ef2b0089e0ea3ffad352f4ec1c3ea29ea7a42704
[linux-3.10.git] / net / ipv4 / ipmr.c
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/system.h>
30 #include <asm/uaccess.h>
31 #include <linux/types.h>
32 #include <linux/capability.h>
33 #include <linux/errno.h>
34 #include <linux/timer.h>
35 #include <linux/mm.h>
36 #include <linux/kernel.h>
37 #include <linux/fcntl.h>
38 #include <linux/stat.h>
39 #include <linux/socket.h>
40 #include <linux/in.h>
41 #include <linux/inet.h>
42 #include <linux/netdevice.h>
43 #include <linux/inetdevice.h>
44 #include <linux/igmp.h>
45 #include <linux/proc_fs.h>
46 #include <linux/seq_file.h>
47 #include <linux/mroute.h>
48 #include <linux/init.h>
49 #include <linux/if_ether.h>
50 #include <linux/slab.h>
51 #include <net/net_namespace.h>
52 #include <net/ip.h>
53 #include <net/protocol.h>
54 #include <linux/skbuff.h>
55 #include <net/route.h>
56 #include <net/sock.h>
57 #include <net/icmp.h>
58 #include <net/udp.h>
59 #include <net/raw.h>
60 #include <linux/notifier.h>
61 #include <linux/if_arp.h>
62 #include <linux/netfilter_ipv4.h>
63 #include <net/ipip.h>
64 #include <net/checksum.h>
65 #include <net/netlink.h>
66 #include <net/fib_rules.h>
67
68 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
69 #define CONFIG_IP_PIMSM 1
70 #endif
71
72 struct mr_table {
73         struct list_head        list;
74 #ifdef CONFIG_NET_NS
75         struct net              *net;
76 #endif
77         u32                     id;
78         struct sock __rcu       *mroute_sk;
79         struct timer_list       ipmr_expire_timer;
80         struct list_head        mfc_unres_queue;
81         struct list_head        mfc_cache_array[MFC_LINES];
82         struct vif_device       vif_table[MAXVIFS];
83         int                     maxvif;
84         atomic_t                cache_resolve_queue_len;
85         int                     mroute_do_assert;
86         int                     mroute_do_pim;
87 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
88         int                     mroute_reg_vif_num;
89 #endif
90 };
91
92 struct ipmr_rule {
93         struct fib_rule         common;
94 };
95
96 struct ipmr_result {
97         struct mr_table         *mrt;
98 };
99
100 /* Big lock, protecting vif table, mrt cache and mroute socket state.
101  * Note that the changes are semaphored via rtnl_lock.
102  */
103
104 static DEFINE_RWLOCK(mrt_lock);
105
106 /*
107  *      Multicast router control variables
108  */
109
110 #define VIF_EXISTS(_mrt, _idx) ((_mrt)->vif_table[_idx].dev != NULL)
111
112 /* Special spinlock for queue of unresolved entries */
113 static DEFINE_SPINLOCK(mfc_unres_lock);
114
115 /* We return to original Alan's scheme. Hash table of resolved
116  * entries is changed only in process context and protected
117  * with weak lock mrt_lock. Queue of unresolved entries is protected
118  * with strong spinlock mfc_unres_lock.
119  *
120  * In this case data path is free of exclusive locks at all.
121  */
122
123 static struct kmem_cache *mrt_cachep __read_mostly;
124
125 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
126 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
127                          struct sk_buff *skb, struct mfc_cache *cache,
128                          int local);
129 static int ipmr_cache_report(struct mr_table *mrt,
130                              struct sk_buff *pkt, vifi_t vifi, int assert);
131 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
132                               struct mfc_cache *c, struct rtmsg *rtm);
133 static void ipmr_expire_process(unsigned long arg);
134
135 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
136 #define ipmr_for_each_table(mrt, net) \
137         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
138
139 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
140 {
141         struct mr_table *mrt;
142
143         ipmr_for_each_table(mrt, net) {
144                 if (mrt->id == id)
145                         return mrt;
146         }
147         return NULL;
148 }
149
150 static int ipmr_fib_lookup(struct net *net, struct flowi *flp,
151                            struct mr_table **mrt)
152 {
153         struct ipmr_result res;
154         struct fib_lookup_arg arg = { .result = &res, };
155         int err;
156
157         err = fib_rules_lookup(net->ipv4.mr_rules_ops, flp, 0, &arg);
158         if (err < 0)
159                 return err;
160         *mrt = res.mrt;
161         return 0;
162 }
163
164 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
165                             int flags, struct fib_lookup_arg *arg)
166 {
167         struct ipmr_result *res = arg->result;
168         struct mr_table *mrt;
169
170         switch (rule->action) {
171         case FR_ACT_TO_TBL:
172                 break;
173         case FR_ACT_UNREACHABLE:
174                 return -ENETUNREACH;
175         case FR_ACT_PROHIBIT:
176                 return -EACCES;
177         case FR_ACT_BLACKHOLE:
178         default:
179                 return -EINVAL;
180         }
181
182         mrt = ipmr_get_table(rule->fr_net, rule->table);
183         if (mrt == NULL)
184                 return -EAGAIN;
185         res->mrt = mrt;
186         return 0;
187 }
188
189 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
190 {
191         return 1;
192 }
193
194 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
195         FRA_GENERIC_POLICY,
196 };
197
198 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
199                                struct fib_rule_hdr *frh, struct nlattr **tb)
200 {
201         return 0;
202 }
203
204 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
205                              struct nlattr **tb)
206 {
207         return 1;
208 }
209
210 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
211                           struct fib_rule_hdr *frh)
212 {
213         frh->dst_len = 0;
214         frh->src_len = 0;
215         frh->tos     = 0;
216         return 0;
217 }
218
219 static const struct fib_rules_ops __net_initdata ipmr_rules_ops_template = {
220         .family         = RTNL_FAMILY_IPMR,
221         .rule_size      = sizeof(struct ipmr_rule),
222         .addr_size      = sizeof(u32),
223         .action         = ipmr_rule_action,
224         .match          = ipmr_rule_match,
225         .configure      = ipmr_rule_configure,
226         .compare        = ipmr_rule_compare,
227         .default_pref   = fib_default_rule_pref,
228         .fill           = ipmr_rule_fill,
229         .nlgroup        = RTNLGRP_IPV4_RULE,
230         .policy         = ipmr_rule_policy,
231         .owner          = THIS_MODULE,
232 };
233
234 static int __net_init ipmr_rules_init(struct net *net)
235 {
236         struct fib_rules_ops *ops;
237         struct mr_table *mrt;
238         int err;
239
240         ops = fib_rules_register(&ipmr_rules_ops_template, net);
241         if (IS_ERR(ops))
242                 return PTR_ERR(ops);
243
244         INIT_LIST_HEAD(&net->ipv4.mr_tables);
245
246         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
247         if (mrt == NULL) {
248                 err = -ENOMEM;
249                 goto err1;
250         }
251
252         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
253         if (err < 0)
254                 goto err2;
255
256         net->ipv4.mr_rules_ops = ops;
257         return 0;
258
259 err2:
260         kfree(mrt);
261 err1:
262         fib_rules_unregister(ops);
263         return err;
264 }
265
266 static void __net_exit ipmr_rules_exit(struct net *net)
267 {
268         struct mr_table *mrt, *next;
269
270         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
271                 list_del(&mrt->list);
272                 kfree(mrt);
273         }
274         fib_rules_unregister(net->ipv4.mr_rules_ops);
275 }
276 #else
277 #define ipmr_for_each_table(mrt, net) \
278         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
279
280 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
281 {
282         return net->ipv4.mrt;
283 }
284
285 static int ipmr_fib_lookup(struct net *net, struct flowi *flp,
286                            struct mr_table **mrt)
287 {
288         *mrt = net->ipv4.mrt;
289         return 0;
290 }
291
292 static int __net_init ipmr_rules_init(struct net *net)
293 {
294         net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
295         return net->ipv4.mrt ? 0 : -ENOMEM;
296 }
297
298 static void __net_exit ipmr_rules_exit(struct net *net)
299 {
300         kfree(net->ipv4.mrt);
301 }
302 #endif
303
304 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
305 {
306         struct mr_table *mrt;
307         unsigned int i;
308
309         mrt = ipmr_get_table(net, id);
310         if (mrt != NULL)
311                 return mrt;
312
313         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
314         if (mrt == NULL)
