,
234 	.nlgroup	= RTNLGRP_IPV4_RULE,
235 	.policy		= ipmr_rule_policy,
236 	.owner		= THIS_MODULE,
237 };
238 
239 static int __net_init ipmr_rules_init(struct net *net)
240 {
241 	struct fib_rules_ops *ops;
242 	struct mr_table *mrt;
243 	int err;
244 
245 	ops = fib_rules_register(&ipmr_rules_ops_template, net);
246 	if (IS_ERR(ops))
247 		return PTR_ERR(ops);
248 
249 	INIT_LIST_HEAD(&net->ipv4.mr_tables);
250 
251 	mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
252 	if (mrt == NULL) {
253 		err = -ENOMEM;
254 		goto err1;
255 	}
256 
257 	err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
258 	if (err < 0)
259 		goto err2;
260 
261 	net->ipv4.mr_rules_ops = ops;
262 	return 0;
263 
264 err2:
265 	kfree(mrt);
266 err1:
267 	fib_rules_unregister(ops);
268 	return err;
269 }
270 
271 static void __net_exit ipmr_rules_exit(struct net *net)
272 {
273 	struct mr_table *mrt, *next;
274 
275 	list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
276 		list_del(&mrt->list);
277 		ipmr_free_table(mrt);
278 	}
279 	fib_rules_unregister(net->ipv4.mr_rules_ops);
280 }
281 #else
282 #define ipmr_for_each_table(mrt, net) \
283 	for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
284 
285 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
286 {
287 	return net->ipv4.mrt;
288 }
289 
290 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
291 			   struct mr_table **mrt)
292 {
293 	*mrt = net->ipv4.mrt;
294 	return 0;
295 }
296 
297 static int __net_init ipmr_rules_init(struct net *net)
298 {
299 	net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
300 	return net->ipv4.mrt ? 0 : -ENOMEM;
301 }
302 
303 static void __net_exit ipmr_rules_exit(struct net *net)
304 {
305 	ipmr_free_table(net->ipv4.mrt);
306 }
307 #endif
308 
309 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
310 {
311 	struct mr_table *mrt;
312 	unsigned int i;
313 
314 	mrt = ipmr_get_table(net, id);
315 	if (mrt != NULL)
316 		return mrt;
317 
318 	mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
319 	if (mrt == NULL)
320 		return NULL;
321 	write_pnet(&mrt->net, net);
322 	mrt->id = id;
323 
324 	/* Forwarding cache */
325 	for (i = 0; i < MFC_LINES; i++)
326 		INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
327 
328 	INIT_LIST_HEAD(&mrt->mfc_unres_queue);
329 
330 	setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
331 		    (unsigned long)mrt);
332 
333 #ifdef CONFIG_IP_PIMSM
334 	mrt->mroute_reg_vif_num = -1;
335 #endif
336 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
337 	list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
338 #endif
339 	return mrt;
340 }
341 
342 static void ipmr_free_table(struct mr_table *mrt)
343 {
344 	del_timer_sync(&mrt->ipmr_expire_timer);
345 	mroute_clean_tables(mrt);
346 	kfree(mrt);
347 }
348 
349 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
350 
351 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
352 {
353 	struct net *net = dev_net(dev);
354 
355 	dev_close(dev);
356 
357 	dev = __dev_get_by_name(net, "tunl0");
358 	if (dev) {
359 		const struct net_device_ops *ops = dev->netdev_ops;
360 		struct ifreq ifr;
361 		struct ip_tunnel_parm p;
362 
363 		memset(&p, 0, sizeof(p));
364 		p.iph.daddr = v->vifc_rmt_addr.s_addr;
365 		p.iph.saddr = v->vifc_lcl_addr.s_addr;
366 		p.iph.version = 4;
367 		p.iph.ihl = 5;
368 		p.iph.protocol = IPPROTO_IPIP;
369 		sprintf(p.name, "dvmrp%d", v->vifc_vifi);
370 		ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
371 
372 		if (ops->ndo_do_ioctl) {
373 			mm_segment_t oldfs = get_fs();
374 
375 			set_fs(KERNEL_DS);
376 			ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
377 			set_fs(oldfs);
378 		}
379 	}
380 }
381 
382 static
383 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
384 {
385 	struct net_device  *dev;
386 
387 	dev = __dev_get_by_name(net, "tunl0");
388 
389 	if (dev) {
390 		const struct net_device_ops *ops = dev->netdev_ops;
391 		int err;
392 		struct ifreq ifr;
393 		struct ip_tunnel_parm p;
394 		struct in_device  *in_dev;
395 
396 		memset(&p, 0, sizeof(p));
397 		p.iph.daddr = v->vifc_rmt_addr.s_addr;
398 		p.iph.saddr = v->vifc_lcl_addr.s_addr;
399 		p.iph.version = 4;
400 		p.iph.ihl = 5;
401 		p.iph.protocol = IPPROTO_IPIP;
402 		sprintf(p.name, "dvmrp%d", v->vifc_vifi);
403 		ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
404 
405 		if (ops->ndo_do_ioctl) {
406 			mm_segment_t oldfs = get_fs();
407 
408 			set_fs(KERNEL_DS);
409 			err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
410 			set_fs(oldfs);
411 		} else {
412 			err = -EOPNOTSUPP;
413 		}
414 		dev = NULL;
415 
416 		if (err == 0 &&
417 		    (dev = __dev_get_by_name(net, p.name)) != NULL) {
418 			dev->flags |= IFF_MULTICAST;
419 
420 			in_dev = __in_dev_get_rtnl(dev);
421 			if (in_dev == NULL)
422 				goto failure;
423 
424 			ipv4_devconf_setall(in_dev);
425 			IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
426 
427 			if (dev_open(dev))
428 				goto failure;
429 			dev_hold(dev);
430 		}
431 	}
432 	return dev;
433 
434 failure:
435 	/* allow the register to be completed before unregistering. */
436 	rtnl_unlock();
437 	rtnl_lock();
438 
439 	unregister_netdevice(dev);
440 	return NULL;
441 }
442 
443 #ifdef CONFIG_IP_PIMSM
444 
445 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
446 {
447 	struct net *net = dev_net(dev);
448 	struct mr_table *mrt;
449 	struct flowi4 fl4 = {
450 		.flowi4_oif	= dev->ifindex,
451 		.flowi4_iif	= skb->skb_iif,
452 		.flowi4_mark	= skb->mark,
453 	};
454 	int err;
455 
456 	err = ipmr_fib_lookup(net, &fl4, &mrt);
457 	if (err < 0) {
458 		kfree_skb(skb);
459 		return err;
460 	}
461 
462 	read_lock(&mrt_lock);
463 	dev->stats.tx_bytes += skb->len;
464 	dev->stats.tx_packets++;
465 	ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
466 	read_unlock(&mrt_lock);
467 	kfree_skb(skb);
468 	return NETDEV_TX_OK;
469 }
470 
471 static const struct net_device_ops reg_vif_netdev_ops = {
472 	.ndo_start_xmit	= reg_vif_xmit,
473 };
474 
475 static void reg_vif_setup(struct net_device *dev)
476 {
477 	dev->type		= ARPHRD_PIMREG;
478 	dev->mtu		= ETH_DATA_LEN - sizeof(struct iphdr) - 8;
479 	dev->flags		= IFF_NOARP;
480 	dev->netdev_ops		= &reg_vif_netdev_ops,
481 	dev->destructor		= free_netdev;
482 	dev->features		|= NETIF_F_NETNS_LOCAL;
483 }
484 
485 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
486 {
487 	struct net_device *dev;
488 	struct in_device *in_dev;
489 	char name[IFNAMSIZ];
490 
491 	if (mrt->id == RT_TABLE_DEFAULT)
492 		sprintf(name, "pimreg");
493 	else
494 		sprintf(name, "pimreg%u", mrt->id);
495 
496 	dev = alloc_netdev(0, name, reg_vif_setup);
497 
498 	if (dev == NULL)
499 		return NULL;
500 
501 	dev_net_set(dev, net);
502 
503 	if (register_netdevice(dev)) {
504 		free_netdev(dev);
505 		return NULL;
506 	}
507 	dev->iflink = 0;
508 
509 	rcu_read_lock();
510 	in_dev = __in_dev_get_rcu(dev);
511 	if (!in_dev) {
512 		rcu_read_unlock();
513 		goto failure;
514 	}
515 
516 	ipv4_devconf_setall(in_dev);
517 	IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
518 	rcu_read_unlock();
519 
520 	if (dev_open(dev))
521 		goto failure;
522 
523 	dev_hold(dev);
524 
525 	return dev;
526 
527 failure:
528 	/* allow the register to be completed before unregistering. */
529 	rtnl_unlock();
530 	rtnl_lock();
531 
532 	unregister_netdevice(dev);
533 	return NULL;
534 }
535 #endif
536 
537 /**
538  *	vif_delete - Delete a VIF entry
539  *	@notify: Set to 1, if the caller is a notifier_call
540  */
541 
542 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
543 		      struct list_head *head)
544 {
545 	struct vif_device *v;
546 	struct net_device *dev;
547 	struct in_device *in_dev;
548 
549 	if (vifi < 0 || vifi >= mrt->maxvif)
550 		return -EADDRNOTAVAIL;
551 
552 	v = &mrt->vif_table[vifi];
553 
554 	write_lock_bh(&mrt_lock);
555 	dev = v->dev;
556 	v->dev = NULL;
557 
558 	if (!dev) {
559 		write_unlock_bh(&mrt_lock);
560 		return -EADDRNOTAVAIL;
561 	}
562 
563 #ifdef CONFIG_IP_PIMSM
564 	if (vifi == mrt->mroute_reg_vif_num)
565 		mrt->mroute_reg_vif_num = -1;
566 #endif
567 
568 	if (vifi + 1 == mrt->maxvif) {
569 		int tmp;
570 
571 		for (tmp = vifi - 1; tmp >= 0; tmp--) {
572 			if (VIF_EXISTS(mrt, tmp))
573 				break;
574 		}
575 		mrt->maxvif = tmp+1;
576 	}
577 
578 	write_unlock_bh(&mrt_lock);
579 
580 	dev_set_allmulti(dev, -1);
581 
582 	in_dev = __in_dev_get_rtnl(dev);
583 	if (in_dev) {
584 		IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
585 		ip_rt_multicast_event(in_dev);
586 	}
587 
588 	if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
589 		unregister_netdevice_queue(dev, head);
590 
591 	dev_put(dev);
592 	return 0;
593 }
594 
595 static void ipmr_cache_free_rcu(struct rcu_head *head)
596 {
597 	struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
598 
599 	kmem_cache_free(mrt_cachep, c);
600 }
601 
602 static inline void ipmr_cache_free(struct mfc_cache *c)
603 {
604 	call_rcu(&c->rcu, ipmr_cache_free_rcu);
605 }
606 
607 /* Destroy an unresolved cache entry, killing queued skbs
608  * and reporting error to netlink readers.
