ipv4: fix multicast losses
[linux-2.6.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 <linux/compat.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 int ip_mr_forward(struct net *net, struct mr_table *mrt,
128                          struct sk_buff *skb, struct mfc_cache *cache,
129                          int local);
130 static int ipmr_cache_report(struct mr_table *mrt,
131                              struct sk_buff *pkt, vifi_t vifi, int assert);
132 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
133                               struct mfc_cache *c, struct rtmsg *rtm);
134 static void ipmr_expire_process(unsigned long arg);
135
136 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
137 #define ipmr_for_each_table(mrt, net) \
138         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
139
140 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
141 {
142         struct mr_table *mrt;
143
144         ipmr_for_each_table(mrt, net) {
145                 if (mrt->id == id)
146                         return mrt;
147         }
148         return NULL;
149 }
150
151 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
152                            struct mr_table **mrt)
153 {
154         struct ipmr_result res;
155         struct fib_lookup_arg arg = { .result = &res, };
156         int err;
157
158         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
159                                flowi4_to_flowi(flp4), 0, &arg);
160         if (err < 0)
161                 return err;
162         *mrt = res.mrt;
163         return 0;
164 }
165
166 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
167                             int flags, struct fib_lookup_arg *arg)
168 {
169         struct ipmr_result *res = arg->result;
170         struct mr_table *mrt;
171
172         switch (rule->action) {
173         case FR_ACT_TO_TBL:
174                 break;
175         case FR_ACT_UNREACHABLE:
176                 return -ENETUNREACH;
177         case FR_ACT_PROHIBIT:
178                 return -EACCES;
179         case FR_ACT_BLACKHOLE:
180         default:
181                 return -EINVAL;
182         }
183
184         mrt = ipmr_get_table(rule->fr_net, rule->table);
185         if (mrt == NULL)
186                 return -EAGAIN;
187         res->mrt = mrt;
188         return 0;
189 }
190
191 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
192 {
193         return 1;
194 }
195
196 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
197         FRA_GENERIC_POLICY,
198 };
199
200 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
201                                struct fib_rule_hdr *frh, struct nlattr **tb)
202 {
203         return 0;
204 }
205
206 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
207                              struct nlattr **tb)
208 {
209         return 1;
210 }
211
212 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
213                           struct fib_rule_hdr *frh)
214 {
215         frh->dst_len = 0;
216         frh->src_len = 0;
217         frh->tos     = 0;
218         return 0;
219 }
220
221 static const struct fib_rules_ops __net_initdata ipmr_rules_ops_template = {
222         .family         = RTNL_FAMILY_IPMR,
223         .rule_size      = sizeof(struct ipmr_rule),
224         .addr_size      = sizeof(u32),
225         .action         = ipmr_rule_action,
226         .match          = ipmr_rule_match,
227         .configure      = ipmr_rule_configure,
228         .compare        = ipmr_rule_compare,
229         .default_pref   = fib_default_rule_pref,
230         .fill           = ipmr_rule_fill,
231         .nlgroup        = RTNLGRP_IPV4_RULE,
232         .policy         = ipmr_rule_policy,
233         .owner          = THIS_MODULE,
234 };
235
236 static int __net_init ipmr_rules_init(struct net *net)
237 {
238         struct fib_rules_ops *ops;
239         struct mr_table *mrt;
240         int err;
241
242         ops = fib_rules_register(&ipmr_rules_ops_template, net);
243         if (IS_ERR(ops))
244                 return PTR_ERR(ops);
245
246         INIT_LIST_HEAD(&net->ipv4.mr_tables);
247
248         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
249         if (mrt == NULL) {
250                 err = -ENOMEM;
251                 goto err1;
252         }
253
254         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
255         if (err < 0)
256                 goto err2;
257
258         net->ipv4.mr_rules_ops = ops;
259         return 0;
260
261 err2:
262         kfree(mrt);
263 err1:
264         fib_rules_unregister(ops);
265         return err;
266 }
267
268 static void __net_exit ipmr_rules_exit(struct net *net)
269 {
270         struct mr_table *mrt, *next;
271
272         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
273                 list_del(&mrt->list);
274                 kfree(mrt);
275         }
276         fib_rules_unregister(net->ipv4.mr_rules_ops);
277 }
278 #else
279 #define ipmr_for_each_table(mrt, net) \
280         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
281
282 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
283 {
284         return net->ipv4.mrt;
285 }
286
287 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
288                            struct mr_table **mrt)
289 {
290         *mrt = net->ipv4.mrt;
291         return 0;
292 }
293
294 static int __net_init ipmr_rules_init(struct net *net)
295 {
296         net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
297         return net->ipv4.mrt ? 0 : -ENOMEM;
298 }
299
300 static void __net_exit ipmr_rules_exit(struct net *net)
301 {
302         kfree(net->ipv4.mrt);
303 }
304 #endif
305
306 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
307 {
308         struct mr_table *mrt;
309         unsigned int i;
310
311         mrt = ipmr_get_table(net, id);
312         if (mrt != NULL)
313                 return mrt;
314
315         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
316         if (mrt == NULL)
317                 return NULL;
318         write_pnet(&mrt->net, net);
319         mrt->id = id;
320
321         /* Forwarding cache */
322         for (i = 0; i < MFC_LINES; i++)
323                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
324
325         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
326
327         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
328                     (unsigned long)mrt);
329
330 #ifdef CONFIG_IP_PIMSM
331         mrt->mroute_reg_vif_num = -1;
332 #endif
333 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
334         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
335 #endif
336         return mrt;
337 }
338
339 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
340
341 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
342 {
343         struct net *net = dev_net(dev);
344
345         dev_close(dev);
346
347         dev = __dev_get_by_name(net, "tunl0");
348         if (dev) {
349                 const struct net_device_ops *ops = dev->netdev_ops;
350                 struct ifreq ifr;
351                 struct ip_tunnel_parm p;
352
353                 memset(&p, 0, sizeof(p));
354                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
355                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
356                 p.iph.version = 4;
357                 p.iph.ihl = 5;
358                 p.iph.protocol = IPPROTO_IPIP;
359                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
360                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
361
362                 if (ops->ndo_do_ioctl) {
363                         mm_segment_t oldfs = get_fs();
364
365                         set_fs(KERNEL_DS);
366                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
367                         set_fs(oldfs);
368                 }
369         }
370 }
371
372 static
373 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
374 {
375         struct net_device  *dev;
376
377         dev = __dev_get_by_name(net, "tunl0");
378
379         if (dev) {
380                 const struct net_device_ops *ops = dev->netdev_ops;
381                 int err;
382                 struct ifreq ifr;
383                 struct ip_tunnel_parm p;
384                 struct in_device  *in_dev;
385
386                 memset(&p, 0, sizeof(p));
387                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
388                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
389                 p.iph.version = 4;
390                 p.iph.ihl = 5;
391                 p.iph.protocol = IPPROTO_IPIP;
392                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
393                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
394
395                 if (ops->ndo_do_ioctl) {
396                         mm_segment_t oldfs = get_fs();
397
398                         set_fs(KERNEL_DS);
399                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
400                         set_fs(oldfs);
401                 } else {
402                         err = -EOPNOTSUPP;
403                 }
404                 dev = NULL;
405
406                 if (err == 0 &&
407                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
408                         dev->flags |= IFF_MULTICAST;
409
410                         in_dev = __in_dev_get_rtnl(dev);
411                         if (in_dev == NULL)
412                                 goto failure;
413
414                         ipv4_devconf_setall(in_dev);
415                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
416
417                         if (dev_open(dev))
418                                 goto failure;
419                         dev_hold(dev);
420                 }
421         }
422         return dev;
423
424 failure:
425         /* allow the register to be completed before unregistering. */
426         rtnl_unlock();
427         rtnl_lock();
428
429         unregister_netdevice(dev);
430         return NULL;
431 }
432
433 #ifdef CONFIG_IP_PIMSM
434
435 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
436 {
437         struct net *net = dev_net(dev);
438         struct mr_table *mrt;
439         struct flowi4 fl4 = {
440                 .flowi4_oif     = dev->ifindex,
441                 .flowi4_iif     = skb->skb_iif,
442                 .flowi4_mark    = skb->mark,
443         };
444         int err;
445
446         err = ipmr_fib_lookup(net, &fl4, &mrt);
447         if (err < 0) {
448                 kfree_skb(skb);
449                 return err;
450         }
451
452         read_lock(&mrt_lock);
453         dev->stats.tx_bytes += skb->len;
454         dev->stats.tx_packets++;
455         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
456         read_unlock(&mrt_lock);
457         kfree_skb(skb);
458         return NETDEV_TX_OK;
459 }
460
461 static const struct net_device_ops reg_vif_netdev_ops = {
462         .ndo_start_xmit = reg_vif_xmit,
463 };
464
465 static void reg_vif_setup(struct net_device *dev)
466 {
467         dev->type               = ARPHRD_PIMREG;
468         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
469         dev->flags              = IFF_NOARP;
470         dev->netdev_ops         = &reg_vif_netdev_ops,
471         dev->destructor         = free_netdev;
472         dev->features           |= NETIF_F_NETNS_LOCAL;
473 }
474
475 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
476 {
477         struct net_device *dev;
478         struct in_device *in_dev;
479         char name[IFNAMSIZ];
480
481         if (mrt->id == RT_TABLE_DEFAULT)
482                 sprintf(name, "pimreg");
483         else
484                 sprintf(name, "pimreg%u", mrt->id);
485
486         dev = alloc_netdev(0, name, reg_vif_setup);
487
488         if (dev == NULL)
489                 return NULL;
490
491         dev_net_set(dev, net);
492
493         if (register_netdevice(dev)) {
494                 free_netdev(dev);
495                 return NULL;
496         }
497         dev->iflink = 0;
498
499         rcu_read_lock();
500         in_dev = __in_dev_get_rcu(dev);
501         if (!in_dev) {
502                 rcu_read_unlock();
503                 goto failure;
504         }
505
506         ipv4_devconf_setall(in_dev);