315                 return NULL;
316         write_pnet(&mrt->net, net);
317         mrt->id = id;
318
319         /* Forwarding cache */
320         for (i = 0; i < MFC_LINES; i++)
321                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
322
323         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
324
325         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
326                     (unsigned long)mrt);
327
328 #ifdef CONFIG_IP_PIMSM
329         mrt->mroute_reg_vif_num = -1;
330 #endif
331 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
332         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
333 #endif
334         return mrt;
335 }
336
337 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
338
339 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
340 {
341         struct net *net = dev_net(dev);
342
343         dev_close(dev);
344
345         dev = __dev_get_by_name(net, "tunl0");
346         if (dev) {
347                 const struct net_device_ops *ops = dev->netdev_ops;
348                 struct ifreq ifr;
349                 struct ip_tunnel_parm p;
350
351                 memset(&p, 0, sizeof(p));
352                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
353                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
354                 p.iph.version = 4;
355                 p.iph.ihl = 5;
356                 p.iph.protocol = IPPROTO_IPIP;
357                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
358                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
359
360                 if (ops->ndo_do_ioctl) {
361                         mm_segment_t oldfs = get_fs();
362
363                         set_fs(KERNEL_DS);
364                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
365                         set_fs(oldfs);
366                 }
367         }
368 }
369
370 static
371 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
372 {
373         struct net_device  *dev;
374
375         dev = __dev_get_by_name(net, "tunl0");
376
377         if (dev) {
378                 const struct net_device_ops *ops = dev->netdev_ops;
379                 int err;
380                 struct ifreq ifr;
381                 struct ip_tunnel_parm p;
382                 struct in_device  *in_dev;
383
384                 memset(&p, 0, sizeof(p));
385                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
386                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
387                 p.iph.version = 4;
388                 p.iph.ihl = 5;
389                 p.iph.protocol = IPPROTO_IPIP;
390                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
391                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
392
393                 if (ops->ndo_do_ioctl) {
394                         mm_segment_t oldfs = get_fs();
395
396                         set_fs(KERNEL_DS);
397                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
398                         set_fs(oldfs);
399                 } else {
400                         err = -EOPNOTSUPP;
401                 }
402                 dev = NULL;
403
404                 if (err == 0 &&
405                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
406                         dev->flags |= IFF_MULTICAST;
407
408                         in_dev = __in_dev_get_rtnl(dev);
409                         if (in_dev == NULL)
410                                 goto failure;
411
412                         ipv4_devconf_setall(in_dev);
413                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
414
415                         if (dev_open(dev))
416                                 goto failure;
417                         dev_hold(dev);
418                 }
419         }
420         return dev;
421
422 failure:
423         /* allow the register to be completed before unregistering. */
424         rtnl_unlock();
425         rtnl_lock();
426
427         unregister_netdevice(dev);
428         return NULL;
429 }
430
431 #ifdef CONFIG_IP_PIMSM
432
433 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
434 {
435         struct net *net = dev_net(dev);
436         struct mr_table *mrt;
437         struct flowi fl = {
438                 .oif            = dev->ifindex,
439                 .iif            = skb->skb_iif,
440                 .mark           = skb->mark,
441         };
442         int err;
443
444         err = ipmr_fib_lookup(net, &fl, &mrt);
445         if (err < 0) {
446                 kfree_skb(skb);
447                 return err;
448         }
449
450         read_lock(&mrt_lock);
451         dev->stats.tx_bytes += skb->len;
452         dev->stats.tx_packets++;
453         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
454         read_unlock(&mrt_lock);
455         kfree_skb(skb);
456         return NETDEV_TX_OK;
457 }
458
459 static const struct net_device_ops reg_vif_netdev_ops = {
460         .ndo_start_xmit = reg_vif_xmit,
461 };
462
463 static void reg_vif_setup(struct net_device *dev)
464 {
465         dev->type               = ARPHRD_PIMREG;
466         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
467         dev->flags              = IFF_NOARP;
468         dev->netdev_ops         = &reg_vif_netdev_ops,
469         dev->destructor         = free_netdev;
470         dev->features           |= NETIF_F_NETNS_LOCAL;
471 }
472
473 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
474 {
475         struct net_device *dev;
476         struct in_device *in_dev;
477         char name[IFNAMSIZ];
478
479         if (mrt->id == RT_TABLE_DEFAULT)
480                 sprintf(name, "pimreg");
481         else
482                 sprintf(name, "pimreg%u", mrt->id);
483
484         dev = alloc_netdev(0, name, reg_vif_setup);
485
486         if (dev == NULL)
487                 return NULL;
488
489         dev_net_set(dev, net);
490
491         if (register_netdevice(dev)) {
492                 free_netdev(dev);
493                 return NULL;
494         }
495         dev->iflink = 0;
496
497         rcu_read_lock();
498         in_dev = __in_dev_get_rcu(dev);
499         if (!in_dev) {
500                 rcu_read_unlock();
501                 goto failure;
502         }
503
504         ipv4_devconf_setall(in_dev);
505         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
506         rcu_read_unlock();
507
508         if (dev_open(dev))
509                 goto failure;
510
511         dev_hold(dev);
512
513         return dev;
514
515 failure:
516         /* allow the register to be completed before unregistering. */
517         rtnl_unlock();
518         rtnl_lock();
519
520         unregister_netdevice(dev);
521         return NULL;
522 }
523 #endif
524
525 /*
526  *      Delete a VIF entry
527  *      @notify: Set to 1, if the caller is a notifier_call
528  */
529
530 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
531                       struct list_head *head)
532 {
533         struct vif_device *v;
534         struct net_device *dev;
535         struct in_device *in_dev;
536
537         if (vifi < 0 || vifi >= mrt->maxvif)
538                 return -EADDRNOTAVAIL;
539
540         v = &mrt->vif_table[vifi];
541
542         write_lock_bh(&mrt_lock);
543         dev = v->dev;
544         v->dev = NULL;
545
546         if (!dev) {
547                 write_unlock_bh(&mrt_lock);
548                 return -EADDRNOTAVAIL;
549         }
550
551 #ifdef CONFIG_IP_PIMSM
552         if (vifi == mrt->mroute_reg_vif_num)
553                 mrt->mroute_reg_vif_num = -1;
554 #endif
555
556         if (vifi + 1 == mrt->maxvif) {
557                 int tmp;
558
559                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
560                         if (VIF_EXISTS(mrt, tmp))
561                                 break;
562                 }
563                 mrt->maxvif = tmp+1;
564         }
565
566         write_unlock_bh(&mrt_lock);
567
568         dev_set_allmulti(dev, -1);
569
570         in_dev = __in_dev_get_rtnl(dev);
571         if (in_dev) {
572                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
573                 ip_rt_multicast_event(in_dev);
574         }
575
576         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
577                 unregister_netdevice_queue(dev, head);
578
579         dev_put(dev);
580         return 0;
581 }
582
583 static void ipmr_cache_free_rcu(struct rcu_head *head)
584 {
585         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
586
587         kmem_cache_free(mrt_cachep, c);
588 }
589
590 static inline void ipmr_cache_free(struct mfc_cache *c)
591 {
592         call_rcu(&c->rcu, ipmr_cache_free_rcu);
593 }
594
595 /* Destroy an unresolved cache entry, killing queued skbs
596  * and reporting error to netlink readers.
