net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <[email protected]>
   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 flowi *flp,
 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, flp, 0, &arg);
 159         if (err < 0)
 160                 return err;
 161         *mrt = res.mrt;
 162         return 0;
 163 }
 164
 165 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 166                             int flags, struct fib_lookup_arg *arg)
 167 {
 168         struct ipmr_result *res = arg->result;
 169         struct mr_table *mrt;
 170
 171         switch (rule->action) {
 172         case FR_ACT_TO_TBL:
 173                 break;
 174         case FR_ACT_UNREACHABLE:
 175                 return -ENETUNREACH;
 176         case FR_ACT_PROHIBIT:
 177                 return -EACCES;
 178         case FR_ACT_BLACKHOLE:
 179         default:
 180                 return -EINVAL;
 181         }
 182
 183         mrt = ipmr_get_table(rule->fr_net, rule->table);
 184         if (mrt == NULL)
 185                 return -EAGAIN;
 186         res->mrt = mrt;
 187         return 0;
 188 }
 189
 190 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 191 {
 192         return 1;
 193 }
 194
 195 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 196         FRA_GENERIC_POLICY,
 197 };
 198
 199 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 200                                struct fib_rule_hdr *frh, struct nlattr **tb)
 201 {
 202         return 0;
 203 }
 204
 205 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 206                              struct nlattr **tb)
 207 {
 208         return 1;
 209 }
 210
 211 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 212                           struct fib_rule_hdr *frh)
 213 {
 214         frh->dst_len = 0;
 215         frh->src_len = 0;
 216         frh->tos     = 0;
 217         return 0;
 218 }
 219
 220 static const struct fib_rules_ops __net_initdata ipmr_rules_ops_template = {
 221         .family         = RTNL_FAMILY_IPMR,
 222         .rule_size      = sizeof(struct ipmr_rule),
 223         .addr_size      = sizeof(u32),
 224         .action         = ipmr_rule_action,
 225         .match          = ipmr_rule_match,
 226         .configure      = ipmr_rule_configure,
 227         .compare        = ipmr_rule_compare,
 228         .default_pref   = fib_default_rule_pref,
 229         .fill           = ipmr_rule_fill,
 230         .nlgroup        = RTNLGRP_IPV4_RULE,
 231         .policy         = ipmr_rule_policy,
 232         .owner          = THIS_MODULE,
 233 };
 234
 235 static int __net_init ipmr_rules_init(struct net *net)
 236 {
 237         struct fib_rules_ops *ops;
 238         struct mr_table *mrt;
 239         int err;
 240
 241         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 242         if (IS_ERR(ops))
 243                 return PTR_ERR(ops);
 244
 245         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 246
 247         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 248         if (mrt == NULL) {
 249                 err = -ENOMEM;
 250                 goto err1;
 251         }
 252
 253         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 254         if (err < 0)
 255                 goto err2;
 256
 257         net->ipv4.mr_rules_ops = ops;
 258         return 0;
 259
 260 err2:
 261         kfree(mrt);
 262 err1:
 263         fib_rules_unregister(ops);
 264         return err;
 265 }
 266
 267 static void __net_exit ipmr_rules_exit(struct net *net)
 268 {
 269         struct mr_table *mrt, *next;
 270
 271         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 272                 list_del(&mrt->list);
 273                 kfree(mrt);
 274         }
 275         fib_rules_unregister(net->ipv4.mr_rules_ops);
 276 }
 277 #else
 278 #define ipmr_for_each_table(mrt, net) \
 279         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 280
 281 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 282 {
 283         return net->ipv4.mrt;
 284 }
 285
 286 static int ipmr_fib_lookup(struct net *net, struct flowi *flp,
 287                            struct mr_table **mrt)
 288 {
 289         *mrt = net->ipv4.mrt;
 290         return 0;
 291 }
 292
 293 static int __net_init ipmr_rules_init(struct net *net)
 294 {
 295         net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 296         return net->ipv4.mrt ? 0 : -ENOMEM;
 297 }
 298
 299 static void __net_exit ipmr_rules_exit(struct net *net)
 300 {
 301         kfree(net->ipv4.mrt);
 302 }
 303 #endif
 304
 305 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 306 {
 307         struct mr_table *mrt;
 308         unsigned int i;
 309
 310         mrt = ipmr_get_table(net, id);
 311         if (mrt != NULL)
 312                 return mrt;
 313
 314         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 315         if (mrt == NULL)
 316                 return NULL;
 317         write_pnet(&mrt->net, net);
 318         mrt->id = id;
 319
 320         /* Forwarding cache */
 321         for (i = 0; i < MFC_LINES; i++)
 322                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 323
 324         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 325
 326         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 327                     (unsigned long)mrt);
 328
 329 #ifdef CONFIG_IP_PIMSM
 330         mrt->mroute_reg_vif_num = -1;
 331 #endif
 332 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 333         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 334 #endif
 335         return mrt;
 336 }
 337
 338 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 339
 340 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 341 {
 342         struct net *net = dev_net(dev);
 343
 344         dev_close(dev);
 345
 346         dev = __dev_get_by_name(net, "tunl0");
 347         if (dev) {
 348                 const struct net_device_ops *ops = dev->netdev_ops;
 349                 struct ifreq ifr;
 350                 struct ip_tunnel_parm p;
 351
 352                 memset(&p, 0, sizeof(p));
 353                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 354                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 355                 p.iph.version = 4;
 356                 p.iph.ihl = 5;
 357                 p.iph.protocol = IPPROTO_IPIP;
 358                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 359                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 360
 361                 if (ops->ndo_do_ioctl) {
 362                         mm_segment_t oldfs = get_fs();
 363
 364                         set_fs(KERNEL_DS);
 365                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 366                         set_fs(oldfs);
 367                 }
 368         }
 369 }
 370
 371 static
 372 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 373 {
 374         struct net_device  *dev;
 375
 376         dev = __dev_get_by_name(net, "tunl0");
 377
 378         if (dev) {
 379                 const struct net_device_ops *ops = dev->netdev_ops;
 380                 int err;
 381                 struct ifreq ifr;
 382                 struct ip_tunnel_parm p;
 383                 struct in_device  *in_dev;
 384
 385                 memset(&p, 0, sizeof(p));
 386                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 387                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 388                 p.iph.version = 4;
 389                 p.iph.ihl = 5;
 390                 p.iph.protocol = IPPROTO_IPIP;
 391                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 392                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 393
 394                 if (ops->ndo_do_ioctl) {
 395                         mm_segment_t oldfs = get_fs();
 396
 397                         set_fs(KERNEL_DS);
 398                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 399                         set_fs(oldfs);
 400                 } else {
 401                         err = -EOPNOTSUPP;
 402                 }
 403                 dev = NULL;
 404
 405                 if (err == 0 &&
 406                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 407                         dev->flags |= IFF_MULTICAST;
 408
 409                         in_dev = __in_dev_get_rtnl(dev);
 410                         if (in_dev == NULL)
 411                                 goto failure;
 412
 413                         ipv4_devconf_setall(in_dev);
 414                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 415
 416                         if (dev_open(dev))
 417                                 goto failure;
 418                         dev_hold(dev);
 419                 }
 420         }
 421         return dev;
 422
 423 failure:
 424         /* allow the register to be completed before unregistering. */
 425         rtnl_unlock();
 426         rtnl_lock();
 427
 428         unregister_netdevice(dev);
 429         return NULL;
 430 }
 431
 432 #ifdef CONFIG_IP_PIMSM
 433
 434 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 435 {
 436         struct net *net = dev_net(dev);
 437         struct mr_table *mrt;
 438         struct flowi fl = {
 439                 .oif            = dev->ifindex,
 440                 .iif            = skb->skb_iif,
 441                 .mark           = skb->mark,
 442         };
 443         int err;
 444
 445         err = ipmr_fib_lookup(net, &fl, &mrt);
 446         if (err < 0) {
 447                 kfree_skb(skb);
 448                 return err;
 449         }
 450
 451         read_lock(&mrt_lock);
 452         dev->stats.tx_bytes += skb->len;
 453         dev->stats.tx_packets++;
 454         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 455         read_unlock(&mrt_lock);
 456         kfree_skb(skb);
 457         return NETDEV_TX_OK;
 458 }
 459
 460 static const struct net_device_ops reg_vif_netdev_ops = {
 461         .ndo_start_xmit = reg_vif_xmit,
 462 };
 463
 464 static void reg_vif_setup(struct net_device *dev)
 465 {
 466         dev->type               = ARPHRD_PIMREG;
 467         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 468         dev->flags              = IFF_NOARP;
 469         dev->netdev_ops         = &reg_vif_netdev_ops,
 470         dev->destructor         = free_netdev;
 471         dev->features           |= NETIF_F_NETNS_LOCAL;
 472 }
 473
 474 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 475 {
 476         struct net_device *dev;
 477         struct in_device *in_dev;
 478         char name[IFNAMSIZ];
 479
 480         if (mrt->id == RT_TABLE_DEFAULT)
 481                 sprintf(name, "pimreg");
 482         else
 483                 sprintf(name, "pimreg%u", mrt->id);
 484
 485         dev = alloc_netdev(0, name, reg_vif_setup);
 486
 487         if (dev == NULL)
 488                 return NULL;
 489
 490         dev_net_set(dev, net);
 491
 492         if (register_netdevice(dev)) {
 493                 free_netdev(dev);
 494                 return NULL;
 495         }
 496         dev->iflink = 0;
 497
 498         rcu_read_lock();
 499         in_dev = __in_dev_get_rcu(dev);
 500         if (!in_dev) {
 501                 rcu_read_unlock();
 502                 goto failure;
 503         }
 504
 505         ipv4_devconf_setall(in_dev);
 506         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 507         rcu_read_unlock();
 508
 509         if (dev_open(dev))
 510                 goto failure;
 511
 512         dev_hold(dev);
 513
 514         return dev;
 515
 516 failure:
 517         /* allow the register to be completed before unregistering. */
 518         rtnl_unlock();
 519         rtnl_lock();
 520
 521         unregister_netdevice(dev);
 522         return NULL;
 523 }
 524 #endif
 525
 526 /*
 527  *      Delete a VIF entry
 528  *      @notify: Set to 1, if the caller is a notifier_call
 529  */
 530
 531 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 532                       struct list_head *head)