507         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
508         rcu_read_unlock();
509
510         if (dev_open(dev))
511                 goto failure;
512
513         dev_hold(dev);
514
515         return dev;
516
517 failure:
518         /* allow the register to be completed before unregistering. */
519         rtnl_unlock();
520         rtnl_lock();
521
522         unregister_netdevice(dev);
523         return NULL;
524 }
525 #endif
526
527 /*
528  *      Delete a VIF entry
529  *      @notify: Set to 1, if the caller is a notifier_call
530  */
531
532 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
533                       struct list_head *head)
534 {
535         struct vif_device *v;
536         struct net_device *dev;
537         struct in_device *in_dev;
538
539         if (vifi < 0 || vifi >= mrt->maxvif)
540                 return -EADDRNOTAVAIL;
541
542         v = &mrt->vif_table[vifi];
543
544         write_lock_bh(&mrt_lock);
545         dev = v->dev;
546         v->dev = NULL;
547
548         if (!dev) {
549                 write_unlock_bh(&mrt_lock);
550                 return -EADDRNOTAVAIL;
551         }
552
553 #ifdef CONFIG_IP_PIMSM
554         if (vifi == mrt->mroute_reg_vif_num)
555                 mrt->mroute_reg_vif_num = -1;
556 #endif
557
558         if (vifi + 1 == mrt->maxvif) {
559                 int tmp;
560
561                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
562                         if (VIF_EXISTS(mrt, tmp))
563                                 break;
564                 }
565                 mrt->maxvif = tmp+1;
566         }
567
568         write_unlock_bh(&mrt_lock);
569
570         dev_set_allmulti(dev, -1);
571
572         in_dev = __in_dev_get_rtnl(dev);
573         if (in_dev) {
574                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
575                 ip_rt_multicast_event(in_dev);
576         }
577
578         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
579                 unregister_netdevice_queue(dev, head);
580
581         dev_put(dev);
582         return 0;
583 }
584
585 static void ipmr_cache_free_rcu(struct rcu_head *head)
586 {
587         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
588
589         kmem_cache_free(mrt_cachep, c);
590 }
591
592 static inline void ipmr_cache_free(struct mfc_cache *c)
593 {
594         call_rcu(&c->rcu, ipmr_cache_free_rcu);
595 }
596
597 /* Destroy an unresolved cache entry, killing queued skbs
598  * and reporting error to netlink readers.
599  */
600
601 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
602 {
603         struct net *net = read_pnet(&mrt->net);
604         struct sk_buff *skb;
605         struct nlmsgerr *e;
606
607         atomic_dec(&mrt->cache_resolve_queue_len);
608
609         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
610                 if (ip_hdr(skb)->version == 0) {
611                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
612                         nlh->nlmsg_type = NLMSG_ERROR;
613                         nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
614                         skb_trim(skb, nlh->nlmsg_len);
615                         e = NLMSG_DATA(nlh);
616                         e->error = -ETIMEDOUT;
617                         memset(&e->msg, 0, sizeof(e->msg));
618
619                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
620                 } else {
621                         kfree_skb(skb);
622                 }
623         }
624
625         ipmr_cache_free(c);
626 }
627
628
629 /* Timer process for the unresolved queue. */
630
631 static void ipmr_expire_process(unsigned long arg)
632 {
633         struct mr_table *mrt = (struct mr_table *)arg;
634         unsigned long now;
635         unsigned long expires;
636         struct mfc_cache *c, *next;
637
638         if (!spin_trylock(&mfc_unres_lock)) {
639                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
640                 return;
641         }
642
643         if (list_empty(&mrt->mfc_unres_queue))
644                 goto out;
645
646         now = jiffies;
647         expires = 10*HZ;
648
649         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
650                 if (time_after(c->mfc_un.unres.expires, now)) {
651                         unsigned long interval = c->mfc_un.unres.expires - now;
652                         if (interval < expires)
653                                 expires = interval;
654                         continue;
655                 }
656
657                 list_del(&c->list);
658                 ipmr_destroy_unres(mrt, c);
659         }
660
661         if (!list_empty(&mrt->mfc_unres_queue))
662                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
663
664 out:
665         spin_unlock(&mfc_unres_lock);
666 }
667
668 /* Fill oifs list. It is called under write locked mrt_lock. */
669
670 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
671                                    unsigned char *ttls)
672 {
673         int vifi;
674
675         cache->mfc_un.res.minvif = MAXVIFS;
676         cache->mfc_un.res.maxvif = 0;
677         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
678
679         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
680                 if (VIF_EXISTS(mrt, vifi) &&
681                     ttls[vifi] && ttls[vifi] < 255) {
682                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
683                         if (cache->mfc_un.res.minvif > vifi)
684                                 cache->mfc_un.res.minvif = vifi;
685                         if (cache->mfc_un.res.maxvif <= vifi)
686                                 cache->mfc_un.res.maxvif = vifi + 1;
687                 }
688         }
689 }
690
691 static int vif_add(struct net *net, struct mr_table *mrt,
692                    struct vifctl *vifc, int mrtsock)
693 {
694         int vifi = vifc->vifc_vifi;
695         struct vif_device *v = &mrt->vif_table[vifi];
696         struct net_device *dev;
697         struct in_device *in_dev;
698         int err;
699
700         /* Is vif busy ? */
701         if (VIF_EXISTS(mrt, vifi))
702                 return -EADDRINUSE;
703
704         switch (vifc->vifc_flags) {
705 #ifdef CONFIG_IP_PIMSM
706         case VIFF_REGISTER:
707                 /*
708                  * Special Purpose VIF in PIM
709                  * All the packets will be sent to the daemon
710                  */
711                 if (mrt->mroute_reg_vif_num >= 0)
712                         return -EADDRINUSE;
713                 dev = ipmr_reg_vif(net, mrt);
714                 if (!dev)
715                         return -ENOBUFS;
716                 err = dev_set_allmulti(dev, 1);
717                 if (err) {
718                         unregister_netdevice(dev);
719                         dev_put(dev);
720                         return err;
721                 }
722                 break;
723 #endif
724         case VIFF_TUNNEL:
725                 dev = ipmr_new_tunnel(net, vifc);
726                 if (!dev)
727                         return -ENOBUFS;
728                 err = dev_set_allmulti(dev, 1);
729                 if (err) {
730                         ipmr_del_tunnel(dev, vifc);
731                         dev_put(dev);
732                         return err;
733                 }
734                 break;
735
736         case VIFF_USE_IFINDEX:
737         case 0:
738                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
739                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
740                         if (dev && __in_dev_get_rtnl(dev) == NULL) {
741                                 dev_put(dev);
742                                 return -EADDRNOTAVAIL;
743                         }
744                 } else {
745                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
746                 }
747                 if (!dev)
748                         return -EADDRNOTAVAIL;
749                 err = dev_set_allmulti(dev, 1);
750                 if (err) {
751                         dev_put(dev);
752                         return err;
753                 }
754                 break;
755         default:
756                 return -EINVAL;
757         }
758
759         in_dev = __in_dev_get_rtnl(dev);
760         if (!in_dev) {
761                 dev_put(dev);
762                 return -EADDRNOTAVAIL;
763         }
764         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
765         ip_rt_multicast_event(in_dev);
766
767         /* Fill in the VIF structures */
768
769         v->rate_limit = vifc->vifc_rate_limit;
770         v->local = vifc->vifc_lcl_addr.s_addr;
771         v->remote = vifc->vifc_rmt_addr.s_addr;
772         v->flags = vifc->vifc_flags;
773         if (!mrtsock)
774                 v->flags |= VIFF_STATIC;
775         v->threshold = vifc->vifc_threshold;
776         v->bytes_in = 0;
777         v->bytes_out = 0;
778         v->pkt_in = 0;
779         v->pkt_out = 0;
780         v->link = dev->ifindex;
781         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
782                 v->link = dev->iflink;
783
784         /* And finish update writing critical data */
785         write_lock_bh(&mrt_lock);
786         v->dev = dev;
787 #ifdef CONFIG_IP_PIMSM
788         if (v->flags & VIFF_REGISTER)
789                 mrt->mroute_reg_vif_num = vifi;
790 #endif
791         if (vifi+1 > mrt->maxvif)
792                 mrt->maxvif = vifi+1;
793         write_unlock_bh(&mrt_lock);
794         return 0;
795 }
796
797 /* called with rcu_read_lock() */
798 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
799                                          __be32 origin,
800                                          __be32 mcastgrp)
801 {
802         int line = MFC_HASH(mcastgrp, origin);
803         struct mfc_cache *c;
804
805         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
806                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
807                         return c;
808         }
809         return NULL;
810 }
811
812 /*
813  *      Allocate a multicast cache entry
814  */
815 static struct mfc_cache *ipmr_cache_alloc(void)
816 {
817         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
818
819         if (c)
820                 c->mfc_un.res.minvif = MAXVIFS;
821         return c;
822 }
823
824 static struct mfc_cache *ipmr_cache_alloc_unres(void)
825 {
826         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
827
828         if (c) {
829                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
830                 c->mfc_un.unres.expires = jiffies + 10*HZ;
831         }
832         return c;
833 }
834
835 /*
836  *      A cache entry has gone into a resolved state from queued
837  */
838
839 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
840                                struct mfc_cache *uc, struct mfc_cache *c)
841 {
842         struct sk_buff *skb;
843         struct nlmsgerr *e;
844
845         /* Play the pending entries through our router */
846
847         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
848                 if (ip_hdr(skb)->version == 0) {
849                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
850
851                         if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
852                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
853                                                  (u8 *)nlh;
854                         } else {
855                                 nlh->nlmsg_type = NLMSG_ERROR;
856                                 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
857                                 skb_trim(skb, nlh->nlmsg_len);
858                                 e = NLMSG_DATA(nlh);
859                                 e->error = -EMSGSIZE;
860                                 memset(&e->msg, 0, sizeof(e->msg));
861                         }
862
863                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
864                 } else {
865                         ip_mr_forward(net, mrt, skb, c, 0);
866                 }
867         }
868 }
869
870 /*
871  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
872  *      expects the following bizarre scheme.