597  */
598
599 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
600 {
601         struct net *net = read_pnet(&mrt->net);
602         struct sk_buff *skb;
603         struct nlmsgerr *e;
604
605         atomic_dec(&mrt->cache_resolve_queue_len);
606
607         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
608                 if (ip_hdr(skb)->version == 0) {
609                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
610                         nlh->nlmsg_type = NLMSG_ERROR;
611                         nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
612                         skb_trim(skb, nlh->nlmsg_len);
613                         e = NLMSG_DATA(nlh);
614                         e->error = -ETIMEDOUT;
615                         memset(&e->msg, 0, sizeof(e->msg));
616
617                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
618                 } else {
619                         kfree_skb(skb);
620                 }
621         }
622
623         ipmr_cache_free(c);
624 }
625
626
627 /* Timer process for the unresolved queue. */
628
629 static void ipmr_expire_process(unsigned long arg)
630 {
631         struct mr_table *mrt = (struct mr_table *)arg;
632         unsigned long now;
633         unsigned long expires;
634         struct mfc_cache *c, *next;
635
636         if (!spin_trylock(&mfc_unres_lock)) {
637                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
638                 return;
639         }
640
641         if (list_empty(&mrt->mfc_unres_queue))
642                 goto out;
643
644         now = jiffies;
645         expires = 10*HZ;
646
647         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
648                 if (time_after(c->mfc_un.unres.expires, now)) {
649                         unsigned long interval = c->mfc_un.unres.expires - now;
650                         if (interval < expires)
651                                 expires = interval;
652                         continue;
653                 }
654
655                 list_del(&c->list);
656                 ipmr_destroy_unres(mrt, c);
657         }
658
659         if (!list_empty(&mrt->mfc_unres_queue))
660                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
661
662 out:
663         spin_unlock(&mfc_unres_lock);
664 }
665
666 /* Fill oifs list. It is called under write locked mrt_lock. */
667
668 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
669                                    unsigned char *ttls)
670 {
671         int vifi;
672
673         cache->mfc_un.res.minvif = MAXVIFS;
674         cache->mfc_un.res.maxvif = 0;
675         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
676
677         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
678                 if (VIF_EXISTS(mrt, vifi) &&
679                     ttls[vifi] && ttls[vifi] < 255) {
680                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
681                         if (cache->mfc_un.res.minvif > vifi)
682                                 cache->mfc_un.res.minvif = vifi;
683                         if (cache->mfc_un.res.maxvif <= vifi)
684                                 cache->mfc_un.res.maxvif = vifi + 1;
685                 }
686         }
687 }
688
689 static int vif_add(struct net *net, struct mr_table *mrt,
690                    struct vifctl *vifc, int mrtsock)
691 {
692         int vifi = vifc->vifc_vifi;
693         struct vif_device *v = &mrt->vif_table[vifi];
694         struct net_device *dev;
695         struct in_device *in_dev;
696         int err;
697
698         /* Is vif busy ? */
699         if (VIF_EXISTS(mrt, vifi))
700                 return -EADDRINUSE;
701
702         switch (vifc->vifc_flags) {
703 #ifdef CONFIG_IP_PIMSM
704         case VIFF_REGISTER:
705                 /*
706                  * Special Purpose VIF in PIM
707                  * All the packets will be sent to the daemon
708                  */
709                 if (mrt->mroute_reg_vif_num >= 0)
710                         return -EADDRINUSE;
711                 dev = ipmr_reg_vif(net, mrt);
712                 if (!dev)
713                         return -ENOBUFS;
714                 err = dev_set_allmulti(dev, 1);
715                 if (err) {
716                         unregister_netdevice(dev);
717                         dev_put(dev);
718                         return err;
719                 }
720                 break;
721 #endif
722         case VIFF_TUNNEL:
723                 dev = ipmr_new_tunnel(net, vifc);
724                 if (!dev)
725                         return -ENOBUFS;
726                 err = dev_set_allmulti(dev, 1);
727                 if (err) {
728                         ipmr_del_tunnel(dev, vifc);
729                         dev_put(dev);
730                         return err;
731                 }
732                 break;
733
734         case VIFF_USE_IFINDEX:
735         case 0:
736                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
737                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
738                         if (dev && __in_dev_get_rtnl(dev) == NULL) {
739                                 dev_put(dev);
740                                 return -EADDRNOTAVAIL;
741                         }
742                 } else {
743                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
744                 }
745                 if (!dev)
746                         return -EADDRNOTAVAIL;
747                 err = dev_set_allmulti(dev, 1);
748                 if (err) {
749                         dev_put(dev);
750                         return err;
751                 }
752                 break;
753         default:
754                 return -EINVAL;
755         }
756
757         in_dev = __in_dev_get_rtnl(dev);
758         if (!in_dev) {
759                 dev_put(dev);
760                 return -EADDRNOTAVAIL;
761         }
762         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
763         ip_rt_multicast_event(in_dev);
764
765         /* Fill in the VIF structures */
766
767         v->rate_limit = vifc->vifc_rate_limit;
768         v->local = vifc->vifc_lcl_addr.s_addr;
769         v->remote = vifc->vifc_rmt_addr.s_addr;
770         v->flags = vifc->vifc_flags;
771         if (!mrtsock)
772                 v->flags |= VIFF_STATIC;
773         v->threshold = vifc->vifc_threshold;
774         v->bytes_in = 0;
775         v->bytes_out = 0;
776         v->pkt_in = 0;
777         v->pkt_out = 0;
778         v->link = dev->ifindex;
779         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
780                 v->link = dev->iflink;
781
782         /* And finish update writing critical data */
783         write_lock_bh(&mrt_lock);
784         v->dev = dev;
785 #ifdef CONFIG_IP_PIMSM
786         if (v->flags & VIFF_REGISTER)
787                 mrt->mroute_reg_vif_num = vifi;
788 #endif
789         if (vifi+1 > mrt->maxvif)
790                 mrt->maxvif = vifi+1;
791         write_unlock_bh(&mrt_lock);
792         return 0;
793 }
794
795 /* called with rcu_read_lock() */
796 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
797                                          __be32 origin,
798                                          __be32 mcastgrp)
799 {
800         int line = MFC_HASH(mcastgrp, origin);
801         struct mfc_cache *c;
802
803         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
804                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
805                         return c;
806         }
807         return NULL;
808 }
809
810 /*
811  *      Allocate a multicast cache entry
812  */
813 static struct mfc_cache *ipmr_cache_alloc(void)
814 {
815         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
816
817         if (c)
818                 c->mfc_un.res.minvif = MAXVIFS;
819         return c;
820 }
821
822 static struct mfc_cache *ipmr_cache_alloc_unres(void)
823 {
824         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
825
826         if (c) {
827                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
828                 c->mfc_un.unres.expires = jiffies + 10*HZ;
829         }
830         return c;
831 }
832
833 /*
834  *      A cache entry has gone into a resolved state from queued
835  */
836
837 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
838                                struct mfc_cache *uc, struct mfc_cache *c)
839 {
840         struct sk_buff *skb;
841         struct nlmsgerr *e;
842
843         /* Play the pending entries through our router */
844
845         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
846                 if (ip_hdr(skb)->version == 0) {
847                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
848
849                         if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
850                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
851                                                  (u8 *)nlh;
852                         } else {
853                                 nlh->nlmsg_type = NLMSG_ERROR;
854                                 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
855                                 skb_trim(skb, nlh->nlmsg_len);
856                                 e = NLMSG_DATA(nlh);
857                                 e->error = -EMSGSIZE;
858                                 memset(&e->msg, 0, sizeof(e->msg));
859                         }
860
861                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
862                 } else {
863                         ip_mr_forward(net, mrt, skb, c, 0);
864                 }
865         }
866 }
867
868 /*
869  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
870  *      expects the following bizarre scheme.
871  *
872  *      Called under mrt_lock.
873  */
874
875 static int ipmr_cache_report(struct mr_table *mrt,
876                              struct sk_buff *pkt, vifi_t vifi, int assert)
877 {
878         struct sk_buff *skb;
879         const int ihl = ip_hdrlen(pkt);
880         struct igmphdr *igmp;
881         struct igmpmsg *msg;
882         struct sock *mroute_sk;
883         int ret;
884
885 #ifdef CONFIG_IP_PIMSM
886         if (assert == IGMPMSG_WHOLEPKT)
887                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
888         else
889 #endif
890                 skb = alloc_skb(128, GFP_ATOMIC);
891
892         if (!skb)
893                 return -ENOBUFS;
894
895 #ifdef CONFIG_IP_PIMSM
896         if (assert == IGMPMSG_WHOLEPKT) {
897                 /* Ugly, but we have no choice with this interface.
898                  * Duplicate old header, fix ihl, length etc.
899                  * And all this only to mangle msg->im_msgtype and
900                  * to set msg->im_mbz to "mbz" :-)
901                  */
902                 skb_push(skb, sizeof(struct iphdr));
903                 skb_reset_network_header(skb);
904                 skb_reset_transport_header(skb);
905                 msg = (struct igmpmsg *)skb_network_header(skb);
906                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
907                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
908                 msg->im_mbz = 0;
909                 msg->im_vif = mrt->mroute_reg_vif_num;
910                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
911                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
912                                              sizeof(struct iphdr));
913         } else
914 #endif
915         {
916
917         /* Copy the IP header */
918
919         skb->network_header = skb->tail;
920         skb_put(skb, ihl);
921         skb_copy_to_linear_data(skb, pkt->data, ihl);
922         ip_hdr(skb)->protocol = 0;      /* Flag to the kernel this is a route add */
923         msg = (struct igmpmsg *)skb_network_header(skb);
924         msg->im_vif = vifi;
925         skb_dst_set(skb, dst_clone(skb_dst(pkt)));
926
927         /* Add our header */
928
929         igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
930         igmp->type      =
931         msg->im_msgtype = assert;
932         igmp->code      = 0;
933         ip_hdr(skb)->tot_len = htons(skb->len);         /* Fix the length */
934         skb->transport_header = skb->network_header;
935         }
936
937         rcu_read_lock();
938         mroute_sk = rcu_dereference(mrt->mroute_sk);
939         if (mroute_sk == NULL) {
940                 rcu_read_unlock();
941                 kfree_skb(skb);
942                 return -EINVAL;
943         }
944
945         /* Deliver to mrouted */
946
947         ret = sock_queue_rcv_skb(mroute_sk, skb);
948         rcu_read_unlock();
949         if (ret < 0) {
950                 if (net_ratelimit())
951                         printk(KERN_WARNING "mroute: pending queue full, dropping entries.\n");
952                 kfree_skb(skb);
953         }
954
955         return ret;
956 }
957
958 /*
959  *      Queue a packet for resolution. It gets locked cache entry!
