]> Pileus Git - ~andy/linux/blob - net/ipv4/ipmr.c
ip6tnl: advertise tunnel param via rtnl
[~andy/linux] / 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/uaccess.h>
30 #include <linux/types.h>
31 #include <linux/capability.h>
32 #include <linux/errno.h>
33 #include <linux/timer.h>
34 #include <linux/mm.h>
35 #include <linux/kernel.h>
36 #include <linux/fcntl.h>
37 #include <linux/stat.h>
38 #include <linux/socket.h>
39 #include <linux/in.h>
40 #include <linux/inet.h>
41 #include <linux/netdevice.h>
42 #include <linux/inetdevice.h>
43 #include <linux/igmp.h>
44 #include <linux/proc_fs.h>
45 #include <linux/seq_file.h>
46 #include <linux/mroute.h>
47 #include <linux/init.h>
48 #include <linux/if_ether.h>
49 #include <linux/slab.h>
50 #include <net/net_namespace.h>
51 #include <net/ip.h>
52 #include <net/protocol.h>
53 #include <linux/skbuff.h>
54 #include <net/route.h>
55 #include <net/sock.h>
56 #include <net/icmp.h>
57 #include <net/udp.h>
58 #include <net/raw.h>
59 #include <linux/notifier.h>
60 #include <linux/if_arp.h>
61 #include <linux/netfilter_ipv4.h>
62 #include <linux/compat.h>
63 #include <linux/export.h>
64 #include <net/ipip.h>
65 #include <net/checksum.h>
66 #include <net/netlink.h>
67 #include <net/fib_rules.h>
68
69 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
70 #define CONFIG_IP_PIMSM 1
71 #endif
72
73 struct mr_table {
74         struct list_head        list;
75 #ifdef CONFIG_NET_NS
76         struct net              *net;
77 #endif
78         u32                     id;
79         struct sock __rcu       *mroute_sk;
80         struct timer_list       ipmr_expire_timer;
81         struct list_head        mfc_unres_queue;
82         struct list_head        mfc_cache_array[MFC_LINES];
83         struct vif_device       vif_table[MAXVIFS];
84         int                     maxvif;
85         atomic_t                cache_resolve_queue_len;
86         int                     mroute_do_assert;
87         int                     mroute_do_pim;
88 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
89         int                     mroute_reg_vif_num;
90 #endif
91 };
92
93 struct ipmr_rule {
94         struct fib_rule         common;
95 };
96
97 struct ipmr_result {
98         struct mr_table         *mrt;
99 };
100
101 /* Big lock, protecting vif table, mrt cache and mroute socket state.
102  * Note that the changes are semaphored via rtnl_lock.
103  */
104
105 static DEFINE_RWLOCK(mrt_lock);
106
107 /*
108  *      Multicast router control variables
109  */
110
111 #define VIF_EXISTS(_mrt, _idx) ((_mrt)->vif_table[_idx].dev != NULL)
112
113 /* Special spinlock for queue of unresolved entries */
114 static DEFINE_SPINLOCK(mfc_unres_lock);
115
116 /* We return to original Alan's scheme. Hash table of resolved
117  * entries is changed only in process context and protected
118  * with weak lock mrt_lock. Queue of unresolved entries is protected
119  * with strong spinlock mfc_unres_lock.
120  *
121  * In this case data path is free of exclusive locks at all.
122  */
123
124 static struct kmem_cache *mrt_cachep __read_mostly;
125
126 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
127 static void ipmr_free_table(struct mr_table *mrt);
128
129 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
130                          struct sk_buff *skb, struct mfc_cache *cache,
131                          int local);
132 static int ipmr_cache_report(struct mr_table *mrt,
133                              struct sk_buff *pkt, vifi_t vifi, int assert);
134 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
135                               struct mfc_cache *c, struct rtmsg *rtm);
136 static void mroute_clean_tables(struct mr_table *mrt);
137 static void ipmr_expire_process(unsigned long arg);
138
139 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
140 #define ipmr_for_each_table(mrt, net) \
141         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
142
143 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
144 {
145         struct mr_table *mrt;
146
147         ipmr_for_each_table(mrt, net) {
148                 if (mrt->id == id)
149                         return mrt;
150         }
151         return NULL;
152 }
153
154 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
155                            struct mr_table **mrt)
156 {
157         struct ipmr_result res;
158         struct fib_lookup_arg arg = { .result = &res, };
159         int err;
160
161         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
162                                flowi4_to_flowi(flp4), 0, &arg);
163         if (err < 0)
164                 return err;
165         *mrt = res.mrt;
166         return 0;
167 }
168
169 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
170                             int flags, struct fib_lookup_arg *arg)
171 {
172         struct ipmr_result *res = arg->result;
173         struct mr_table *mrt;
174
175         switch (rule->action) {
176         case FR_ACT_TO_TBL:
177                 break;
178         case FR_ACT_UNREACHABLE:
179                 return -ENETUNREACH;
180         case FR_ACT_PROHIBIT:
181                 return -EACCES;
182         case FR_ACT_BLACKHOLE:
183         default:
184                 return -EINVAL;
185         }
186
187         mrt = ipmr_get_table(rule->fr_net, rule->table);
188         if (mrt == NULL)
189                 return -EAGAIN;
190         res->mrt = mrt;
191         return 0;
192 }
193
194 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
195 {
196         return 1;
197 }
198
199 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
200         FRA_GENERIC_POLICY,
201 };
202
203 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
204                                struct fib_rule_hdr *frh, struct nlattr **tb)
205 {
206         return 0;
207 }
208
209 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
210                              struct nlattr **tb)
211 {
212         return 1;
213 }
214
215 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
216                           struct fib_rule_hdr *frh)
217 {
218         frh->dst_len = 0;
219         frh->src_len = 0;
220         frh->tos     = 0;
221         return 0;
222 }
223
224 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
225         .family         = RTNL_FAMILY_IPMR,
226         .rule_size      = sizeof(struct ipmr_rule),
227         .addr_size      = sizeof(u32),
228         .action         = ipmr_rule_action,
229         .match          = ipmr_rule_match,
230         .configure      = ipmr_rule_configure,
231         .compare        = ipmr_rule_compare,
232         .default_pref   = fib_default_rule_pref,
233         .fill           = ipmr_rule_fill,
234         .nlgroup        = RTNLGRP_IPV4_RULE,
235         .policy         = ipmr_rule_policy,
236         .owner          = THIS_MODULE,
237 };
238
239 static int __net_init ipmr_rules_init(struct net *net)
240 {
241         struct fib_rules_ops *ops;
242         struct mr_table *mrt;
243         int err;
244
245         ops = fib_rules_register(&ipmr_rules_ops_template, net);
246         if (IS_ERR(ops))
247                 return PTR_ERR(ops);
248
249         INIT_LIST_HEAD(&net->ipv4.mr_tables);
250
251         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
252         if (mrt == NULL) {
253                 err = -ENOMEM;
254                 goto err1;
255         }
256
257         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
258         if (err < 0)
259                 goto err2;
260
261         net->ipv4.mr_rules_ops = ops;
262         return 0;
263
264 err2:
265         kfree(mrt);
266 err1:
267         fib_rules_unregister(ops);
268         return err;
269 }
270
271 static void __net_exit ipmr_rules_exit(struct net *net)
272 {
273         struct mr_table *mrt, *next;
274
275         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
276                 list_del(&mrt->list);
277                 ipmr_free_table(mrt);
278         }
279         fib_rules_unregister(net->ipv4.mr_rules_ops);
280 }
281 #else
282 #define ipmr_for_each_table(mrt, net) \
283         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
284
285 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
286 {
287         return net->ipv4.mrt;
288 }
289
290 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
291                            struct mr_table **mrt)
292 {
293         *mrt = net->ipv4.mrt;
294         return 0;
295 }
296
297 static int __net_init ipmr_rules_init(struct net *net)
298 {
299         net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
300         return net->ipv4.mrt ? 0 : -ENOMEM;
301 }
302
303 static void __net_exit ipmr_rules_exit(struct net *net)
304 {
305         ipmr_free_table(net->ipv4.mrt);
306 }
307 #endif
308
309 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
310 {
311         struct mr_table *mrt;
312         unsigned int i;
313
314         mrt = ipmr_get_table(net, id);
315         if (mrt != NULL)
316                 return mrt;
317
318         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
319         if (mrt == NULL)
320                 return NULL;
321         write_pnet(&mrt->net, net);
322         mrt->id = id;
323
324         /* Forwarding cache */
325         for (i = 0; i < MFC_LINES; i++)
326                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
327
328         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
329
330         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
331                     (unsigned long)mrt);
332
333 #ifdef CONFIG_IP_PIMSM
334         mrt->mroute_reg_vif_num = -1;
335 #endif
336 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
337         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
338 #endif
339         return mrt;
340 }
341
342 static void ipmr_free_table(struct mr_table *mrt)
343 {
344         del_timer_sync(&mrt->ipmr_expire_timer);
345         mroute_clean_tables(mrt);
346         kfree(mrt);
347 }
348
349 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
350
351 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
352 {
353         struct net *net = dev_net(dev);
354
355         dev_close(dev);
356
357         dev = __dev_get_by_name(net, "tunl0");
358         if (dev) {
359                 const struct net_device_ops *ops = dev->netdev_ops;
360                 struct ifreq ifr;
361                 struct ip_tunnel_parm p;
362
363                 memset(&p, 0, sizeof(p));
364                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
365                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
366                 p.iph.version = 4;
367                 p.iph.ihl = 5;
368                 p.iph.protocol = IPPROTO_IPIP;
369                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
370                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
371
372                 if (ops->ndo_do_ioctl) {
373                         mm_segment_t oldfs = get_fs();
374
375                         set_fs(KERNEL_DS);
376                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
377                         set_fs(oldfs);
378                 }
379         }
380 }
381
382 static
383 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
384 {
385         struct net_device  *dev;
386
387         dev = __dev_get_by_name(net, "tunl0");
388
389         if (dev) {
390                 const struct net_device_ops *ops = dev->netdev_ops;
391                 int err;
392                 struct ifreq ifr;
393                 struct ip_tunnel_parm p;
394                 struct in_device  *in_dev;
395
396                 memset(&p, 0, sizeof(p));
397                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
398                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
399                 p.iph.version = 4;
400                 p.iph.ihl = 5;
401                 p.iph.protocol = IPPROTO_IPIP;
402                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
403                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