873  *
874  *      Called under mrt_lock.
875  */
876
877 static int ipmr_cache_report(struct mr_table *mrt,
878                              struct sk_buff *pkt, vifi_t vifi, int assert)
879 {
880         struct sk_buff *skb;
881         const int ihl = ip_hdrlen(pkt);
882         struct igmphdr *igmp;
883         struct igmpmsg *msg;
884         struct sock *mroute_sk;
885         int ret;
886
887 #ifdef CONFIG_IP_PIMSM
888         if (assert == IGMPMSG_WHOLEPKT)
889                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
890         else
891 #endif
892                 skb = alloc_skb(128, GFP_ATOMIC);
893
894         if (!skb)
895                 return -ENOBUFS;
896
897 #ifdef CONFIG_IP_PIMSM
898         if (assert == IGMPMSG_WHOLEPKT) {
899                 /* Ugly, but we have no choice with this interface.
900                  * Duplicate old header, fix ihl, length etc.
901                  * And all this only to mangle msg->im_msgtype and
902                  * to set msg->im_mbz to "mbz" :-)
903                  */
904                 skb_push(skb, sizeof(struct iphdr));
905                 skb_reset_network_header(skb);
906                 skb_reset_transport_header(skb);
907                 msg = (struct igmpmsg *)skb_network_header(skb);
908                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
909                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
910                 msg->im_mbz = 0;
911                 msg->im_vif = mrt->mroute_reg_vif_num;
912                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
913                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
914                                              sizeof(struct iphdr));
915         } else
916 #endif
917         {
918
919         /* Copy the IP header */
920
921         skb->network_header = skb->tail;
922         skb_put(skb, ihl);
923         skb_copy_to_linear_data(skb, pkt->data, ihl);
924         ip_hdr(skb)->protocol = 0;      /* Flag to the kernel this is a route add */
925         msg = (struct igmpmsg *)skb_network_header(skb);
926         msg->im_vif = vifi;
927         skb_dst_set(skb, dst_clone(skb_dst(pkt)));
928
929         /* Add our header */
930
931         igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
932         igmp->type      =
933         msg->im_msgtype = assert;
934         igmp->code      = 0;
935         ip_hdr(skb)->tot_len = htons(skb->len);         /* Fix the length */
936         skb->transport_header = skb->network_header;
937         }
938
939         rcu_read_lock();
940         mroute_sk = rcu_dereference(mrt->mroute_sk);
941         if (mroute_sk == NULL) {
942                 rcu_read_unlock();
943                 kfree_skb(skb);
944                 return -EINVAL;
945         }
946
947         /* Deliver to mrouted */
948
949         ret = sock_queue_rcv_skb(mroute_sk, skb);
950         rcu_read_unlock();
951         if (ret < 0) {
952                 if (net_ratelimit())
953                         printk(KERN_WARNING "mroute: pending queue full, dropping entries.\n");
954                 kfree_skb(skb);
955         }
956
957         return ret;
958 }
959
960 /*
961  *      Queue a packet for resolution. It gets locked cache entry!
962  */
963
964 static int
965 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
966 {
967         bool found = false;
968         int err;
969         struct mfc_cache *c;
970         const struct iphdr *iph = ip_hdr(skb);
971
972         spin_lock_bh(&mfc_unres_lock);
973         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
974                 if (c->mfc_mcastgrp == iph->daddr &&
975                     c->mfc_origin == iph->saddr) {
976                         found = true;
977                         break;
978                 }
979         }
980
981         if (!found) {
982                 /* Create a new entry if allowable */
983
984                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
985                     (c = ipmr_cache_alloc_unres()) == NULL) {
986                         spin_unlock_bh(&mfc_unres_lock);
987
988                         kfree_skb(skb);
989                         return -ENOBUFS;
990                 }
991
992                 /* Fill in the new cache entry */
993
994                 c->mfc_parent   = -1;
995                 c->mfc_origin   = iph->saddr;
996                 c->mfc_mcastgrp = iph->daddr;
997
998                 /* Reflect first query at mrouted. */
999
1000                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1001                 if (err < 0) {
1002                         /* If the report failed throw the cache entry
1003                            out - Brad Parker
1004                          */
1005                         spin_unlock_bh(&mfc_unres_lock);
1006
1007                         ipmr_cache_free(c);
1008                         kfree_skb(skb);
1009                         return err;
1010                 }
1011
1012                 atomic_inc(&mrt->cache_resolve_queue_len);
1013                 list_add(&c->list, &mrt->mfc_unres_queue);
1014
1015                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1016                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1017         }
1018
1019         /* See if we can append the packet */
1020
1021         if (c->mfc_un.unres.unresolved.qlen > 3) {
1022                 kfree_skb(skb);
1023                 err = -ENOBUFS;
1024         } else {
1025                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1026                 err = 0;
1027         }
1028
1029         spin_unlock_bh(&mfc_unres_lock);
1030         return err;
1031 }
1032
1033 /*
1034  *      MFC cache manipulation by user space mroute daemon
1035  */
1036
1037 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1038 {
1039         int line;
1040         struct mfc_cache *c, *next;
1041
1042         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1043
1044         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1045                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1046                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1047                         list_del_rcu(&c->list);
1048
1049                         ipmr_cache_free(c);
1050                         return 0;
1051                 }
1052         }
1053         return -ENOENT;
1054 }
1055
1056 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1057                         struct mfcctl *mfc, int mrtsock)
1058 {
1059         bool found = false;
1060         int line;
1061         struct mfc_cache *uc, *c;
1062
1063         if (mfc->mfcc_parent >= MAXVIFS)
1064                 return -ENFILE;
1065
1066         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1067
1068         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1069                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1070                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1071                         found = true;
1072                         break;
1073                 }
1074         }
1075
1076         if (found) {
1077                 write_lock_bh(&mrt_lock);
1078                 c->mfc_parent = mfc->mfcc_parent;
1079                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1080                 if (!mrtsock)
1081                         c->mfc_flags |= MFC_STATIC;
1082                 write_unlock_bh(&mrt_lock);
1083                 return 0;
1084         }
1085
1086         if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1087                 return -EINVAL;
1088
1089         c = ipmr_cache_alloc();
1090         if (c == NULL)
1091                 return -ENOMEM;
1092
1093         c->mfc_origin = mfc->mfcc_origin.s_addr;
1094         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1095         c->mfc_parent = mfc->mfcc_parent;
1096         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1097         if (!mrtsock)
1098                 c->mfc_flags |= MFC_STATIC;
1099
1100         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1101
1102         /*
1103          *      Check to see if we resolved a queued list. If so we
1104          *      need to send on the frames and tidy up.