960  */
961
962 static int
963 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
964 {
965         bool found = false;
966         int err;
967         struct mfc_cache *c;
968         const struct iphdr *iph = ip_hdr(skb);
969
970         spin_lock_bh(&mfc_unres_lock);
971         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
972                 if (c->mfc_mcastgrp == iph->daddr &&
973                     c->mfc_origin == iph->saddr) {
974                         found = true;
975                         break;
976                 }
977         }
978
979         if (!found) {
980                 /* Create a new entry if allowable */
981
982                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
983                     (c = ipmr_cache_alloc_unres()) == NULL) {
984                         spin_unlock_bh(&mfc_unres_lock);
985
986                         kfree_skb(skb);
987                         return -ENOBUFS;
988                 }
989
990                 /* Fill in the new cache entry */
991
992                 c->mfc_parent   = -1;
993                 c->mfc_origin   = iph->saddr;
994                 c->mfc_mcastgrp = iph->daddr;
995
996                 /* Reflect first query at mrouted. */
997
998                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
999                 if (err < 0) {
1000                         /* If the report failed throw the cache entry
1001                            out - Brad Parker
1002                          */
1003                         spin_unlock_bh(&mfc_unres_lock);
1004
1005                         ipmr_cache_free(c);
1006                         kfree_skb(skb);
1007                         return err;
1008                 }
1009
1010                 atomic_inc(&mrt->cache_resolve_queue_len);
1011                 list_add(&c->list, &mrt->mfc_unres_queue);
1012
1013                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1014                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1015         }
1016
1017         /* See if we can append the packet */
1018
1019         if (c->mfc_un.unres.unresolved.qlen > 3) {
1020                 kfree_skb(skb);
1021                 err = -ENOBUFS;
1022         } else {
1023                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1024                 err = 0;
1025         }
1026
1027         spin_unlock_bh(&mfc_unres_lock);
1028         return err;
1029 }
1030
1031 /*
1032  *      MFC cache manipulation by user space mroute daemon
1033  */
1034
1035 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1036 {
1037         int line;
1038         struct mfc_cache *c, *next;
1039
1040         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1041
1042         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1043                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1044                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1045                         list_del_rcu(&c->list);
1046
1047                         ipmr_cache_free(c);
1048                         return 0;
1049                 }
1050         }
1051         return -ENOENT;
1052 }
1053
1054 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1055                         struct mfcctl *mfc, int mrtsock)
1056 {
1057         bool found = false;
1058         int line;
1059         struct mfc_cache *uc, *c;
1060
1061         if (mfc->mfcc_parent >= MAXVIFS)
1062                 return -ENFILE;
1063
1064         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1065
1066         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1067                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1068                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1069                         found = true;
1070                         break;
1071                 }
1072         }
1073
1074         if (found) {
1075                 write_lock_bh(&mrt_lock);
1076                 c->mfc_parent = mfc->mfcc_parent;
1077                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1078                 if (!mrtsock)
1079                         c->mfc_flags |= MFC_STATIC;
1080                 write_unlock_bh(&mrt_lock);
1081                 return 0;
1082         }
1083
1084         if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1085                 return -EINVAL;
1086
1087         c = ipmr_cache_alloc();
1088         if (c == NULL)
1089                 return -ENOMEM;
1090
1091         c->mfc_origin = mfc->mfcc_origin.s_addr;
1092         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1093         c->mfc_parent = mfc->mfcc_parent;
1094         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1095         if (!mrtsock)
1096                 c->mfc_flags |= MFC_STATIC;
1097
1098         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1099
1100         /*
1101          *      Check to see if we resolved a queued list. If so we
1102          *      need to send on the frames and tidy up.
1103          */
1104         found = false;
1105         spin_lock_bh(&mfc_unres_lock);
1106         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1107                 if (uc->mfc_origin == c->mfc_origin &&
1108                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1109                         list_del(&uc->list);
1110                         atomic_dec(&mrt->cache_resolve_queue_len);
1111                         found = true;
1112                         break;
1113                 }
1114         }
1115         if (list_empty(&mrt->mfc_unres_queue))
1116                 del_timer(&mrt->ipmr_expire_timer);
1117         spin_unlock_bh(&mfc_unres_lock);
1118
1119         if (found) {
1120                 ipmr_cache_resolve(net, mrt, uc, c);
1121                 ipmr_cache_free(uc);
1122         }
1123         return 0;
1124 }
1125
1126 /*
1127  *      Close the multicast socket, and clear the vif tables etc
1128  */
1129
1130 static void mroute_clean_tables(struct mr_table *mrt)
1131 {
1132         int i;
1133         LIST_HEAD(list);
1134         struct mfc_cache *c, *next;
1135
1136         /* Shut down all active vif entries */
1137
1138         for (i = 0; i < mrt->maxvif; i++) {
1139                 if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1140                         vif_delete(mrt, i, 0, &list);
1141         }
1142         unregister_netdevice_many(&list);
1143
1144         /* Wipe the cache */
1145
1146         for (i = 0; i < MFC_LINES; i++) {
1147                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1148                         if (c->mfc_flags & MFC_STATIC)
1149                                 continue;
1150                         list_del_rcu(&c->list);
1151                         ipmr_cache_free(c);
1152                 }
1153         }
1154
1155         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1156                 spin_lock_bh(&mfc_unres_lock);
1157                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1158                         list_del(&c->list);
1159                         ipmr_destroy_unres(mrt, c);
1160                 }
1161                 spin_unlock_bh(&mfc_unres_lock);
1162         }
1163 }
1164
1165 /* called from ip_ra_control(), before an RCU grace period,
1166  * we dont need to call synchronize_rcu() here
1167  */
1168 static void mrtsock_destruct(struct sock *sk)
1169 {
1170         struct net *net = sock_net(sk);
1171         struct mr_table *mrt;
1172
1173         rtnl_lock();
1174         ipmr_for_each_table(mrt, net) {
1175                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1176                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1177                         rcu_assign_pointer(mrt->mroute_sk, NULL);
1178                         mroute_clean_tables(mrt);
1179                 }
1180         }
1181         rtnl_unlock();
1182 }
1183
1184 /*
1185  *      Socket options and virtual interface manipulation. The whole
1186  *      virtual interface system is a complete heap, but unfortunately
1187  *      that's how BSD mrouted happens to think. Maybe one day with a proper
1188  *      MOSPF/PIM router set up we can clean this up.
1189  */
1190
1191 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1192 {
1193         int ret;
1194         struct vifctl vif;
1195         struct mfcctl mfc;
1196         struct net *net = sock_net(sk);
1197         struct mr_table *mrt;
1198
1199         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1200         if (mrt == NULL)
1201                 return -ENOENT;
1202
1203         if (optname != MRT_INIT) {
1204                 if (sk != rcu_dereference_raw(mrt->mroute_sk) &&
1205                     !capable(CAP_NET_ADMIN))
1206                         return -EACCES;
1207         }
1208
1209         switch (optname) {
1210         case MRT_INIT:
1211                 if (sk->sk_type != SOCK_RAW ||
1212                     inet_sk(sk)->inet_num != IPPROTO_IGMP)
1213                         return -EOPNOTSUPP;
1214                 if (optlen != sizeof(int))
1215                         return -ENOPROTOOPT;
1216
1217                 rtnl_lock();
1218                 if (rtnl_dereference(mrt->mroute_sk)) {
1219                         rtnl_unlock();
1220                         return -EADDRINUSE;
1221                 }
1222
1223                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1224                 if (ret == 0) {
1225                         rcu_assign_pointer(mrt->mroute_sk, sk);
1226                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1227                 }
1228                 rtnl_unlock();
1229                 return ret;
1230         case MRT_DONE:
1231                 if (sk != rcu_dereference_raw(mrt->mroute_sk))
1232                         return -EACCES;
1233                 return ip_ra_control(sk, 0, NULL);
1234         case MRT_ADD_VIF:
1235         case MRT_DEL_VIF:
1236                 if (optlen != sizeof(vif))
1237                         return -EINVAL;
1238                 if (copy_from_user(&vif, optval, sizeof(vif)))
1239                         return -EFAULT;
1240                 if (vif.vifc_vifi >= MAXVIFS)
1241                         return -ENFILE;
1242                 rtnl_lock();
1243                 if (optname == MRT_ADD_VIF) {
1244                         ret = vif_add(net, mrt, &vif,
1245                                       sk == rtnl_dereference(mrt->mroute_sk));
1246                 } else {
1247                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1248                 }
1249                 rtnl_unlock();
1250                 return ret;
1251
1252                 /*
1253                  *      Manipulate the forwarding caches. These live
1254                  *      in a sort of kernel/user symbiosis.
1255                  */
1256         case MRT_ADD_MFC:
1257         case MRT_DEL_MFC:
1258                 if (optlen != sizeof(mfc))
1259                         return -EINVAL;
1260                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1261                         return -EFAULT;
1262                 rtnl_lock();
1263                 if (optname == MRT_DEL_MFC)
1264                         ret = ipmr_mfc_delete(mrt, &mfc);
1265                 else
1266                         ret = ipmr_mfc_add(net, mrt, &mfc,
1267                                            sk == rtnl_dereference(mrt->mroute_sk));
1268                 rtnl_unlock();
1269                 return ret;
1270                 /*
1271                  *      Control PIM assert.
1272                  */
1273         case MRT_ASSERT:
1274         {
1275                 int v;
1276                 if (get_user(v, (int __user *)optval))
1277                         return -EFAULT;
1278                 mrt->mroute_do_assert = (v) ? 1 : 0;
1279                 return 0;
1280         }
1281 #ifdef CONFIG_IP_PIMSM
1282         case MRT_PIM:
1283         {
1284                 int v;
1285
1286                 if (get_user(v, (int __user *)optval))
1287                         return -EFAULT;
1288                 v = (v) ? 1 : 0;
1289
1290                 rtnl_lock();
1291                 ret = 0;
1292                 if (v != mrt->mroute_do_pim) {
1293                         mrt->mroute_do_pim = v;
1294                         mrt->mroute_do_assert = v;
1295                 }
1296                 rtnl_unlock();
1297                 return ret;
1298         }
1299 #endif
1300 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1301         case MRT_TABLE:
1302         {
1303                 u32 v;
1304
1305                 if (optlen != sizeof(u32))
1306                         return -EINVAL;
1307                 if (get_user(v, (u32 __user *)optval))
1308                         return -EFAULT;
1309
1310                 rtnl_lock();
1311                 ret = 0;
1312                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1313                         ret = -EBUSY;
1314                 } else {
1315                         if (!ipmr_new_table(net, v))
1316                                 ret = -ENOMEM;
1317                         raw_sk(sk)->ipmr_table = v;
1318                 }
1319                 rtnl_unlock();
1320                 return ret;
1321         }
1322 #endif
1323         /*
1324          *      Spurious command, or MRT_VERSION which you cannot
1325          *      set.