 533 {
 534         struct vif_device *v;
 535         struct net_device *dev;
 536         struct in_device *in_dev;
 537
 538         if (vifi < 0 || vifi >= mrt->maxvif)
 539                 return -EADDRNOTAVAIL;
 540
 541         v = &mrt->vif_table[vifi];
 542
 543         write_lock_bh(&mrt_lock);
 544         dev = v->dev;
 545         v->dev = NULL;
 546
 547         if (!dev) {
 548                 write_unlock_bh(&mrt_lock);
 549                 return -EADDRNOTAVAIL;
 550         }
 551
 552 #ifdef CONFIG_IP_PIMSM
 553         if (vifi == mrt->mroute_reg_vif_num)
 554                 mrt->mroute_reg_vif_num = -1;
 555 #endif
 556
 557         if (vifi + 1 == mrt->maxvif) {
 558                 int tmp;
 559
 560                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 561                         if (VIF_EXISTS(mrt, tmp))
 562                                 break;
 563                 }
 564                 mrt->maxvif = tmp+1;
 565         }
 566
 567         write_unlock_bh(&mrt_lock);
 568
 569         dev_set_allmulti(dev, -1);
 570
 571         in_dev = __in_dev_get_rtnl(dev);
 572         if (in_dev) {
 573                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 574                 ip_rt_multicast_event(in_dev);
 575         }
 576
 577         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 578                 unregister_netdevice_queue(dev, head);
 579
 580         dev_put(dev);
 581         return 0;
 582 }
 583
 584 static void ipmr_cache_free_rcu(struct rcu_head *head)
 585 {
 586         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 587
 588         kmem_cache_free(mrt_cachep, c);
 589 }
 590
 591 static inline void ipmr_cache_free(struct mfc_cache *c)
 592 {
 593         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 594 }
 595
 596 /* Destroy an unresolved cache entry, killing queued skbs
 597  * and reporting error to netlink readers.
 598  */
 599
 600 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 601 {
 602         struct net *net = read_pnet(&mrt->net);
 603         struct sk_buff *skb;
 604         struct nlmsgerr *e;
 605
 606         atomic_dec(&mrt->cache_resolve_queue_len);
 607
 608         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 609                 if (ip_hdr(skb)->version == 0) {
 610                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 611                         nlh->nlmsg_type = NLMSG_ERROR;
 612                         nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 613                         skb_trim(skb, nlh->nlmsg_len);
 614                         e = NLMSG_DATA(nlh);
 615                         e->error = -ETIMEDOUT;
 616                         memset(&e->msg, 0, sizeof(e->msg));
 617
 618                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
 619                 } else {
 620                         kfree_skb(skb);
 621                 }
 622         }
 623
 624         ipmr_cache_free(c);
 625 }
 626
 627
 628 /* Timer process for the unresolved queue. */
 629
 630 static void ipmr_expire_process(unsigned long arg)
 631 {
 632         struct mr_table *mrt = (struct mr_table *)arg;
 633         unsigned long now;
 634         unsigned long expires;
 635         struct mfc_cache *c, *next;
 636
 637         if (!spin_trylock(&mfc_unres_lock)) {
 638                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 639                 return;
 640         }
 641
 642         if (list_empty(&mrt->mfc_unres_queue))
 643                 goto out;
 644
 645         now = jiffies;
 646         expires = 10*HZ;
 647
 648         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 649                 if (time_after(c->mfc_un.unres.expires, now)) {
 650                         unsigned long interval = c->mfc_un.unres.expires - now;
 651                         if (interval < expires)
 652                                 expires = interval;
 653                         continue;
 654                 }
 655
 656                 list_del(&c->list);
 657                 ipmr_destroy_unres(mrt, c);
 658         }
 659
 660         if (!list_empty(&mrt->mfc_unres_queue))
 661                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 662
 663 out:
 664         spin_unlock(&mfc_unres_lock);
 665 }
 666
 667 /* Fill oifs list. It is called under write locked mrt_lock. */
 668
 669 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 670                                    unsigned char *ttls)
 671 {
 672         int vifi;
 673
 674         cache->mfc_un.res.minvif = MAXVIFS;
 675         cache->mfc_un.res.maxvif = 0;
 676         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 677
 678         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 679                 if (VIF_EXISTS(mrt, vifi) &&
 680                     ttls[vifi] && ttls[vifi] < 255) {
 681                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 682                         if (cache->mfc_un.res.minvif > vifi)
 683                                 cache->mfc_un.res.minvif = vifi;
 684                         if (cache->mfc_un.res.maxvif <= vifi)
 685                                 cache->mfc_un.res.maxvif = vifi + 1;
 686                 }
 687         }
 688 }
 689
 690 static int vif_add(struct net *net, struct mr_table *mrt,
 691                    struct vifctl *vifc, int mrtsock)
 692 {
 693         int vifi = vifc->vifc_vifi;
 694         struct vif_device *v = &mrt->vif_table[vifi];
 695         struct net_device *dev;
 696         struct in_device *in_dev;
 697         int err;
 698
 699         /* Is vif busy ? */
 700         if (VIF_EXISTS(mrt, vifi))
 701                 return -EADDRINUSE;
 702
 703         switch (vifc->vifc_flags) {
 704 #ifdef CONFIG_IP_PIMSM
 705         case VIFF_REGISTER:
 706                 /*
 707                  * Special Purpose VIF in PIM
 708                  * All the packets will be sent to the daemon
 709                  */
 710                 if (mrt->mroute_reg_vif_num >= 0)
 711                         return -EADDRINUSE;
 712                 dev = ipmr_reg_vif(net, mrt);
 713                 if (!dev)
 714                         return -ENOBUFS;
 715                 err = dev_set_allmulti(dev, 1);
 716                 if (err) {
 717                         unregister_netdevice(dev);
 718                         dev_put(dev);
 719                         return err;
 720                 }
 721                 break;
 722 #endif
 723         case VIFF_TUNNEL:
 724                 dev = ipmr_new_tunnel(net, vifc);
 725                 if (!dev)
 726                         return -ENOBUFS;
 727                 err = dev_set_allmulti(dev, 1);
 728                 if (err) {
 729                         ipmr_del_tunnel(dev, vifc);
 730                         dev_put(dev);
 731                         return err;
 732                 }
 733                 break;
 734
 735         case VIFF_USE_IFINDEX:
 736         case 0:
 737                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 738                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 739                         if (dev && __in_dev_get_rtnl(dev) == NULL) {
 740                                 dev_put(dev);
 741                                 return -EADDRNOTAVAIL;
 742                         }
 743                 } else {
 744                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 745                 }
 746                 if (!dev)
 747                         return -EADDRNOTAVAIL;
 748                 err = dev_set_allmulti(dev, 1);
 749                 if (err) {
 750                         dev_put(dev);
 751                         return err;
 752                 }
 753                 break;
 754         default:
 755                 return -EINVAL;
 756         }
 757
 758         in_dev = __in_dev_get_rtnl(dev);
 759         if (!in_dev) {
 760                 dev_put(dev);
 761                 return -EADDRNOTAVAIL;
 762         }
 763         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 764         ip_rt_multicast_event(in_dev);
 765
 766         /* Fill in the VIF structures */
 767
 768         v->rate_limit = vifc->vifc_rate_limit;
 769         v->local = vifc->vifc_lcl_addr.s_addr;
 770         v->remote = vifc->vifc_rmt_addr.s_addr;
 771         v->flags = vifc->vifc_flags;
 772         if (!mrtsock)
 773                 v->flags |= VIFF_STATIC;
 774         v->threshold = vifc->vifc_threshold;
 775         v->bytes_in = 0;
 776         v->bytes_out = 0;
 777         v->pkt_in = 0;
 778         v->pkt_out = 0;
 779         v->link = dev->ifindex;
 780         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 781                 v->link = dev->iflink;
 782
 783         /* And finish update writing critical data */
 784         write_lock_bh(&mrt_lock);
 785         v->dev = dev;
 786 #ifdef CONFIG_IP_PIMSM
 787         if (v->flags & VIFF_REGISTER)
 788                 mrt->mroute_reg_vif_num = vifi;
 789 #endif
 790         if (vifi+1 > mrt->maxvif)
 791                 mrt->maxvif = vifi+1;
 792         write_unlock_bh(&mrt_lock);
 793         return 0;
 794 }
 795
 796 /* called with rcu_read_lock() */
 797 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 798                                          __be32 origin,
 799                                          __be32 mcastgrp)
 800 {
 801         int line = MFC_HASH(mcastgrp, origin);
 802         struct mfc_cache *c;
 803
 804         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 805                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 806                         return c;
 807         }
 808         return NULL;
 809 }
 810
 811 /*
 812  *      Allocate a multicast cache entry
 813  */
 814 static struct mfc_cache *ipmr_cache_alloc(void)
 815 {
 816         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 817
 818         if (c)
 819                 c->mfc_un.res.minvif = MAXVIFS;
 820         return c;
 821 }
 822
 823 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 824 {
 825         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 826
 827         if (c) {
 828                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 829                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 830         }
 831         return c;
 832 }
 833
 834 /*
 835  *      A cache entry has gone into a resolved state from queued
 836  */
 837
 838 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 839                                struct mfc_cache *uc, struct mfc_cache *c)
 840 {
 841         struct sk_buff *skb;
 842         struct nlmsgerr *e;
 843
 844         /* Play the pending entries through our router */
 845
 846         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 847                 if (ip_hdr(skb)->version == 0) {
 848                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 849
 850                         if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
 851                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 852                                                  (u8 *)nlh;
 853                         } else {
 854                                 nlh->nlmsg_type = NLMSG_ERROR;
 855                                 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 856                                 skb_trim(skb, nlh->nlmsg_len);
 857                                 e = NLMSG_DATA(nlh);
 858                                 e->error = -EMSGSIZE;
 859                                 memset(&e->msg, 0, sizeof(e->msg));
 860                         }
 861
 862                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
 863                 } else {
 864                         ip_mr_forward(net, mrt, skb, c, 0);
 865                 }
 866         }
 867 }
 868
 869 /*
 870  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 871  *      expects the following bizarre scheme.