1105          */
1106         found = false;
1107         spin_lock_bh(&mfc_unres_lock);
1108         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1109                 if (uc->mfc_origin == c->mfc_origin &&
1110                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1111                         list_del(&uc->list);
1112                         atomic_dec(&mrt->cache_resolve_queue_len);
1113                         found = true;
1114                         break;
1115                 }
1116         }
1117         if (list_empty(&mrt->mfc_unres_queue))
1118                 del_timer(&mrt->ipmr_expire_timer);
1119         spin_unlock_bh(&mfc_unres_lock);
1120
1121         if (found) {
1122                 ipmr_cache_resolve(net, mrt, uc, c);
1123                 ipmr_cache_free(uc);
1124         }
1125         return 0;
1126 }
1127
1128 /*
1129  *      Close the multicast socket, and clear the vif tables etc
1130  */
1131
1132 static void mroute_clean_tables(struct mr_table *mrt)
1133 {
1134         int i;
1135         LIST_HEAD(list);
1136         struct mfc_cache *c, *next;
1137
1138         /* Shut down all active vif entries */
1139
1140         for (i = 0; i < mrt->maxvif; i++) {
1141                 if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1142                         vif_delete(mrt, i, 0, &list);
1143         }
1144         unregister_netdevice_many(&list);
1145
1146         /* Wipe the cache */
1147
1148         for (i = 0; i < MFC_LINES; i++) {
1149                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1150                         if (c->mfc_flags & MFC_STATIC)
1151                                 continue;
1152                         list_del_rcu(&c->list);
1153                         ipmr_cache_free(c);
1154                 }
1155         }
1156
1157         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1158                 spin_lock_bh(&mfc_unres_lock);
1159                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1160                         list_del(&c->list);
1161                         ipmr_destroy_unres(mrt, c);
1162                 }
1163                 spin_unlock_bh(&mfc_unres_lock);
1164         }
1165 }
1166
1167 /* called from ip_ra_control(), before an RCU grace period,
1168  * we dont need to call synchronize_rcu() here
1169  */
1170 static void mrtsock_destruct(struct sock *sk)
1171 {
1172         struct net *net = sock_net(sk);
1173         struct mr_table *mrt;
1174
1175         rtnl_lock();
1176         ipmr_for_each_table(mrt, net) {
1177                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1178                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1179                         rcu_assign_pointer(mrt->mroute_sk, NULL);
1180                         mroute_clean_tables(mrt);
1181                 }
1182         }
1183         rtnl_unlock();
1184 }
1185
1186 /*
1187  *      Socket options and virtual interface manipulation. The whole
1188  *      virtual interface system is a complete heap, but unfortunately
1189  *      that's how BSD mrouted happens to think. Maybe one day with a proper
1190  *      MOSPF/PIM router set up we can clean this up.
1191  */
1192
1193 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1194 {
1195         int ret;
1196         struct vifctl vif;
1197         struct mfcctl mfc;
1198         struct net *net = sock_net(sk);
1199         struct mr_table *mrt;
1200
1201         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1202         if (mrt == NULL)
1203                 return -ENOENT;
1204
1205         if (optname != MRT_INIT) {
1206                 if (sk != rcu_dereference_raw(mrt->mroute_sk) &&
1207                     !capable(CAP_NET_ADMIN))
1208                         return -EACCES;
1209         }
1210
1211         switch (optname) {
1212         case MRT_INIT:
1213                 if (sk->sk_type != SOCK_RAW ||
1214                     inet_sk(sk)->inet_num != IPPROTO_IGMP)
1215                         return -EOPNOTSUPP;
1216                 if (optlen != sizeof(int))
1217                         return -ENOPROTOOPT;
1218
1219                 rtnl_lock();
1220                 if (rtnl_dereference(mrt->mroute_sk)) {
1221                         rtnl_unlock();
1222                         return -EADDRINUSE;
1223                 }
1224
1225                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1226                 if (ret == 0) {
1227                         rcu_assign_pointer(mrt->mroute_sk, sk);
1228                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1229                 }
1230                 rtnl_unlock();
1231                 return ret;
1232         case MRT_DONE:
1233                 if (sk != rcu_dereference_raw(mrt->mroute_sk))
1234                         return -EACCES;
1235                 return ip_ra_control(sk, 0, NULL);
1236         case MRT_ADD_VIF:
1237         case MRT_DEL_VIF:
1238                 if (optlen != sizeof(vif))
1239                         return -EINVAL;
1240                 if (copy_from_user(&vif, optval, sizeof(vif)))
1241                         return -EFAULT;
1242                 if (vif.vifc_vifi >= MAXVIFS)
1243                         return -ENFILE;
1244                 rtnl_lock();
1245                 if (optname == MRT_ADD_VIF) {
1246                         ret = vif_add(net, mrt, &vif,
1247                                       sk == rtnl_dereference(mrt->mroute_sk));
1248                 } else {
1249                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1250                 }
1251                 rtnl_unlock();
1252                 return ret;
1253
1254                 /*
1255                  *      Manipulate the forwarding caches. These live
1256                  *      in a sort of kernel/user symbiosis.
1257                  */
1258         case MRT_ADD_MFC:
1259         case MRT_DEL_MFC:
1260                 if (optlen != sizeof(mfc))
1261                         return -EINVAL;
1262                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1263                         return -EFAULT;
1264                 rtnl_lock();
1265                 if (optname == MRT_DEL_MFC)
1266                         ret = ipmr_mfc_delete(mrt, &mfc);
1267                 else
1268                         ret = ipmr_mfc_add(net, mrt, &mfc,
1269                                            sk == rtnl_dereference(mrt->mroute_sk));
1270                 rtnl_unlock();
1271                 return ret;
1272                 /*
1273                  *      Control PIM assert.
1274                  */
1275         case MRT_ASSERT:
1276         {
1277                 int v;
1278                 if (get_user(v, (int __user *)optval))
1279                         return -EFAULT;
1280                 mrt->mroute_do_assert = (v) ? 1 : 0;
1281                 return 0;
1282         }
1283 #ifdef CONFIG_IP_PIMSM
1284         case MRT_PIM:
1285         {
1286                 int v;
1287
1288                 if (get_user(v, (int __user *)optval))
1289                         return -EFAULT;
1290                 v = (v) ? 1 : 0;
1291
1292                 rtnl_lock();
1293                 ret = 0;
1294                 if (v != mrt->mroute_do_pim) {
1295                         mrt->mroute_do_pim = v;
1296                         mrt->mroute_do_assert = v;
1297                 }
1298                 rtnl_unlock();
1299                 return ret;
1300         }
1301 #endif
1302 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1303         case MRT_TABLE:
1304         {
1305                 u32 v;
1306
1307                 if (optlen != sizeof(u32))
1308                         return -EINVAL;
1309                 if (get_user(v, (u32 __user *)optval))
1310                         return -EFAULT;
1311
1312                 rtnl_lock();
1313                 ret = 0;
1314                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1315                         ret = -EBUSY;
1316                 } else {
1317                         if (!ipmr_new_table(net, v))
1318                                 ret = -ENOMEM;
1319                         raw_sk(sk)->ipmr_table = v;
1320                 }
1321                 rtnl_unlock();
1322                 return ret;
1323         }
1324 #endif
1325         /*
1326          *      Spurious command, or MRT_VERSION which you cannot
1327          *      set.
1328          */
1329         default:
1330                 return -ENOPROTOOPT;
1331         }
1332 }
1333
1334 /*
1335  *      Getsock opt support for the multicast routing system.
1336  */
1337
1338 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1339 {
1340         int olr;
1341         int val;
1342         struct net *net = sock_net(sk);
1343         struct mr_table *mrt;
1344
1345         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1346         if (mrt == NULL)
1347                 return -ENOENT;
1348
1349         if (optname != MRT_VERSION &&
1350 #ifdef CONFIG_IP_PIMSM
1351            optname != MRT_PIM &&
1352 #endif
1353            optname != MRT_ASSERT)
1354                 return -ENOPROTOOPT;
1355
1356         if (get_user(olr, optlen))
1357                 return -EFAULT;
1358
1359         olr = min_t(unsigned int, olr, sizeof(int));
1360         if (olr < 0)
1361                 return -EINVAL;
1362
1363         if (put_user(olr, optlen))
1364                 return -EFAULT;
1365         if (optname == MRT_VERSION)
1366                 val = 0x0305;
1367 #ifdef CONFIG_IP_PIMSM
1368         else if (optname == MRT_PIM)
1369                 val = mrt->mroute_do_pim;
1370 #endif
1371         else
1372                 val = mrt->mroute_do_assert;
1373         if (copy_to_user(optval, &val, olr))
1374                 return -EFAULT;
1375         return 0;
1376 }
1377
1378 /*
1379  *      The IP multicast ioctl support routines.