404
405                 if (ops->ndo_do_ioctl) {
406                         mm_segment_t oldfs = get_fs();
407
408                         set_fs(KERNEL_DS);
409                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
410                         set_fs(oldfs);
411                 } else {
412                         err = -EOPNOTSUPP;
413                 }
414                 dev = NULL;
415
416                 if (err == 0 &&
417                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
418                         dev->flags |= IFF_MULTICAST;
419
420                         in_dev = __in_dev_get_rtnl(dev);
421                         if (in_dev == NULL)
422                                 goto failure;
423
424                         ipv4_devconf_setall(in_dev);
425                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
426
427                         if (dev_open(dev))
428                                 goto failure;
429                         dev_hold(dev);
430                 }
431         }
432         return dev;
433
434 failure:
435         /* allow the register to be completed before unregistering. */
436         rtnl_unlock();
437         rtnl_lock();
438
439         unregister_netdevice(dev);
440         return NULL;
441 }
442
443 #ifdef CONFIG_IP_PIMSM
444
445 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
446 {
447         struct net *net = dev_net(dev);
448         struct mr_table *mrt;
449         struct flowi4 fl4 = {
450                 .flowi4_oif     = dev->ifindex,
451                 .flowi4_iif     = skb->skb_iif,
452                 .flowi4_mark    = skb->mark,
453         };
454         int err;
455
456         err = ipmr_fib_lookup(net, &fl4, &mrt);
457         if (err < 0) {
458                 kfree_skb(skb);
459                 return err;
460         }
461
462         read_lock(&mrt_lock);
463         dev->stats.tx_bytes += skb->len;
464         dev->stats.tx_packets++;
465         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
466         read_unlock(&mrt_lock);
467         kfree_skb(skb);
468         return NETDEV_TX_OK;
469 }
470
471 static const struct net_device_ops reg_vif_netdev_ops = {
472         .ndo_start_xmit = reg_vif_xmit,
473 };
474
475 static void reg_vif_setup(struct net_device *dev)
476 {
477         dev->type               = ARPHRD_PIMREG;
478         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
479         dev->flags              = IFF_NOARP;
480         dev->netdev_ops         = &reg_vif_netdev_ops,
481         dev->destructor         = free_netdev;
482         dev->features           |= NETIF_F_NETNS_LOCAL;
483 }
484
485 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
486 {
487         struct net_device *dev;
488         struct in_device *in_dev;
489         char name[IFNAMSIZ];
490
491         if (mrt->id == RT_TABLE_DEFAULT)
492                 sprintf(name, "pimreg");
493         else
494                 sprintf(name, "pimreg%u", mrt->id);
495
496         dev = alloc_netdev(0, name, reg_vif_setup);
497
498         if (dev == NULL)
499                 return NULL;
500
501         dev_net_set(dev, net);
502
503         if (register_netdevice(dev)) {
504                 free_netdev(dev);
505                 return NULL;
506         }
507         dev->iflink = 0;
508
509         rcu_read_lock();
510         in_dev = __in_dev_get_rcu(dev);
511         if (!in_dev) {
512                 rcu_read_unlock();
513                 goto failure;
514         }
515
516         ipv4_devconf_setall(in_dev);
517         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
518         rcu_read_unlock();
519
520         if (dev_open(dev))
521                 goto failure;
522
523         dev_hold(dev);
524
525         return dev;
526
527 failure:
528         /* allow the register to be completed before unregistering. */
529         rtnl_unlock();
530         rtnl_lock();
531
532         unregister_netdevice(dev);
533         return NULL;
534 }
535 #endif
536
537 /**
538  *      vif_delete - Delete a VIF entry
539  *      @notify: Set to 1, if the caller is a notifier_call
540  */
541
542 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
543                       struct list_head *head)
544 {
545         struct vif_device *v;
546         struct net_device *dev;
547         struct in_device *in_dev;
548
549         if (vifi < 0 || vifi >= mrt->maxvif)
550                 return -EADDRNOTAVAIL;
551
552         v = &mrt->vif_table[vifi];
553
554         write_lock_bh(&mrt_lock);
555         dev = v->dev;
556         v->dev = NULL;
557
558         if (!dev) {
559                 write_unlock_bh(&mrt_lock);
560                 return -EADDRNOTAVAIL;
561         }
562
563 #ifdef CONFIG_IP_PIMSM
564         if (vifi == mrt->mroute_reg_vif_num)
565                 mrt->mroute_reg_vif_num = -1;
566 #endif
567
568         if (vifi + 1 == mrt->maxvif) {
569                 int tmp;
570
571                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
572                         if (VIF_EXISTS(mrt, tmp))
573                                 break;
574                 }
575                 mrt->maxvif = tmp+1;
576         }
577
578         write_unlock_bh(&mrt_lock);
579
580         dev_set_allmulti(dev, -1);
581
582         in_dev = __in_dev_get_rtnl(dev);
583         if (in_dev) {
584                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
585                 ip_rt_multicast_event(in_dev);
586         }
587
588         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
589                 unregister_netdevice_queue(dev, head);
590
591         dev_put(dev);
592         return 0;
593 }
594
595 static void ipmr_cache_free_rcu(struct rcu_head *head)
596 {
597         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
598
599         kmem_cache_free(mrt_cachep, c);
600 }
601
602 static inline void ipmr_cache_free(struct mfc_cache *c)
603 {
604         call_rcu(&c->rcu, ipmr_cache_free_rcu);
605 }
606
607 /* Destroy an unresolved cache entry, killing queued skbs
608  * and reporting error to netlink readers.
609  */
610
611 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
612 {
613         struct net *net = read_pnet(&mrt->net);
614         struct sk_buff *skb;
615         struct nlmsgerr *e;
616
617         atomic_dec(&mrt->cache_resolve_queue_len);
618
619         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
620                 if (ip_hdr(skb)->version == 0) {
621                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
622                         nlh->nlmsg_type = NLMSG_ERROR;
623                         nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
624                         skb_trim(skb, nlh->nlmsg_len);
625                         e = NLMSG_DATA(nlh);
626                         e->error = -ETIMEDOUT;
627                         memset(&e->msg, 0, sizeof(e->msg));
628
629                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
630                 } else {
631                         kfree_skb(skb);
632                 }
633         }
634
635         ipmr_cache_free(c);
636 }
637
638
639 /* Timer process for the unresolved queue. */
640
641 static void ipmr_expire_process(unsigned long arg)
642 {
643         struct mr_table *mrt = (struct mr_table *)arg;
644         unsigned long now;
645         unsigned long expires;
646         struct mfc_cache *c, *next;
647
648         if (!spin_trylock(&mfc_unres_lock)) {
649                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
650                 return;
651         }
652
653         if (list_empty(&mrt->mfc_unres_queue))
654                 goto out;
655
656         now = jiffies;
657         expires = 10*HZ;
658
659         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
660                 if (time_after(c->mfc_un.unres.expires, now)) {
661                         unsigned long interval = c->mfc_un.unres.expires - now;
662                         if (interval < expires)
663                                 expires = interval;
664                         continue;
665                 }
666
667                 list_del(&c->list);
668                 ipmr_destroy_unres(mrt, c);
669         }
670
671         if (!list_empty(&mrt->mfc_unres_queue))
672                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
673
674 out:
675         spin_unlock(&mfc_unres_lock);
676 }
677
678 /* Fill oifs list. It is called under write locked mrt_lock. */
679
680 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
681                                    unsigned char *ttls)
682 {
683         int vifi;
684
685         cache->mfc_un.res.minvif = MAXVIFS;
686         cache->mfc_un.res.maxvif = 0;
687         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
688
689         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
690                 if (VIF_EXISTS(mrt, vifi) &&
691                     ttls[vifi] && ttls[vifi] < 255) {
692                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
693                         if (cache->mfc_un.res.minvif > vifi)
694                                 cache->mfc_un.res.minvif = vifi;
695                         if (cache->mfc_un.res.maxvif <= vifi)
696                                 cache->mfc_un.res.maxvif = vifi + 1;
697                 }
698         }
699 }
700
701 static int vif_add(struct net *net, struct mr_table *mrt,
702                    struct vifctl *vifc, int mrtsock)
703 {
704         int vifi = vifc->vifc_vifi;
705         struct vif_device *v = &mrt->vif_table[vifi];
706         struct net_device *dev;
707         struct in_device *in_dev;
708         int err;
709
710         /* Is vif busy ? */
711         if (VIF_EXISTS(mrt, vifi))
712                 return -EADDRINUSE;
713
714         switch (vifc->vifc_flags) {
715 #ifdef CONFIG_IP_PIMSM
716         case VIFF_REGISTER:
717                 /*
718                  * Special Purpose VIF in PIM
719                  * All the packets will be sent to the daemon
720                  */
721                 if (mrt->mroute_reg_vif_num >= 0)
722                         return -EADDRINUSE;
723                 dev = ipmr_reg_vif(net, mrt);
724                 if (!dev)
725                         return -ENOBUFS;
726                 err = dev_set_allmulti(dev, 1);
727                 if (err) {
728                         unregister_netdevice(dev);
729                         dev_put(dev);
730                         return err;
731                 }
732                 break;
733 #endif
734         case VIFF_TUNNEL:
735                 dev = ipmr_new_tunnel(net, vifc);
736                 if (!dev)
737                         return -ENOBUFS;
738                 err = dev_set_allmulti(dev, 1);
739                 if (err) {
740                         ipmr_del_tunnel(dev, vifc);
741                         dev_put(dev);
742                         return err;
743                 }
744                 break;
745
746         case VIFF_USE_IFINDEX:
747         case 0:
748                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
749                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
750                         if (dev && __in_dev_get_rtnl(dev) == NULL) {
751                                 dev_put(dev);
752                                 return -EADDRNOTAVAIL;
753                         }
754                 } else {
755                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
756                 }
757                 if (!dev)
758                         return -EADDRNOTAVAIL;
759                 err = dev_set_allmulti(dev, 1);
760                 if (err) {
761                         dev_put(dev);
762                         return err;
763                 }
764                 break;
765         default:
766                 return -EINVAL;
767         }
768
769         in_dev = __in_dev_get_rtnl(dev);
770         if (!in_dev) {
771                 dev_put(dev);
772                 return -EADDRNOTAVAIL;
773         }