609  */
610 
611 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
612 {
613 	struct net *net = read_pnet(&mrt->net);
614 	struct sk_buff *skb;
615 	struct nlmsgerr *e;
616 
617 	atomic_dec(&mrt->cache_resolve_queue_len);
618 
619 	while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
620 		if (ip_hdr(skb)->version == 0) {
621 			struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
622 			nlh->nlmsg_type = NLMSG_ERROR;
623 			nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
624 			skb_trim(skb, nlh->nlmsg_len);
625 			e = NLMSG_DATA(nlh);
626 			e->error = -ETIMEDOUT;
627 			memset(&e->msg, 0, sizeof(e->msg));
628 
629 			rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
630 		} else {
631 			kfree_skb(skb);
632 		}
633 	}
634 
635 	ipmr_cache_free(c);
636 }
637 
638 
639 /* Timer process for the unresolved queue. */
640 
641 static void ipmr_expire_process(unsigned long arg)
642 {
643 	struct mr_table *mrt = (struct mr_table *)arg;
644 	unsigned long now;
645 	unsigned long expires;
646 	struct mfc_cache *c, *next;
647 
648 	if (!spin_trylock(&mfc_unres_lock)) {
649 		mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
650 		return;
651 	}
652 
653 	if (list_empty(&mrt->mfc_unres_queue))
654 		goto out;
655 
656 	now = jiffies;
657 	expires = 10*HZ;
658 
659 	list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
660 		if (time_after(c->mfc_un.unres.expires, now)) {
661 			unsigned long interval = c->mfc_un.unres.expires - now;
662 			if (interval < expires)
663 				expires = interval;
664 			continue;
665 		}
666 
667 		list_del(&c->list);
668 		ipmr_destroy_unres(mrt, c);
669 	}
670 
671 	if (!list_empty(&mrt->mfc_unres_queue))
672 		mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
673 
674 out:
675 	spin_unlock(&mfc_unres_lock);
676 }
677 
678 /* Fill oifs list. It is called under write locked mrt_lock. */
679 
680 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
681 				   unsigned char *ttls)
682 {
683 	int vifi;
684 
685 	cache->mfc_un.res.minvif = MAXVIFS;
686 	cache->mfc_un.res.maxvif = 0;
687 	memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
688 
689 	for (vifi = 0; vifi < mrt->maxvif; vifi++) {
690 		if (VIF_EXISTS(mrt, vifi) &&
691 		    ttls[vifi] && ttls[vifi] < 255) {
692 			cache->mfc_un.res.ttls[vifi] = ttls[vifi];
693 			if (cache->mfc_un.res.minvif > vifi)
694 				cache->mfc_un.res.minvif = vifi;
695 			if (cache->mfc_un.res.maxvif <= vifi)
696 				cache->mfc_un.res.maxvif = vifi + 1;
697 		}
698 	}
699 }
700 
701 static int vif_add(struct net *net, struct mr_table *mrt,
702 		   struct vifctl *vifc, int mrtsock)
703 {
704 	int vifi = vifc->vifc_vifi;
705 	struct vif_device *v = &mrt->vif_table[vifi];
706 	struct net_device *dev;
707 	struct in_device *in_dev;
708 	int err;
709 
710 	/* Is vif busy ? */
711 	if (VIF_EXISTS(mrt, vifi))
712 		return -EADDRINUSE;
713 
714 	switch (vifc->vifc_flags) {
715 #ifdef CONFIG_IP_PIMSM
716 	case VIFF_REGISTER:
717 		/*
718 		 * Special Purpose VIF in PIM
719 		 * All the packets will be sent to the daemon
720 		 */
721 		if (mrt->mroute_reg_vif_num >= 0)
722 			return -EADDRINUSE;
723 		dev = ipmr_reg_vif(net, mrt);
724 		if (!dev)
725 			return -ENOBUFS;
726 		err = dev_set_allmulti(dev, 1);
727 		if (err) {
728 			unregister_netdevice(dev);
729 			dev_put(dev);
730 			return err;
731 		}
732 		break;
733 #endif
734 	case VIFF_TUNNEL:
735 		dev = ipmr_new_tunnel(net, vifc);
736 		if (!dev)
737 			return -ENOBUFS;
738 		err = dev_set_allmulti(dev, 1);
739 		if (err) {
740 			ipmr_del_tunnel(dev, vifc);
741 			dev_put(dev);
742 			return err;
743 		}
744 		break;
745 
746 	case VIFF_USE_IFINDEX:
747 	case 0:
748 		if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
749 			dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
750 			if (dev && __in_dev_get_rtnl(dev) == NULL) {
751 				dev_put(dev);
752 				return -EADDRNOTAVAIL;
753 			}
754 		} else {
755 			dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
756 		}
757 		if (!dev)
758 			return -EADDRNOTAVAIL;
759 		err = dev_set_allmulti(dev, 1);
760 		if (err) {
761 			dev_put(dev);
762 			return err;
763 		}
764 		break;
765 	default:
766 		return -EINVAL;
767 	}
768 
769 	in_dev = __in_dev_get_rtnl(dev);
770 	if (!in_dev) {
771 		dev_put(dev);
772 		return -EADDRNOTAVAIL;
773 	}
774 	IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
775 	ip_rt_multicast_event(in_dev);
776 
777 	/* Fill in the VIF structures */
778 
779 	v->rate_limit = vifc->vifc_rate_limit;
780 	v->local = vifc->vifc_lcl_addr.s_addr;
781 	v->remote = vifc->vifc_rmt_addr.s_addr;
782 	v->flags = vifc->vifc_flags;
783 	if (!mrtsock)
784 		v->flags |= VIFF_STATIC;
785 	v->threshold = vifc->vifc_threshold;
786 	v->bytes_in = 0;
787 	v->bytes_out = 0;
788 	v->pkt_in = 0;
789 	v->pkt_out = 0;
790 	v->link = dev->ifindex;
791 	if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
792 		v->link = dev->iflink;
793 
794 	/* And finish update writing critical data */
795 	write_lock_bh(&mrt_lock);
796 	v->dev = dev;
797 #ifdef CONFIG_IP_PIMSM
798 	if (v->flags & VIFF_REGISTER)
799 		mrt->mroute_reg_vif_num = vifi;
800 #endif
801 	if (vifi+1 > mrt->maxvif)
802 		mrt->maxvif = vifi+1;
803 	write_unlock_bh(&mrt_lock);
804 	return 0;
805 }
806 
807 /* called with rcu_read_lock() */
808 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
809 					 __be32 origin,
810 					 __be32 mcastgrp)
811 {
812 	int line = MFC_HASH(mcastgrp, origin);
813 	struct mfc_cache *c;
814 
815 	list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
816 		if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
817 			return c;
818 	}
819 	return NULL;
820 }
821 
822 /*
823  *	Allocate a multicast cache entry
824  */
825 static struct mfc_cache *ipmr_cache_alloc(void)
826 {
827 	struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
828 
829 	if (c)
830 		c->mfc_un.res.minvif = MAXVIFS;
831 	return c;
832 }
833 
834 static struct mfc_cache *ipmr_cache_alloc_unres(void)
835 {
836 	struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
837 
838 	if (c) {
839 		skb_queue_head_init(&c->mfc_un.unres.unresolved);
840 		c->mfc_un.unres.expires = jiffies + 10*HZ;
841 	}
842 	return c;
843 }
844 
845 /*
846  *	A cache entry has gone into a resolved state from queued
847  */
848 
849 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
850 			       struct mfc_cache *uc, struct mfc_cache *c)
851 {
852 	struct sk_buff *skb;
853 	struct nlmsgerr *e;
854 
855 	/* Play the pending entries through our router */
856 
857 	while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
858 		if (ip_hdr(skb)->version == 0) {
859 			struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
860 
861 			if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
862 				nlh->nlmsg_len = skb_tail_pointer(skb) -
863 						 (u8 *)nlh;
864 			} else {
865 				nlh->nlmsg_type = NLMSG_ERROR;
866 				nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
867 				skb_trim(skb, nlh->nlmsg_len);
868 				e = NLMSG_DATA(nlh);
869 				e->error = -EMSGSIZE;
870 				memset(&e->msg, 0, sizeof(e->msg));
871 			}
872 
873 			rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
874 		} else {
875 			ip_mr_forward(net, mrt, skb, c, 0);
876 		}
877 	}
878 }
879 
880 /*
881  *	Bounce a cache query up to mrouted. We could use netlink for this but mrouted
882  *	expects the following bizarre scheme.