1380  */
1381
1382 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1383 {
1384         struct sioc_sg_req sr;
1385         struct sioc_vif_req vr;
1386         struct vif_device *vif;
1387         struct mfc_cache *c;
1388         struct net *net = sock_net(sk);
1389         struct mr_table *mrt;
1390
1391         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1392         if (mrt == NULL)
1393                 return -ENOENT;
1394
1395         switch (cmd) {
1396         case SIOCGETVIFCNT:
1397                 if (copy_from_user(&vr, arg, sizeof(vr)))
1398                         return -EFAULT;
1399                 if (vr.vifi >= mrt->maxvif)
1400                         return -EINVAL;
1401                 read_lock(&mrt_lock);
1402                 vif = &mrt->vif_table[vr.vifi];
1403                 if (VIF_EXISTS(mrt, vr.vifi)) {
1404                         vr.icount = vif->pkt_in;
1405                         vr.ocount = vif->pkt_out;
1406                         vr.ibytes = vif->bytes_in;
1407                         vr.obytes = vif->bytes_out;
1408                         read_unlock(&mrt_lock);
1409
1410                         if (copy_to_user(arg, &vr, sizeof(vr)))
1411                                 return -EFAULT;
1412                         return 0;
1413                 }
1414                 read_unlock(&mrt_lock);
1415                 return -EADDRNOTAVAIL;
1416         case SIOCGETSGCNT:
1417                 if (copy_from_user(&sr, arg, sizeof(sr)))
1418                         return -EFAULT;
1419
1420                 rcu_read_lock();
1421                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1422                 if (c) {
1423                         sr.pktcnt = c->mfc_un.res.pkt;
1424                         sr.bytecnt = c->mfc_un.res.bytes;
1425                         sr.wrong_if = c->mfc_un.res.wrong_if;
1426                         rcu_read_unlock();
1427
1428                         if (copy_to_user(arg, &sr, sizeof(sr)))
1429                                 return -EFAULT;
1430                         return 0;
1431                 }
1432                 rcu_read_unlock();
1433                 return -EADDRNOTAVAIL;
1434         default:
1435                 return -ENOIOCTLCMD;
1436         }
1437 }
1438
1439 #ifdef CONFIG_COMPAT
1440 struct compat_sioc_sg_req {
1441         struct in_addr src;
1442         struct in_addr grp;
1443         compat_ulong_t pktcnt;
1444         compat_ulong_t bytecnt;
1445         compat_ulong_t wrong_if;
1446 };
1447
1448 struct compat_sioc_vif_req {
1449         vifi_t  vifi;           /* Which iface */
1450         compat_ulong_t icount;
1451         compat_ulong_t ocount;
1452         compat_ulong_t ibytes;
1453         compat_ulong_t obytes;
1454 };
1455
1456 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1457 {
1458         struct compat_sioc_sg_req sr;
1459         struct compat_sioc_vif_req vr;
1460         struct vif_device *vif;
1461         struct mfc_cache *c;
1462         struct net *net = sock_net(sk);
1463         struct mr_table *mrt;
1464
1465         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1466         if (mrt == NULL)
1467                 return -ENOENT;
1468
1469         switch (cmd) {
1470         case SIOCGETVIFCNT:
1471                 if (copy_from_user(&vr, arg, sizeof(vr)))
1472                         return -EFAULT;
1473                 if (vr.vifi >= mrt->maxvif)
1474                         return -EINVAL;
1475                 read_lock(&mrt_lock);
1476                 vif = &mrt->vif_table[vr.vifi];
1477                 if (VIF_EXISTS(mrt, vr.vifi)) {
1478                         vr.icount = vif->pkt_in;
1479                         vr.ocount = vif->pkt_out;
1480                         vr.ibytes = vif->bytes_in;
1481                         vr.obytes = vif->bytes_out;
1482                         read_unlock(&mrt_lock);
1483
1484                         if (copy_to_user(arg, &vr, sizeof(vr)))
1485                                 return -EFAULT;
1486                         return 0;
1487                 }
1488                 read_unlock(&mrt_lock);
1489                 return -EADDRNOTAVAIL;
1490         case SIOCGETSGCNT:
1491                 if (copy_from_user(&sr, arg, sizeof(sr)))
1492                         return -EFAULT;
1493
1494                 rcu_read_lock();
1495                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1496                 if (c) {
1497                         sr.pktcnt = c->mfc_un.res.pkt;
1498                         sr.bytecnt = c->mfc_un.res.bytes;
1499                         sr.wrong_if = c->mfc_un.res.wrong_if;
1500                         rcu_read_unlock();
1501
1502                         if (copy_to_user(arg, &sr, sizeof(sr)))
1503                                 return -EFAULT;
1504                         return 0;
1505                 }
1506                 rcu_read_unlock();
1507                 return -EADDRNOTAVAIL;
1508         default:
1509                 return -ENOIOCTLCMD;
1510         }
1511 }
1512 #endif
1513
1514
1515 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1516 {
1517         struct net_device *dev = ptr;
1518         struct net *net = dev_net(dev);
1519         struct mr_table *mrt;
1520         struct vif_device *v;
1521         int ct;
1522         LIST_HEAD(list);
1523
1524         if (event != NETDEV_UNREGISTER)
1525                 return NOTIFY_DONE;
1526
1527         ipmr_for_each_table(mrt, net) {
1528                 v = &mrt->vif_table[0];
1529                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1530                         if (v->dev == dev)
1531                                 vif_delete(mrt, ct, 1, &list);
1532                 }
1533         }
1534         unregister_netdevice_many(&list);
1535         return NOTIFY_DONE;
1536 }
1537
1538
1539 static struct notifier_block ip_mr_notifier = {
1540         .notifier_call = ipmr_device_event,
1541 };
1542
1543 /*
1544  *      Encapsulate a packet by attaching a valid IPIP header to it.
1545  *      This avoids tunnel drivers and other mess and gives us the speed so
1546  *      important for multicast video.
1547  */
1548
1549 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1550 {
1551         struct iphdr *iph;
1552         const struct iphdr *old_iph = ip_hdr(skb);
1553
1554         skb_push(skb, sizeof(struct iphdr));
1555         skb->transport_header = skb->network_header;
1556         skb_reset_network_header(skb);
1557         iph = ip_hdr(skb);
1558
1559         iph->version    =       4;
1560         iph->tos        =       old_iph->tos;
1561         iph->ttl        =       old_iph->ttl;
1562         iph->frag_off   =       0;
1563         iph->daddr      =       daddr;
1564         iph->saddr      =       saddr;
1565         iph->protocol   =       IPPROTO_IPIP;
1566         iph->ihl        =       5;
1567         iph->tot_len    =       htons(skb->len);
1568         ip_select_ident(iph, skb_dst(skb), NULL);
1569         ip_send_check(iph);
1570
1571         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1572         nf_reset(skb);
1573 }
1574
1575 static inline int ipmr_forward_finish(struct sk_buff *skb)
1576 {
1577         struct ip_options *opt = &(IPCB(skb)->opt);
1578
1579         IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1580
1581         if (unlikely(opt->optlen))
1582                 ip_forward_options(skb);
1583
1584         return dst_output(skb);
1585 }
1586
1587 /*
1588  *      Processing handlers for ipmr_forward
1589  */
1590
1591 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1592                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1593 {
1594         const struct iphdr *iph = ip_hdr(skb);
1595         struct vif_device *vif = &mrt->vif_table[vifi];
1596         struct net_device *dev;
1597         struct rtable *rt;
1598         struct flowi4 fl4;
1599         int    encap = 0;
1600
1601         if (vif->dev == NULL)
1602                 goto out_free;
1603
1604 #ifdef CONFIG_IP_PIMSM
1605         if (vif->flags & VIFF_REGISTER) {
1606                 vif->pkt_out++;
1607                 vif->bytes_out += skb->len;
1608                 vif->dev->stats.tx_bytes += skb->len;
1609                 vif->dev->stats.tx_packets++;
1610                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1611                 goto out_free;
1612         }
1613 #endif
1614
1615         if (vif->flags & VIFF_TUNNEL) {
1616                 rt = ip_route_output_ports(net, &fl4, NULL,
1617                                            vif->remote, vif->local,
1618                                            0, 0,
1619                                            IPPROTO_IPIP,
1620                                            RT_TOS(iph->tos), vif->link);
1621                 if (IS_ERR(rt))
1622                         goto out_free;
1623                 encap = sizeof(struct iphdr);
1624         } else {
1625                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1626                                            0, 0,
1627                                            IPPROTO_IPIP,
1628                                            RT_TOS(iph->tos), vif->link);
1629                 if (IS_ERR(rt))
1630                         goto out_free;
1631         }
1632
1633         dev = rt->dst.dev;
1634
1635         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1636                 /* Do not fragment multicasts. Alas, IPv4 does not
1637                  * allow to send ICMP, so that packets will disappear
1638                  * to blackhole.
1639                  */
1640
1641                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1642                 ip_rt_put(rt);
1643                 goto out_free;
1644         }
1645
1646         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1647
1648         if (skb_cow(skb, encap)) {
1649                 ip_rt_put(rt);
1650                 goto out_free;
1651         }
1652
1653         vif->pkt_out++;
1654         vif->bytes_out += skb->len;
1655
1656         skb_dst_drop(skb);
1657         skb_dst_set(skb, &rt->dst);
1658         ip_decrease_ttl(ip_hdr(skb));
1659
1660         /* FIXME: forward and output firewalls used to be called here.
1661          * What do we do with netfilter? -- RR
1662          */
1663         if (vif->flags & VIFF_TUNNEL) {
1664                 ip_encap(skb, vif->local, vif->remote);
1665                 /* FIXME: extra output firewall step used to be here. --RR */
1666                 vif->dev->stats.tx_packets++;
1667                 vif->dev->stats.tx_bytes += skb->len;
1668         }
1669
1670         IPCB(skb)->flags |= IPSKB_FORWARDED;
1671
1672         /*
1673          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1674          * not only before forwarding, but after forwarding on all output
1675          * interfaces. It is clear, if mrouter runs a multicasting
1676          * program, it should receive packets not depending to what interface
1677          * program is joined.
1678          * If we will not make it, the program will have to join on all
1679          * interfaces. On the other hand, multihoming host (or router, but
1680          * not mrouter) cannot join to more than one interface - it will
1681          * result in receiving multiple packets.