 872  *
 873  *      Called under mrt_lock.
 874  */
 875
 876 static int ipmr_cache_report(struct mr_table *mrt,
 877                              struct sk_buff *pkt, vifi_t vifi, int assert)
 878 {
 879         struct sk_buff *skb;
 880         const int ihl = ip_hdrlen(pkt);
 881         struct igmphdr *igmp;
 882         struct igmpmsg *msg;
 883         struct sock *mroute_sk;
 884         int ret;
 885
 886 #ifdef CONFIG_IP_PIMSM
 887         if (assert == IGMPMSG_WHOLEPKT)
 888                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 889         else
 890 #endif
 891                 skb = alloc_skb(128, GFP_ATOMIC);
 892
 893         if (!skb)
 894                 return -ENOBUFS;
 895
 896 #ifdef CONFIG_IP_PIMSM
 897         if (assert == IGMPMSG_WHOLEPKT) {
 898                 /* Ugly, but we have no choice with this interface.
 899                  * Duplicate old header, fix ihl, length etc.
 900                  * And all this only to mangle msg->im_msgtype and
 901                  * to set msg->im_mbz to "mbz" :-)
 902                  */
 903                 skb_push(skb, sizeof(struct iphdr));
 904                 skb_reset_network_header(skb);
 905                 skb_reset_transport_header(skb);
 906                 msg = (struct igmpmsg *)skb_network_header(skb);
 907                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 908                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 909                 msg->im_mbz = 0;
 910                 msg->im_vif = mrt->mroute_reg_vif_num;
 911                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 912                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 913                                              sizeof(struct iphdr));
 914         } else
 915 #endif
 916         {
 917
 918         /* Copy the IP header */
 919
 920         skb->network_header = skb->tail;
 921         skb_put(skb, ihl);
 922         skb_copy_to_linear_data(skb, pkt->data, ihl);
 923         ip_hdr(skb)->protocol = 0;      /* Flag to the kernel this is a route add */
 924         msg = (struct igmpmsg *)skb_network_header(skb);
 925         msg->im_vif = vifi;
 926         skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 927
 928         /* Add our header */
 929
 930         igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 931         igmp->type      =
 932         msg->im_msgtype = assert;
 933         igmp->code      = 0;
 934         ip_hdr(skb)->tot_len = htons(skb->len);         /* Fix the length */
 935         skb->transport_header = skb->network_header;
 936         }
 937
 938         rcu_read_lock();
 939         mroute_sk = rcu_dereference(mrt->mroute_sk);
 940         if (mroute_sk == NULL) {
 941                 rcu_read_unlock();
 942                 kfree_skb(skb);
 943                 return -EINVAL;
 944         }
 945
 946         /* Deliver to mrouted */
 947
 948         ret = sock_queue_rcv_skb(mroute_sk, skb);
 949         rcu_read_unlock();
 950         if (ret < 0) {
 951                 if (net_ratelimit())
 952                         printk(KERN_WARNING "mroute: pending queue full, dropping entries.\n");
 953                 kfree_skb(skb);
 954         }
 955
 956         return ret;
 957 }
 958
 959 /*
 960  *      Queue a packet for resolution. It gets locked cache entry!
 961  */
 962
 963 static int
 964 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
 965 {
 966         bool found = false;
 967         int err;
 968         struct mfc_cache *c;
 969         const struct iphdr *iph = ip_hdr(skb);
 970
 971         spin_lock_bh(&mfc_unres_lock);
 972         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
 973                 if (c->mfc_mcastgrp == iph->daddr &&
 974                     c->mfc_origin == iph->saddr) {
 975                         found = true;
 976                         break;
 977                 }
 978         }
 979
 980         if (!found) {
 981                 /* Create a new entry if allowable */
 982
 983                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
 984                     (c = ipmr_cache_alloc_unres()) == NULL) {
 985                         spin_unlock_bh(&mfc_unres_lock);
 986
 987                         kfree_skb(skb);
 988                         return -ENOBUFS;
 989                 }
 990
 991                 /* Fill in the new cache entry */
 992
 993                 c->mfc_parent   = -1;
 994                 c->mfc_origin   = iph->saddr;
 995                 c->mfc_mcastgrp = iph->daddr;
 996
 997                 /* Reflect first query at mrouted. */
 998
 999                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1000                 if (err < 0) {
1001                         /* If the report failed throw the cache entry
1002                            out - Brad Parker
1003                          */
1004                         spin_unlock_bh(&mfc_unres_lock);
1005
1006                         ipmr_cache_free(c);
1007                         kfree_skb(skb);
1008                         return err;
1009                 }
1010
1011                 atomic_inc(&mrt->cache_resolve_queue_len);
1012                 list_add(&c->list, &mrt->mfc_unres_queue);
1013
1014                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1015                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1016         }
1017
1018         /* See if we can append the packet */
1019
1020         if (c->mfc_un.unres.unresolved.qlen > 3) {
1021                 kfree_skb(skb);
1022                 err = -ENOBUFS;
1023         } else {
1024                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1025                 err = 0;
1026         }
1027
1028         spin_unlock_bh(&mfc_unres_lock);
1029         return err;
1030 }
1031
1032 /*
1033  *      MFC cache manipulation by user space mroute daemon
1034  */
1035
1036 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1037 {
1038         int line;
1039         struct mfc_cache *c, *next;
1040
1041         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1042
1043         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1044                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1045                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1046                         list_del_rcu(&c->list);
1047
1048                         ipmr_cache_free(c);
1049                         return 0;
1050                 }
1051         }
1052         return -ENOENT;
1053 }
1054
1055 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1056                         struct mfcctl *mfc, int mrtsock)
1057 {
1058         bool found = false;
1059         int line;
1060         struct mfc_cache *uc, *c;
1061
1062         if (mfc->mfcc_parent >= MAXVIFS)
1063                 return -ENFILE;
1064
1065         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1066
1067         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1068                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1069                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1070                         found = true;
1071                         break;
1072                 }
1073         }
1074
1075         if (found) {
1076                 write_lock_bh(&mrt_lock);
1077                 c->mfc_parent = mfc->mfcc_parent;
1078                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1079                 if (!mrtsock)
1080                         c->mfc_flags |= MFC_STATIC;
1081                 write_unlock_bh(&mrt_lock);
1082                 return 0;
1083         }
1084
1085         if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1086                 return -EINVAL;
1087
1088         c = ipmr_cache_alloc();
1089         if (c == NULL)
1090                 return -ENOMEM;
1091
1092         c->mfc_origin = mfc->mfcc_origin.s_addr;
1093         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1094         c->mfc_parent = mfc->mfcc_parent;
1095         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1096         if (!mrtsock)
1097                 c->mfc_flags |= MFC_STATIC;
1098
1099         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1100
1101         /*
1102          *      Check to see if we resolved a queued list. If so we
1103          *      need to send on the frames and tidy up.