774         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
775         ip_rt_multicast_event(in_dev);
776
777         /* Fill in the VIF structures */
778
779         v->rate_limit = vifc->vifc_rate_limit;
780         v->local = vifc->vifc_lcl_addr.s_addr;
781         v->remote = vifc->vifc_rmt_addr.s_addr;
782         v->flags = vifc->vifc_flags;
783         if (!mrtsock)
784                 v->flags |= VIFF_STATIC;
785         v->threshold = vifc->vifc_threshold;
786         v->bytes_in = 0;
787         v->bytes_out = 0;
788         v->pkt_in = 0;
789         v->pkt_out = 0;
790         v->link = dev->ifindex;
791         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
792                 v->link = dev->iflink;
793
794         /* And finish update writing critical data */
795         write_lock_bh(&mrt_lock);
796         v->dev = dev;
797 #ifdef CONFIG_IP_PIMSM
798         if (v->flags & VIFF_REGISTER)
799                 mrt->mroute_reg_vif_num = vifi;
800 #endif
801         if (vifi+1 > mrt->maxvif)
802                 mrt->maxvif = vifi+1;
803         write_unlock_bh(&mrt_lock);
804         return 0;
805 }
806
807 /* called with rcu_read_lock() */
808 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
809                                          __be32 origin,
810                                          __be32 mcastgrp)
811 {
812         int line = MFC_HASH(mcastgrp, origin);
813         struct mfc_cache *c;
814
815         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
816                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
817                         return c;
818         }
819         return NULL;
820 }
821
822 /*
823  *      Allocate a multicast cache entry
824  */
825 static struct mfc_cache *ipmr_cache_alloc(void)
826 {
827         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
828
829         if (c)
830                 c->mfc_un.res.minvif = MAXVIFS;
831         return c;
832 }
833
834 static struct mfc_cache *ipmr_cache_alloc_unres(void)
835 {
836         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
837
838         if (c) {
839                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
840                 c->mfc_un.unres.expires = jiffies + 10*HZ;
841         }
842         return c;
843 }
844
845 /*
846  *      A cache entry has gone into a resolved state from queued
847  */
848
849 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
850                                struct mfc_cache *uc, struct mfc_cache *c)
851 {
852         struct sk_buff *skb;
853         struct nlmsgerr *e;
854
855         /* Play the pending entries through our router */
856
857         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
858                 if (ip_hdr(skb)->version == 0) {
859                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
860
861                         if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
862                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
863                                                  (u8 *)nlh;
864                         } else {
865                                 nlh->nlmsg_type = NLMSG_ERROR;
866                                 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
867                                 skb_trim(skb, nlh->nlmsg_len);
868                                 e = NLMSG_DATA(nlh);
869                                 e->error = -EMSGSIZE;
870                                 memset(&e->msg, 0, sizeof(e->msg));
871                         }
872
873                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
874                 } else {
875                         ip_mr_forward(net, mrt, skb, c, 0);
876                 }
877         }
878 }
879
880 /*
881  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
882  *      expects the following bizarre scheme.
883  *
884  *      Called under mrt_lock.
885  */
886
887 static int ipmr_cache_report(struct mr_table *mrt,
888                              struct sk_buff *pkt, vifi_t vifi, int assert)
889 {
890         struct sk_buff *skb;
891         const int ihl = ip_hdrlen(pkt);
892         struct igmphdr *igmp;
893         struct igmpmsg *msg;
894         struct sock *mroute_sk;
895         int ret;
896
897 #ifdef CONFIG_IP_PIMSM
898         if (assert == IGMPMSG_WHOLEPKT)
899                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
900         else
901 #endif
902                 skb = alloc_skb(128, GFP_ATOMIC);
903
904         if (!skb)
905                 return -ENOBUFS;
906
907 #ifdef CONFIG_IP_PIMSM
908         if (assert == IGMPMSG_WHOLEPKT) {
909                 /* Ugly, but we have no choice with this interface.
910                  * Duplicate old header, fix ihl, length etc.
911                  * And all this only to mangle msg->im_msgtype and
912                  * to set msg->im_mbz to "mbz" :-)
913                  */
914                 skb_push(skb, sizeof(struct iphdr));
915                 skb_reset_network_header(skb);
916                 skb_reset_transport_header(skb);
917                 msg = (struct igmpmsg *)skb_network_header(skb);
918                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
919                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
920                 msg->im_mbz = 0;
921                 msg->im_vif = mrt->mroute_reg_vif_num;
922                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
923                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
924                                              sizeof(struct iphdr));
925         } else
926 #endif
927         {
928
929         /* Copy the IP header */
930
931         skb->network_header = skb->tail;
932         skb_put(skb, ihl);
933         skb_copy_to_linear_data(skb, pkt->data, ihl);
934         ip_hdr(skb)->protocol = 0;      /* Flag to the kernel this is a route add */
935         msg = (struct igmpmsg *)skb_network_header(skb);
936         msg->im_vif = vifi;
937         skb_dst_set(skb, dst_clone(skb_dst(pkt)));
938
939         /* Add our header */
940
941         igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
942         igmp->type      =
943         msg->im_msgtype = assert;
944         igmp->code      = 0;
945         ip_hdr(skb)->tot_len = htons(skb->len);         /* Fix the length */
946         skb->transport_header = skb->network_header;
947         }
948
949         rcu_read_lock();
950         mroute_sk = rcu_dereference(mrt->mroute_sk);
951         if (mroute_sk == NULL) {
952                 rcu_read_unlock();
953                 kfree_skb(skb);
954                 return -EINVAL;
955         }
956
957         /* Deliver to mrouted */
958
959         ret = sock_queue_rcv_skb(mroute_sk, skb);
960         rcu_read_unlock();
961         if (ret < 0) {
962                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
963                 kfree_skb(skb);
964         }
965
966         return ret;
967 }
968
969 /*
970  *      Queue a packet for resolution. It gets locked cache entry!
971  */
972
973 static int
974 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
975 {
976         bool found = false;
977         int err;
978         struct mfc_cache *c;
979         const struct iphdr *iph = ip_hdr(skb);
980
981         spin_lock_bh(&mfc_unres_lock);
982         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
983                 if (c->mfc_mcastgrp == iph->daddr &&
984                     c->mfc_origin == iph->saddr) {
985                         found = true;
986                         break;
987                 }
988         }
989
990         if (!found) {
991                 /* Create a new entry if allowable */
992
993                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
994                     (c = ipmr_cache_alloc_unres()) == NULL) {
995                         spin_unlock_bh(&mfc_unres_lock);
996
997                         kfree_skb(skb);
998                         return -ENOBUFS;
999                 }
1000
1001                 /* Fill in the new cache entry */
1002
1003                 c->mfc_parent   = -1;
1004                 c->mfc_origin   = iph->saddr;
1005                 c->mfc_mcastgrp = iph->daddr;
1006
1007                 /* Reflect first query at mrouted. */
1008
1009                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1010                 if (err < 0) {
1011                         /* If the report failed throw the cache entry
1012                            out - Brad Parker
1013                          */
1014                         spin_unlock_bh(&mfc_unres_lock);
1015
1016                         ipmr_cache_free(c);
1017                         kfree_skb(skb);
1018                         return err;
1019                 }
1020
1021                 atomic_inc(&mrt->cache_resolve_queue_len);
1022                 list_add(&c->list, &mrt->mfc_unres_queue);
1023
1024                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1025                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1026         }
1027
1028         /* See if we can append the packet */
1029
1030         if (c->mfc_un.unres.unresolved.qlen > 3) {
1031                 kfree_skb(skb);
1032                 err = -ENOBUFS;
1033         } else {
1034                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1035                 err = 0;
1036         }
1037
1038         spin_unlock_bh(&mfc_unres_lock);
1039         return err;
1040 }
1041
1042 /*
1043  *      MFC cache manipulation by user space mroute daemon
1044  */
1045
1046 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1047 {
1048         int line;
1049         struct mfc_cache *c, *next;
1050
1051         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1052
1053         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1054                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1055                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1056                         list_del_rcu(&c->list);
1057
1058                         ipmr_cache_free(c);
1059                         return 0;
1060                 }
1061         }
1062         return -ENOENT;
1063 }
1064
1065 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1066                         struct mfcctl *mfc, int mrtsock)
1067 {
1068         bool found = false;
1069         int line;
1070         struct mfc_cache *uc, *c;
1071
1072         if (mfc->mfcc_parent >= MAXVIFS)
1073                 return -ENFILE;
1074
1075         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1076
1077         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1078                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1079                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1080                         found = true;
1081                         break;
1082                 }
1083         }
1084
1085         if (found) {
1086                 write_lock_bh(&mrt_lock);
1087                 c->mfc_parent = mfc->mfcc_parent;
1088                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1089                 if (!mrtsock)
1090                         c->mfc_flags |= MFC_STATIC;
1091                 write_unlock_bh(&mrt_lock);
1092                 return 0;
1093         }
1094
1095         if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1096                 return -EINVAL;
1097
1098         c = ipmr_cache_alloc();
1099         if (c == NULL)
1100                 return -ENOMEM;
1101
1102         c->mfc_origin = mfc->mfcc_origin.s_addr;
1103         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1104         c->mfc_parent = mfc->mfcc_parent;
1105         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1106         if (!mrtsock)
1107                 c->mfc_flags |= MFC_STATIC;
1108
1109         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1110
1111         /*
1112          *      Check to see if we resolved a queued list. If so we
1113          *      need to send on the frames and tidy up.