1682          */
1683         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1684                 ipmr_forward_finish);
1685         return;
1686
1687 out_free:
1688         kfree_skb(skb);
1689 }
1690
1691 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1692 {
1693         int ct;
1694
1695         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1696                 if (mrt->vif_table[ct].dev == dev)
1697                         break;
1698         }
1699         return ct;
1700 }
1701
1702 /* "local" means that we should preserve one skb (for local delivery) */
1703
1704 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1705                          struct sk_buff *skb, struct mfc_cache *cache,
1706                          int local)
1707 {
1708         int psend = -1;
1709         int vif, ct;
1710
1711         vif = cache->mfc_parent;
1712         cache->mfc_un.res.pkt++;
1713         cache->mfc_un.res.bytes += skb->len;
1714
1715         /*
1716          * Wrong interface: drop packet and (maybe) send PIM assert.
1717          */
1718         if (mrt->vif_table[vif].dev != skb->dev) {
1719                 int true_vifi;
1720
1721                 if (rt_is_output_route(skb_rtable(skb))) {
1722                         /* It is our own packet, looped back.
1723                          * Very complicated situation...
1724                          *
1725                          * The best workaround until routing daemons will be
1726                          * fixed is not to redistribute packet, if it was
1727                          * send through wrong interface. It means, that
1728                          * multicast applications WILL NOT work for
1729                          * (S,G), which have default multicast route pointing
1730                          * to wrong oif. In any case, it is not a good
1731                          * idea to use multicasting applications on router.
1732                          */
1733                         goto dont_forward;
1734                 }
1735
1736                 cache->mfc_un.res.wrong_if++;
1737                 true_vifi = ipmr_find_vif(mrt, skb->dev);
1738
1739                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1740                     /* pimsm uses asserts, when switching from RPT to SPT,
1741                      * so that we cannot check that packet arrived on an oif.
1742                      * It is bad, but otherwise we would need to move pretty
1743                      * large chunk of pimd to kernel. Ough... --ANK
1744                      */
1745                     (mrt->mroute_do_pim ||
1746                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1747                     time_after(jiffies,
1748                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1749                         cache->mfc_un.res.last_assert = jiffies;
1750                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1751                 }
1752                 goto dont_forward;
1753         }
1754
1755         mrt->vif_table[vif].pkt_in++;
1756         mrt->vif_table[vif].bytes_in += skb->len;
1757
1758         /*
1759          *      Forward the frame
1760          */
1761         for (ct = cache->mfc_un.res.maxvif - 1;
1762              ct >= cache->mfc_un.res.minvif; ct--) {
1763                 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1764                         if (psend != -1) {
1765                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1766
1767                                 if (skb2)
1768                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1769                                                         psend);
1770                         }
1771                         psend = ct;
1772                 }
1773         }
1774         if (psend != -1) {
1775                 if (local) {
1776                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1777
1778                         if (skb2)
1779                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1780                 } else {
1781                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1782                         return 0;
1783                 }
1784         }
1785
1786 dont_forward:
1787         if (!local)
1788                 kfree_skb(skb);
1789         return 0;
1790 }
1791
1792 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1793 {
1794         struct rtable *rt = skb_rtable(skb);
1795         struct iphdr *iph = ip_hdr(skb);
1796         struct flowi4 fl4 = {
1797                 .daddr = iph->daddr,
1798                 .saddr = iph->saddr,
1799                 .flowi4_tos = iph->tos,
1800                 .flowi4_oif = rt->rt_oif,
1801                 .flowi4_iif = rt->rt_iif,
1802                 .flowi4_mark = rt->rt_mark,
1803         };
1804         struct mr_table *mrt;
1805         int err;
1806
1807         err = ipmr_fib_lookup(net, &fl4, &mrt);
1808         if (err)
1809                 return ERR_PTR(err);
1810         return mrt;
1811 }
1812
1813 /*
1814  *      Multicast packets for forwarding arrive here
1815  *      Called with rcu_read_lock();
1816  */
1817
1818 int ip_mr_input(struct sk_buff *skb)
1819 {
1820         struct mfc_cache *cache;
1821         struct net *net = dev_net(skb->dev);
1822         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1823         struct mr_table *mrt;
1824
1825         /* Packet is looped back after forward, it should not be
1826          * forwarded second time, but still can be delivered locally.
1827          */
1828         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1829                 goto dont_forward;
1830
1831         mrt = ipmr_rt_fib_lookup(net, skb);
1832         if (IS_ERR(mrt)) {
1833                 kfree_skb(skb);
1834                 return PTR_ERR(mrt);
1835         }
1836         if (!local) {
1837                 if (IPCB(skb)->opt.router_alert) {
1838                         if (ip_call_ra_chain(skb))
1839                                 return 0;
1840                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1841                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1842                          * Cisco IOS <= 11.2(8)) do not put router alert
1843                          * option to IGMP packets destined to routable
1844                          * groups. It is very bad, because it means
1845                          * that we can forward NO IGMP messages.
1846                          */
1847                         struct sock *mroute_sk;
1848
1849                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1850                         if (mroute_sk) {
1851                                 nf_reset(skb);
1852                                 raw_rcv(mroute_sk, skb);
1853                                 return 0;
1854                         }
1855                     }
1856         }
1857
1858         /* already under rcu_read_lock() */
1859         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1860
1861         /*
1862          *      No usable cache entry
1863          */
1864         if (cache == NULL) {
1865                 int vif;
1866
1867                 if (local) {
1868                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1869                         ip_local_deliver(skb);
1870                         if (skb2 == NULL)
1871                                 return -ENOBUFS;
1872                         skb = skb2;
1873                 }
1874
1875                 read_lock(&mrt_lock);
1876                 vif = ipmr_find_vif(mrt, skb->dev);
1877                 if (vif >= 0) {
1878                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1879                         read_unlock(&mrt_lock);
1880
1881                         return err2;
1882                 }
1883                 read_unlock(&mrt_lock);
1884                 kfree_skb(skb);
1885                 return -ENODEV;
1886         }
1887
1888         read_lock(&mrt_lock);
1889         ip_mr_forward(net, mrt, skb, cache, local);
1890         read_unlock(&mrt_lock);
1891
1892         if (local)
1893                 return ip_local_deliver(skb);
1894
1895         return 0;
1896
1897 dont_forward:
1898         if (local)
1899                 return ip_local_deliver(skb);
1900         kfree_skb(skb);
1901         return 0;
1902 }
1903
1904 #ifdef CONFIG_IP_PIMSM
1905 /* called with rcu_read_lock() */
1906 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1907                      unsigned int pimlen)
1908 {
1909         struct net_device *reg_dev = NULL;
1910         struct iphdr *encap;
1911
1912         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1913         /*
1914          * Check that:
1915          * a. packet is really sent to a multicast group
1916          * b. packet is not a NULL-REGISTER
1917          * c. packet is not truncated
1918          */
1919         if (!ipv4_is_multicast(encap->daddr) ||
1920             encap->tot_len == 0 ||
1921             ntohs(encap->tot_len) + pimlen > skb->len)
1922                 return 1;
1923
1924         read_lock(&mrt_lock);
1925         if (mrt->mroute_reg_vif_num >= 0)
1926                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1927         read_unlock(&mrt_lock);
1928
1929         if (reg_dev == NULL)
1930                 return 1;
1931
1932         skb->mac_header = skb->network_header;
1933         skb_pull(skb, (u8 *)encap - skb->data);
1934         skb_reset_network_header(skb);
1935         skb->protocol = htons(ETH_P_IP);
1936         skb->ip_summed = CHECKSUM_NONE;
1937         skb->pkt_type = PACKET_HOST;
1938
1939         skb_tunnel_rx(skb, reg_dev);
1940
1941         netif_rx(skb);
1942
1943         return NET_RX_SUCCESS;
1944 }
1945 #endif
1946
1947 #ifdef CONFIG_IP_PIMSM_V1
1948 /*
1949  * Handle IGMP messages of PIMv1
1950  */
1951
1952 int pim_rcv_v1(struct sk_buff *skb)
1953 {
1954         struct igmphdr *pim;
1955         struct net *net = dev_net(skb->dev);
1956         struct mr_table *mrt;
1957
1958         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1959                 goto drop;
1960
1961         pim = igmp_hdr(skb);
1962
1963         mrt = ipmr_rt_fib_lookup(net, skb);
1964         if (IS_ERR(mrt))
1965                 goto drop;
1966         if (!mrt->mroute_do_pim ||