1104          */
1105         found = false;
1106         spin_lock_bh(&mfc_unres_lock);
1107         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1108                 if (uc->mfc_origin == c->mfc_origin &&
1109                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1110                         list_del(&uc->list);
1111                         atomic_dec(&mrt->cache_resolve_queue_len);
1112                         found = true;
1113                         break;
1114                 }
1115         }
1116         if (list_empty(&mrt->mfc_unres_queue))
1117                 del_timer(&mrt->ipmr_expire_timer);
1118         spin_unlock_bh(&mfc_unres_lock);
1119
1120         if (found) {
1121                 ipmr_cache_resolve(net, mrt, uc, c);
1122                 ipmr_cache_free(uc);
1123         }
1124         return 0;
1125 }
1126
1127 /*
1128  *      Close the multicast socket, and clear the vif tables etc
1129  */
1130
1131 static void mroute_clean_tables(struct mr_table *mrt)
1132 {
1133         int i;
1134         LIST_HEAD(list);
1135         struct mfc_cache *c, *next;
1136
1137         /* Shut down all active vif entries */
1138
1139         for (i = 0; i < mrt->maxvif; i++) {
1140                 if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1141                         vif_delete(mrt, i, 0, &list);
1142         }
1143         unregister_netdevice_many(&list);
1144
1145         /* Wipe the cache */
1146
1147         for (i = 0; i < MFC_LINES; i++) {
1148                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1149                         if (c->mfc_flags & MFC_STATIC)
1150                                 continue;
1151                         list_del_rcu(&c->list);
1152                         ipmr_cache_free(c);
1153                 }
1154         }
1155
1156         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1157                 spin_lock_bh(&mfc_unres_lock);
1158                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1159                         list_del(&c->list);
1160                         ipmr_destroy_unres(mrt, c);
1161                 }
1162                 spin_unlock_bh(&mfc_unres_lock);
1163         }
1164 }
1165
1166 /* called from ip_ra_control(), before an RCU grace period,
1167  * we dont need to call synchronize_rcu() here
1168  */
1169 static void mrtsock_destruct(struct sock *sk)
1170 {
1171         struct net *net = sock_net(sk);
1172         struct mr_table *mrt;
1173
1174         rtnl_lock();
1175         ipmr_for_each_table(mrt, net) {
1176                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1177                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1178                         rcu_assign_pointer(mrt->mroute_sk, NULL);
1179                         mroute_clean_tables(mrt);
1180                 }
1181         }
1182         rtnl_unlock();
1183 }
1184
1185 /*
1186  *      Socket options and virtual interface manipulation. The whole
1187  *      virtual interface system is a complete heap, but unfortunately
1188  *      that's how BSD mrouted happens to think. Maybe one day with a proper
1189  *      MOSPF/PIM router set up we can clean this up.
1190  */
1191
1192 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1193 {
1194         int ret;
1195         struct vifctl vif;
1196         struct mfcctl mfc;
1197         struct net *net = sock_net(sk);
1198         struct mr_table *mrt;
1199
1200         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1201         if (mrt == NULL)
1202                 return -ENOENT;
1203
1204         if (optname != MRT_INIT) {
1205                 if (sk != rcu_dereference_raw(mrt->mroute_sk) &&
1206                     !capable(CAP_NET_ADMIN))
1207                         return -EACCES;
1208         }
1209
1210         switch (optname) {
1211         case MRT_INIT:
1212                 if (sk->sk_type != SOCK_RAW ||
1213                     inet_sk(sk)->inet_num != IPPROTO_IGMP)
1214                         return -EOPNOTSUPP;
1215                 if (optlen != sizeof(int))
1216                         return -ENOPROTOOPT;
1217
1218                 rtnl_lock();
1219                 if (rtnl_dereference(mrt->mroute_sk)) {
1220                         rtnl_unlock();
1221                         return -EADDRINUSE;
1222                 }
1223
1224                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1225                 if (ret == 0) {
1226                         rcu_assign_pointer(mrt->mroute_sk, sk);
1227                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1228                 }
1229                 rtnl_unlock();
1230                 return ret;
1231         case MRT_DONE:
1232                 if (sk != rcu_dereference_raw(mrt->mroute_sk))
1233                         return -EACCES;
1234                 return ip_ra_control(sk, 0, NULL);
1235         case MRT_ADD_VIF:
1236         case MRT_DEL_VIF:
1237                 if (optlen != sizeof(vif))
1238                         return -EINVAL;
1239                 if (copy_from_user(&vif, optval, sizeof(vif)))
1240                         return -EFAULT;
1241                 if (vif.vifc_vifi >= MAXVIFS)
1242                         return -ENFILE;
1243                 rtnl_lock();
1244                 if (optname == MRT_ADD_VIF) {
1245                         ret = vif_add(net, mrt, &vif,
1246                                       sk == rtnl_dereference(mrt->mroute_sk));
1247                 } else {
1248                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1249                 }
1250                 rtnl_unlock();
1251                 return ret;
1252
1253                 /*
1254                  *      Manipulate the forwarding caches. These live
1255                  *      in a sort of kernel/user symbiosis.
1256                  */
1257         case MRT_ADD_MFC:
1258         case MRT_DEL_MFC:
1259                 if (optlen != sizeof(mfc))
1260                         return -EINVAL;
1261                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1262                         return -EFAULT;
1263                 rtnl_lock();
1264                 if (optname == MRT_DEL_MFC)
1265                         ret = ipmr_mfc_delete(mrt, &mfc);
1266                 else
1267                         ret = ipmr_mfc_add(net, mrt, &mfc,
1268                                            sk == rtnl_dereference(mrt->mroute_sk));
1269                 rtnl_unlock();
1270                 return ret;
1271                 /*
1272                  *      Control PIM assert.
1273                  */
1274         case MRT_ASSERT:
1275         {
1276                 int v;
1277                 if (get_user(v, (int __user *)optval))
1278                         return -EFAULT;
1279                 mrt->mroute_do_assert = (v) ? 1 : 0;
1280                 return 0;
1281         }
1282 #ifdef CONFIG_IP_PIMSM
1283         case MRT_PIM:
1284         {
1285                 int v;
1286
1287                 if (get_user(v, (int __user *)optval))
1288                         return -EFAULT;
1289                 v = (v) ? 1 : 0;
1290
1291                 rtnl_lock();
1292                 ret = 0;
1293                 if (v != mrt->mroute_do_pim) {
1294                         mrt->mroute_do_pim = v;
1295                         mrt->mroute_do_assert = v;
1296                 }
1297                 rtnl_unlock();
1298                 return ret;
1299         }
1300 #endif
1301 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1302         case MRT_TABLE:
1303         {
1304                 u32 v;
1305
1306                 if (optlen != sizeof(u32))
1307                         return -EINVAL;
1308                 if (get_user(v, (u32 __user *)optval))
1309                         return -EFAULT;
1310
1311                 rtnl_lock();
1312                 ret = 0;
1313                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1314                         ret = -EBUSY;
1315                 } else {
1316                         if (!ipmr_new_table(net, v))
1317                                 ret = -ENOMEM;
1318                         raw_sk(sk)->ipmr_table = v;
1319                 }
1320                 rtnl_unlock();
1321                 return ret;
1322         }
1323 #endif
1324         /*
1325          *      Spurious command, or MRT_VERSION which you cannot
1326          *      set.
1327          */
1328         default:
1329                 return -ENOPROTOOPT;
1330         }
1331 }
1332
1333 /*
1334  *      Getsock opt support for the multicast routing system.
1335  */
1336
1337 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1338 {
1339         int olr;
1340         int val;
1341         struct net *net = sock_net(sk);
1342         struct mr_table *mrt;
1343
1344         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1345         if (mrt == NULL)
1346                 return -ENOENT;
1347
1348         if (optname != MRT_VERSION &&
1349 #ifdef CONFIG_IP_PIMSM
1350            optname != MRT_PIM &&
1351 #endif
1352            optname != MRT_ASSERT)
1353                 return -ENOPROTOOPT;
1354
1355         if (get_user(olr, optlen))
1356                 return -EFAULT;
1357
1358         olr = min_t(unsigned int, olr, sizeof(int));
1359         if (olr < 0)
1360                 return -EINVAL;
1361
1362         if (put_user(olr, optlen))
1363                 return -EFAULT;
1364         if (optname == MRT_VERSION)
1365                 val = 0x0305;
1366 #ifdef CONFIG_IP_PIMSM
1367         else if (optname == MRT_PIM)
1368                 val = mrt->mroute_do_pim;
1369 #endif
1370         else
1371                 val = mrt->mroute_do_assert;
1372         if (copy_to_user(optval, &val, olr))
1373                 return -EFAULT;
1374         return 0;
1375 }
1376
1377 /*
1378  *      The IP multicast ioctl support routines.