883  *
884  *	Called under mrt_lock.
885  */
886 
887 static int ipmr_cache_report(struct mr_table *mrt,
888 			     struct sk_buff *pkt, vifi_t vifi, int assert)
889 {
890 	struct sk_buff *skb;
891 	const int ihl = ip_hdrlen(pkt);
892 	struct igmphdr *igmp;
893 	struct igmpmsg *msg;
894 	struct sock *mroute_sk;
895 	int ret;
896 
897 #ifdef CONFIG_IP_PIMSM
898 	if (assert == IGMPMSG_WHOLEPKT)
899 		skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
900 	else
901 #endif
902 		skb = alloc_skb(128, GFP_ATOMIC);
903 
904 	if (!skb)
905 		return -ENOBUFS;
906 
907 #ifdef CONFIG_IP_PIMSM
908 	if (assert == IGMPMSG_WHOLEPKT) {
909 		/* Ugly, but we have no choice with this interface.
910 		 * Duplicate old header, fix ihl, length etc.
911 		 * And all this only to mangle msg->im_msgtype and
912 		 * to set msg->im_mbz to "mbz" :-)
913 		 */
914 		skb_push(skb, sizeof(struct iphdr));
915 		skb_reset_network_header(skb);
916 		skb_reset_transport_header(skb);
917 		msg = (struct igmpmsg *)skb_network_header(skb);
918 		memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
919 		msg->im_msgtype = IGMPMSG_WHOLEPKT;
920 		msg->im_mbz = 0;
921 		msg->im_vif = mrt->mroute_reg_vif_num;
922 		ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
923 		ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
924 					     sizeof(struct iphdr));
925 	} else
926 #endif
927 	{
928 
929 	/* Copy the IP header */
930 
931 	skb->network_header = skb->tail;
932 	skb_put(skb, ihl);
933 	skb_copy_to_linear_data(skb, pkt->data, ihl);
934 	ip_hdr(skb)->protocol = 0;	/* Flag to the kernel this is a route add */
935 	msg = (struct igmpmsg *)skb_network_header(skb);
936 	msg->im_vif = vifi;
937 	skb_dst_set(skb, dst_clone(skb_dst(pkt)));
938 
939 	/* Add our header */
940 
941 	igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
942 	igmp->type	=
943 	msg->im_msgtype = assert;
944 	igmp->code	= 0;
945 	ip_hdr(skb)->tot_len = htons(skb->len);		/* Fix the length */
946 	skb->transport_header = skb->network_header;
947 	}
948 
949 	rcu_read_lock();
950 	mroute_sk = rcu_dereference(mrt->mroute_sk);
951 	if (mroute_sk == NULL) {
952 		rcu_read_unlock();
953 		kfree_skb(skb);
954 		return -EINVAL;
955 	}
956 
957 	/* Deliver to mrouted */
958 
959 	ret = sock_queue_rcv_skb(mroute_sk, skb);
960 	rcu_read_unlock();
961 	if (ret < 0) {
962 		net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
963 		kfree_skb(skb);
964 	}
965 
966 	return ret;
967 }
968 
969 /*
970  *	Queue a packet for resolution. It gets locked cache entry!
971  */
972 
973 static int
974 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
975 {
976 	bool found = false;
977 	int err;
978 	struct mfc_cache *c;
979 	const struct iphdr *iph = ip_hdr(skb);
980 
981 	spin_lock_bh(&mfc_unres_lock);
982 	list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
983 		if (c->mfc_mcastgrp == iph->daddr &&
984 		    c->mfc_origin == iph->saddr) {
985 			found = true;
986 			break;
987 		}
988 	}
989 
990 	if (!found) {
991 		/* Create a new entry if allowable */
992 
993 		if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
994 		    (c = ipmr_cache_alloc_unres()) == NULL) {
995 			spin_unlock_bh(&mfc_unres_lock);
996 
997 			kfree_skb(skb);
998 			return -ENOBUFS;
999 		}
1000 
1001 		/* Fill in the new cache entry */
1002 
1003 		c->mfc_parent	= -1;
1004 		c->mfc_origin	= iph->saddr;
1005 		c->mfc_mcastgrp	= iph->daddr;
1006 
1007 		/* Reflect first query at mrouted. */
1008 
1009 		err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1010 		if (err < 0) {
1011 			/* If the report failed throw the cache entry
1012 			   out - Brad Parker
1013 			 */
1014 			spin_unlock_bh(&mfc_unres_lock);
1015 
1016 			ipmr_cache_free(c);
1017 			kfree_skb(skb);
1018 			return err;
1019 		}
1020 
1021 		atomic_inc(&mrt->cache_resolve_queue_len);
1022 		list_add(&c->list, &mrt->mfc_unres_queue);
1023 
1024 		if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1025 			mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1026 	}
1027 
1028 	/* See if we can append the packet */
1029 
1030 	if (c->mfc_un.unres.unresolved.qlen > 3) {
1031 		kfree_skb(skb);
1032 		err = -ENOBUFS;
1033 	} else {
1034 		skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1035 		err = 0;
1036 	}
1037 
1038 	spin_unlock_bh(&mfc_unres_lock);
1039 	return err;
1040 }
1041 
1042 /*
1043  *	MFC cache manipulation by user space mroute daemon
1044  */
1045 
1046 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1047 {
1048 	int line;
1049 	struct mfc_cache *c, *next;
1050 
1051 	line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1052 
1053 	list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1054 		if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1055 		    c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1056 			list_del_rcu(&c->list);
1057 
1058 			ipmr_cache_free(c);
1059 			return 0;
1060 		}
1061 	}
1062 	return -ENOENT;
1063 }
1064 
1065 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1066 			struct mfcctl *mfc, int mrtsock)
1067 {
1068 	bool found = false;
1069 	int line;
1070 	struct mfc_cache *uc, *c;
1071 
1072 	if (mfc->mfcc_parent >= MAXVIFS)
1073 		return -ENFILE;
1074 
1075 	line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1076 
1077 	list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1078 		if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1079 		    c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1080 			found = true;
1081 			break;
1082 		}
1083 	}
1084 
1085 	if (found) {
1086 		write_lock_bh(&mrt_lock);
1087 		c->mfc_parent = mfc->mfcc_parent;
1088 		ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1089 		if (!mrtsock)
1090 			c->mfc_flags |= MFC_STATIC;
1091 		write_unlock_bh(&mrt_lock);
1092 		return 0;
1093 	}
1094 
1095 	if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1096 		return -EINVAL;
1097 
1098 	c = ipmr_cache_alloc();
1099 	if (c == NULL)
1100 		return -ENOMEM;
1101 
1102 	c->mfc_origin = mfc->mfcc_origin.s_addr;
1103 	c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1104 	c->mfc_parent = mfc->mfcc_parent;
1105 	ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1106 	if (!mrtsock)
1107 		c->mfc_flags |= MFC_STATIC;
1108 
1109 	list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1110 
1111 	/*
1112 	 *	Check to see if we resolved a queued list. If so we
1113 	 *	need to send on the frames and tidy up.
1114 	 */
1115 	found = false;
1116 	spin_lock_bh(&mfc_unres_lock);
1117 	list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1118 		if (uc->mfc_origin == c->mfc_origin &&
1119 		    uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1120 			list_del(&uc->list);
1121 			atomic_dec(&mrt->cache_resolve_queue_len);
1122 			found = true;
1123 			break;
1124 		}
1125 	}
1126 	if (list_empty(&mrt->mfc_unres_queue))
1127 		del_timer(&mrt->ipmr_expire_timer);
1128 	spin_unlock_bh(&mfc_unres_lock);
1129 
1130 	if (found) {
1131 		ipmr_cache_resolve(net, mrt, uc, c);
1132 		ipmr_cache_free(uc);
1133 	}
1134 	return 0;
1135 }
1136 
1137 /*
1138  *	Close the multicast socket, and clear the vif tables etc
1139  */
1140 
1141 static void mroute_clean_tables(struct mr_table *mrt)
1142 {
1143 	int i;
1144 	LIST_HEAD(list);
1145 	struct mfc_cache *c, *next;
1146 
1147 	/* Shut down all active vif entries */
1148 
1149 	for (i = 0; i < mrt->maxvif; i++) {
1150 		if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1151 			vif_delete(mrt, i, 0, &list);
1152 	}
1153 	unregister_netdevice_many(&list);
1154 
1155 	/* Wipe the cache */
1156 
1157 	for (i = 0; i < MFC_LINES; i++) {
1158 		list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1159 			if (c->mfc_flags & MFC_STATIC)
1160 				continue;
1161 			list_del_rcu(&c->list);
1162 			ipmr_cache_free(c);
1163 		}
1164 	}
1165 
1166 	if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1167 		spin_lock_bh(&mfc_unres_lock);
1168 		list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1169 			list_del(&c->list);
1170 			ipmr_destroy_unres(mrt, c);
1171 		}
1172 		spin_unlock_bh(&mfc_unres_lock);
1173 	}
1174 }
1175 
1176 /* called from ip_ra_control(), before an RCU grace period,
1177  * we dont need to call synchronize_rcu() here
1178  */
1179 static void mrtsock_destruct(struct sock *sk)
1180 {
1181 	struct net *net = sock_net(sk);
1182 	struct mr_table *mrt;
1183 
1184 	rtnl_lock();
1185 	ipmr_for_each_table(mrt, net) {
1186 		if (sk == rtnl_dereference(mrt->mroute_sk)) {
1187 			IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1188 			RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1189 			mroute_clean_tables(mrt);
1190 		}
1191 	}
1192 	rtnl_unlock();
1193 }
1194 
1195 /*
1196  *	Socket options and virtual interface manipulation. The whole
1197  *	virtual interface system is a complete heap, but unfortunately
1198  *	that's how BSD mrouted happens to think. Maybe one day with a proper
1199  *	MOSPF/PIM router set up we can clean this up.