1967             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
1968                 goto drop;
1969
1970         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1971 drop:
1972                 kfree_skb(skb);
1973         }
1974         return 0;
1975 }
1976 #endif
1977
1978 #ifdef CONFIG_IP_PIMSM_V2
1979 static int pim_rcv(struct sk_buff *skb)
1980 {
1981         struct pimreghdr *pim;
1982         struct net *net = dev_net(skb->dev);
1983         struct mr_table *mrt;
1984
1985         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1986                 goto drop;
1987
1988         pim = (struct pimreghdr *)skb_transport_header(skb);
1989         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
1990             (pim->flags & PIM_NULL_REGISTER) ||
1991             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
1992              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
1993                 goto drop;
1994
1995         mrt = ipmr_rt_fib_lookup(net, skb);
1996         if (IS_ERR(mrt))
1997                 goto drop;
1998         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1999 drop:
2000                 kfree_skb(skb);
2001         }
2002         return 0;
2003 }
2004 #endif
2005
2006 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2007                               struct mfc_cache *c, struct rtmsg *rtm)
2008 {
2009         int ct;
2010         struct rtnexthop *nhp;
2011         u8 *b = skb_tail_pointer(skb);
2012         struct rtattr *mp_head;
2013
2014         /* If cache is unresolved, don't try to parse IIF and OIF */
2015         if (c->mfc_parent >= MAXVIFS)
2016                 return -ENOENT;
2017
2018         if (VIF_EXISTS(mrt, c->mfc_parent))
2019                 RTA_PUT(skb, RTA_IIF, 4, &mrt->vif_table[c->mfc_parent].dev->ifindex);
2020
2021         mp_head = (struct rtattr *)skb_put(skb, RTA_LENGTH(0));
2022
2023         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2024                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2025                         if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
2026                                 goto rtattr_failure;
2027                         nhp = (struct rtnexthop *)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
2028                         nhp->rtnh_flags = 0;
2029                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2030                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2031                         nhp->rtnh_len = sizeof(*nhp);
2032                 }
2033         }
2034         mp_head->rta_type = RTA_MULTIPATH;
2035         mp_head->rta_len = skb_tail_pointer(skb) - (u8 *)mp_head;
2036         rtm->rtm_type = RTN_MULTICAST;
2037         return 1;
2038
2039 rtattr_failure:
2040         nlmsg_trim(skb, b);
2041         return -EMSGSIZE;
2042 }
2043
2044 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2045                    __be32 saddr, __be32 daddr,
2046                    struct rtmsg *rtm, int nowait)
2047 {
2048         struct mfc_cache *cache;
2049         struct mr_table *mrt;
2050         int err;
2051
2052         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2053         if (mrt == NULL)
2054                 return -ENOENT;
2055
2056         rcu_read_lock();
2057         cache = ipmr_cache_find(mrt, saddr, daddr);
2058
2059         if (cache == NULL) {
2060                 struct sk_buff *skb2;
2061                 struct iphdr *iph;
2062                 struct net_device *dev;
2063                 int vif = -1;
2064
2065                 if (nowait) {
2066                         rcu_read_unlock();
2067                         return -EAGAIN;
2068                 }
2069
2070                 dev = skb->dev;
2071                 read_lock(&mrt_lock);
2072                 if (dev)
2073                         vif = ipmr_find_vif(mrt, dev);
2074                 if (vif < 0) {
2075                         read_unlock(&mrt_lock);
2076                         rcu_read_unlock();
2077                         return -ENODEV;
2078                 }
2079                 skb2 = skb_clone(skb, GFP_ATOMIC);
2080                 if (!skb2) {
2081                         read_unlock(&mrt_lock);
2082                         rcu_read_unlock();
2083                         return -ENOMEM;
2084                 }
2085
2086                 skb_push(skb2, sizeof(struct iphdr));
2087                 skb_reset_network_header(skb2);
2088                 iph = ip_hdr(skb2);
2089                 iph->ihl = sizeof(struct iphdr) >> 2;
2090                 iph->saddr = saddr;
2091                 iph->daddr = daddr;
2092                 iph->version = 0;
2093                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2094                 read_unlock(&mrt_lock);
2095                 rcu_read_unlock();
2096                 return err;
2097         }
2098
2099         read_lock(&mrt_lock);
2100         if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2101                 cache->mfc_flags |= MFC_NOTIFY;
2102         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2103         read_unlock(&mrt_lock);
2104         rcu_read_unlock();
2105         return err;
2106 }
2107
2108 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2109                             u32 pid, u32 seq, struct mfc_cache *c)
2110 {
2111         struct nlmsghdr *nlh;
2112         struct rtmsg *rtm;
2113
2114         nlh = nlmsg_put(skb, pid, seq, RTM_NEWROUTE, sizeof(*rtm), NLM_F_MULTI);
2115         if (nlh == NULL)
2116                 return -EMSGSIZE;
2117
2118         rtm = nlmsg_data(nlh);
2119         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2120         rtm->rtm_dst_len  = 32;
2121         rtm->rtm_src_len  = 32;
2122         rtm->rtm_tos      = 0;
2123         rtm->rtm_table    = mrt->id;
2124         NLA_PUT_U32(skb, RTA_TABLE, mrt->id);
2125         rtm->rtm_type     = RTN_MULTICAST;
2126         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2127         rtm->rtm_protocol = RTPROT_UNSPEC;
2128         rtm->rtm_flags    = 0;
2129
2130         NLA_PUT_BE32(skb, RTA_SRC, c->mfc_origin);
2131         NLA_PUT_BE32(skb, RTA_DST, c->mfc_mcastgrp);
2132
2133         if (__ipmr_fill_mroute(mrt, skb, c, rtm) < 0)
2134                 goto nla_put_failure;
2135
2136         return nlmsg_end(skb, nlh);
2137
2138 nla_put_failure:
2139         nlmsg_cancel(skb, nlh);
2140         return -EMSGSIZE;
2141 }
2142
2143 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2144 {
2145         struct net *net = sock_net(skb->sk);
2146         struct mr_table *mrt;
2147         struct mfc_cache *mfc;
2148         unsigned int t = 0, s_t;
2149         unsigned int h = 0, s_h;
2150         unsigned int e = 0, s_e;
2151
2152         s_t = cb->args[0];
2153         s_h = cb->args[1];
2154         s_e = cb->args[2];
2155
2156         rcu_read_lock();
2157         ipmr_for_each_table(mrt, net) {
2158                 if (t < s_t)
2159                         goto next_table;
2160                 if (t > s_t)
2161                         s_h = 0;
2162                 for (h = s_h; h < MFC_LINES; h++) {
2163                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2164                                 if (e < s_e)
2165                                         goto next_entry;
2166                                 if (ipmr_fill_mroute(mrt, skb,
2167                                                      NETLINK_CB(cb->skb).pid,
2168                                                      cb->nlh->nlmsg_seq,
2169                                                      mfc) < 0)
2170                                         goto done;
2171 next_entry:
2172                                 e++;
2173                         }
2174                         e = s_e = 0;
2175                 }
2176                 s_h = 0;
2177 next_table:
2178                 t++;
2179         }
2180 done:
2181         rcu_read_unlock();
2182
2183         cb->args[2] = e;
2184         cb->args[1] = h;
2185         cb->args[0] = t;
2186
2187         return skb->len;
2188 }
2189
2190 #ifdef CONFIG_PROC_FS
2191 /*
2192  *      The /proc interfaces to multicast routing :
2193  *      /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2194  */
2195 struct ipmr_vif_iter {
2196         struct seq_net_private p;
2197         struct mr_table *mrt;
2198         int ct;
2199 };
2200
2201 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2202                                            struct ipmr_vif_iter *iter,
2203                                            loff_t pos)
2204 {
2205         struct mr_table *mrt = iter->mrt;
2206
2207         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2208                 if (!VIF_EXISTS(mrt, iter->ct))
2209                         continue;
2210                 if (pos-- == 0)
2211                         return &mrt->vif_table[iter->ct];
2212         }
2213         return NULL;
2214 }
2215
2216 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2217         __acquires(mrt_lock)
2218 {
2219         struct ipmr_vif_iter *iter = seq->private;
2220         struct net *net = seq_file_net(seq);
2221         struct mr_table *mrt;
2222
2223         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2224         if (mrt == NULL)
2225                 return ERR_PTR(-ENOENT);
2226
2227         iter->mrt = mrt;
2228
2229         read_lock(&mrt_lock);
2230         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2231                 : SEQ_START_TOKEN;
2232 }
2233
2234 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2235 {
2236         struct ipmr_vif_iter *iter = seq->private;
2237         struct net *net = seq_file_net(seq);
2238         struct mr_table *mrt = iter->mrt;
2239
2240         ++*pos;
2241         if (v == SEQ_START_TOKEN)
2242                 return ipmr_vif_seq_idx(net, iter, 0);
2243
2244         while (++iter->ct < mrt->maxvif) {
2245                 if (!VIF_EXISTS(mrt, iter->ct))
2246                         continue;
2247                 return &mrt->vif_table[iter->ct];
2248         }
2249         return NULL;
2250 }
2251
2252 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2253         __releases(mrt_lock)
2254 {
2255         read_unlock(&mrt_lock);
2256 }
2257
2258 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2259 {