1379  */
1380
1381 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1382 {
1383         struct sioc_sg_req sr;
1384         struct sioc_vif_req vr;
1385         struct vif_device *vif;
1386         struct mfc_cache *c;
1387         struct net *net = sock_net(sk);
1388         struct mr_table *mrt;
1389
1390         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1391         if (mrt == NULL)
1392                 return -ENOENT;
1393
1394         switch (cmd) {
1395         case SIOCGETVIFCNT:
1396                 if (copy_from_user(&vr, arg, sizeof(vr)))
1397                         return -EFAULT;
1398                 if (vr.vifi >= mrt->maxvif)
1399                         return -EINVAL;
1400                 read_lock(&mrt_lock);
1401                 vif = &mrt->vif_table[vr.vifi];
1402                 if (VIF_EXISTS(mrt, vr.vifi)) {
1403                         vr.icount = vif->pkt_in;
1404                         vr.ocount = vif->pkt_out;
1405                         vr.ibytes = vif->bytes_in;
1406                         vr.obytes = vif->bytes_out;
1407                         read_unlock(&mrt_lock);
1408
1409                         if (copy_to_user(arg, &vr, sizeof(vr)))
1410                                 return -EFAULT;
1411                         return 0;
1412                 }
1413                 read_unlock(&mrt_lock);
1414                 return -EADDRNOTAVAIL;
1415         case SIOCGETSGCNT:
1416                 if (copy_from_user(&sr, arg, sizeof(sr)))
1417                         return -EFAULT;
1418
1419                 rcu_read_lock();
1420                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1421                 if (c) {
1422                         sr.pktcnt = c->mfc_un.res.pkt;
1423                         sr.bytecnt = c->mfc_un.res.bytes;
1424                         sr.wrong_if = c->mfc_un.res.wrong_if;
1425                         rcu_read_unlock();
1426
1427                         if (copy_to_user(arg, &sr, sizeof(sr)))
1428                                 return -EFAULT;
1429                         return 0;
1430                 }
1431                 rcu_read_unlock();
1432                 return -EADDRNOTAVAIL;
1433         default:
1434                 return -ENOIOCTLCMD;
1435         }
1436 }
1437
1438 #ifdef CONFIG_COMPAT
1439 struct compat_sioc_sg_req {
1440         struct in_addr src;
1441         struct in_addr grp;
1442         compat_ulong_t pktcnt;
1443         compat_ulong_t bytecnt;
1444         compat_ulong_t wrong_if;
1445 };
1446
1447 struct compat_sioc_vif_req {
1448         vifi_t  vifi;           /* Which iface */
1449         compat_ulong_t icount;
1450         compat_ulong_t ocount;
1451         compat_ulong_t ibytes;
1452         compat_ulong_t obytes;
1453 };
1454
1455 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1456 {
1457         struct compat_sioc_sg_req sr;
1458         struct compat_sioc_vif_req vr;
1459         struct vif_device *vif;
1460         struct mfc_cache *c;
1461         struct net *net = sock_net(sk);
1462         struct mr_table *mrt;
1463
1464         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1465         if (mrt == NULL)
1466                 return -ENOENT;
1467
1468         switch (cmd) {
1469         case SIOCGETVIFCNT:
1470                 if (copy_from_user(&vr, arg, sizeof(vr)))
1471                         return -EFAULT;
1472                 if (vr.vifi >= mrt->maxvif)
1473                         return -EINVAL;
1474                 read_lock(&mrt_lock);
1475                 vif = &mrt->vif_table[vr.vifi];
1476                 if (VIF_EXISTS(mrt, vr.vifi)) {
1477                         vr.icount = vif->pkt_in;
1478                         vr.ocount = vif->pkt_out;
1479                         vr.ibytes = vif->bytes_in;
1480                         vr.obytes = vif->bytes_out;
1481                         read_unlock(&mrt_lock);
1482
1483                         if (copy_to_user(arg, &vr, sizeof(vr)))
1484                                 return -EFAULT;
1485                         return 0;
1486                 }
1487                 read_unlock(&mrt_lock);
1488                 return -EADDRNOTAVAIL;
1489         case SIOCGETSGCNT:
1490                 if (copy_from_user(&sr, arg, sizeof(sr)))
1491                         return -EFAULT;
1492
1493                 rcu_read_lock();
1494                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1495                 if (c) {
1496                         sr.pktcnt = c->mfc_un.res.pkt;
1497                         sr.bytecnt = c->mfc_un.res.bytes;
1498                         sr.wrong_if = c->mfc_un.res.wrong_if;
1499                         rcu_read_unlock();
1500
1501                         if (copy_to_user(arg, &sr, sizeof(sr)))
1502                                 return -EFAULT;
1503                         return 0;
1504                 }
1505                 rcu_read_unlock();
1506                 return -EADDRNOTAVAIL;
1507         default:
1508                 return -ENOIOCTLCMD;
1509         }
1510 }
1511 #endif
1512
1513
1514 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1515 {
1516         struct net_device *dev = ptr;
1517         struct net *net = dev_net(dev);
1518         struct mr_table *mrt;
1519         struct vif_device *v;
1520         int ct;
1521         LIST_HEAD(list);
1522
1523         if (event != NETDEV_UNREGISTER)
1524                 return NOTIFY_DONE;
1525
1526         ipmr_for_each_table(mrt, net) {
1527                 v = &mrt->vif_table[0];
1528                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1529                         if (v->dev == dev)
1530                                 vif_delete(mrt, ct, 1, &list);
1531                 }
1532         }
1533         unregister_netdevice_many(&list);
1534         return NOTIFY_DONE;
1535 }
1536
1537
1538 static struct notifier_block ip_mr_notifier = {
1539         .notifier_call = ipmr_device_event,
1540 };
1541
1542 /*
1543  *      Encapsulate a packet by attaching a valid IPIP header to it.
1544  *      This avoids tunnel drivers and other mess and gives us the speed so
1545  *      important for multicast video.
1546  */
1547
1548 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1549 {
1550         struct iphdr *iph;
1551         struct iphdr *old_iph = ip_hdr(skb);
1552
1553         skb_push(skb, sizeof(struct iphdr));
1554         skb->transport_header = skb->network_header;
1555         skb_reset_network_header(skb);
1556         iph = ip_hdr(skb);
1557
1558         iph->version    =       4;
1559         iph->tos        =       old_iph->tos;
1560         iph->ttl        =       old_iph->ttl;
1561         iph->frag_off   =       0;
1562         iph->daddr      =       daddr;
1563         iph->saddr      =       saddr;
1564         iph->protocol   =       IPPROTO_IPIP;
1565         iph->ihl        =       5;
1566         iph->tot_len    =       htons(skb->len);
1567         ip_select_ident(iph, skb_dst(skb), NULL);
1568         ip_send_check(iph);
1569
1570         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1571         nf_reset(skb);
1572 }
1573
1574 static inline int ipmr_forward_finish(struct sk_buff *skb)
1575 {
1576         struct ip_options *opt = &(IPCB(skb)->opt);
1577
1578         IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1579
1580         if (unlikely(opt->optlen))
1581                 ip_forward_options(skb);
1582
1583         return dst_output(skb);
1584 }
1585
1586 /*
1587  *      Processing handlers for ipmr_forward
1588  */
1589
1590 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1591                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1592 {
1593         const struct iphdr *iph = ip_hdr(skb);
1594         struct vif_device *vif = &mrt->vif_table[vifi];
1595         struct net_device *dev;
1596         struct rtable *rt;
1597         int    encap = 0;
1598
1599         if (vif->dev == NULL)
1600                 goto out_free;
1601
1602 #ifdef CONFIG_IP_PIMSM
1603         if (vif->flags & VIFF_REGISTER) {
1604                 vif->pkt_out++;
1605                 vif->bytes_out += skb->len;
1606                 vif->dev->stats.tx_bytes += skb->len;
1607                 vif->dev->stats.tx_packets++;
1608                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1609                 goto out_free;
1610         }
1611 #endif
1612
1613         if (vif->flags & VIFF_TUNNEL) {
1614                 struct flowi fl = {
1615                         .oif = vif->link,
1616                         .fl4_dst = vif->remote,
1617                         .fl4_src = vif->local,
1618                         .fl4_tos = RT_TOS(iph->tos),
1619                         .proto = IPPROTO_IPIP
1620                 };
1621
1622                 if (ip_route_output_key(net, &rt, &fl))
1623                         goto out_free;
1624                 encap = sizeof(struct iphdr);
1625         } else {
1626                 struct flowi fl = {
1627                         .oif = vif->link,
1628                         .fl4_dst = iph->daddr,
1629                         .fl4_tos = RT_TOS(iph->tos),
1630                         .proto = IPPROTO_IPIP
1631                 };
1632
1633                 if (ip_route_output_key(net, &rt, &fl))
1634                         goto out_free;
1635         }
1636
1637         dev = rt->dst.dev;
1638
1639         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1640                 /* Do not fragment multicasts. Alas, IPv4 does not
1641                  * allow to send ICMP, so that packets will disappear
1642                  * to blackhole.
1643                  */
1644
1645                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1646                 ip_rt_put(rt);
1647                 goto out_free;
1648         }
1649
1650         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1651
1652         if (skb_cow(skb, encap)) {
1653                 ip_rt_put(rt);
1654                 goto out_free;
1655         }
1656
1657         vif->pkt_out++;
1658         vif->bytes_out += skb->len;
1659
1660         skb_dst_drop(skb);
1661         skb_dst_set(skb, &rt->dst);
1662         ip_decrease_ttl(ip_hdr(skb));
1663
1664         /* FIXME: forward and output firewalls used to be called here.