1200  */
1201 
1202 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1203 {
1204 	int ret;
1205 	struct vifctl vif;
1206 	struct mfcctl mfc;
1207 	struct net *net = sock_net(sk);
1208 	struct mr_table *mrt;
1209 
1210 	mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1211 	if (mrt == NULL)
1212 		return -ENOENT;
1213 
1214 	if (optname != MRT_INIT) {
1215 		if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1216 		    !capable(CAP_NET_ADMIN))
1217 			return -EACCES;
1218 	}
1219 
1220 	switch (optname) {
1221 	case MRT_INIT:
1222 		if (sk->sk_type != SOCK_RAW ||
1223 		    inet_sk(sk)->inet_num != IPPROTO_IGMP)
1224 			return -EOPNOTSUPP;
1225 		if (optlen != sizeof(int))
1226 			return -ENOPROTOOPT;
1227 
1228 		rtnl_lock();
1229 		if (rtnl_dereference(mrt->mroute_sk)) {
1230 			rtnl_unlock();
1231 			return -EADDRINUSE;
1232 		}
1233 
1234 		ret = ip_ra_control(sk, 1, mrtsock_destruct);
1235 		if (ret == 0) {
1236 			rcu_assign_pointer(mrt->mroute_sk, sk);
1237 			IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1238 		}
1239 		rtnl_unlock();
1240 		return ret;
1241 	case MRT_DONE:
1242 		if (sk != rcu_access_pointer(mrt->mroute_sk))
1243 			return -EACCES;
1244 		return ip_ra_control(sk, 0, NULL);
1245 	case MRT_ADD_VIF:
1246 	case MRT_DEL_VIF:
1247 		if (optlen != sizeof(vif))
1248 			return -EINVAL;
1249 		if (copy_from_user(&vif, optval, sizeof(vif)))
1250 			return -EFAULT;
1251 		if (vif.vifc_vifi >= MAXVIFS)
1252 			return -ENFILE;
1253 		rtnl_lock();
1254 		if (optname == MRT_ADD_VIF) {
1255 			ret = vif_add(net, mrt, &vif,
1256 				      sk == rtnl_dereference(mrt->mroute_sk));
1257 		} else {
1258 			ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1259 		}
1260 		rtnl_unlock();
1261 		return ret;
1262 
1263 		/*
1264 		 *	Manipulate the forwarding caches. These live
1265 		 *	in a sort of kernel/user symbiosis.
1266 		 */
1267 	case MRT_ADD_MFC:
1268 	case MRT_DEL_MFC:
1269 		if (optlen != sizeof(mfc))
1270 			return -EINVAL;
1271 		if (copy_from_user(&mfc, optval, sizeof(mfc)))
1272 			return -EFAULT;
1273 		rtnl_lock();
1274 		if (optname == MRT_DEL_MFC)
1275 			ret = ipmr_mfc_delete(mrt, &mfc);
1276 		else
1277 			ret = ipmr_mfc_add(net, mrt, &mfc,
1278 					   sk == rtnl_dereference(mrt->mroute_sk));
1279 		rtnl_unlock();
1280 		return ret;
1281 		/*
1282 		 *	Control PIM assert.
1283 		 */
1284 	case MRT_ASSERT:
1285 	{
1286 		int v;
1287 		if (get_user(v, (int __user *)optval))
1288 			return -EFAULT;
1289 		mrt->mroute_do_assert = (v) ? 1 : 0;
1290 		return 0;
1291 	}
1292 #ifdef CONFIG_IP_PIMSM
1293 	case MRT_PIM:
1294 	{
1295 		int v;
1296 
1297 		if (get_user(v, (int __user *)optval))
1298 			return -EFAULT;
1299 		v = (v) ? 1 : 0;
1300 
1301 		rtnl_lock();
1302 		ret = 0;
1303 		if (v != mrt->mroute_do_pim) {
1304 			mrt->mroute_do_pim = v;
1305 			mrt->mroute_do_assert = v;
1306 		}
1307 		rtnl_unlock();
1308 		return ret;
1309 	}
1310 #endif
1311 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1312 	case MRT_TABLE:
1313 	{
1314 		u32 v;
1315 
1316 		if (optlen != sizeof(u32))
1317 			return -EINVAL;
1318 		if (get_user(v, (u32 __user *)optval))
1319 			return -EFAULT;
1320 
1321 		/* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
1322 		if (v != RT_TABLE_DEFAULT && v >= 1000000000)
1323 			return -EINVAL;
1324 
1325 		rtnl_lock();
1326 		ret = 0;
1327 		if (sk == rtnl_dereference(mrt->mroute_sk)) {
1328 			ret = -EBUSY;
1329 		} else {
1330 			if (!ipmr_new_table(net, v))
1331 				ret = -ENOMEM;
1332 			raw_sk(sk)->ipmr_table = v;
1333 		}
1334 		rtnl_unlock();
1335 		return ret;
1336 	}
1337 #endif
1338 	/*
1339 	 *	Spurious command, or MRT_VERSION which you cannot
1340 	 *	set.
1341 	 */
1342 	default:
1343 		return -ENOPROTOOPT;
1344 	}
1345 }
1346 
1347 /*
1348  *	Getsock opt support for the multicast routing system.
1349  */
1350 
1351 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1352 {
1353 	int olr;
1354 	int val;
1355 	struct net *net = sock_net(sk);
1356 	struct mr_table *mrt;
1357 
1358 	mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1359 	if (mrt == NULL)
1360 		return -ENOENT;
1361 
1362 	if (optname != MRT_VERSION &&
1363 #ifdef CONFIG_IP_PIMSM
1364 	   optname != MRT_PIM &&
1365 #endif
1366 	   optname != MRT_ASSERT)
1367 		return -ENOPROTOOPT;
1368 
1369 	if (get_user(olr, optlen))
1370 		return -EFAULT;
1371 
1372 	olr = min_t(unsigned int, olr, sizeof(int));
1373 	if (olr < 0)
1374 		return -EINVAL;
1375 
1376 	if (put_user(olr, optlen))
1377 		return -EFAULT;
1378 	if (optname == MRT_VERSION)
1379 		val = 0x0305;
1380 #ifdef CONFIG_IP_PIMSM
1381 	else if (optname == MRT_PIM)
1382 		val = mrt->mroute_do_pim;
1383 #endif
1384 	else
1385 		val = mrt->mroute_do_assert;
1386 	if (copy_to_user(optval, &val, olr))
1387 		return -EFAULT;
1388 	return 0;
1389 }
1390 
1391 /*
1392  *	The IP multicast ioctl support routines.