1326          */
1327         default:
1328                 return -ENOPROTOOPT;
1329         }
1330 }
1331
1332 /*
1333  *      Getsock opt support for the multicast routing system.
1334  */
1335
1336 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1337 {
1338         int olr;
1339         int val;
1340         struct net *net = sock_net(sk);
1341         struct mr_table *mrt;
1342
1343         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1344         if (mrt == NULL)
1345                 return -ENOENT;
1346
1347         if (optname != MRT_VERSION &&
1348 #ifdef CONFIG_IP_PIMSM
1349            optname != MRT_PIM &&
1350 #endif
1351            optname != MRT_ASSERT)
1352                 return -ENOPROTOOPT;
1353
1354         if (get_user(olr, optlen))
1355                 return -EFAULT;
1356
1357         olr = min_t(unsigned int, olr, sizeof(int));
1358         if (olr < 0)
1359                 return -EINVAL;
1360
1361         if (put_user(olr, optlen))
1362                 return -EFAULT;
1363         if (optname == MRT_VERSION)
1364                 val = 0x0305;
1365 #ifdef CONFIG_IP_PIMSM
1366         else if (optname == MRT_PIM)
1367                 val = mrt->mroute_do_pim;
1368 #endif
1369         else
1370                 val = mrt->mroute_do_assert;
1371         if (copy_to_user(optval, &val, olr))
1372                 return -EFAULT;
1373         return 0;
1374 }
1375
1376 /*
1377  *      The IP multicast ioctl support routines.
1378  */
1379
1380 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1381 {
1382         struct sioc_sg_req sr;
1383         struct sioc_vif_req vr;
1384         struct vif_device *vif;
1385         struct mfc_cache *c;
1386         struct net *net = sock_net(sk);
1387         struct mr_table *mrt;
1388
1389         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1390         if (mrt == NULL)
1391                 return -ENOENT;
1392
1393         switch (cmd) {
1394         case SIOCGETVIFCNT:
1395                 if (copy_from_user(&vr, arg, sizeof(vr)))
1396                         return -EFAULT;
1397                 if (vr.vifi >= mrt->maxvif)
1398                         return -EINVAL;
1399                 read_lock(&mrt_lock);
1400                 vif = &mrt->vif_table[vr.vifi];
1401                 if (VIF_EXISTS(mrt, vr.vifi)) {
1402                         vr.icount = vif->pkt_in;
1403                         vr.ocount = vif->pkt_out;
1404                         vr.ibytes = vif->bytes_in;
1405                         vr.obytes = vif->bytes_out;
1406                         read_unlock(&mrt_lock);
1407
1408                         if (copy_to_user(arg, &vr, sizeof(vr)))
1409                                 return -EFAULT;
1410                         return 0;
1411                 }
1412                 read_unlock(&mrt_lock);
1413                 return -EADDRNOTAVAIL;
1414         case SIOCGETSGCNT:
1415                 if (copy_from_user(&sr, arg, sizeof(sr)))
1416                         return -EFAULT;
1417
1418                 rcu_read_lock();
1419                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1420                 if (c) {
1421                         sr.pktcnt = c->mfc_un.res.pkt;
1422                         sr.bytecnt = c->mfc_un.res.bytes;
1423                         sr.wrong_if = c->mfc_un.res.wrong_if;
1424                         rcu_read_unlock();
1425
1426                         if (copy_to_user(arg, &sr, sizeof(sr)))
1427                                 return -EFAULT;
1428                         return 0;
1429                 }
1430                 rcu_read_unlock();
1431                 return -EADDRNOTAVAIL;
1432         default:
1433                 return -ENOIOCTLCMD;
1434         }
1435 }
1436
1437
1438 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1439 {
1440         struct net_device *dev = ptr;
1441         struct net *net = dev_net(dev);
1442         struct mr_table *mrt;
1443         struct vif_device *v;
1444         int ct;
1445         LIST_HEAD(list);
1446
1447         if (event != NETDEV_UNREGISTER)
1448                 return NOTIFY_DONE;
1449
1450         ipmr_for_each_table(mrt, net) {
1451                 v = &mrt->vif_table[0];
1452                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1453                         if (v->dev == dev)
1454                                 vif_delete(mrt, ct, 1, &list);
1455                 }
1456         }
1457         unregister_netdevice_many(&list);
1458         return NOTIFY_DONE;
1459 }
1460
1461
1462 static struct notifier_block ip_mr_notifier = {
1463         .notifier_call = ipmr_device_event,
1464 };
1465
1466 /*
1467  *      Encapsulate a packet by attaching a valid IPIP header to it.
1468  *      This avoids tunnel drivers and other mess and gives us the speed so
1469  *      important for multicast video.
1470  */
1471
1472 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1473 {
1474         struct iphdr *iph;
1475         struct iphdr *old_iph = ip_hdr(skb);
1476
1477         skb_push(skb, sizeof(struct iphdr));
1478         skb->transport_header = skb->network_header;
1479         skb_reset_network_header(skb);
1480         iph = ip_hdr(skb);
1481
1482         iph->version    =       4;
1483         iph->tos        =       old_iph->tos;
1484         iph->ttl        =       old_iph->ttl;
1485         iph->frag_off   =       0;
1486         iph->daddr      =       daddr;
1487         iph->saddr      =       saddr;
1488         iph->protocol   =       IPPROTO_IPIP;
1489         iph->ihl        =       5;
1490         iph->tot_len    =       htons(skb->len);
1491         ip_select_ident(iph, skb_dst(skb), NULL);
1492         ip_send_check(iph);
1493
1494         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1495         nf_reset(skb);
1496 }
1497
1498 static inline int ipmr_forward_finish(struct sk_buff *skb)
1499 {
1500         struct ip_options *opt = &(IPCB(skb)->opt);
1501
1502         IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1503
1504         if (unlikely(opt->optlen))
1505                 ip_forward_options(skb);
1506
1507         return dst_output(skb);
1508 }
1509
1510 /*
1511  *      Processing handlers for ipmr_forward
1512  */
1513
1514 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1515                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1516 {
1517         const struct iphdr *iph = ip_hdr(skb);
1518         struct vif_device *vif = &mrt->vif_table[vifi];
1519         struct net_device *dev;
1520         struct rtable *rt;
1521         int    encap = 0;
1522
1523         if (vif->dev == NULL)
1524                 goto out_free;
1525
1526 #ifdef CONFIG_IP_PIMSM
1527         if (vif->flags & VIFF_REGISTER) {
1528                 vif->pkt_out++;
1529                 vif->bytes_out += skb->len;
1530                 vif->dev->stats.tx_bytes += skb->len;
1531                 vif->dev->stats.tx_packets++;
1532                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1533                 goto out_free;
1534         }
1535 #endif
1536
1537         if (vif->flags & VIFF_TUNNEL) {
1538                 struct flowi fl = {
1539                         .oif = vif->link,
1540                         .nl_u = {
1541                                 .ip4_u = {
1542                                         .daddr = vif->remote,
1543                                         .saddr = vif->local,
1544                                         .tos = RT_TOS(iph->tos)
1545                                 }
1546                         },
1547                         .proto = IPPROTO_IPIP
1548                 };
1549
1550                 if (ip_route_output_key(net, &rt, &fl))
1551                         goto out_free;
1552                 encap = sizeof(struct iphdr);
1553         } else {
1554                 struct flowi fl = {
1555                         .oif = vif->link,
1556                         .nl_u = {
1557                                 .ip4_u = {
1558                                         .daddr = iph->daddr,
1559                                         .tos = RT_TOS(iph->tos)
1560                                 }
1561                         },
1562                         .proto = IPPROTO_IPIP
1563                 };
1564
1565                 if (ip_route_output_key(net, &rt, &fl))
1566                         goto out_free;
1567         }
1568
1569         dev = rt->dst.dev;
1570
1571         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1572                 /* Do not fragment multicasts. Alas, IPv4 does not
1573                  * allow to send ICMP, so that packets will disappear
1574                  * to blackhole.
1575                  */
1576
1577                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1578                 ip_rt_put(rt);
1579                 goto out_free;
1580         }
1581
1582         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1583
1584         if (skb_cow(skb, encap)) {
1585                 ip_rt_put(rt);
1586                 goto out_free;
1587         }
1588
1589         vif->pkt_out++;
1590         vif->bytes_out += skb->len;
1591
1592         skb_dst_drop(skb);
1593         skb_dst_set(skb, &rt->dst);
1594         ip_decrease_ttl(ip_hdr(skb));
1595
1596         /* FIXME: forward and output firewalls used to be called here.