1114          */
1115         found = false;
1116         spin_lock_bh(&mfc_unres_lock);
1117         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1118                 if (uc->mfc_origin == c->mfc_origin &&
1119                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1120                         list_del(&uc->list);
1121                         atomic_dec(&mrt->cache_resolve_queue_len);
1122                         found = true;
1123                         break;
1124                 }
1125         }
1126         if (list_empty(&mrt->mfc_unres_queue))
1127                 del_timer(&mrt->ipmr_expire_timer);
1128         spin_unlock_bh(&mfc_unres_lock);
1129
1130         if (found) {
1131                 ipmr_cache_resolve(net, mrt, uc, c);
1132                 ipmr_cache_free(uc);
1133         }
1134         return 0;
1135 }
1136
1137 /*
1138  *      Close the multicast socket, and clear the vif tables etc
1139  */
1140
1141 static void mroute_clean_tables(struct mr_table *mrt)
1142 {
1143         int i;
1144         LIST_HEAD(list);
1145         struct mfc_cache *c, *next;
1146
1147         /* Shut down all active vif entries */
1148
1149         for (i = 0; i < mrt->maxvif; i++) {
1150                 if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1151                         vif_delete(mrt, i, 0, &list);
1152         }
1153         unregister_netdevice_many(&list);
1154
1155         /* Wipe the cache */
1156
1157         for (i = 0; i < MFC_LINES; i++) {
1158                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1159                         if (c->mfc_flags & MFC_STATIC)
1160                                 continue;
1161                         list_del_rcu(&c->list);
1162                         ipmr_cache_free(c);
1163                 }
1164         }
1165
1166         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1167                 spin_lock_bh(&mfc_unres_lock);
1168                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1169                         list_del(&c->list);
1170                         ipmr_destroy_unres(mrt, c);
1171                 }
1172                 spin_unlock_bh(&mfc_unres_lock);
1173         }
1174 }
1175
1176 /* called from ip_ra_control(), before an RCU grace period,
1177  * we dont need to call synchronize_rcu() here
1178  */
1179 static void mrtsock_destruct(struct sock *sk)
1180 {
1181         struct net *net = sock_net(sk);
1182         struct mr_table *mrt;
1183
1184         rtnl_lock();
1185         ipmr_for_each_table(mrt, net) {
1186                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1187                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1188                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1189                         mroute_clean_tables(mrt);
1190                 }
1191         }
1192         rtnl_unlock();
1193 }
1194
1195 /*
1196  *      Socket options and virtual interface manipulation. The whole
1197  *      virtual interface system is a complete heap, but unfortunately
1198  *      that's how BSD mrouted happens to think. Maybe one day with a proper
1199  *      MOSPF/PIM router set up we can clean this up.
1200  */
1201
1202 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1203 {
1204         int ret;
1205         struct vifctl vif;
1206         struct mfcctl mfc;
1207         struct net *net = sock_net(sk);
1208         struct mr_table *mrt;
1209
1210         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1211         if (mrt == NULL)
1212                 return -ENOENT;
1213
1214         if (optname != MRT_INIT) {
1215                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1216                     !capable(CAP_NET_ADMIN))
1217                         return -EACCES;
1218         }
1219
1220         switch (optname) {
1221         case MRT_INIT:
1222                 if (sk->sk_type != SOCK_RAW ||
1223                     inet_sk(sk)->inet_num != IPPROTO_IGMP)
1224                         return -EOPNOTSUPP;
1225                 if (optlen != sizeof(int))
1226                         return -ENOPROTOOPT;
1227
1228                 rtnl_lock();
1229                 if (rtnl_dereference(mrt->mroute_sk)) {
1230                         rtnl_unlock();
1231                         return -EADDRINUSE;
1232                 }
1233
1234                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1235                 if (ret == 0) {
1236                         rcu_assign_pointer(mrt->mroute_sk, sk);
1237                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1238                 }
1239                 rtnl_unlock();
1240                 return ret;
1241         case MRT_DONE:
1242                 if (sk != rcu_access_pointer(mrt->mroute_sk))
1243                         return -EACCES;
1244                 return ip_ra_control(sk, 0, NULL);
1245         case MRT_ADD_VIF:
1246         case MRT_DEL_VIF:
1247                 if (optlen != sizeof(vif))
1248                         return -EINVAL;
1249                 if (copy_from_user(&vif, optval, sizeof(vif)))
1250                         return -EFAULT;
1251                 if (vif.vifc_vifi >= MAXVIFS)
1252                         return -ENFILE;
1253                 rtnl_lock();
1254                 if (optname == MRT_ADD_VIF) {
1255                         ret = vif_add(net, mrt, &vif,
1256                                       sk == rtnl_dereference(mrt->mroute_sk));
1257                 } else {
1258                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1259                 }
1260                 rtnl_unlock();
1261                 return ret;
1262
1263                 /*
1264                  *      Manipulate the forwarding caches. These live
1265                  *      in a sort of kernel/user symbiosis.
1266                  */
1267         case MRT_ADD_MFC:
1268         case MRT_DEL_MFC:
1269                 if (optlen != sizeof(mfc))
1270                         return -EINVAL;
1271                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1272                         return -EFAULT;
1273                 rtnl_lock();
1274                 if (optname == MRT_DEL_MFC)
1275                         ret = ipmr_mfc_delete(mrt, &mfc);
1276                 else
1277                         ret = ipmr_mfc_add(net, mrt, &mfc,
1278                                            sk == rtnl_dereference(mrt->mroute_sk));
1279                 rtnl_unlock();
1280                 return ret;
1281                 /*
1282                  *      Control PIM assert.
1283                  */
1284         case MRT_ASSERT:
1285         {
1286                 int v;
1287                 if (get_user(v, (int __user *)optval))
1288                         return -EFAULT;
1289                 mrt->mroute_do_assert = (v) ? 1 : 0;
1290                 return 0;
1291         }
1292 #ifdef CONFIG_IP_PIMSM
1293         case MRT_PIM:
1294         {
1295                 int v;
1296
1297                 if (get_user(v, (int __user *)optval))
1298                         return -EFAULT;
1299                 v = (v) ? 1 : 0;
1300
1301                 rtnl_lock();
1302                 ret = 0;
1303                 if (v != mrt->mroute_do_pim) {
1304                         mrt->mroute_do_pim = v;
1305                         mrt->mroute_do_assert = v;
1306                 }
1307                 rtnl_unlock();
1308                 return ret;
1309         }
1310 #endif
1311 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1312         case MRT_TABLE:
1313         {
1314                 u32 v;
1315
1316                 if (optlen != sizeof(u32))
1317                         return -EINVAL;
1318                 if (get_user(v, (u32 __user *)optval))
1319                         return -EFAULT;
1320
1321                 rtnl_lock();
1322                 ret = 0;
1323                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1324                         ret = -EBUSY;
1325                 } else {
1326                         if (!ipmr_new_table(net, v))
1327                                 ret = -ENOMEM;
1328                         raw_sk(sk)->ipmr_table = v;
1329                 }
1330                 rtnl_unlock();
1331                 return ret;
1332         }
1333 #endif
1334         /*
1335          *      Spurious command, or MRT_VERSION which you cannot
1336          *      set.
1337          */
1338         default:
1339                 return -ENOPROTOOPT;
1340         }
1341 }
1342
1343 /*
1344  *      Getsock opt support for the multicast routing system.