1393  */
1394 
1395 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1396 {
1397 	struct sioc_sg_req sr;
1398 	struct sioc_vif_req vr;
1399 	struct vif_device *vif;
1400 	struct mfc_cache *c;
1401 	struct net *net = sock_net(sk);
1402 	struct mr_table *mrt;
1403 
1404 	mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1405 	if (mrt == NULL)
1406 		return -ENOENT;
1407 
1408 	switch (cmd) {
1409 	case SIOCGETVIFCNT:
1410 		if (copy_from_user(&vr, arg, sizeof(vr)))
1411 			return -EFAULT;
1412 		if (vr.vifi >= mrt->maxvif)
1413 			return -EINVAL;
1414 		read_lock(&mrt_lock);
1415 		vif = &mrt->vif_table[vr.vifi];
1416 		if (VIF_EXISTS(mrt, vr.vifi)) {
1417 			vr.icount = vif->pkt_in;
1418 			vr.ocount = vif->pkt_out;
1419 			vr.ibytes = vif->bytes_in;
1420 			vr.obytes = vif->bytes_out;
1421 			read_unlock(&mrt_lock);
1422 
1423 			if (copy_to_user(arg, &vr, sizeof(vr)))
1424 				return -EFAULT;
1425 			return 0;
1426 		}
1427 		read_unlock(&mrt_lock);
1428 		return -EADDRNOTAVAIL;
1429 	case SIOCGETSGCNT:
1430 		if (copy_from_user(&sr, arg, sizeof(sr)))
1431 			return -EFAULT;
1432 
1433 		rcu_read_lock();
1434 		c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1435 		if (c) {
1436 			sr.pktcnt = c->mfc_un.res.pkt;
1437 			sr.bytecnt = c->mfc_un.res.bytes;
1438 			sr.wrong_if = c->mfc_un.res.wrong_if;
1439 			rcu_read_unlock();
1440 
1441 			if (copy_to_user(arg, &sr, sizeof(sr)))
1442 				return -EFAULT;
1443 			return 0;
1444 		}
1445 		rcu_read_unlock();
1446 		return -EADDRNOTAVAIL;
1447 	default:
1448 		return -ENOIOCTLCMD;
1449 	}
1450 }
1451 
1452 #ifdef CONFIG_COMPAT
1453 struct compat_sioc_sg_req {
1454 	struct in_addr src;
1455 	struct in_addr grp;
1456 	compat_ulong_t pktcnt;
1457 	compat_ulong_t bytecnt;
1458 	compat_ulong_t wrong_if;
1459 };
1460 
1461 struct compat_sioc_vif_req {
1462 	vifi_t	vifi;		/* Which iface */
1463 	compat_ulong_t icount;
1464 	compat_ulong_t ocount;
1465 	compat_ulong_t ibytes;
1466 	compat_ulong_t obytes;
1467 };
1468 
1469 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1470 {
1471 	struct compat_sioc_sg_req sr;
1472 	struct compat_sioc_vif_req vr;
1473 	struct vif_device *vif;
1474 	struct mfc_cache *c;
1475 	struct net *net = sock_net(sk);
1476 	struct mr_table *mrt;
1477 
1478 	mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1479 	if (mrt == NULL)
1480 		return -ENOENT;
1481 
1482 	switch (cmd) {
1483 	case SIOCGETVIFCNT:
1484 		if (copy_from_user(&vr, arg, sizeof(vr)))
1485 			return -EFAULT;
1486 		if (vr.vifi >= mrt->maxvif)
1487 			return -EINVAL;
1488 		read_lock(&mrt_lock);
1489 		vif = &mrt->vif_table[vr.vifi];
1490 		if (VIF_EXISTS(mrt, vr.vifi)) {
1491 			vr.icount = vif->pkt_in;
1492 			vr.ocount = vif->pkt_out;
1493 			vr.ibytes = vif->bytes_in;
1494 			vr.obytes = vif->bytes_out;
1495 			read_unlock(&mrt_lock);
1496 
1497 			if (copy_to_user(arg, &vr, sizeof(vr)))
1498 				return -EFAULT;
1499 			return 0;
1500 		}
1501 		read_unlock(&mrt_lock);
1502 		return -EADDRNOTAVAIL;
1503 	case SIOCGETSGCNT:
1504 		if (copy_from_user(&sr, arg, sizeof(sr)))
1505 			return -EFAULT;
1506 
1507 		rcu_read_lock();
1508 		c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1509 		if (c) {
1510 			sr.pktcnt = c->mfc_un.res.pkt;
1511 			sr.bytecnt = c->mfc_un.res.bytes;
1512 			sr.wrong_if = c->mfc_un.res.wrong_if;
1513 			rcu_read_unlock();
1514 
1515 			if (copy_to_user(arg, &sr, sizeof(sr)))
1516 				return -EFAULT;
1517 			return 0;
1518 		}
1519 		rcu_read_unlock();
1520 		return -EADDRNOTAVAIL;
1521 	default:
1522 		return -ENOIOCTLCMD;
1523 	}
1524 }
1525 #endif
1526 
1527 
1528 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1529 {
1530 	struct net_device *dev = ptr;
1531 	struct net *net = dev_net(dev);
1532 	struct mr_table *mrt;
1533 	struct vif_device *v;
1534 	int ct;
1535 
1536 	if (event != NETDEV_UNREGISTER)
1537 		return NOTIFY_DONE;
1538 
1539 	ipmr_for_each_table(mrt, net) {
1540 		v = &mrt->vif_table[0];
1541 		for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1542 			if (v->dev == dev)
1543 				vif_delete(mrt, ct, 1, NULL);
1544 		}
1545 	}
1546 	return NOTIFY_DONE;
1547 }
1548 
1549 
1550 static struct notifier_block ip_mr_notifier = {
1551 	.notifier_call = ipmr_device_event,
1552 };
1553 
1554 /*
1555  *	Encapsulate a packet by attaching a valid IPIP header to it.
1556  *	This avoids tunnel drivers and other mess and gives us the speed so
1557  *	important for multicast video.
1558  */
1559 
1560 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1561 {
1562 	struct iphdr *iph;
1563 	const struct iphdr *old_iph = ip_hdr(skb);
1564 
1565 	skb_push(skb, sizeof(struct iphdr));
1566 	skb->transport_header = skb->network_header;
1567 	skb_reset_network_header(skb);
1568 	iph = ip_hdr(skb);
1569 
1570 	iph->version	=	4;
1571 	iph->tos	=	old_iph->tos;
1572 	iph->ttl	=	old_iph->ttl;
1573 	iph->frag_off	=	0;
1574 	iph->daddr	=	daddr;
1575 	iph->saddr	=	saddr;
1576 	iph->protocol	=	IPPROTO_IPIP;
1577 	iph->ihl	=	5;
1578 	iph->tot_len	=	htons(skb->len);
1579 	ip_select_ident(iph, skb_dst(skb), NULL);
1580 	ip_send_check(iph);
1581 
1582 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1583 	nf_reset(skb);
1584 }
1585 
1586 static inline int ipmr_forward_finish(struct sk_buff *skb)
1587 {
1588 	struct ip_options *opt = &(IPCB(skb)->opt);
1589 
1590 	IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1591 	IP_ADD_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTOCTETS, skb->len);
1592 
1593 	if (unlikely(opt->optlen))
1594 		ip_forward_options(skb);
1595 
1596 	return dst_output(skb);
1597 }
1598 
1599 /*
1600  *	Processing handlers for ipmr_forward
1601  */
1602 
1603 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1604 			    struct sk_buff *skb, struct mfc_cache *c, int vifi)
1605 {
1606 	const struct iphdr *iph = ip_hdr(skb);
1607 	struct vif_device *vif = &mrt->vif_table[vifi];
1608 	struct net_device *dev;
1609 	struct rtable *rt;
1610 	struct flowi4 fl4;
1611 	int    encap = 0;
1612 
1613 	if (vif->dev == NULL)
1614 		goto out_free;
1615 
1616 #ifdef CONFIG_IP_PIMSM
1617 	if (vif->flags & VIFF_REGISTER) {
1618 		vif->pkt_out++;
1619 		vif->bytes_out += skb->len;
1620 		vif->dev->stats.tx_bytes += skb->len;
1621 		vif->dev->stats.tx_packets++;
1622 		ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1623 		goto out_free;
1624 	}
1625 #endif
1626 
1627 	if (vif->flags & VIFF_TUNNEL) {
1628 		rt = ip_route_output_ports(net, &fl4, NULL,
1629 					   vif->remote, vif->local,
1630 					   0, 0,
1631 					   IPPROTO_IPIP,
1632 					   RT_TOS(iph->tos), vif->link);
1633 		if (IS_ERR(rt))
1634 			goto out_free;
1635 		encap = sizeof(struct iphdr);
1636 	} else {
1637 		rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1638 					   0, 0,
1639 					   IPPROTO_IPIP,
1640 					   RT_TOS(iph->tos), vif->link);
1641 		if (IS_ERR(rt))
1642 			goto out_free;
1643 	}
1644 
1645 	dev = rt->dst.dev;
1646 
1647 	if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1648 		/* Do not fragment multicasts. Alas, IPv4 does not
1649 		 * allow to send ICMP, so that packets will disappear
1650 		 * to blackhole.
1651 		 */
1652 
1653 		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1654 		ip_rt_put(rt);
1655 		goto out_free;
1656 	}
1657 
1658 	encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1659 
1660 	if (skb_cow(skb, encap)) {
1661 		ip_rt_put(rt);
1662 		goto out_free;
1663 	}
1664 
1665 	vif->pkt_out++;
1666 	vif->bytes_out += skb->len;
1667 
1668 	skb_dst_drop(skb);
1669 	skb_dst_set(skb, &rt->dst);
1670 	ip_decrease_ttl(ip_hdr(skb));
1671 
1672 	/* FIXME: forward and output firewalls used to be called here.
1673 	 * What do we do with netfilter? -- RR
1674 	 */
1675 	if (vif->flags & VIFF_TUNNEL) {
1676 		ip_encap(skb, vif->local, vif->remote);
1677 		/* FIXME: extra output firewall step used to be here. --RR */
1678 		vif->dev->stats.tx_packets++;
1679 		vif->dev->stats.tx_bytes += skb->len;
1680 	}
1681 
1682 	IPCB(skb)->flags |= IPSKB_FORWARDED;
1683 
1684 	/*
1685 	 * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1686 	 * not only before forwarding, but after forwarding on all output
1687 	 * interfaces. It is clear, if mrouter runs a multicasting
1688 	 * program, it should receive packets not depending to what interface
1689 	 * program is joined.
1690 	 * If we will not make it, the program will have to join on all
1691 	 * interfaces. On the other hand, multihoming host (or router, but
1692 	 * not mrouter) cannot join to more than one interface - it will
1693 	 * result in receiving multiple packets.
1694 	 */
1695 	NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1696 		ipmr_forward_finish);
1697 	return;
1698 
1699 out_free:
1700 	kfree_skb(skb);
1701 }
1702 
1703 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1704 {
1705 	int ct;
1706 
1707 	for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1708 		if (mrt->vif_table[ct].dev == dev)
1709 			break;
1710 	}
1711 	return ct;
1712 }
1713 
1714 /* "local" means that we should preserve one skb (for local delivery) */
1715 
1716 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1717 			 struct sk_buff *skb, struct mfc_cache *cache,
1718 			 int local)
1719 {
1720 	int psend = -1;
1721 	int vif, ct;
1722 
1723 	vif = cache->mfc_parent;
1724 	cache->mfc_un.res.pkt++;
1725 	cache->mfc_un.res.bytes += skb->len;
1726 
1727 	/*
1728 	 * Wrong interface: drop packet and (maybe) send PIM assert.