1597          * What do we do with netfilter? -- RR
1598          */
1599         if (vif->flags & VIFF_TUNNEL) {
1600                 ip_encap(skb, vif->local, vif->remote);
1601                 /* FIXME: extra output firewall step used to be here. --RR */
1602                 vif->dev->stats.tx_packets++;
1603                 vif->dev->stats.tx_bytes += skb->len;
1604         }
1605
1606         IPCB(skb)->flags |= IPSKB_FORWARDED;
1607
1608         /*
1609          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1610          * not only before forwarding, but after forwarding on all output
1611          * interfaces. It is clear, if mrouter runs a multicasting
1612          * program, it should receive packets not depending to what interface
1613          * program is joined.
1614          * If we will not make it, the program will have to join on all
1615          * interfaces. On the other hand, multihoming host (or router, but
1616          * not mrouter) cannot join to more than one interface - it will
1617          * result in receiving multiple packets.
1618          */
1619         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1620                 ipmr_forward_finish);
1621         return;
1622
1623 out_free:
1624         kfree_skb(skb);
1625 }
1626
1627 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1628 {
1629         int ct;
1630
1631         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1632                 if (mrt->vif_table[ct].dev == dev)
1633                         break;
1634         }
1635         return ct;
1636 }
1637
1638 /* "local" means that we should preserve one skb (for local delivery) */
1639
1640 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1641                          struct sk_buff *skb, struct mfc_cache *cache,
1642                          int local)
1643 {
1644         int psend = -1;
1645         int vif, ct;
1646
1647         vif = cache->mfc_parent;
1648         cache->mfc_un.res.pkt++;
1649         cache->mfc_un.res.bytes += skb->len;
1650
1651         /*
1652          * Wrong interface: drop packet and (maybe) send PIM assert.
1653          */
1654         if (mrt->vif_table[vif].dev != skb->dev) {
1655                 int true_vifi;
1656
1657                 if (rt_is_output_route(skb_rtable(skb))) {
1658                         /* It is our own packet, looped back.
1659                          * Very complicated situation...
1660                          *
1661                          * The best workaround until routing daemons will be
1662                          * fixed is not to redistribute packet, if it was
1663                          * send through wrong interface. It means, that
1664                          * multicast applications WILL NOT work for
1665                          * (S,G), which have default multicast route pointing
1666                          * to wrong oif. In any case, it is not a good
1667                          * idea to use multicasting applications on router.
1668                          */
1669                         goto dont_forward;
1670                 }
1671
1672                 cache->mfc_un.res.wrong_if++;
1673                 true_vifi = ipmr_find_vif(mrt, skb->dev);
1674
1675                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1676                     /* pimsm uses asserts, when switching from RPT to SPT,
1677                      * so that we cannot check that packet arrived on an oif.
1678                      * It is bad, but otherwise we would need to move pretty
1679                      * large chunk of pimd to kernel. Ough... --ANK
1680                      */
1681                     (mrt->mroute_do_pim ||
1682                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1683                     time_after(jiffies,
1684                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1685                         cache->mfc_un.res.last_assert = jiffies;
1686                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1687                 }
1688                 goto dont_forward;
1689         }
1690
1691         mrt->vif_table[vif].pkt_in++;
1692         mrt->vif_table[vif].bytes_in += skb->len;
1693
1694         /*
1695          *      Forward the frame
1696          */
1697         for (ct = cache->mfc_un.res.maxvif - 1;
1698              ct >= cache->mfc_un.res.minvif; ct--) {
1699                 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1700                         if (psend != -1) {
1701                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1702
1703                                 if (skb2)
1704                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1705                                                         psend);
1706                         }
1707                         psend = ct;
1708                 }
1709         }
1710         if (psend != -1) {
1711                 if (local) {
1712                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1713
1714                         if (skb2)
1715                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1716                 } else {
1717                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1718                         return 0;
1719                 }
1720         }
1721
1722 dont_forward:
1723         if (!local)
1724                 kfree_skb(skb);
1725         return 0;
1726 }
1727
1728
1729 /*
1730  *      Multicast packets for forwarding arrive here
1731  *      Called with rcu_read_lock();
1732  */
1733
1734 int ip_mr_input(struct sk_buff *skb)
1735 {
1736         struct mfc_cache *cache;
1737         struct net *net = dev_net(skb->dev);
1738         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1739         struct mr_table *mrt;
1740         int err;
1741
1742         /* Packet is looped back after forward, it should not be
1743          * forwarded second time, but still can be delivered locally.
1744          */
1745         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1746                 goto dont_forward;
1747
1748         err = ipmr_fib_lookup(net, &skb_rtable(skb)->fl, &mrt);
1749         if (err < 0) {
1750                 kfree_skb(skb);
1751                 return err;
1752         }
1753
1754         if (!local) {
1755                 if (IPCB(skb)->opt.router_alert) {
1756                         if (ip_call_ra_chain(skb))
1757                                 return 0;
1758                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1759                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1760                          * Cisco IOS <= 11.2(8)) do not put router alert
1761                          * option to IGMP packets destined to routable
1762                          * groups. It is very bad, because it means
1763                          * that we can forward NO IGMP messages.
1764                          */
1765                         struct sock *mroute_sk;
1766
1767                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1768                         if (mroute_sk) {
1769                                 nf_reset(skb);
1770                                 raw_rcv(mroute_sk, skb);
1771                                 return 0;
1772                         }
1773                     }
1774         }
1775
1776         /* already under rcu_read_lock() */
1777         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1778
1779         /*
1780          *      No usable cache entry
1781          */
1782         if (cache == NULL) {
1783                 int vif;
1784
1785                 if (local) {
1786                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1787                         ip_local_deliver(skb);
1788                         if (skb2 == NULL)
1789                                 return -ENOBUFS;
1790                         skb = skb2;
1791                 }
1792
1793                 read_lock(&mrt_lock);
1794                 vif = ipmr_find_vif(mrt, skb->dev);
1795                 if (vif >= 0) {
1796                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1797                         read_unlock(&mrt_lock);
1798
1799                         return err2;
1800                 }
1801                 read_unlock(&mrt_lock);
1802                 kfree_skb(skb);
1803                 return -ENODEV;
1804         }
1805
1806         read_lock(&mrt_lock);
1807         ip_mr_forward(net, mrt, skb, cache, local);
1808         read_unlock(&mrt_lock);
1809
1810         if (local)
1811                 return ip_local_deliver(skb);
1812
1813         return 0;
1814
1815 dont_forward:
1816         if (local)
1817                 return ip_local_deliver(skb);
1818         kfree_skb(skb);
1819         return 0;
1820 }
1821
1822 #ifdef CONFIG_IP_PIMSM
1823 /* called with rcu_read_lock() */
1824 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1825                      unsigned int pimlen)
1826 {
1827         struct net_device *reg_dev = NULL;
1828         struct iphdr *encap;
1829
1830         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1831         /*
1832          * Check that:
1833          * a. packet is really sent to a multicast group
1834          * b. packet is not a NULL-REGISTER
1835          * c. packet is not truncated
1836          */
1837         if (!ipv4_is_multicast(encap->daddr) ||
1838             encap->tot_len == 0 ||
1839             ntohs(encap->tot_len) + pimlen > skb->len)
1840                 return 1;
1841
1842         read_lock(&mrt_lock);
1843         if (mrt->mroute_reg_vif_num >= 0)
1844                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1845         read_unlock(&mrt_lock);
1846
1847         if (reg_dev == NULL)
1848                 return 1;
1849
1850         skb->mac_header = skb->network_header;
1851         skb_pull(skb, (u8 *)encap - skb->data);
1852         skb_reset_network_header(skb);
1853         skb->protocol = htons(ETH_P_IP);
1854         skb->ip_summed = CHECKSUM_NONE;
1855         skb->pkt_type = PACKET_HOST;
1856
1857         skb_tunnel_rx(skb, reg_dev);
1858
1859         netif_rx(skb);
1860
1861         return NET_RX_SUCCESS;
1862 }
1863 #endif
1864
1865 #ifdef CONFIG_IP_PIMSM_V1
1866 /*
1867  * Handle IGMP messages of PIMv1
1868  */
1869
1870 int pim_rcv_v1(struct sk_buff *skb)
1871 {
1872         struct igmphdr *pim;
1873         struct net *net = dev_net(skb->dev);
1874         struct mr_table *mrt;
1875