2260         struct ipmr_vif_iter *iter = seq->private;
2261         struct mr_table *mrt = iter->mrt;
2262
2263         if (v == SEQ_START_TOKEN) {
2264                 seq_puts(seq,
2265                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2266         } else {
2267                 const struct vif_device *vif = v;
2268                 const char *name =  vif->dev ? vif->dev->name : "none";
2269
2270                 seq_printf(seq,
2271                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2272                            vif - mrt->vif_table,
2273                            name, vif->bytes_in, vif->pkt_in,
2274                            vif->bytes_out, vif->pkt_out,
2275                            vif->flags, vif->local, vif->remote);
2276         }
2277         return 0;
2278 }
2279
2280 static const struct seq_operations ipmr_vif_seq_ops = {
2281         .start = ipmr_vif_seq_start,
2282         .next  = ipmr_vif_seq_next,
2283         .stop  = ipmr_vif_seq_stop,
2284         .show  = ipmr_vif_seq_show,
2285 };
2286
2287 static int ipmr_vif_open(struct inode *inode, struct file *file)
2288 {
2289         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2290                             sizeof(struct ipmr_vif_iter));
2291 }
2292
2293 static const struct file_operations ipmr_vif_fops = {
2294         .owner   = THIS_MODULE,
2295         .open    = ipmr_vif_open,
2296         .read    = seq_read,
2297         .llseek  = seq_lseek,
2298         .release = seq_release_net,
2299 };
2300
2301 struct ipmr_mfc_iter {
2302         struct seq_net_private p;
2303         struct mr_table *mrt;
2304         struct list_head *cache;
2305         int ct;
2306 };
2307
2308
2309 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2310                                           struct ipmr_mfc_iter *it, loff_t pos)
2311 {
2312         struct mr_table *mrt = it->mrt;
2313         struct mfc_cache *mfc;
2314
2315         rcu_read_lock();
2316         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2317                 it->cache = &mrt->mfc_cache_array[it->ct];
2318                 list_for_each_entry_rcu(mfc, it->cache, list)
2319                         if (pos-- == 0)
2320                                 return mfc;
2321         }
2322         rcu_read_unlock();
2323
2324         spin_lock_bh(&mfc_unres_lock);
2325         it->cache = &mrt->mfc_unres_queue;
2326         list_for_each_entry(mfc, it->cache, list)
2327                 if (pos-- == 0)
2328                         return mfc;
2329         spin_unlock_bh(&mfc_unres_lock);
2330
2331         it->cache = NULL;
2332         return NULL;
2333 }
2334
2335
2336 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2337 {
2338         struct ipmr_mfc_iter *it = seq->private;
2339         struct net *net = seq_file_net(seq);
2340         struct mr_table *mrt;
2341
2342         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2343         if (mrt == NULL)
2344                 return ERR_PTR(-ENOENT);
2345
2346         it->mrt = mrt;
2347         it->cache = NULL;
2348         it->ct = 0;
2349         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2350                 : SEQ_START_TOKEN;
2351 }
2352
2353 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2354 {
2355         struct mfc_cache *mfc = v;
2356         struct ipmr_mfc_iter *it = seq->private;
2357         struct net *net = seq_file_net(seq);
2358         struct mr_table *mrt = it->mrt;
2359
2360         ++*pos;
2361
2362         if (v == SEQ_START_TOKEN)
2363                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2364
2365         if (mfc->list.next != it->cache)
2366                 return list_entry(mfc->list.next, struct mfc_cache, list);
2367
2368         if (it->cache == &mrt->mfc_unres_queue)
2369                 goto end_of_list;
2370
2371         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2372
2373         while (++it->ct < MFC_LINES) {
2374                 it->cache = &mrt->mfc_cache_array[it->ct];
2375                 if (list_empty(it->cache))
2376                         continue;
2377                 return list_first_entry(it->cache, struct mfc_cache, list);
2378         }
2379
2380         /* exhausted cache_array, show unresolved */
2381         rcu_read_unlock();
2382         it->cache = &mrt->mfc_unres_queue;
2383         it->ct = 0;
2384
2385         spin_lock_bh(&mfc_unres_lock);
2386         if (!list_empty(it->cache))
2387                 return list_first_entry(it->cache, struct mfc_cache, list);
2388
2389 end_of_list:
2390         spin_unlock_bh(&mfc_unres_lock);
2391         it->cache = NULL;
2392
2393         return NULL;
2394 }
2395
2396 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2397 {
2398         struct ipmr_mfc_iter *it = seq->private;
2399         struct mr_table *mrt = it->mrt;
2400
2401         if (it->cache == &mrt->mfc_unres_queue)
2402                 spin_unlock_bh(&mfc_unres_lock);
2403         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2404                 rcu_read_unlock();
2405 }
2406
2407 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2408 {
2409         int n;
2410
2411         if (v == SEQ_START_TOKEN) {
2412                 seq_puts(seq,
2413                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2414         } else {
2415                 const struct mfc_cache *mfc = v;
2416                 const struct ipmr_mfc_iter *it = seq->private;
2417                 const struct mr_table *mrt = it->mrt;
2418
2419                 seq_printf(seq, "%08X %08X %-3hd",
2420                            (__force u32) mfc->mfc_mcastgrp,
2421                            (__force u32) mfc->mfc_origin,
2422                            mfc->mfc_parent);
2423
2424                 if (it->cache != &mrt->mfc_unres_queue) {
2425                         seq_printf(seq, " %8lu %8lu %8lu",
2426                                    mfc->mfc_un.res.pkt,
2427                                    mfc->mfc_un.res.bytes,
2428                                    mfc->mfc_un.res.wrong_if);
2429                         for (n = mfc->mfc_un.res.minvif;
2430                              n < mfc->mfc_un.res.maxvif; n++) {
2431                                 if (VIF_EXISTS(mrt, n) &&
2432                                     mfc->mfc_un.res.ttls[n] < 255)
2433                                         seq_printf(seq,
2434                                            " %2d:%-3d",
2435                                            n, mfc->mfc_un.res.ttls[n]);
2436                         }
2437                 } else {
2438                         /* unresolved mfc_caches don't contain
2439                          * pkt, bytes and wrong_if values
2440                          */
2441                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2442                 }
2443                 seq_putc(seq, '\n');
2444         }
2445         return 0;
2446 }
2447
2448 static const struct seq_operations ipmr_mfc_seq_ops = {
2449         .start = ipmr_mfc_seq_start,
2450         .next  = ipmr_mfc_seq_next,
2451         .stop  = ipmr_mfc_seq_stop,
2452         .show  = ipmr_mfc_seq_show,
2453 };
2454
2455 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2456 {
2457         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2458                             sizeof(struct ipmr_mfc_iter));
2459 }
2460
2461 static const struct file_operations ipmr_mfc_fops = {
2462         .owner   = THIS_MODULE,
2463         .open    = ipmr_mfc_open,
2464         .read    = seq_read,
2465         .llseek  = seq_lseek,
2466         .release = seq_release_net,
2467 };
2468 #endif
2469
2470 #ifdef CONFIG_IP_PIMSM_V2
2471 static const struct net_protocol pim_protocol = {
2472         .handler        =       pim_rcv,
2473         .netns_ok       =       1,
2474 };
2475 #endif
2476
2477
2478 /*
2479  *      Setup for IP multicast routing
2480  */
2481 static int __net_init ipmr_net_init(struct net *net)
2482 {
2483         int err;
2484
2485         err = ipmr_rules_init(net);
2486         if (err < 0)
2487                 goto fail;
2488
2489 #ifdef CONFIG_PROC_FS
2490         err = -ENOMEM;
2491         if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2492                 goto proc_vif_fail;
2493         if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2494                 goto proc_cache_fail;
2495 #endif
2496         return 0;
2497
2498 #ifdef CONFIG_PROC_FS
2499 proc_cache_fail:
2500         proc_net_remove(net, "ip_mr_vif");
2501 proc_vif_fail:
2502         ipmr_rules_exit(net);
2503 #endif
2504 fail:
2505         return err;
2506 }
2507
2508 static void __net_exit ipmr_net_exit(struct net *net)
2509 {
2510 #ifdef CONFIG_PROC_FS
2511         proc_net_remove(net, "ip_mr_cache");
2512         proc_net_remove(net, "ip_mr_vif");
2513 #endif
2514         ipmr_rules_exit(net);
2515 }
2516
2517 static struct pernet_operations ipmr_net_ops = {
2518         .init = ipmr_net_init,
2519         .exit = ipmr_net_exit,
2520 };
2521
2522 int __init ip_mr_init(void)
2523 {
2524         int err;
2525
2526         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2527                                        sizeof(struct mfc_cache),
2528                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2529                                        NULL);
2530         if (!mrt_cachep)
2531                 return -ENOMEM;
2532
2533         err = register_pernet_subsys(&ipmr_net_ops);
2534         if (err)
2535                 goto reg_pernet_fail;
2536
2537         err = register_netdevice_notifier(&ip_mr_notifier);
2538         if (err)
2539                 goto reg_notif_fail;
2540 #ifdef CONFIG_IP_PIMSM_V2
2541         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2542                 printk(KERN_ERR "ip_mr_init: can't add PIM protocol\n");
2543                 err = -EAGAIN;
2544                 goto add_proto_fail;
2545         }
2546 #endif
2547         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE, NULL, ipmr_rtm_dumproute);
2548         return 0;
2549
2550 #ifdef CONFIG_IP_PIMSM_V2
2551 add_proto_fail:
2552         unregister_netdevice_notifier(&ip_mr_notifier);
2553 #endif
2554 reg_notif_fail:
2555         unregister_pernet_subsys(&ipmr_net_ops);
2556 reg_pernet_fail:
2557         kmem_cache_destroy(mrt_cachep);
2558         return err;
2559 }