1729 	 */
1730 	if (mrt->vif_table[vif].dev != skb->dev) {
1731 		int true_vifi;
1732 
1733 		if (rt_is_output_route(skb_rtable(skb))) {
1734 			/* It is our own packet, looped back.
1735 			 * Very complicated situation...
1736 			 *
1737 			 * The best workaround until routing daemons will be
1738 			 * fixed is not to redistribute packet, if it was
1739 			 * send through wrong interface. It means, that
1740 			 * multicast applications WILL NOT work for
1741 			 * (S,G), which have default multicast route pointing
1742 			 * to wrong oif. In any case, it is not a good
1743 			 * idea to use multicasting applications on router.
1744 			 */
1745 			goto dont_forward;
1746 		}
1747 
1748 		cache->mfc_un.res.wrong_if++;
1749 		true_vifi = ipmr_find_vif(mrt, skb->dev);
1750 
1751 		if (true_vifi >= 0 && mrt->mroute_do_assert &&
1752 		    /* pimsm uses asserts, when switching from RPT to SPT,
1753 		     * so that we cannot check that packet arrived on an oif.
1754 		     * It is bad, but otherwise we would need to move pretty
1755 		     * large chunk of pimd to kernel. Ough... --ANK
1756 		     */
1757 		    (mrt->mroute_do_pim ||
1758 		     cache->mfc_un.res.ttls[true_vifi] < 255) &&
1759 		    time_after(jiffies,
1760 			       cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1761 			cache->mfc_un.res.last_assert = jiffies;
1762 			ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1763 		}
1764 		goto dont_forward;
1765 	}
1766 
1767 	mrt->vif_table[vif].pkt_in++;
1768 	mrt->vif_table[vif].bytes_in += skb->len;
1769 
1770 	/*
1771 	 *	Forward the frame
1772 	 */
1773 	for (ct = cache->mfc_un.res.maxvif - 1;
1774 	     ct >= cache->mfc_un.res.minvif; ct--) {
1775 		if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1776 			if (psend != -1) {
1777 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1778 
1779 				if (skb2)
1780 					ipmr_queue_xmit(net, mrt, skb2, cache,
1781 							psend);
1782 			}
1783 			psend = ct;
1784 		}
1785 	}
1786 	if (psend != -1) {
1787 		if (local) {
1788 			struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1789 
1790 			if (skb2)
1791 				ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1792 		} else {
1793 			ipmr_queue_xmit(net, mrt, skb, cache, psend);
1794 			return 0;
1795 		}
1796 	}
1797 
1798 dont_forward:
1799 	if (!local)
1800 		kfree_skb(skb);
1801 	return 0;
1802 }
1803 
1804 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1805 {
1806 	struct rtable *rt = skb_rtable(skb);
1807 	struct iphdr *iph = ip_hdr(skb);
1808 	struct flowi4 fl4 = {
1809 		.daddr = iph->daddr,
1810 		.saddr = iph->saddr,
1811 		.flowi4_tos = RT_TOS(iph->tos),
1812 		.flowi4_oif = (rt_is_output_route(rt) ?
1813 			       skb->dev->ifindex : 0),
1814 		.flowi4_iif = (rt_is_output_route(rt) ?
1815 			       LOOPBACK_IFINDEX :
1816 			       skb->dev->ifindex),
1817 		.flowi4_mark = skb->mark,
1818 	};
1819 	struct mr_table *mrt;
1820 	int err;
1821 
1822 	err = ipmr_fib_lookup(net, &fl4, &mrt);
1823 	if (err)
1824 		return ERR_PTR(err);
1825 	return mrt;
1826 }
1827 
1828 /*
1829  *	Multicast packets for forwarding arrive here
1830  *	Called with rcu_read_lock();
1831  */
1832 
1833 int ip_mr_input(struct sk_buff *skb)
1834 {
1835 	struct mfc_cache *cache;
1836 	struct net *net = dev_net(skb->dev);
1837 	int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1838 	struct mr_table *mrt;
1839 
1840 	/* Packet is looped back after forward, it should not be
1841 	 * forwarded second time, but still can be delivered locally.
1842 	 */
1843 	if (IPCB(skb)->flags & IPSKB_FORWARDED)
1844 		goto dont_forward;
1845 
1846 	mrt = ipmr_rt_fib_lookup(net, skb);
1847 	if (IS_ERR(mrt)) {
1848 		kfree_skb(skb);
1849 		return PTR_ERR(mrt);
1850 	}
1851 	if (!local) {
1852 		if (IPCB(skb)->opt.router_alert) {
1853 			if (ip_call_ra_chain(skb))
1854 				return 0;
1855 		} else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1856 			/* IGMPv1 (and broken IGMPv2 implementations sort of
1857 			 * Cisco IOS <= 11.2(8)) do not put router alert
1858 			 * option to IGMP packets destined to routable
1859 			 * groups. It is very bad, because it means
1860 			 * that we can forward NO IGMP messages.
1861 			 */
1862 			struct sock *mroute_sk;
1863 
1864 			mroute_sk = rcu_dereference(mrt->mroute_sk);
1865 			if (mroute_sk) {
1866 				nf_reset(skb);
1867 				raw_rcv(mroute_sk, skb);
1868 				return 0;
1869 			}
1870 		    }
1871 	}
1872 
1873 	/* already under rcu_read_lock() */
1874 	cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1875 
1876 	/*
1877 	 *	No usable cache entry
1878 	 */
1879 	if (cache == NULL) {
1880 		int vif;
1881 
1882 		if (local) {
1883 			struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1884 			ip_local_deliver(skb);
1885 			if (skb2 == NULL)
1886 				return -ENOBUFS;
1887 			skb = skb2;
1888 		}
1889 
1890 		read_lock(&mrt_lock);
1891 		vif = ipmr_find_vif(mrt, skb->dev);
1892 		if (vif >= 0) {
1893 			int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1894 			read_unlock(&mrt_lock);
1895 
1896 			return err2;
1897 		}
1898 		read_unlock(&mrt_lock);
1899 		kfree_skb(skb);
1900 		return -ENODEV;
1901 	}
1902 
1903 	read_lock(&mrt_lock);
1904 	ip_mr_forward(net, mrt, skb, cache, local);
1905 	read_unlock(&mrt_lock);
1906 
1907 	if (local)
1908 		return ip_local_deliver(skb);
1909 
1910 	return 0;
1911 
1912 dont_forward:
1913 	if (local)
1914 		return ip_local_deliver(skb);
1915 	kfree_skb(skb);
1916 	return 0;
1917 }
1918 
1919 #ifdef CONFIG_IP_PIMSM
1920 /* called with rcu_read_lock() */
1921 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1922 		     unsigned int pimlen)
1923 {
1924 	struct net_device *reg_dev = NULL;
1925 	struct iphdr *encap;
1926 
1927 	encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1928 	/*
1929 	 * Check that:
1930 	 * a. packet is really sent to a multicast group
1931 	 * b. packet is not a NULL-REGISTER
1932 	 * c. packet is not truncated
1933 	 */
1934 	if (!ipv4_is_multicast(encap->daddr) ||
1935 	    encap->tot_len == 0 ||
1936 	    ntohs(encap->tot_len) + pimlen > skb->len)
1937 		return 1;
1938 
1939 	read_lock(&mrt_lock);
1940 	if (mrt->mroute_reg_vif_num >= 0)
1941 		reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1942 	read_unlock(&mrt_lock);
1943 
1944 	if (reg_dev == NULL)
1945 		return 1;
1946 
1947 	skb->mac_header = skb->network_header;
1948 	skb_pull(skb, (u8 *)encap - skb->data);
1949 	skb_reset_network_header(skb);
1950 	skb->protocol = htons(ETH_P_IP);
1951 	skb->ip_summed = CHECKSUM_NONE;
1952 	skb->pkt_type = PACKET_HOST;
1953 
1954 	skb_tunnel_rx(skb, reg_dev);
1955 
1956 	netif_rx(skb);
1957 
1958 	return NET_RX_SUCCESS;
1959 }
1960 #endif
1961 
1962 #ifdef CONFIG_IP_PIMSM_V1
1963 /*
1964  * Handle IGMP messages of PIMv1
1965  */
1966 
1967 int pim_rcv_v1(struct sk_buff *skb)
1968 {
1969 	struct igmphdr *pim;
1970 	struct net *net = dev_net(skb->dev);
1971 	struct mr_table *mrt;
1972 
1973 	if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1974 		goto drop;
1975 
1976 	pim = igmp_hdr(skb);
1977 
1978 	mrt = ipmr_rt_fib_lookup(net, skb);
1979 	if (IS_ERR(mrt))
1980 		goto drop;
1981 	if (!mrt->mroute_do_pim ||
1982 	    pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
1983 		goto drop;
1984 
1985 	if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1986 drop:
1987 		kfree_skb(skb);
1988 	}
1989 	return 0;
1990 }
1991 #endif
1992 
1993 #ifdef CONFIG_IP_PIMSM_V2
1994 static int pim_rcv(struct sk_buff *skb)
1995 {
1996 	struct pimreghdr *pim;
1997 	struct net *net = dev_net(skb->dev);
1998 	struct mr_table *mrt;
1999 
2000 	if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2001 		goto drop;
2002 
2003 	pim = (struct pimreghdr *)skb_transport_header(skb);
2004 	if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2005 	    (pim->flags & PIM_NULL_REGISTER) ||