1345  */
1346
1347 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1348 {
1349         int olr;
1350         int val;
1351         struct net *net = sock_net(sk);
1352         struct mr_table *mrt;
1353
1354         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1355         if (mrt == NULL)
1356                 return -ENOENT;
1357
1358         if (optname != MRT_VERSION &&
1359 #ifdef CONFIG_IP_PIMSM
1360            optname != MRT_PIM &&
1361 #endif
1362            optname != MRT_ASSERT)
1363                 return -ENOPROTOOPT;
1364
1365         if (get_user(olr, optlen))
1366                 return -EFAULT;
1367
1368         olr = min_t(unsigned int, olr, sizeof(int));
1369         if (olr < 0)
1370                 return -EINVAL;
1371
1372         if (put_user(olr, optlen))
1373                 return -EFAULT;
1374         if (optname == MRT_VERSION)
1375                 val = 0x0305;
1376 #ifdef CONFIG_IP_PIMSM
1377         else if (optname == MRT_PIM)
1378                 val = mrt->mroute_do_pim;
1379 #endif
1380         else
1381                 val = mrt->mroute_do_assert;
1382         if (copy_to_user(optval, &val, olr))
1383                 return -EFAULT;
1384         return 0;
1385 }
1386
1387 /*
1388  *      The IP multicast ioctl support routines.
1389  */
1390
1391 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1392 {
1393         struct sioc_sg_req sr;
1394         struct sioc_vif_req vr;
1395         struct vif_device *vif;
1396         struct mfc_cache *c;
1397         struct net *net = sock_net(sk);
1398         struct mr_table *mrt;
1399
1400         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1401         if (mrt == NULL)
1402                 return -ENOENT;
1403
1404         switch (cmd) {
1405         case SIOCGETVIFCNT:
1406                 if (copy_from_user(&vr, arg, sizeof(vr)))
1407                         return -EFAULT;
1408                 if (vr.vifi >= mrt->maxvif)
1409                         return -EINVAL;
1410                 read_lock(&mrt_lock);
1411                 vif = &mrt->vif_table[vr.vifi];
1412                 if (VIF_EXISTS(mrt, vr.vifi)) {
1413                         vr.icount = vif->pkt_in;
1414                         vr.ocount = vif->pkt_out;
1415                         vr.ibytes = vif->bytes_in;
1416                         vr.obytes = vif->bytes_out;
1417                         read_unlock(&mrt_lock);
1418
1419                         if (copy_to_user(arg, &vr, sizeof(vr)))
1420                                 return -EFAULT;
1421                         return 0;
1422                 }
1423                 read_unlock(&mrt_lock);
1424                 return -EADDRNOTAVAIL;
1425         case SIOCGETSGCNT:
1426                 if (copy_from_user(&sr, arg, sizeof(sr)))
1427                         return -EFAULT;
1428
1429                 rcu_read_lock();
1430                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1431                 if (c) {
1432                         sr.pktcnt = c->mfc_un.res.pkt;
1433                         sr.bytecnt = c->mfc_un.res.bytes;
1434                         sr.wrong_if = c->mfc_un.res.wrong_if;
1435                         rcu_read_unlock();
1436
1437                         if (copy_to_user(arg, &sr, sizeof(sr)))
1438                                 return -EFAULT;
1439                         return 0;
1440                 }
1441                 rcu_read_unlock();
1442                 return -EADDRNOTAVAIL;
1443         default:
1444                 return -ENOIOCTLCMD;
1445         }
1446 }
1447
1448 #ifdef CONFIG_COMPAT
1449 struct compat_sioc_sg_req {
1450         struct in_addr src;
1451         struct in_addr grp;
1452         compat_ulong_t pktcnt;
1453         compat_ulong_t bytecnt;
1454         compat_ulong_t wrong_if;
1455 };
1456
1457 struct compat_sioc_vif_req {
1458         vifi_t  vifi;           /* Which iface */
1459         compat_ulong_t icount;
1460         compat_ulong_t ocount;
1461         compat_ulong_t ibytes;
1462         compat_ulong_t obytes;
1463 };
1464
1465 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1466 {
1467         struct compat_sioc_sg_req sr;
1468         struct compat_sioc_vif_req vr;
1469         struct vif_device *vif;
1470         struct mfc_cache *c;
1471         struct net *net = sock_net(sk);
1472         struct mr_table *mrt;
1473
1474         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1475         if (mrt == NULL)
1476                 return -ENOENT;
1477
1478         switch (cmd) {
1479         case SIOCGETVIFCNT:
1480                 if (copy_from_user(&vr, arg, sizeof(vr)))
1481                         return -EFAULT;
1482                 if (vr.vifi >= mrt->maxvif)
1483                         return -EINVAL;
1484                 read_lock(&mrt_lock);
1485                 vif = &mrt->vif_table[vr.vifi];
1486                 if (VIF_EXISTS(mrt, vr.vifi)) {
1487                         vr.icount = vif->pkt_in;
1488                         vr.ocount = vif->pkt_out;
1489                         vr.ibytes = vif->bytes_in;
1490                         vr.obytes = vif->bytes_out;
1491                         read_unlock(&mrt_lock);
1492
1493                         if (copy_to_user(arg, &vr, sizeof(vr)))
1494                                 return -EFAULT;
1495                         return 0;
1496                 }
1497                 read_unlock(&mrt_lock);
1498                 return -EADDRNOTAVAIL;
1499         case SIOCGETSGCNT:
1500                 if (copy_from_user(&sr, arg, sizeof(sr)))
1501                         return -EFAULT;
1502
1503                 rcu_read_lock();
1504                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1505                 if (c) {
1506                         sr.pktcnt = c->mfc_un.res.pkt;
1507                         sr.bytecnt = c->mfc_un.res.bytes;
1508                         sr.wrong_if = c->mfc_un.res.wrong_if;
1509                         rcu_read_unlock();
1510
1511                         if (copy_to_user(arg, &sr, sizeof(sr)))
1512                                 return -EFAULT;
1513                         return 0;
1514                 }
1515                 rcu_read_unlock();
1516                 return -EADDRNOTAVAIL;
1517         default:
1518                 return -ENOIOCTLCMD;
1519         }
1520 }
1521 #endif
1522
1523
1524 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1525 {
1526         struct net_device *dev = ptr;
1527         struct net *net = dev_net(dev);
1528         struct mr_table *mrt;
1529         struct vif_device *v;
1530         int ct;
1531
1532         if (event != NETDEV_UNREGISTER)
1533                 return NOTIFY_DONE;
1534
1535         ipmr_for_each_table(mrt, net) {
1536                 v = &mrt->vif_table[0];
1537                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1538                         if (v->dev == dev)
1539                                 vif_delete(mrt, ct, 1, NULL);
1540                 }
1541         }
1542         return NOTIFY_DONE;
1543 }
1544
1545
1546 static struct notifier_block ip_mr_notifier = {
1547         .notifier_call = ipmr_device_event,
1548 };
1549
1550 /*
1551  *      Encapsulate a packet by attaching a valid IPIP header to it.
1552  *      This avoids tunnel drivers and other mess and gives us the speed so
1553  *      important for multicast video.
1554  */
1555
1556 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1557 {
1558         struct iphdr *iph;
1559         const struct iphdr *old_iph = ip_hdr(skb);
1560
1561         skb_push(skb, sizeof(struct iphdr));
1562         skb->transport_header = skb->network_header;
1563         skb_reset_network_header(skb);
1564         iph = ip_hdr(skb);
1565
1566         iph->version    =       4;
1567         iph->tos        =       old_iph->tos;
1568         iph->ttl        =       old_iph->ttl;
1569         iph->frag_off   =       0;
1570         iph->daddr      =       daddr;
1571         iph->saddr      =       saddr;
1572         iph->protocol   =       IPPROTO_IPIP;
1573         iph->ihl        =       5;
1574         iph->tot_len    =       htons(skb->len);
1575         ip_select_ident(iph, skb_dst(skb), NULL);
1576         ip_send_check(iph);
1577
1578         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1579         nf_reset(skb);
1580 }
1581
1582 static inline int ipmr_forward_finish(struct sk_buff *skb)
1583 {
1584         struct ip_options *opt = &(IPCB(skb)->opt);
1585
1586         IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1587         IP_ADD_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTOCTETS, skb->len);
1588
1589         if (unlikely(opt->optlen))
1590                 ip_forward_options(skb);
1591
1592         return dst_output(skb);
1593 }
1594
1595 /*
1596  *      Processing handlers for ipmr_forward
1597  */
1598
1599 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1600                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1601 {
1602         const struct iphdr *iph = ip_hdr(skb);
1603         struct vif_device *vif = &mrt->vif_table[vifi];
1604         struct net_device *dev;
1605         struct rtable *rt;
1606         struct flowi4 fl4;
1607         int    encap = 0;
1608
1609         if (vif->dev == NULL)
1610                 goto out_free;
1611
1612 #ifdef CONFIG_IP_PIMSM
1613         if (vif->flags & VIFF_REGISTER) {
1614                 vif->pkt_out++;
1615                 vif->bytes_out += skb->len;
1616                 vif->dev->stats.tx_bytes += skb->len;
1617                 vif->dev->stats.tx_packets++;
1618                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1619                 goto out_free;
1620         }
1621 #endif
1622
1623         if (vif->flags & VIFF_TUNNEL) {
1624                 rt = ip_route_output_ports(net, &fl4, NULL,
1625                                            vif->remote, vif->local,
1626                                            0, 0,
1627                                            IPPROTO_IPIP,
1628                                            RT_TOS(iph->tos), vif->link);
1629                 if (IS_ERR(rt))
1630                         goto out_free;
1631                 encap = sizeof(struct iphdr);
1632         } else {
1633                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1634                                            0, 0,
1635                                            IPPROTO_IPIP,
1636                                            RT_TOS(iph->tos), vif->link);
1637                 if (IS_ERR(rt))
1638                         goto out_free;
1639         }
1640
1641         dev = rt->dst.dev;
1642
1643         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1644                 /* Do not fragment multicasts. Alas, IPv4 does not
1645                  * allow to send ICMP, so that packets will disappear
1646                  * to blackhole.