1876         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1877                 goto drop;
1878
1879         pim = igmp_hdr(skb);
1880
1881         if (ipmr_fib_lookup(net, &skb_rtable(skb)->fl, &mrt) < 0)
1882                 goto drop;
1883
1884         if (!mrt->mroute_do_pim ||
1885             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
1886                 goto drop;
1887
1888         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1889 drop:
1890                 kfree_skb(skb);
1891         }
1892         return 0;
1893 }
1894 #endif
1895
1896 #ifdef CONFIG_IP_PIMSM_V2
1897 static int pim_rcv(struct sk_buff *skb)
1898 {
1899         struct pimreghdr *pim;
1900         struct net *net = dev_net(skb->dev);
1901         struct mr_table *mrt;
1902
1903         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1904                 goto drop;
1905
1906         pim = (struct pimreghdr *)skb_transport_header(skb);
1907         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
1908             (pim->flags & PIM_NULL_REGISTER) ||
1909             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
1910              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
1911                 goto drop;
1912
1913         if (ipmr_fib_lookup(net, &skb_rtable(skb)->fl, &mrt) < 0)
1914                 goto drop;
1915
1916         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1917 drop:
1918                 kfree_skb(skb);
1919         }
1920         return 0;
1921 }
1922 #endif
1923
1924 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
1925                               struct mfc_cache *c, struct rtmsg *rtm)
1926 {
1927         int ct;
1928         struct rtnexthop *nhp;
1929         u8 *b = skb_tail_pointer(skb);
1930         struct rtattr *mp_head;
1931
1932         /* If cache is unresolved, don't try to parse IIF and OIF */
1933         if (c->mfc_parent >= MAXVIFS)
1934                 return -ENOENT;
1935
1936         if (VIF_EXISTS(mrt, c->mfc_parent))
1937                 RTA_PUT(skb, RTA_IIF, 4, &mrt->vif_table[c->mfc_parent].dev->ifindex);
1938
1939         mp_head = (struct rtattr *)skb_put(skb, RTA_LENGTH(0));
1940
1941         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
1942                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
1943                         if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
1944                                 goto rtattr_failure;
1945                         nhp = (struct rtnexthop *)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
1946                         nhp->rtnh_flags = 0;
1947                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
1948                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
1949                         nhp->rtnh_len = sizeof(*nhp);
1950                 }
1951         }
1952         mp_head->rta_type = RTA_MULTIPATH;
1953         mp_head->rta_len = skb_tail_pointer(skb) - (u8 *)mp_head;
1954         rtm->rtm_type = RTN_MULTICAST;
1955         return 1;
1956
1957 rtattr_failure:
1958         nlmsg_trim(skb, b);
1959         return -EMSGSIZE;
1960 }
1961
1962 int ipmr_get_route(struct net *net,
1963                    struct sk_buff *skb, struct rtmsg *rtm, int nowait)
1964 {
1965         int err;
1966         struct mr_table *mrt;
1967         struct mfc_cache *cache;
1968         struct rtable *rt = skb_rtable(skb);
1969
1970         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
1971         if (mrt == NULL)
1972                 return -ENOENT;
1973
1974         rcu_read_lock();
1975         cache = ipmr_cache_find(mrt, rt->rt_src, rt->rt_dst);
1976
1977         if (cache == NULL) {
1978                 struct sk_buff *skb2;
1979                 struct iphdr *iph;
1980                 struct net_device *dev;
1981                 int vif = -1;
1982
1983                 if (nowait) {
1984                         rcu_read_unlock();
1985                         return -EAGAIN;
1986                 }
1987
1988                 dev = skb->dev;
1989                 read_lock(&mrt_lock);
1990                 if (dev)
1991                         vif = ipmr_find_vif(mrt, dev);
1992                 if (vif < 0) {
1993                         read_unlock(&mrt_lock);
1994                         rcu_read_unlock();
1995                         return -ENODEV;
1996                 }
1997                 skb2 = skb_clone(skb, GFP_ATOMIC);
1998                 if (!skb2) {
1999                         read_unlock(&mrt_lock);
2000                         rcu_read_unlock();
2001                         return -ENOMEM;
2002                 }
2003
2004                 skb_push(skb2, sizeof(struct iphdr));
2005                 skb_reset_network_header(skb2);
2006                 iph = ip_hdr(skb2);
2007                 iph->ihl = sizeof(struct iphdr) >> 2;
2008                 iph->saddr = rt->rt_src;
2009                 iph->daddr = rt->rt_dst;
2010                 iph->version = 0;
2011                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2012                 read_unlock(&mrt_lock);
2013                 rcu_read_unlock();
2014                 return err;
2015         }
2016
2017         read_lock(&mrt_lock);
2018         if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2019                 cache->mfc_flags |= MFC_NOTIFY;
2020         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2021         read_unlock(&mrt_lock);
2022         rcu_read_unlock();
2023         return err;
2024 }
2025
2026 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2027                             u32 pid, u32 seq, struct mfc_cache *c)
2028 {
2029         struct nlmsghdr *nlh;
2030         struct rtmsg *rtm;
2031
2032         nlh = nlmsg_put(skb, pid, seq, RTM_NEWROUTE, sizeof(*rtm), NLM_F_MULTI);
2033         if (nlh == NULL)
2034                 return -EMSGSIZE;
2035
2036         rtm = nlmsg_data(nlh);
2037         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2038         rtm->rtm_dst_len  = 32;
2039         rtm->rtm_src_len  = 32;
2040         rtm->rtm_tos      = 0;
2041         rtm->rtm_table    = mrt->id;
2042         NLA_PUT_U32(skb, RTA_TABLE, mrt->id);
2043         rtm->rtm_type     = RTN_MULTICAST;
2044         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2045         rtm->rtm_protocol = RTPROT_UNSPEC;
2046         rtm->rtm_flags    = 0;
2047
2048         NLA_PUT_BE32(skb, RTA_SRC, c->mfc_origin);
2049         NLA_PUT_BE32(skb, RTA_DST, c->mfc_mcastgrp);
2050
2051         if (__ipmr_fill_mroute(mrt, skb, c, rtm) < 0)
2052                 goto nla_put_failure;
2053
2054         return nlmsg_end(skb, nlh);
2055
2056 nla_put_failure:
2057         nlmsg_cancel(skb, nlh);
2058         return -EMSGSIZE;
2059 }
2060
2061 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2062 {
2063         struct net *net = sock_net(skb->sk);
2064         struct mr_table *mrt;
2065         struct mfc_cache *mfc;
2066         unsigned int t = 0, s_t;
2067         unsigned int h = 0, s_h;
2068         unsigned int e = 0, s_e;
2069
2070         s_t = cb->args[0];
2071         s_h = cb->args[1];
2072         s_e = cb->args[2];
2073
2074         rcu_read_lock();
2075         ipmr_for_each_table(mrt, net) {
2076                 if (t < s_t)
2077                         goto next_table;
2078                 if (t > s_t)
2079                         s_h = 0;
2080                 for (h = s_h; h < MFC_LINES; h++) {
2081                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2082                                 if (e < s_e)
2083                                         goto next_entry;
2084                                 if (ipmr_fill_mroute(mrt, skb,
2085                                                      NETLINK_CB(cb->skb).pid,
2086                                                      cb->nlh->nlmsg_seq,
2087                                                      mfc) < 0)
2088                                         goto done;
2089 next_entry:
2090                                 e++;
2091                         }
2092                         e = s_e = 0;
2093                 }
2094                 s_h = 0;
2095 next_table:
2096                 t++;
2097         }
2098 done:
2099         rcu_read_unlock();
2100
2101         cb->args[2] = e;
2102         cb->args[1] = h;
2103         cb->args[0] = t;
2104
2105         return skb->len;
2106 }
2107
2108 #ifdef CONFIG_PROC_FS
2109 /*
2110  *      The /proc interfaces to multicast routing :
2111  *      /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2112  */
2113 struct ipmr_vif_iter {
2114         struct seq_net_private p;
2115         struct mr_table *mrt;
2116         int ct;
2117 };
2118
2119 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2120                                            struct ipmr_vif_iter *iter,
2121                                            loff_t pos)
2122 {
2123         struct mr_table *mrt = iter->mrt;
2124
2125         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2126                 if (!VIF_EXISTS(mrt, iter->ct))
2127                         continue;
2128                 if (pos-- == 0)
2129                         return &mrt->vif_table[iter->ct];
2130         }
2131         return NULL;
2132 }
2133
2134 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2135         __acquires(mrt_lock)
2136 {
2137         struct ipmr_vif_iter *iter = seq->private;
2138         struct net *net = seq_file_net(seq);
2139         struct mr_table *mrt;
2140
2141         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2142         if (mrt == NULL)
2143                 return ERR_PTR(-ENOENT);
2144
2145         iter->mrt = mrt;
2146
2147         read_lock(&mrt_lock);
2148         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2149                 : SEQ_START_TOKEN;
2150 }
2151
2152 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2153 {
2154         struct ipmr_vif_iter *iter = seq->private;
2155         struct net *net = seq_file_net(seq);
2156         struct mr_table *mrt = iter->mrt;
2157
2158         ++*pos;
2159         if (v == SEQ_START_TOKEN)
2160                 return ipmr_vif_seq_idx(net, iter, 0);
2161
2162         while (++iter->ct < mrt->maxvif) {
2163                 if (!VIF_EXISTS(mrt, iter->ct))
2164                         continue;
2165                 return &mrt->vif_table[iter->ct];
2166         }
2167         return NULL;
2168 }
2169
2170 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2171         __releases(mrt_lock)