1665          * What do we do with netfilter? -- RR
1666          */
1667         if (vif->flags & VIFF_TUNNEL) {
1668                 ip_encap(skb, vif->local, vif->remote);
1669                 /* FIXME: extra output firewall step used to be here. --RR */
1670                 vif->dev->stats.tx_packets++;
1671                 vif->dev->stats.tx_bytes += skb->len;
1672         }
1673
1674         IPCB(skb)->flags |= IPSKB_FORWARDED;
1675
1676         /*
1677          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1678          * not only before forwarding, but after forwarding on all output
1679          * interfaces. It is clear, if mrouter runs a multicasting
1680          * program, it should receive packets not depending to what interface
1681          * program is joined.
1682          * If we will not make it, the program will have to join on all
1683          * interfaces. On the other hand, multihoming host (or router, but
1684          * not mrouter) cannot join to more than one interface - it will
1685          * result in receiving multiple packets.
1686          */
1687         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1688                 ipmr_forward_finish);
1689         return;
1690
1691 out_free:
1692         kfree_skb(skb);
1693 }
1694
1695 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1696 {
1697         int ct;
1698
1699         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1700                 if (mrt->vif_table[ct].dev == dev)
1701                         break;
1702         }
1703         return ct;
1704 }
1705
1706 /* "local" means that we should preserve one skb (for local delivery) */
1707
1708 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1709                          struct sk_buff *skb, struct mfc_cache *cache,
1710                          int local)
1711 {
1712         int psend = -1;
1713         int vif, ct;
1714
1715         vif = cache->mfc_parent;
1716         cache->mfc_un.res.pkt++;
1717         cache->mfc_un.res.bytes += skb->len;
1718
1719         /*
1720          * Wrong interface: drop packet and (maybe) send PIM assert.
1721          */
1722         if (mrt->vif_table[vif].dev != skb->dev) {
1723                 int true_vifi;
1724
1725                 if (rt_is_output_route(skb_rtable(skb))) {
1726                         /* It is our own packet, looped back.
1727                          * Very complicated situation...
1728                          *
1729                          * The best workaround until routing daemons will be
1730                          * fixed is not to redistribute packet, if it was
1731                          * send through wrong interface. It means, that
1732                          * multicast applications WILL NOT work for
1733                          * (S,G), which have default multicast route pointing
1734                          * to wrong oif. In any case, it is not a good
1735                          * idea to use multicasting applications on router.
1736                          */
1737                         goto dont_forward;
1738                 }
1739
1740                 cache->mfc_un.res.wrong_if++;
1741                 true_vifi = ipmr_find_vif(mrt, skb->dev);
1742
1743                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1744                     /* pimsm uses asserts, when switching from RPT to SPT,
1745                      * so that we cannot check that packet arrived on an oif.
1746                      * It is bad, but otherwise we would need to move pretty
1747                      * large chunk of pimd to kernel. Ough... --ANK
1748                      */
1749                     (mrt->mroute_do_pim ||
1750                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1751                     time_after(jiffies,
1752                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1753                         cache->mfc_un.res.last_assert = jiffies;
1754                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1755                 }
1756                 goto dont_forward;
1757         }
1758
1759         mrt->vif_table[vif].pkt_in++;
1760         mrt->vif_table[vif].bytes_in += skb->len;
1761
1762         /*
1763          *      Forward the frame
1764          */
1765         for (ct = cache->mfc_un.res.maxvif - 1;
1766              ct >= cache->mfc_un.res.minvif; ct--) {
1767                 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1768                         if (psend != -1) {
1769                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1770
1771                                 if (skb2)
1772                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1773                                                         psend);
1774                         }
1775                         psend = ct;
1776                 }
1777         }
1778         if (psend != -1) {
1779                 if (local) {
1780                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1781
1782                         if (skb2)
1783                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1784                 } else {
1785                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1786                         return 0;
1787                 }
1788         }
1789
1790 dont_forward:
1791         if (!local)
1792                 kfree_skb(skb);
1793         return 0;
1794 }
1795
1796
1797 /*
1798  *      Multicast packets for forwarding arrive here
1799  *      Called with rcu_read_lock();
1800  */
1801
1802 int ip_mr_input(struct sk_buff *skb)
1803 {
1804         struct mfc_cache *cache;
1805         struct net *net = dev_net(skb->dev);
1806         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1807         struct mr_table *mrt;
1808         int err;
1809
1810         /* Packet is looped back after forward, it should not be
1811          * forwarded second time, but still can be delivered locally.
1812          */
1813         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1814                 goto dont_forward;
1815
1816         err = ipmr_fib_lookup(net, &skb_rtable(skb)->fl, &mrt);
1817         if (err < 0) {
1818                 kfree_skb(skb);
1819                 return err;
1820         }
1821
1822         if (!local) {
1823                 if (IPCB(skb)->opt.router_alert) {
1824                         if (ip_call_ra_chain(skb))
1825                                 return 0;
1826                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1827                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1828                          * Cisco IOS <= 11.2(8)) do not put router alert
1829                          * option to IGMP packets destined to routable
1830                          * groups. It is very bad, because it means
1831                          * that we can forward NO IGMP messages.
1832                          */
1833                         struct sock *mroute_sk;
1834
1835                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1836                         if (mroute_sk) {
1837                                 nf_reset(skb);
1838                                 raw_rcv(mroute_sk, skb);
1839                                 return 0;
1840                         }
1841                     }
1842         }
1843
1844         /* already under rcu_read_lock() */
1845         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1846
1847         /*
1848          *      No usable cache entry
1849          */
1850         if (cache == NULL) {
1851                 int vif;
1852
1853                 if (local) {
1854                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1855                         ip_local_deliver(skb);
1856                         if (skb2 == NULL)
1857                                 return -ENOBUFS;
1858                         skb = skb2;
1859                 }
1860
1861                 read_lock(&mrt_lock);
1862                 vif = ipmr_find_vif(mrt, skb->dev);
1863                 if (vif >= 0) {
1864                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1865                         read_unlock(&mrt_lock);
1866
1867                         return err2;
1868                 }
1869                 read_unlock(&mrt_lock);
1870                 kfree_skb(skb);
1871                 return -ENODEV;
1872         }
1873
1874         read_lock(&mrt_lock);
1875         ip_mr_forward(net, mrt, skb, cache, local);
1876         read_unlock(&mrt_lock);
1877
1878         if (local)
1879                 return ip_local_deliver(skb);
1880
1881         return 0;
1882
1883 dont_forward:
1884         if (local)
1885                 return ip_local_deliver(skb);
1886         kfree_skb(skb);
1887         return 0;
1888 }
1889
1890 #ifdef CONFIG_IP_PIMSM
1891 /* called with rcu_read_lock() */
1892 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1893                      unsigned int pimlen)
1894 {
1895         struct net_device *reg_dev = NULL;
1896         struct iphdr *encap;
1897
1898         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1899         /*
1900          * Check that:
1901          * a. packet is really sent to a multicast group
1902          * b. packet is not a NULL-REGISTER
1903          * c. packet is not truncated
1904          */
1905         if (!ipv4_is_multicast(encap->daddr) ||
1906             encap->tot_len == 0 ||
1907             ntohs(encap->tot_len) + pimlen > skb->len)
1908                 return 1;
1909
1910         read_lock(&mrt_lock);
1911         if (mrt->mroute_reg_vif_num >= 0)
1912                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1913         read_unlock(&mrt_lock);
1914
1915         if (reg_dev == NULL)
1916                 return 1;
1917
1918         skb->mac_header = skb->network_header;
1919         skb_pull(skb, (u8 *)encap - skb->data);
1920         skb_reset_network_header(skb);
1921         skb->protocol = htons(ETH_P_IP);
1922         skb->ip_summed = CHECKSUM_NONE;
1923         skb->pkt_type = PACKET_HOST;
1924
1925         skb_tunnel_rx(skb, reg_dev);
1926
1927         netif_rx(skb);
1928
1929         return NET_RX_SUCCESS;
1930 }
1931 #endif
1932
1933 #ifdef CONFIG_IP_PIMSM_V1
1934 /*
1935  * Handle IGMP messages of PIMv1
1936  */