1647                  */
1648
1649                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1650                 ip_rt_put(rt);
1651                 goto out_free;
1652         }
1653
1654         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1655
1656         if (skb_cow(skb, encap)) {
1657                 ip_rt_put(rt);
1658                 goto out_free;
1659         }
1660
1661         vif->pkt_out++;
1662         vif->bytes_out += skb->len;
1663
1664         skb_dst_drop(skb);
1665         skb_dst_set(skb, &rt->dst);
1666         ip_decrease_ttl(ip_hdr(skb));
1667
1668         /* FIXME: forward and output firewalls used to be called here.
1669          * What do we do with netfilter? -- RR
1670          */
1671         if (vif->flags & VIFF_TUNNEL) {
1672                 ip_encap(skb, vif->local, vif->remote);
1673                 /* FIXME: extra output firewall step used to be here. --RR */
1674                 vif->dev->stats.tx_packets++;
1675                 vif->dev->stats.tx_bytes += skb->len;
1676         }
1677
1678         IPCB(skb)->flags |= IPSKB_FORWARDED;
1679
1680         /*
1681          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1682          * not only before forwarding, but after forwarding on all output
1683          * interfaces. It is clear, if mrouter runs a multicasting
1684          * program, it should receive packets not depending to what interface
1685          * program is joined.
1686          * If we will not make it, the program will have to join on all
1687          * interfaces. On the other hand, multihoming host (or router, but
1688          * not mrouter) cannot join to more than one interface - it will
1689          * result in receiving multiple packets.
1690          */
1691         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1692                 ipmr_forward_finish);
1693         return;
1694
1695 out_free:
1696         kfree_skb(skb);
1697 }
1698
1699 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1700 {
1701         int ct;
1702
1703         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1704                 if (mrt->vif_table[ct].dev == dev)
1705                         break;
1706         }
1707         return ct;
1708 }
1709
1710 /* "local" means that we should preserve one skb (for local delivery) */
1711
1712 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1713                          struct sk_buff *skb, struct mfc_cache *cache,
1714                          int local)
1715 {
1716         int psend = -1;
1717         int vif, ct;
1718
1719         vif = cache->mfc_parent;
1720         cache->mfc_un.res.pkt++;
1721         cache->mfc_un.res.bytes += skb->len;
1722
1723         /*
1724          * Wrong interface: drop packet and (maybe) send PIM assert.
1725          */
1726         if (mrt->vif_table[vif].dev != skb->dev) {
1727                 int true_vifi;
1728
1729                 if (rt_is_output_route(skb_rtable(skb))) {
1730                         /* It is our own packet, looped back.
1731                          * Very complicated situation...
1732                          *
1733                          * The best workaround until routing daemons will be
1734                          * fixed is not to redistribute packet, if it was
1735                          * send through wrong interface. It means, that
1736                          * multicast applications WILL NOT work for
1737                          * (S,G), which have default multicast route pointing
1738                          * to wrong oif. In any case, it is not a good
1739                          * idea to use multicasting applications on router.
1740                          */
1741                         goto dont_forward;
1742                 }
1743
1744                 cache->mfc_un.res.wrong_if++;
1745                 true_vifi = ipmr_find_vif(mrt, skb->dev);
1746
1747                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1748                     /* pimsm uses asserts, when switching from RPT to SPT,
1749                      * so that we cannot check that packet arrived on an oif.
1750                      * It is bad, but otherwise we would need to move pretty
1751                      * large chunk of pimd to kernel. Ough... --ANK
1752                      */
1753                     (mrt->mroute_do_pim ||
1754                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1755                     time_after(jiffies,
1756                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1757                         cache->mfc_un.res.last_assert = jiffies;
1758                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1759                 }
1760                 goto dont_forward;
1761         }
1762
1763         mrt->vif_table[vif].pkt_in++;
1764         mrt->vif_table[vif].bytes_in += skb->len;
1765
1766         /*
1767          *      Forward the frame
1768          */
1769         for (ct = cache->mfc_un.res.maxvif - 1;
1770              ct >= cache->mfc_un.res.minvif; ct--) {
1771                 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1772                         if (psend != -1) {
1773                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1774
1775                                 if (skb2)
1776                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1777                                                         psend);
1778                         }
1779                         psend = ct;
1780                 }
1781         }
1782         if (psend != -1) {
1783                 if (local) {
1784                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1785
1786                         if (skb2)
1787                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1788                 } else {
1789                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1790                         return 0;
1791                 }
1792         }
1793
1794 dont_forward:
1795         if (!local)
1796                 kfree_skb(skb);
1797         return 0;
1798 }
1799
1800 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1801 {
1802         struct rtable *rt = skb_rtable(skb);
1803         struct iphdr *iph = ip_hdr(skb);
1804         struct flowi4 fl4 = {
1805                 .daddr = iph->daddr,
1806                 .saddr = iph->saddr,
1807                 .flowi4_tos = RT_TOS(iph->tos),
1808                 .flowi4_oif = (rt_is_output_route(rt) ?
1809                                skb->dev->ifindex : 0),
1810                 .flowi4_iif = (rt_is_output_route(rt) ?
1811                                LOOPBACK_IFINDEX :
1812                                skb->dev->ifindex),
1813                 .flowi4_mark = skb->mark,
1814         };
1815         struct mr_table *mrt;
1816         int err;
1817
1818         err = ipmr_fib_lookup(net, &fl4, &mrt);
1819         if (err)
1820                 return ERR_PTR(err);
1821         return mrt;
1822 }
1823
1824 /*
1825  *      Multicast packets for forwarding arrive here
1826  *      Called with rcu_read_lock();
1827  */
1828
1829 int ip_mr_input(struct sk_buff *skb)
1830 {
1831         struct mfc_cache *cache;
1832         struct net *net = dev_net(skb->dev);
1833         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1834         struct mr_table *mrt;
1835
1836         /* Packet is looped back after forward, it should not be
1837          * forwarded second time, but still can be delivered locally.
1838          */
1839         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1840                 goto dont_forward;
1841
1842         mrt = ipmr_rt_fib_lookup(net, skb);
1843         if (IS_ERR(mrt)) {
1844                 kfree_skb(skb);
1845                 return PTR_ERR(mrt);
1846         }
1847         if (!local) {
1848                 if (IPCB(skb)->opt.router_alert) {
1849                         if (ip_call_ra_chain(skb))
1850                                 return 0;
1851                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1852                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1853                          * Cisco IOS <= 11.2(8)) do not put router alert
1854                          * option to IGMP packets destined to routable
1855                          * groups. It is very bad, because it means
1856                          * that we can forward NO IGMP messages.
1857                          */
1858                         struct sock *mroute_sk;
1859
1860                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1861                         if (mroute_sk) {
1862                                 nf_reset(skb);
1863                                 raw_rcv(mroute_sk, skb);
1864                                 return 0;
1865                         }
1866                     }
1867         }
1868
1869         /* already under rcu_read_lock() */
1870         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1871
1872         /*
1873          *      No usable cache entry
1874          */
1875         if (cache == NULL) {
1876                 int vif;
1877
1878                 if (local) {
1879                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1880                         ip_local_deliver(skb);
1881                         if (skb2 == NULL)
1882                                 return -ENOBUFS;
1883                         skb = skb2;
1884                 }
1885
1886                 read_lock(&mrt_lock);
1887                 vif = ipmr_find_vif(mrt, skb->dev);
1888                 if (vif >= 0) {
1889                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1890                         read_unlock(&mrt_lock);
1891
1892                         return err2;
1893                 }
1894                 read_unlock(&mrt_lock);
1895                 kfree_skb(skb);
1896                 return -ENODEV;
1897         }
1898
1899         read_lock(&mrt_lock);
1900         ip_mr_forward(net, mrt, skb, cache, local);
1901         read_unlock(&mrt_lock);
1902
1903         if (local)
1904                 return ip_local_deliver(skb);
1905
1906         return 0;
1907
1908 dont_forward:
1909         if (local)
1910                 return ip_local_deliver(skb);
1911         kfree_skb(skb);
1912         return 0;
1913 }
1914
1915 #ifdef CONFIG_IP_PIMSM
1916 /* called with rcu_read_lock() */
1917 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1918                      unsigned int pimlen)
1919 {
1920         struct net_device *reg_dev = NULL;
1921         struct iphdr *encap;
1922
1923         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1924         /*
1925          * Check that:
1926          * a. packet is really sent to a multicast group
1927          * b. packet is not a NULL-REGISTER
1928          * c. packet is not truncated
1929          */
1930         if (!ipv4_is_multicast(encap->daddr) ||
1931             encap->tot_len == 0 ||