2006 	    (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2007 	     csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2008 		goto drop;
2009 
2010 	mrt = ipmr_rt_fib_lookup(net, skb);
2011 	if (IS_ERR(mrt))
2012 		goto drop;
2013 	if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2014 drop:
2015 		kfree_skb(skb);
2016 	}
2017 	return 0;
2018 }
2019 #endif
2020 
2021 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2022 			      struct mfc_cache *c, struct rtmsg *rtm)
2023 {
2024 	int ct;
2025 	struct rtnexthop *nhp;
2026 	struct nlattr *mp_attr;
2027 
2028 	/* If cache is unresolved, don't try to parse IIF and OIF */
2029 	if (c->mfc_parent >= MAXVIFS)
2030 		return -ENOENT;
2031 
2032 	if (VIF_EXISTS(mrt, c->mfc_parent) &&
2033 	    nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2034 		return -EMSGSIZE;
2035 
2036 	if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2037 		return -EMSGSIZE;
2038 
2039 	for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2040 		if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2041 			if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2042 				nla_nest_cancel(skb, mp_attr);
2043 				return -EMSGSIZE;
2044 			}
2045 
2046 			nhp->rtnh_flags = 0;
2047 			nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2048 			nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2049 			nhp->rtnh_len = sizeof(*nhp);
2050 		}
2051 	}
2052 
2053 	nla_nest_end(skb, mp_attr);
2054 
2055 	rtm->rtm_type = RTN_MULTICAST;
2056 	return 1;
2057 }
2058 
2059 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2060 		   __be32 saddr, __be32 daddr,
2061 		   struct rtmsg *rtm, int nowait)
2062 {
2063 	struct mfc_cache *cache;
2064 	struct mr_table *mrt;
2065 	int err;
2066 
2067 	mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2068 	if (mrt == NULL)
2069 		return -ENOENT;
2070 
2071 	rcu_read_lock();
2072 	cache = ipmr_cache_find(mrt, saddr, daddr);
2073 
2074 	if (cache == NULL) {
2075 		struct sk_buff *skb2;
2076 		struct iphdr *iph;
2077 		struct net_device *dev;
2078 		int vif = -1;
2079 
2080 		if (nowait) {
2081 			rcu_read_unlock();
2082 			return -EAGAIN;
2083 		}
2084 
2085 		dev = skb->dev;
2086 		read_lock(&mrt_lock);
2087 		if (dev)
2088 			vif = ipmr_find_vif(mrt, dev);
2089 		if (vif < 0) {
2090 			read_unlock(&mrt_lock);
2091 			rcu_read_unlock();
2092 			return -ENODEV;
2093 		}
2094 		skb2 = skb_clone(skb, GFP_ATOMIC);
2095 		if (!skb2) {
2096 			read_unlock(&mrt_lock);
2097 			rcu_read_unlock();
2098 			return -ENOMEM;
2099 		}
2100 
2101 		skb_push(skb2, sizeof(struct iphdr));
2102 		skb_reset_network_header(skb2);
2103 		iph = ip_hdr(skb2);
2104 		iph->ihl = sizeof(struct iphdr) >> 2;
2105 		iph->saddr = saddr;
2106 		iph->daddr = daddr;
2107 		iph->version = 0;
2108 		err = ipmr_cache_unresolved(mrt, vif, skb2);
2109 		read_unlock(&mrt_lock);
2110 		rcu_read_unlock();
2111 		return err;
2112 	}
2113 
2114 	read_lock(&mrt_lock);
2115 	if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2116 		cache->mfc_flags |= MFC_NOTIFY;
2117 	err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2118 	read_unlock(&mrt_lock);
2119 	rcu_read_unlock();
2120 	return err;
2121 }
2122 
2123 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2124 			    u32 portid, u32 seq, struct mfc_cache *c)
2125 {
2126 	struct nlmsghdr *nlh;
2127 	struct rtmsg *rtm;
2128 
2129 	nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*rtm), NLM_F_MULTI);
2130 	if (nlh == NULL)
2131 		return -EMSGSIZE;
2132 
2133 	rtm = nlmsg_data(nlh);
2134 	rtm->rtm_family   = RTNL_FAMILY_IPMR;
2135 	rtm->rtm_dst_len  = 32;
2136 	rtm->rtm_src_len  = 32;
2137 	rtm->rtm_tos      = 0;
2138 	rtm->rtm_table    = mrt->id;
2139 	if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2140 		goto nla_put_failure;
2141 	rtm->rtm_type     = RTN_MULTICAST;
2142 	rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2143 	rtm->rtm_protocol = RTPROT_UNSPEC;
2144 	rtm->rtm_flags    = 0;
2145 
2146 	if (nla_put_be32(skb, RTA_SRC, c->mfc_origin) ||
2147 	    nla_put_be32(skb, RTA_DST, c->mfc_mcastgrp))
2148 		goto nla_put_failure;
2149 	if (__ipmr_fill_mroute(mrt, skb, c, rtm) < 0)
2150 		goto nla_put_failure;
2151 
2152 	return nlmsg_end(skb, nlh);
2153 
2154 nla_put_failure:
2155 	nlmsg_cancel(skb, nlh);
2156 	return -EMSGSIZE;
2157 }
2158 
2159 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2160 {
2161 	struct net *net = sock_net(skb->sk);
2162 	struct mr_table *mrt;
2163 	struct mfc_cache *mfc;
2164 	unsigned int t = 0, s_t;
2165 	unsigned int h = 0, s_h;
2166 	unsigned int e = 0, s_e;
2167 
2168 	s_t = cb->args[0];
2169 	s_h = cb->args[1];
2170 	s_e = cb->args[2];
2171 
2172 	rcu_read_lock();
2173 	ipmr_for_each_table(mrt, net) {
2174 		if (t < s_t)
2175 			goto next_table;
2176 		if (t > s_t)
2177 			s_h = 0;
2178 		for (h = s_h; h < MFC_LINES; h++) {
2179 			list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2180 				if (e < s_e)
2181 					goto next_entry;
2182 				if (ipmr_fill_mroute(mrt, skb,
2183 						     NETLINK_CB(cb->skb).portid,
2184 						     cb->nlh->nlmsg_seq,
2185 						     mfc) < 0)
2186 					goto done;
2187 next_entry:
2188 				e++;
2189 			}
2190 			e = s_e = 0;
2191 		}
2192 		s_h = 0;
2193 next_table:
2194 		t++;
2195 	}
2196 done:
2197 	rcu_read_unlock();
2198 
2199 	cb->args[2] = e;
2200 	cb->args[1] = h;
2201 	cb->args[0] = t;
2202 
2203 	return skb->len;
2204 }
2205 
2206 #ifdef CONFIG_PROC_FS
2207 /*
2208  *	The /proc interfaces to multicast routing :
2209  *	/proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2210  */
2211 struct ipmr_vif_iter {
2212 	struct seq_net_private p;
2213 	struct mr_table *mrt;
2214 	int ct;
2215 };
2216 
2217 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2218 					   struct ipmr_vif_iter *iter,
2219 					   loff_t pos)
2220 {
2221 	struct mr_table *mrt = iter->mrt;
2222 
2223 	for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2224 		if (!VIF_EXISTS(mrt, iter->ct))
2225 			continue;
2226 		if (pos-- == 0)
2227 			return &mrt->vif_table[iter->ct];
2228 	}
2229 	return NULL;
2230 }
2231 
2232 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2233 	__acquires(mrt_lock)
2234 {
2235 	struct ipmr_vif_iter *iter = seq->private;
2236 	struct net *net = seq_file_net(seq);
2237 	struct mr_table *mrt;
2238 
2239 	mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2240 	if (mrt == NULL)
2241 		return ERR_PTR(-ENOENT);
2242 
2243 	iter->mrt = mrt;
2244 
2245 	read_lock(&mrt_lock);
2246 	return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2247 		: SEQ_START_TOKEN;
2248 }
2249 
2250 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2251 {
2252 	struct ipmr_vif_iter *iter = seq->private;
2253 	struct net *net = seq_file_net(seq);
2254 	struct mr_table *mrt = iter->mrt;
2255 
2256 	++*pos;
2257 	if (v == SEQ_START_TOKEN)
2258 		return ipmr_vif_seq_idx(net, iter, 0);
2259 
2260 	while (++iter->ct < mrt->maxvif) {
2261 		if (!VIF_EXISTS(mrt, iter->ct))
2262 			continue;
2263 		return &mrt->vif_table[iter->ct];
2264 	}
2265 	return NULL;
2266 }
2267 
2268 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2269 	__releases(mrt_lock)
2270 {
2271 	read_unlock(&mrt_lock);
2272 }
2273 
2274 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2275 {
2276 	struct ipmr_vif_iter *iter = seq->private;
2277 	struct mr_table *mrt = iter->mrt;
2278 
2279 	if (v == SEQ_START_TOKEN) {
2280 		seq_puts(seq,
2281 			 "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2282 	} else {
2283 		const struct vif_device *vif = v;
2284 		const char *name =  vif->dev ? vif->dev->name : "none";
2285 
2286 		seq_printf(seq,
2287 			   "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2288 			   vif - mrt->vif_table,
2289 			   name, vif->bytes_in, vif->pkt_in,
2290 			   vif->bytes_out, vif->pkt_out,