1937
1938 int pim_rcv_v1(struct sk_buff *skb)
1939 {
1940         struct igmphdr *pim;
1941         struct net *net = dev_net(skb->dev);
1942         struct mr_table *mrt;
1943
1944         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1945                 goto drop;
1946
1947         pim = igmp_hdr(skb);
1948
1949         if (ipmr_fib_lookup(net, &skb_rtable(skb)->fl, &mrt) < 0)
1950                 goto drop;
1951
1952         if (!mrt->mroute_do_pim ||
1953             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
1954                 goto drop;
1955
1956         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1957 drop:
1958                 kfree_skb(skb);
1959         }
1960         return 0;
1961 }
1962 #endif
1963
1964 #ifdef CONFIG_IP_PIMSM_V2
1965 static int pim_rcv(struct sk_buff *skb)
1966 {
1967         struct pimreghdr *pim;
1968         struct net *net = dev_net(skb->dev);
1969         struct mr_table *mrt;
1970
1971         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1972                 goto drop;
1973
1974         pim = (struct pimreghdr *)skb_transport_header(skb);
1975         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
1976             (pim->flags & PIM_NULL_REGISTER) ||
1977             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
1978              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
1979                 goto drop;
1980
1981         if (ipmr_fib_lookup(net, &skb_rtable(skb)->fl, &mrt) < 0)
1982                 goto drop;
1983
1984         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1985 drop:
1986                 kfree_skb(skb);
1987         }
1988         return 0;
1989 }
1990 #endif
1991
1992 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
1993                               struct mfc_cache *c, struct rtmsg *rtm)
1994 {
1995         int ct;
1996         struct rtnexthop *nhp;
1997         u8 *b = skb_tail_pointer(skb);
1998         struct rtattr *mp_head;
1999
2000         /* If cache is unresolved, don't try to parse IIF and OIF */
2001         if (c->mfc_parent >= MAXVIFS)
2002                 return -ENOENT;
2003
2004         if (VIF_EXISTS(mrt, c->mfc_parent))
2005                 RTA_PUT(skb, RTA_IIF, 4, &mrt->vif_table[c->mfc_parent].dev->ifindex);
2006
2007         mp_head = (struct rtattr *)skb_put(skb, RTA_LENGTH(0));
2008
2009         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2010                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2011                         if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
2012                                 goto rtattr_failure;
2013                         nhp = (struct rtnexthop *)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
2014                         nhp->rtnh_flags = 0;
2015                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2016                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2017                         nhp->rtnh_len = sizeof(*nhp);
2018                 }
2019         }
2020         mp_head->rta_type = RTA_MULTIPATH;
2021         mp_head->rta_len = skb_tail_pointer(skb) - (u8 *)mp_head;
2022         rtm->rtm_type = RTN_MULTICAST;
2023         return 1;
2024
2025 rtattr_failure:
2026         nlmsg_trim(skb, b);
2027         return -EMSGSIZE;
2028 }
2029
2030 int ipmr_get_route(struct net *net,
2031                    struct sk_buff *skb, struct rtmsg *rtm, int nowait)
2032 {
2033         int err;
2034         struct mr_table *mrt;
2035         struct mfc_cache *cache;
2036         struct rtable *rt = skb_rtable(skb);
2037
2038         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2039         if (mrt == NULL)
2040                 return -ENOENT;
2041
2042         rcu_read_lock();
2043         cache = ipmr_cache_find(mrt, rt->rt_src, rt->rt_dst);
2044
2045         if (cache == NULL) {
2046                 struct sk_buff *skb2;
2047                 struct iphdr *iph;
2048                 struct net_device *dev;
2049                 int vif = -1;
2050
2051                 if (nowait) {
2052                         rcu_read_unlock();
2053                         return -EAGAIN;
2054                 }
2055
2056                 dev = skb->dev;
2057                 read_lock(&mrt_lock);
2058                 if (dev)
2059                         vif = ipmr_find_vif(mrt, dev);
2060                 if (vif < 0) {
2061                         read_unlock(&mrt_lock);
2062                         rcu_read_unlock();
2063                         return -ENODEV;
2064                 }
2065                 skb2 = skb_clone(skb, GFP_ATOMIC);
2066                 if (!skb2) {
2067                         read_unlock(&mrt_lock);
2068                         rcu_read_unlock();
2069                         return -ENOMEM;
2070                 }
2071
2072                 skb_push(skb2, sizeof(struct iphdr));
2073                 skb_reset_network_header(skb2);
2074                 iph = ip_hdr(skb2);
2075                 iph->ihl = sizeof(struct iphdr) >> 2;
2076                 iph->saddr = rt->rt_src;
2077                 iph->daddr = rt->rt_dst;
2078                 iph->version = 0;
2079                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2080                 read_unlock(&mrt_lock);
2081                 rcu_read_unlock();
2082                 return err;
2083         }
2084
2085         read_lock(&mrt_lock);
2086         if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2087                 cache->mfc_flags |= MFC_NOTIFY;
2088         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2089         read_unlock(&mrt_lock);
2090         rcu_read_unlock();
2091         return err;
2092 }
2093
2094 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2095                             u32 pid, u32 seq, struct mfc_cache *c)
2096 {
2097         struct nlmsghdr *nlh;
2098         struct rtmsg *rtm;
2099
2100         nlh = nlmsg_put(skb, pid, seq, RTM_NEWROUTE, sizeof(*rtm), NLM_F_MULTI);
2101         if (nlh == NULL)
2102                 return -EMSGSIZE;
2103
2104         rtm = nlmsg_data(nlh);
2105         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2106         rtm->rtm_dst_len  = 32;
2107         rtm->rtm_src_len  = 32;
2108         rtm->rtm_tos      = 0;
2109         rtm->rtm_table    = mrt->id;
2110         NLA_PUT_U32(skb, RTA_TABLE, mrt->id);
2111         rtm->rtm_type     = RTN_MULTICAST;
2112         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2113         rtm->rtm_protocol = RTPROT_UNSPEC;
2114         rtm->rtm_flags    = 0;
2115
2116         NLA_PUT_BE32(skb, RTA_SRC, c->mfc_origin);
2117         NLA_PUT_BE32(skb, RTA_DST, c->mfc_mcastgrp);
2118
2119         if (__ipmr_fill_mroute(mrt, skb, c, rtm) < 0)
2120                 goto nla_put_failure;
2121
2122         return nlmsg_end(skb, nlh);
2123
2124 nla_put_failure:
2125         nlmsg_cancel(skb, nlh);
2126         return -EMSGSIZE;
2127 }
2128
2129 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2130 {
2131         struct net *net = sock_net(skb->sk);
2132         struct mr_table *mrt;
2133         struct mfc_cache *mfc;
2134         unsigned int t = 0, s_t;
2135         unsigned int h = 0, s_h;
2136         unsigned int e = 0, s_e;
2137
2138         s_t = cb->args[0];
2139         s_h = cb->args[1];
2140         s_e = cb->args[2];
2141
2142         rcu_read_lock();
2143         ipmr_for_each_table(mrt, net) {
2144                 if (t < s_t)
2145                         goto next_table;
2146                 if (t > s_t)
2147                         s_h = 0;
2148                 for (h = s_h; h < MFC_LINES; h++) {
2149                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2150                                 if (e < s_e)
2151                                         goto next_entry;
2152                                 if (ipmr_fill_mroute(mrt, skb,
2153                                                      NETLINK_CB(cb->skb).pid,
2154                                                      cb->nlh->nlmsg_seq,
2155                                                      mfc) < 0)
2156                                         goto done;
2157 next_entry:
2158                                 e++;
2159                         }
2160                         e = s_e = 0;
2161                 }
2162                 s_h = 0;
2163 next_table:
2164                 t++;
2165         }
2166 done:
2167         rcu_read_unlock();
2168
2169         cb->args[2] = e;
2170         cb->args[1] = h;
2171         cb->args[0] = t;
2172
2173         return skb->len;
2174 }
2175
2176 #ifdef CONFIG_PROC_FS
2177 /*
2178  *      The /proc interfaces to multicast routing :
2179  *      /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2180  */
2181 struct ipmr_vif_iter {
2182         struct seq_net_private p;
2183         struct mr_table *mrt;
2184         int ct;
2185 };
2186
2187 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2188                                            struct ipmr_vif_iter *iter,
2189                                            loff_t pos)
2190 {
2191         struct mr_table *mrt = iter->mrt;
2192
2193         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2194                 if (!VIF_EXISTS(mrt, iter->ct))
2195                         continue;
2196                 if (pos-- == 0)
2197                         return &mrt->vif_table[iter->ct];
2198         }
2199         return NULL;
2200 }
2201
2202 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2203         __acquires(mrt_lock)
2204 {
2205         struct ipmr_vif_iter *iter = seq->private;
2206         struct net *net = seq_file_net(seq);
2207         struct mr_table *mrt;
2208
2209         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2210         if (mrt == NULL)
2211                 return ERR_PTR(-ENOENT);
2212
2213         iter->mrt = mrt;
2214
2215         read_lock(&mrt_lock);
2216         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2217                 : SEQ_START_TOKEN;
2218 }
2219
2220 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2221 {
2222         struct ipmr_vif_iter *iter = seq->private;
2223         struct net *net = seq_file_net(seq);
2224         struct mr_table *mrt = iter->mrt;
2225
2226         ++*pos;
2227         if (v == SEQ_START_TOKEN)