2172 {
2173         read_unlock(&mrt_lock);
2174 }
2175
2176 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2177 {
2178         struct ipmr_vif_iter *iter = seq->private;
2179         struct mr_table *mrt = iter->mrt;
2180
2181         if (v == SEQ_START_TOKEN) {
2182                 seq_puts(seq,
2183                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2184         } else {
2185                 const struct vif_device *vif = v;
2186                 const char *name =  vif->dev ? vif->dev->name : "none";
2187
2188                 seq_printf(seq,
2189                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2190                            vif - mrt->vif_table,
2191                            name, vif->bytes_in, vif->pkt_in,
2192                            vif->bytes_out, vif->pkt_out,
2193                            vif->flags, vif->local, vif->remote);
2194         }
2195         return 0;
2196 }
2197
2198 static const struct seq_operations ipmr_vif_seq_ops = {
2199         .start = ipmr_vif_seq_start,
2200         .next  = ipmr_vif_seq_next,
2201         .stop  = ipmr_vif_seq_stop,
2202         .show  = ipmr_vif_seq_show,
2203 };
2204
2205 static int ipmr_vif_open(struct inode *inode, struct file *file)
2206 {
2207         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2208                             sizeof(struct ipmr_vif_iter));
2209 }
2210
2211 static const struct file_operations ipmr_vif_fops = {
2212         .owner   = THIS_MODULE,
2213         .open    = ipmr_vif_open,
2214         .read    = seq_read,
2215         .llseek  = seq_lseek,
2216         .release = seq_release_net,
2217 };
2218
2219 struct ipmr_mfc_iter {
2220         struct seq_net_private p;
2221         struct mr_table *mrt;
2222         struct list_head *cache;
2223         int ct;
2224 };
2225
2226
2227 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2228                                           struct ipmr_mfc_iter *it, loff_t pos)
2229 {
2230         struct mr_table *mrt = it->mrt;
2231         struct mfc_cache *mfc;
2232
2233         rcu_read_lock();
2234         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2235                 it->cache = &mrt->mfc_cache_array[it->ct];
2236                 list_for_each_entry_rcu(mfc, it->cache, list)
2237                         if (pos-- == 0)
2238                                 return mfc;
2239         }
2240         rcu_read_unlock();
2241
2242         spin_lock_bh(&mfc_unres_lock);
2243         it->cache = &mrt->mfc_unres_queue;
2244         list_for_each_entry(mfc, it->cache, list)
2245                 if (pos-- == 0)
2246                         return mfc;
2247         spin_unlock_bh(&mfc_unres_lock);
2248
2249         it->cache = NULL;
2250         return NULL;
2251 }
2252
2253
2254 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2255 {
2256         struct ipmr_mfc_iter *it = seq->private;
2257         struct net *net = seq_file_net(seq);
2258         struct mr_table *mrt;
2259
2260         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2261         if (mrt == NULL)
2262                 return ERR_PTR(-ENOENT);
2263
2264         it->mrt = mrt;
2265         it->cache = NULL;
2266         it->ct = 0;
2267         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2268                 : SEQ_START_TOKEN;
2269 }
2270
2271 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2272 {
2273         struct mfc_cache *mfc = v;
2274         struct ipmr_mfc_iter *it = seq->private;
2275         struct net *net = seq_file_net(seq);
2276         struct mr_table *mrt = it->mrt;
2277
2278         ++*pos;
2279
2280         if (v == SEQ_START_TOKEN)
2281                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2282
2283         if (mfc->list.next != it->cache)
2284                 return list_entry(mfc->list.next, struct mfc_cache, list);
2285
2286         if (it->cache == &mrt->mfc_unres_queue)
2287                 goto end_of_list;
2288
2289         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2290
2291         while (++it->ct < MFC_LINES) {
2292                 it->cache = &mrt->mfc_cache_array[it->ct];
2293                 if (list_empty(it->cache))
2294                         continue;
2295                 return list_first_entry(it->cache, struct mfc_cache, list);
2296         }
2297
2298         /* exhausted cache_array, show unresolved */
2299         rcu_read_unlock();
2300         it->cache = &mrt->mfc_unres_queue;
2301         it->ct = 0;
2302
2303         spin_lock_bh(&mfc_unres_lock);
2304         if (!list_empty(it->cache))
2305                 return list_first_entry(it->cache, struct mfc_cache, list);
2306
2307 end_of_list:
2308         spin_unlock_bh(&mfc_unres_lock);
2309         it->cache = NULL;
2310
2311         return NULL;
2312 }
2313
2314 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2315 {
2316         struct ipmr_mfc_iter *it = seq->private;
2317         struct mr_table *mrt = it->mrt;
2318
2319         if (it->cache == &mrt->mfc_unres_queue)
2320                 spin_unlock_bh(&mfc_unres_lock);
2321         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2322                 rcu_read_unlock();
2323 }
2324
2325 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2326 {
2327         int n;
2328
2329         if (v == SEQ_START_TOKEN) {
2330                 seq_puts(seq,
2331                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2332         } else {
2333                 const struct mfc_cache *mfc = v;
2334                 const struct ipmr_mfc_iter *it = seq->private;
2335                 const struct mr_table *mrt = it->mrt;
2336
2337                 seq_printf(seq, "%08X %08X %-3hd",
2338                            (__force u32) mfc->mfc_mcastgrp,
2339                            (__force u32) mfc->mfc_origin,
2340                            mfc->mfc_parent);
2341
2342                 if (it->cache != &mrt->mfc_unres_queue) {
2343                         seq_printf(seq, " %8lu %8lu %8lu",
2344                                    mfc->mfc_un.res.pkt,
2345                                    mfc->mfc_un.res.bytes,
2346                                    mfc->mfc_un.res.wrong_if);
2347                         for (n = mfc->mfc_un.res.minvif;
2348                              n < mfc->mfc_un.res.maxvif; n++) {
2349                                 if (VIF_EXISTS(mrt, n) &&
2350                                     mfc->mfc_un.res.ttls[n] < 255)
2351                                         seq_printf(seq,
2352                                            " %2d:%-3d",
2353                                            n, mfc->mfc_un.res.ttls[n]);
2354                         }
2355                 } else {
2356                         /* unresolved mfc_caches don't contain
2357                          * pkt, bytes and wrong_if values
2358                          */
2359                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2360                 }
2361                 seq_putc(seq, '\n');
2362         }
2363         return 0;
2364 }
2365
2366 static const struct seq_operations ipmr_mfc_seq_ops = {
2367         .start = ipmr_mfc_seq_start,
2368         .next  = ipmr_mfc_seq_next,
2369         .stop  = ipmr_mfc_seq_stop,
2370         .show  = ipmr_mfc_seq_show,
2371 };
2372
2373 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2374 {
2375         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2376                             sizeof(struct ipmr_mfc_iter));
2377 }
2378
2379 static const struct file_operations ipmr_mfc_fops = {
2380         .owner   = THIS_MODULE,
2381         .open    = ipmr_mfc_open,
2382         .read    = seq_read,
2383         .llseek  = seq_lseek,
2384         .release = seq_release_net,
2385 };
2386 #endif
2387
2388 #ifdef CONFIG_IP_PIMSM_V2
2389 static const struct net_protocol pim_protocol = {
2390         .handler        =       pim_rcv,
2391         .netns_ok       =       1,
2392 };
2393 #endif
2394
2395
2396 /*
2397  *      Setup for IP multicast routing
2398  */
2399 static int __net_init ipmr_net_init(struct net *net)
2400 {
2401         int err;
2402
2403         err = ipmr_rules_init(net);
2404         if (err < 0)
2405                 goto fail;
2406
2407 #ifdef CONFIG_PROC_FS
2408         err = -ENOMEM;
2409         if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2410                 goto proc_vif_fail;
2411         if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2412                 goto proc_cache_fail;
2413 #endif
2414         return 0;
2415
2416 #ifdef CONFIG_PROC_FS
2417 proc_cache_fail:
2418         proc_net_remove(net, "ip_mr_vif");
2419 proc_vif_fail:
2420         ipmr_rules_exit(net);
2421 #endif
2422 fail:
2423         return err;
2424 }
2425
2426 static void __net_exit ipmr_net_exit(struct net *net)
2427 {
2428 #ifdef CONFIG_PROC_FS
2429         proc_net_remove(net, "ip_mr_cache");
2430         proc_net_remove(net, "ip_mr_vif");
2431 #endif
2432         ipmr_rules_exit(net);
2433 }
2434
2435 static struct pernet_operations ipmr_net_ops = {
2436         .init = ipmr_net_init,
2437         .exit = ipmr_net_exit,
2438 };
2439
2440 int __init ip_mr_init(void)
2441 {
2442         int err;
2443
2444         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2445                                        sizeof(struct mfc_cache),
2446                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2447                                        NULL);
2448         if (!mrt_cachep)
2449                 return -ENOMEM;
2450
2451         err = register_pernet_subsys(&ipmr_net_ops);
2452         if (err)
2453                 goto reg_pernet_fail;
2454
2455         err = register_netdevice_notifier(&ip_mr_notifier);
2456         if (err)
2457                 goto reg_notif_fail;
2458 #ifdef CONFIG_IP_PIMSM_V2
2459         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2460                 printk(KERN_ERR "ip_mr_init: can't add PIM protocol\n");
2461                 err = -EAGAIN;
2462                 goto add_proto_fail;
2463         }
2464 #endif
2465         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE, NULL, ipmr_rtm_dumproute);
2466         return 0;
2467
2468 #ifdef CONFIG_IP_PIMSM_V2
2469 add_proto_fail:
2470         unregister_netdevice_notifier(&ip_mr_notifier);
2471 #endif
2472 reg_notif_fail:
2473         unregister_pernet_subsys(&ipmr_net_ops);
2474 reg_pernet_fail:
2475         kmem_cache_destroy(mrt_cachep);
2476         return err;
2477 }