1932             ntohs(encap->tot_len) + pimlen > skb->len)
1933                 return 1;
1934
1935         read_lock(&mrt_lock);
1936         if (mrt->mroute_reg_vif_num >= 0)
1937                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1938         read_unlock(&mrt_lock);
1939
1940         if (reg_dev == NULL)
1941                 return 1;
1942
1943         skb->mac_header = skb->network_header;
1944         skb_pull(skb, (u8 *)encap - skb->data);
1945         skb_reset_network_header(skb);
1946         skb->protocol = htons(ETH_P_IP);
1947         skb->ip_summed = CHECKSUM_NONE;
1948         skb->pkt_type = PACKET_HOST;
1949
1950         skb_tunnel_rx(skb, reg_dev);
1951
1952         netif_rx(skb);
1953
1954         return NET_RX_SUCCESS;
1955 }
1956 #endif
1957
1958 #ifdef CONFIG_IP_PIMSM_V1
1959 /*
1960  * Handle IGMP messages of PIMv1
1961  */
1962
1963 int pim_rcv_v1(struct sk_buff *skb)
1964 {
1965         struct igmphdr *pim;
1966         struct net *net = dev_net(skb->dev);
1967         struct mr_table *mrt;
1968
1969         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1970                 goto drop;
1971
1972         pim = igmp_hdr(skb);
1973
1974         mrt = ipmr_rt_fib_lookup(net, skb);
1975         if (IS_ERR(mrt))
1976                 goto drop;
1977         if (!mrt->mroute_do_pim ||
1978             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
1979                 goto drop;
1980
1981         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1982 drop:
1983                 kfree_skb(skb);
1984         }
1985         return 0;
1986 }
1987 #endif
1988
1989 #ifdef CONFIG_IP_PIMSM_V2
1990 static int pim_rcv(struct sk_buff *skb)
1991 {
1992         struct pimreghdr *pim;
1993         struct net *net = dev_net(skb->dev);
1994         struct mr_table *mrt;
1995
1996         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1997                 goto drop;
1998
1999         pim = (struct pimreghdr *)skb_transport_header(skb);
2000         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2001             (pim->flags & PIM_NULL_REGISTER) ||
2002             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2003              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2004                 goto drop;
2005
2006         mrt = ipmr_rt_fib_lookup(net, skb);
2007         if (IS_ERR(mrt))
2008                 goto drop;
2009         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2010 drop:
2011                 kfree_skb(skb);
2012         }
2013         return 0;
2014 }
2015 #endif
2016
2017 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2018                               struct mfc_cache *c, struct rtmsg *rtm)
2019 {
2020         int ct;
2021         struct rtnexthop *nhp;
2022         struct nlattr *mp_attr;
2023
2024         /* If cache is unresolved, don't try to parse IIF and OIF */
2025         if (c->mfc_parent >= MAXVIFS)
2026                 return -ENOENT;
2027
2028         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2029             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2030                 return -EMSGSIZE;
2031
2032         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2033                 return -EMSGSIZE;
2034
2035         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2036                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2037                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2038                                 nla_nest_cancel(skb, mp_attr);
2039                                 return -EMSGSIZE;
2040                         }
2041
2042                         nhp->rtnh_flags = 0;
2043                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2044                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2045                         nhp->rtnh_len = sizeof(*nhp);
2046                 }
2047         }
2048
2049         nla_nest_end(skb, mp_attr);
2050
2051         rtm->rtm_type = RTN_MULTICAST;
2052         return 1;
2053 }
2054
2055 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2056                    __be32 saddr, __be32 daddr,
2057                    struct rtmsg *rtm, int nowait)
2058 {
2059         struct mfc_cache *cache;
2060         struct mr_table *mrt;
2061         int err;
2062
2063         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2064         if (mrt == NULL)
2065                 return -ENOENT;
2066
2067         rcu_read_lock();
2068         cache = ipmr_cache_find(mrt, saddr, daddr);
2069
2070         if (cache == NULL) {
2071                 struct sk_buff *skb2;
2072                 struct iphdr *iph;
2073                 struct net_device *dev;
2074                 int vif = -1;
2075
2076                 if (nowait) {
2077                         rcu_read_unlock();
2078                         return -EAGAIN;
2079                 }
2080
2081                 dev = skb->dev;
2082                 read_lock(&mrt_lock);
2083                 if (dev)
2084                         vif = ipmr_find_vif(mrt, dev);
2085                 if (vif < 0) {
2086                         read_unlock(&mrt_lock);
2087                         rcu_read_unlock();
2088                         return -ENODEV;
2089                 }
2090                 skb2 = skb_clone(skb, GFP_ATOMIC);
2091                 if (!skb2) {
2092                         read_unlock(&mrt_lock);
2093                         rcu_read_unlock();
2094                         return -ENOMEM;
2095                 }
2096
2097                 skb_push(skb2, sizeof(struct iphdr));
2098                 skb_reset_network_header(skb2);
2099                 iph = ip_hdr(skb2);
2100                 iph->ihl = sizeof(struct iphdr) >> 2;
2101                 iph->saddr = saddr;
2102                 iph->daddr = daddr;
2103                 iph->version = 0;
2104                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2105                 read_unlock(&mrt_lock);
2106                 rcu_read_unlock();
2107                 return err;
2108         }
2109
2110         read_lock(&mrt_lock);
2111         if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2112                 cache->mfc_flags |= MFC_NOTIFY;
2113         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2114         read_unlock(&mrt_lock);
2115         rcu_read_unlock();
2116         return err;
2117 }
2118
2119 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2120                             u32 portid, u32 seq, struct mfc_cache *c)
2121 {
2122         struct nlmsghdr *nlh;
2123         struct rtmsg *rtm;
2124
2125         nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*rtm), NLM_F_MULTI);
2126         if (nlh == NULL)
2127                 return -EMSGSIZE;
2128
2129         rtm = nlmsg_data(nlh);
2130         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2131         rtm->rtm_dst_len  = 32;
2132         rtm->rtm_src_len  = 32;
2133         rtm->rtm_tos      = 0;
2134         rtm->rtm_table    = mrt->id;
2135         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2136                 goto nla_put_failure;
2137         rtm->rtm_type     = RTN_MULTICAST;
2138         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2139         rtm->rtm_protocol = RTPROT_UNSPEC;
2140         rtm->rtm_flags    = 0;
2141
2142         if (nla_put_be32(skb, RTA_SRC, c->mfc_origin) ||
2143             nla_put_be32(skb, RTA_DST, c->mfc_mcastgrp))
2144                 goto nla_put_failure;
2145         if (__ipmr_fill_mroute(mrt, skb, c, rtm) < 0)
2146                 goto nla_put_failure;
2147
2148         return nlmsg_end(skb, nlh);
2149
2150 nla_put_failure:
2151         nlmsg_cancel(skb, nlh);
2152         return -EMSGSIZE;
2153 }
2154
2155 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2156 {
2157         struct net *net = sock_net(skb->sk);
2158         struct mr_table *mrt;
2159         struct mfc_cache *mfc;
2160         unsigned int t = 0, s_t;
2161         unsigned int h = 0, s_h;
2162         unsigned int e = 0, s_e;
2163
2164         s_t = cb->args[0];
2165         s_h = cb->args[1];
2166         s_e = cb->args[2];
2167
2168         rcu_read_lock();
2169         ipmr_for_each_table(mrt, net) {
2170                 if (t < s_t)
2171                         goto next_table;
2172                 if (t > s_t)
2173                         s_h = 0;
2174                 for (h = s_h; h < MFC_LINES; h++) {
2175                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2176                                 if (e < s_e)
2177                                         goto next_entry;
2178                                 if (ipmr_fill_mroute(mrt, skb,
2179                                                      NETLINK_CB(cb->skb).portid,
2180                                                      cb->nlh->nlmsg_seq,
2181                                                      mfc) < 0)
2182                                         goto done;
2183 next_entry:
2184                                 e++;
2185                         }
2186                         e = s_e = 0;
2187                 }
2188                 s_h = 0;
2189 next_table:
2190                 t++;
2191         }
2192 done:
2193         rcu_read_unlock();
2194
2195         cb->args[2] = e;
2196         cb->args[1] = h;
2197         cb->args[0] = t;
2198
2199         return skb->len;
2200 }
2201
2202 #ifdef CONFIG_PROC_FS
2203 /*
2204  *      The /proc interfaces to multicast routing :
2205  *      /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2206  */
2207 struct ipmr_vif_iter {
2208         struct seq_net_private p;
2209         struct mr_table *mrt;
2210         int ct;
2211 };
2212
2213 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2214                                            struct ipmr_vif_iter *iter,
2215                                            loff_t pos)
2216 {
2217         struct mr_table *mrt = iter->mrt;
2218
2219         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2220                 if (!VIF_EXISTS(mrt, iter->ct))
2221                         continue;
2222                 if (pos-- == 0)
2223                         return &mrt->vif_table[iter->ct];
2224         }
2225         return NULL;
2226 }
2227
2228 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2229         __acquires(mrt_lock)
2230 {
2231         struct ipmr_vif_iter *iter = seq->private;
2232         struct net *net = seq_file_net(seq);
2233         struct mr_table *mrt;
2234
2235         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2236         if (mrt == NULL)
2237                 return ERR_PTR(-ENOENT);
2238
2239         iter->mrt = mrt;
2240
2241         read_lock(&mrt_lock);
2242         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2243                 : SEQ_START_TOKEN;
2244 }
2245
2246 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2247 {
2248         struct ipmr_vif_iter *iter = seq->private;
2249         struct net *net = seq_file_net(seq);