2291 			   vif->flags, vif->local, vif->remote);
2292 	}
2293 	return 0;
2294 }
2295 
2296 static const struct seq_operations ipmr_vif_seq_ops = {
2297 	.start = ipmr_vif_seq_start,
2298 	.next  = ipmr_vif_seq_next,
2299 	.stop  = ipmr_vif_seq_stop,
2300 	.show  = ipmr_vif_seq_show,
2301 };
2302 
2303 static int ipmr_vif_open(struct inode *inode, struct file *file)
2304 {
2305 	return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2306 			    sizeof(struct ipmr_vif_iter));
2307 }
2308 
2309 static const struct file_operations ipmr_vif_fops = {
2310 	.owner	 = THIS_MODULE,
2311 	.open    = ipmr_vif_open,
2312 	.read    = seq_read,
2313 	.llseek  = seq_lseek,
2314 	.release = seq_release_net,
2315 };
2316 
2317 struct ipmr_mfc_iter {
2318 	struct seq_net_private p;
2319 	struct mr_table *mrt;
2320 	struct list_head *cache;
2321 	int ct;
2322 };
2323 
2324 
2325 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2326 					  struct ipmr_mfc_iter *it, loff_t pos)
2327 {
2328 	struct mr_table *mrt = it->mrt;
2329 	struct mfc_cache *mfc;
2330 
2331 	rcu_read_lock();
2332 	for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2333 		it->cache = &mrt->mfc_cache_array[it->ct];
2334 		list_for_each_entry_rcu(mfc, it->cache, list)
2335 			if (pos-- == 0)
2336 				return mfc;
2337 	}
2338 	rcu_read_unlock();
2339 
2340 	spin_lock_bh(&mfc_unres_lock);
2341 	it->cache = &mrt->mfc_unres_queue;
2342 	list_for_each_entry(mfc, it->cache, list)
2343 		if (pos-- == 0)
2344 			return mfc;
2345 	spin_unlock_bh(&mfc_unres_lock);
2346 
2347 	it->cache = NULL;
2348 	return NULL;
2349 }
2350 
2351 
2352 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2353 {
2354 	struct ipmr_mfc_iter *it = seq->private;
2355 	struct net *net = seq_file_net(seq);
2356 	struct mr_table *mrt;
2357 
2358 	mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2359 	if (mrt == NULL)
2360 		return ERR_PTR(-ENOENT);
2361 
2362 	it->mrt = mrt;
2363 	it->cache = NULL;
2364 	it->ct = 0;
2365 	return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2366 		: SEQ_START_TOKEN;
2367 }
2368 
2369 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2370 {
2371 	struct mfc_cache *mfc = v;
2372 	struct ipmr_mfc_iter *it = seq->private;
2373 	struct net *net = seq_file_net(seq);
2374 	struct mr_table *mrt = it->mrt;
2375 
2376 	++*pos;
2377 
2378 	if (v == SEQ_START_TOKEN)
2379 		return ipmr_mfc_seq_idx(net, seq->private, 0);
2380 
2381 	if (mfc->list.next != it->cache)
2382 		return list_entry(mfc->list.next, struct mfc_cache, list);
2383 
2384 	if (it->cache == &mrt->mfc_unres_queue)
2385 		goto end_of_list;
2386 
2387 	BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2388 
2389 	while (++it->ct < MFC_LINES) {
2390 		it->cache = &mrt->mfc_cache_array[it->ct];
2391 		if (list_empty(it->cache))
2392 			continue;
2393 		return list_first_entry(it->cache, struct mfc_cache, list);
2394 	}
2395 
2396 	/* exhausted cache_array, show unresolved */
2397 	rcu_read_unlock();
2398 	it->cache = &mrt->mfc_unres_queue;
2399 	it->ct = 0;
2400 
2401 	spin_lock_bh(&mfc_unres_lock);
2402 	if (!list_empty(it->cache))
2403 		return list_first_entry(it->cache, struct mfc_cache, list);
2404 
2405 end_of_list:
2406 	spin_unlock_bh(&mfc_unres_lock);
2407 	it->cache = NULL;
2408 
2409 	return NULL;
2410 }
2411 
2412 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2413 {
2414 	struct ipmr_mfc_iter *it = seq->private;
2415 	struct mr_table *mrt = it->mrt;
2416 
2417 	if (it->cache == &mrt->mfc_unres_queue)
2418 		spin_unlock_bh(&mfc_unres_lock);
2419 	else if (it->cache == &mrt->mfc_cache_array[it->ct])
2420 		rcu_read_unlock();
2421 }
2422 
2423 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2424 {
2425 	int n;
2426 
2427 	if (v == SEQ_START_TOKEN) {
2428 		seq_puts(seq,
2429 		 "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2430 	} else {
2431 		const struct mfc_cache *mfc = v;
2432 		const struct ipmr_mfc_iter *it = seq->private;
2433 		const struct mr_table *mrt = it->mrt;
2434 
2435 		seq_printf(seq, "%08X %08X %-3hd",
2436 			   (__force u32) mfc->mfc_mcastgrp,
2437 			   (__force u32) mfc->mfc_origin,
2438 			   mfc->mfc_parent);
2439 
2440 		if (it->cache != &mrt->mfc_unres_queue) {
2441 			seq_printf(seq, " %8lu %8lu %8lu",
2442 				   mfc->mfc_un.res.pkt,
2443 				   mfc->mfc_un.res.bytes,
2444 				   mfc->mfc_un.res.wrong_if);
2445 			for (n = mfc->mfc_un.res.minvif;
2446 			     n < mfc->mfc_un.res.maxvif; n++) {
2447 				if (VIF_EXISTS(mrt, n) &&
2448 				    mfc->mfc_un.res.ttls[n] < 255)
2449 					seq_printf(seq,
2450 					   " %2d:%-3d",
2451 					   n, mfc->mfc_un.res.ttls[n]);
2452 			}
2453 		} else {
2454 			/* unresolved mfc_caches don't contain
2455 			 * pkt, bytes and wrong_if values
2456 			 */
2457 			seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2458 		}
2459 		seq_putc(seq, '\n');
2460 	}
2461 	return 0;
2462 }
2463 
2464 static const struct seq_operations ipmr_mfc_seq_ops = {
2465 	.start = ipmr_mfc_seq_start,
2466 	.next  = ipmr_mfc_seq_next,
2467 	.stop  = ipmr_mfc_seq_stop,
2468 	.show  = ipmr_mfc_seq_show,
2469 };
2470 
2471 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2472 {
2473 	return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2474 			    sizeof(struct ipmr_mfc_iter));
2475 }
2476 
2477 static const struct file_operations ipmr_mfc_fops = {
2478 	.owner	 = THIS_MODULE,
2479 	.open    = ipmr_mfc_open,
2480 	.read    = seq_read,
2481 	.llseek  = seq_lseek,
2482 	.release = seq_release_net,
2483 };
2484 #endif
2485 
2486 #ifdef CONFIG_IP_PIMSM_V2
2487 static const struct net_protocol pim_protocol = {
2488 	.handler	=	pim_rcv,
2489 	.netns_ok	=	1,
2490 };
2491 #endif
2492 
2493 
2494 /*
2495  *	Setup for IP multicast routing
2496  */
2497 static int __net_init ipmr_net_init(struct net *net)
2498 {
2499 	int err;
2500 
2501 	err = ipmr_rules_init(net);
2502 	if (err < 0)
2503 		goto fail;
2504 
2505 #ifdef CONFIG_PROC_FS
2506 	err = -ENOMEM;
2507 	if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2508 		goto proc_vif_fail;
2509 	if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2510 		goto proc_cache_fail;
2511 #endif
2512 	return 0;
2513 
2514 #ifdef CONFIG_PROC_FS
2515 proc_cache_fail:
2516 	proc_net_remove(net, "ip_mr_vif");
2517 proc_vif_fail:
2518 	ipmr_rules_exit(net);
2519 #endif
2520 fail:
2521 	return err;
2522 }
2523 
2524 static void __net_exit ipmr_net_exit(struct net *net)
2525 {
2526 #ifdef CONFIG_PROC_FS
2527 	proc_net_remove(net, "ip_mr_cache");
2528 	proc_net_remove(net, "ip_mr_vif");
2529 #endif
2530 	ipmr_rules_exit(net);
2531 }
2532 
2533 static struct pernet_operations ipmr_net_ops = {
2534 	.init = ipmr_net_init,
2535 	.exit = ipmr_net_exit,
2536 };
2537 
2538 int __init ip_mr_init(void)
2539 {
2540 	int err;
2541 
2542 	mrt_cachep = kmem_cache_create("ip_mrt_cache",
2543 				       sizeof(struct mfc_cache),
2544 				       0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2545 				       NULL);
2546 	if (!mrt_cachep)
2547 		return -ENOMEM;
2548 
2549 	err = register_pernet_subsys(&ipmr_net_ops);
2550 	if (err)
2551 		goto reg_pernet_fail;
2552 
2553 	err = register_netdevice_notifier(&ip_mr_notifier);
2554 	if (err)
2555 		goto reg_notif_fail;
2556 #ifdef CONFIG_IP_PIMSM_V2
2557 	if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2558 		pr_err("%s: can't add PIM protocol\n", __func__);
2559 		err = -EAGAIN;
2560 		goto add_proto_fail;
2561 	}
2562 #endif
2563 	rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2564 		      NULL, ipmr_rtm_dumproute, NULL);
2565 	return 0;
2566 
2567 #ifdef CONFIG_IP_PIMSM_V2
2568 add_proto_fail:
2569 	unregister_netdevice_notifier(&ip_mr_notifier);
2570 #endif
2571 reg_notif_fail:
2572 	unregister_pernet_subsys(&ipmr_net_ops);
2573 reg_pernet_fail:
2574 	kmem_cache_destroy(mrt_cachep);
2575 	return err;
2576 }
2577