2250         struct mr_table *mrt = iter->mrt;
2251
2252         ++*pos;
2253         if (v == SEQ_START_TOKEN)
2254                 return ipmr_vif_seq_idx(net, iter, 0);
2255
2256         while (++iter->ct < mrt->maxvif) {
2257                 if (!VIF_EXISTS(mrt, iter->ct))
2258                         continue;
2259                 return &mrt->vif_table[iter->ct];
2260         }
2261         return NULL;
2262 }
2263
2264 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2265         __releases(mrt_lock)
2266 {
2267         read_unlock(&mrt_lock);
2268 }
2269
2270 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2271 {
2272         struct ipmr_vif_iter *iter = seq->private;
2273         struct mr_table *mrt = iter->mrt;
2274
2275         if (v == SEQ_START_TOKEN) {
2276                 seq_puts(seq,
2277                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2278         } else {
2279                 const struct vif_device *vif = v;
2280                 const char *name =  vif->dev ? vif->dev->name : "none";
2281
2282                 seq_printf(seq,
2283                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2284                            vif - mrt->vif_table,
2285                            name, vif->bytes_in, vif->pkt_in,
2286                            vif->bytes_out, vif->pkt_out,
2287                            vif->flags, vif->local, vif->remote);
2288         }
2289         return 0;
2290 }
2291
2292 static const struct seq_operations ipmr_vif_seq_ops = {
2293         .start = ipmr_vif_seq_start,
2294         .next  = ipmr_vif_seq_next,
2295         .stop  = ipmr_vif_seq_stop,
2296         .show  = ipmr_vif_seq_show,
2297 };
2298
2299 static int ipmr_vif_open(struct inode *inode, struct file *file)
2300 {
2301         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2302                             sizeof(struct ipmr_vif_iter));
2303 }
2304
2305 static const struct file_operations ipmr_vif_fops = {
2306         .owner   = THIS_MODULE,
2307         .open    = ipmr_vif_open,
2308         .read    = seq_read,
2309         .llseek  = seq_lseek,
2310         .release = seq_release_net,
2311 };
2312
2313 struct ipmr_mfc_iter {
2314         struct seq_net_private p;
2315         struct mr_table *mrt;
2316         struct list_head *cache;
2317         int ct;
2318 };
2319
2320
2321 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2322                                           struct ipmr_mfc_iter *it, loff_t pos)
2323 {
2324         struct mr_table *mrt = it->mrt;
2325         struct mfc_cache *mfc;
2326
2327         rcu_read_lock();
2328         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2329                 it->cache = &mrt->mfc_cache_array[it->ct];
2330                 list_for_each_entry_rcu(mfc, it->cache, list)
2331                         if (pos-- == 0)
2332                                 return mfc;
2333         }
2334         rcu_read_unlock();
2335
2336         spin_lock_bh(&mfc_unres_lock);
2337         it->cache = &mrt->mfc_unres_queue;
2338         list_for_each_entry(mfc, it->cache, list)
2339                 if (pos-- == 0)
2340                         return mfc;
2341         spin_unlock_bh(&mfc_unres_lock);
2342
2343         it->cache = NULL;
2344         return NULL;
2345 }
2346
2347
2348 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2349 {
2350         struct ipmr_mfc_iter *it = seq->private;
2351         struct net *net = seq_file_net(seq);
2352         struct mr_table *mrt;
2353
2354         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2355         if (mrt == NULL)
2356                 return ERR_PTR(-ENOENT);
2357
2358         it->mrt = mrt;
2359         it->cache = NULL;
2360         it->ct = 0;
2361         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2362                 : SEQ_START_TOKEN;
2363 }
2364
2365 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2366 {
2367         struct mfc_cache *mfc = v;
2368         struct ipmr_mfc_iter *it = seq->private;
2369         struct net *net = seq_file_net(seq);
2370         struct mr_table *mrt = it->mrt;
2371
2372         ++*pos;
2373
2374         if (v == SEQ_START_TOKEN)
2375                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2376
2377         if (mfc->list.next != it->cache)
2378                 return list_entry(mfc->list.next, struct mfc_cache, list);
2379
2380         if (it->cache == &mrt->mfc_unres_queue)
2381                 goto end_of_list;
2382
2383         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2384
2385         while (++it->ct < MFC_LINES) {
2386                 it->cache = &mrt->mfc_cache_array[it->ct];
2387                 if (list_empty(it->cache))
2388                         continue;
2389                 return list_first_entry(it->cache, struct mfc_cache, list);
2390         }
2391
2392         /* exhausted cache_array, show unresolved */
2393         rcu_read_unlock();
2394         it->cache = &mrt->mfc_unres_queue;
2395         it->ct = 0;
2396
2397         spin_lock_bh(&mfc_unres_lock);
2398         if (!list_empty(it->cache))
2399                 return list_first_entry(it->cache, struct mfc_cache, list);
2400
2401 end_of_list:
2402         spin_unlock_bh(&mfc_unres_lock);
2403         it->cache = NULL;
2404
2405         return NULL;
2406 }
2407
2408 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2409 {
2410         struct ipmr_mfc_iter *it = seq->private;
2411         struct mr_table *mrt = it->mrt;
2412
2413         if (it->cache == &mrt->mfc_unres_queue)
2414                 spin_unlock_bh(&mfc_unres_lock);
2415         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2416                 rcu_read_unlock();
2417 }
2418
2419 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2420 {
2421         int n;
2422
2423         if (v == SEQ_START_TOKEN) {
2424                 seq_puts(seq,
2425                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2426         } else {
2427                 const struct mfc_cache *mfc = v;
2428                 const struct ipmr_mfc_iter *it = seq->private;
2429                 const struct mr_table *mrt = it->mrt;
2430
2431                 seq_printf(seq, "%08X %08X %-3hd",
2432                            (__force u32) mfc->mfc_mcastgrp,
2433                            (__force u32) mfc->mfc_origin,
2434                            mfc->mfc_parent);
2435
2436                 if (it->cache != &mrt->mfc_unres_queue) {
2437                         seq_printf(seq, " %8lu %8lu %8lu",
2438                                    mfc->mfc_un.res.pkt,
2439                                    mfc->mfc_un.res.bytes,
2440                                    mfc->mfc_un.res.wrong_if);
2441                         for (n = mfc->mfc_un.res.minvif;
2442                              n < mfc->mfc_un.res.maxvif; n++) {
2443                                 if (VIF_EXISTS(mrt, n) &&
2444                                     mfc->mfc_un.res.ttls[n] < 255)
2445                                         seq_printf(seq,
2446                                            " %2d:%-3d",
2447                                            n, mfc->mfc_un.res.ttls[n]);
2448                         }
2449                 } else {
2450                         /* unresolved mfc_caches don't contain
2451                          * pkt, bytes and wrong_if values
2452                          */
2453                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2454                 }
2455                 seq_putc(seq, '\n');
2456         }
2457         return 0;
2458 }
2459
2460 static const struct seq_operations ipmr_mfc_seq_ops = {
2461         .start = ipmr_mfc_seq_start,
2462         .next  = ipmr_mfc_seq_next,
2463         .stop  = ipmr_mfc_seq_stop,
2464         .show  = ipmr_mfc_seq_show,
2465 };
2466
2467 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2468 {
2469         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2470                             sizeof(struct ipmr_mfc_iter));
2471 }
2472
2473 static const struct file_operations ipmr_mfc_fops = {
2474         .owner   = THIS_MODULE,
2475         .open    = ipmr_mfc_open,
2476         .read    = seq_read,
2477         .llseek  = seq_lseek,
2478         .release = seq_release_net,
2479 };
2480 #endif
2481
2482 #ifdef CONFIG_IP_PIMSM_V2
2483 static const struct net_protocol pim_protocol = {
2484         .handler        =       pim_rcv,
2485         .netns_ok       =       1,
2486 };
2487 #endif
2488
2489
2490 /*
2491  *      Setup for IP multicast routing
2492  */
2493 static int __net_init ipmr_net_init(struct net *net)
2494 {
2495         int err;
2496
2497         err = ipmr_rules_init(net);
2498         if (err < 0)
2499                 goto fail;
2500
2501 #ifdef CONFIG_PROC_FS
2502         err = -ENOMEM;
2503         if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2504                 goto proc_vif_fail;
2505         if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2506                 goto proc_cache_fail;
2507 #endif
2508         return 0;
2509
2510 #ifdef CONFIG_PROC_FS
2511 proc_cache_fail:
2512         proc_net_remove(net, "ip_mr_vif");
2513 proc_vif_fail:
2514         ipmr_rules_exit(net);
2515 #endif
2516 fail:
2517         return err;
2518 }
2519
2520 static void __net_exit ipmr_net_exit(struct net *net)
2521 {
2522 #ifdef CONFIG_PROC_FS
2523         proc_net_remove(net, "ip_mr_cache");
2524         proc_net_remove(net, "ip_mr_vif");
2525 #endif
2526         ipmr_rules_exit(net);
2527 }
2528
2529 static struct pernet_operations ipmr_net_ops = {
2530         .init = ipmr_net_init,
2531         .exit = ipmr_net_exit,
2532 };
2533
2534 int __init ip_mr_init(void)
2535 {
2536         int err;
2537
2538         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2539                                        sizeof(struct mfc_cache),
2540                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2541                                        NULL);
2542         if (!mrt_cachep)
2543                 return -ENOMEM;
2544
2545         err = register_pernet_subsys(&ipmr_net_ops);
2546         if (err)
2547                 goto reg_pernet_fail;
2548
2549         err = register_netdevice_notifier(&ip_mr_notifier);
2550         if (err)
2551                 goto reg_notif_fail;
2552 #ifdef CONFIG_IP_PIMSM_V2
2553         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2554                 pr_err("%s: can't add PIM protocol\n", __func__);
2555                 err = -EAGAIN;
2556                 goto add_proto_fail;
2557         }
2558 #endif
2559         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2560                       NULL, ipmr_rtm_dumproute, NULL);
2561         return 0;
2562
2563 #ifdef CONFIG_IP_PIMSM_V2
2564 add_proto_fail:
2565         unregister_netdevice_notifier(&ip_mr_notifier);
2566 #endif
2567 reg_notif_fail:
2568         unregister_pernet_subsys(&ipmr_net_ops);
2569 reg_pernet_fail:
2570         kmem_cache_destroy(mrt_cachep);
2571         return err;
2572 }