2228                 return ipmr_vif_seq_idx(net, iter, 0);
2229
2230         while (++iter->ct < mrt->maxvif) {
2231                 if (!VIF_EXISTS(mrt, iter->ct))
2232                         continue;
2233                 return &mrt->vif_table[iter->ct];
2234         }
2235         return NULL;
2236 }
2237
2238 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2239         __releases(mrt_lock)
2240 {
2241         read_unlock(&mrt_lock);
2242 }
2243
2244 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2245 {
2246         struct ipmr_vif_iter *iter = seq->private;
2247         struct mr_table *mrt = iter->mrt;
2248
2249         if (v == SEQ_START_TOKEN) {
2250                 seq_puts(seq,
2251                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2252         } else {
2253                 const struct vif_device *vif = v;
2254                 const char *name =  vif->dev ? vif->dev->name : "none";
2255
2256                 seq_printf(seq,
2257                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2258                            vif - mrt->vif_table,
2259                            name, vif->bytes_in, vif->pkt_in,
2260                            vif->bytes_out, vif->pkt_out,
2261                            vif->flags, vif->local, vif->remote);
2262         }
2263         return 0;
2264 }
2265
2266 static const struct seq_operations ipmr_vif_seq_ops = {
2267         .start = ipmr_vif_seq_start,
2268         .next  = ipmr_vif_seq_next,
2269         .stop  = ipmr_vif_seq_stop,
2270         .show  = ipmr_vif_seq_show,
2271 };
2272
2273 static int ipmr_vif_open(struct inode *inode, struct file *file)
2274 {
2275         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2276                             sizeof(struct ipmr_vif_iter));
2277 }
2278
2279 static const struct file_operations ipmr_vif_fops = {
2280         .owner   = THIS_MODULE,
2281         .open    = ipmr_vif_open,
2282         .read    = seq_read,
2283         .llseek  = seq_lseek,
2284         .release = seq_release_net,
2285 };
2286
2287 struct ipmr_mfc_iter {
2288         struct seq_net_private p;
2289         struct mr_table *mrt;
2290         struct list_head *cache;
2291         int ct;
2292 };
2293
2294
2295 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2296                                           struct ipmr_mfc_iter *it, loff_t pos)
2297 {
2298         struct mr_table *mrt = it->mrt;
2299         struct mfc_cache *mfc;
2300
2301         rcu_read_lock();
2302         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2303                 it->cache = &mrt->mfc_cache_array[it->ct];
2304                 list_for_each_entry_rcu(mfc, it->cache, list)
2305                         if (pos-- == 0)
2306                                 return mfc;
2307         }
2308         rcu_read_unlock();
2309
2310         spin_lock_bh(&mfc_unres_lock);
2311         it->cache = &mrt->mfc_unres_queue;
2312         list_for_each_entry(mfc, it->cache, list)
2313                 if (pos-- == 0)
2314                         return mfc;
2315         spin_unlock_bh(&mfc_unres_lock);
2316
2317         it->cache = NULL;
2318         return NULL;
2319 }
2320
2321
2322 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2323 {
2324         struct ipmr_mfc_iter *it = seq->private;
2325         struct net *net = seq_file_net(seq);
2326         struct mr_table *mrt;
2327
2328         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2329         if (mrt == NULL)
2330                 return ERR_PTR(-ENOENT);
2331
2332         it->mrt = mrt;
2333         it->cache = NULL;
2334         it->ct = 0;
2335         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2336                 : SEQ_START_TOKEN;
2337 }
2338
2339 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2340 {
2341         struct mfc_cache *mfc = v;
2342         struct ipmr_mfc_iter *it = seq->private;
2343         struct net *net = seq_file_net(seq);
2344         struct mr_table *mrt = it->mrt;
2345
2346         ++*pos;
2347
2348         if (v == SEQ_START_TOKEN)
2349                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2350
2351         if (mfc->list.next != it->cache)
2352                 return list_entry(mfc->list.next, struct mfc_cache, list);
2353
2354         if (it->cache == &mrt->mfc_unres_queue)
2355                 goto end_of_list;
2356
2357         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2358
2359         while (++it->ct < MFC_LINES) {
2360                 it->cache = &mrt->mfc_cache_array[it->ct];
2361                 if (list_empty(it->cache))
2362                         continue;
2363                 return list_first_entry(it->cache, struct mfc_cache, list);
2364         }
2365
2366         /* exhausted cache_array, show unresolved */
2367         rcu_read_unlock();
2368         it->cache = &mrt->mfc_unres_queue;
2369         it->ct = 0;
2370
2371         spin_lock_bh(&mfc_unres_lock);
2372         if (!list_empty(it->cache))
2373                 return list_first_entry(it->cache, struct mfc_cache, list);
2374
2375 end_of_list:
2376         spin_unlock_bh(&mfc_unres_lock);
2377         it->cache = NULL;
2378
2379         return NULL;
2380 }
2381
2382 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2383 {
2384         struct ipmr_mfc_iter *it = seq->private;
2385         struct mr_table *mrt = it->mrt;
2386
2387         if (it->cache == &mrt->mfc_unres_queue)
2388                 spin_unlock_bh(&mfc_unres_lock);
2389         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2390                 rcu_read_unlock();
2391 }
2392
2393 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2394 {
2395         int n;
2396
2397         if (v == SEQ_START_TOKEN) {
2398                 seq_puts(seq,
2399                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2400         } else {
2401                 const struct mfc_cache *mfc = v;
2402                 const struct ipmr_mfc_iter *it = seq->private;
2403                 const struct mr_table *mrt = it->mrt;
2404
2405                 seq_printf(seq, "%08X %08X %-3hd",
2406                            (__force u32) mfc->mfc_mcastgrp,
2407                            (__force u32) mfc->mfc_origin,
2408                            mfc->mfc_parent);
2409
2410                 if (it->cache != &mrt->mfc_unres_queue) {
2411                         seq_printf(seq, " %8lu %8lu %8lu",
2412                                    mfc->mfc_un.res.pkt,
2413                                    mfc->mfc_un.res.bytes,
2414                                    mfc->mfc_un.res.wrong_if);
2415                         for (n = mfc->mfc_un.res.minvif;
2416                              n < mfc->mfc_un.res.maxvif; n++) {
2417                                 if (VIF_EXISTS(mrt, n) &&
2418                                     mfc->mfc_un.res.ttls[n] < 255)
2419                                         seq_printf(seq,
2420                                            " %2d:%-3d",
2421                                            n, mfc->mfc_un.res.ttls[n]);
2422                         }
2423                 } else {
2424                         /* unresolved mfc_caches don't contain
2425                          * pkt, bytes and wrong_if values
2426                          */
2427                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2428                 }
2429                 seq_putc(seq, '\n');
2430         }
2431         return 0;
2432 }
2433
2434 static const struct seq_operations ipmr_mfc_seq_ops = {
2435         .start = ipmr_mfc_seq_start,
2436         .next  = ipmr_mfc_seq_next,
2437         .stop  = ipmr_mfc_seq_stop,
2438         .show  = ipmr_mfc_seq_show,
2439 };
2440
2441 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2442 {
2443         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2444                             sizeof(struct ipmr_mfc_iter));
2445 }
2446
2447 static const struct file_operations ipmr_mfc_fops = {
2448         .owner   = THIS_MODULE,
2449         .open    = ipmr_mfc_open,
2450         .read    = seq_read,
2451         .llseek  = seq_lseek,
2452         .release = seq_release_net,
2453 };
2454 #endif
2455
2456 #ifdef CONFIG_IP_PIMSM_V2
2457 static const struct net_protocol pim_protocol = {
2458         .handler        =       pim_rcv,
2459         .netns_ok       =       1,
2460 };
2461 #endif
2462
2463
2464 /*
2465  *      Setup for IP multicast routing
2466  */
2467 static int __net_init ipmr_net_init(struct net *net)
2468 {
2469         int err;
2470
2471         err = ipmr_rules_init(net);
2472         if (err < 0)
2473                 goto fail;
2474
2475 #ifdef CONFIG_PROC_FS
2476         err = -ENOMEM;
2477         if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2478                 goto proc_vif_fail;
2479         if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2480                 goto proc_cache_fail;
2481 #endif
2482         return 0;
2483
2484 #ifdef CONFIG_PROC_FS
2485 proc_cache_fail:
2486         proc_net_remove(net, "ip_mr_vif");
2487 proc_vif_fail:
2488         ipmr_rules_exit(net);
2489 #endif
2490 fail:
2491         return err;
2492 }
2493
2494 static void __net_exit ipmr_net_exit(struct net *net)
2495 {
2496 #ifdef CONFIG_PROC_FS
2497         proc_net_remove(net, "ip_mr_cache");
2498         proc_net_remove(net, "ip_mr_vif");
2499 #endif
2500         ipmr_rules_exit(net);
2501 }
2502
2503 static struct pernet_operations ipmr_net_ops = {
2504         .init = ipmr_net_init,
2505         .exit = ipmr_net_exit,
2506 };
2507
2508 int __init ip_mr_init(void)
2509 {
2510         int err;
2511
2512         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2513                                        sizeof(struct mfc_cache),
2514                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2515                                        NULL);
2516         if (!mrt_cachep)
2517                 return -ENOMEM;
2518
2519         err = register_pernet_subsys(&ipmr_net_ops);
2520         if (err)
2521                 goto reg_pernet_fail;
2522
2523         err = register_netdevice_notifier(&ip_mr_notifier);
2524         if (err)
2525                 goto reg_notif_fail;
2526 #ifdef CONFIG_IP_PIMSM_V2
2527         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2528                 printk(KERN_ERR "ip_mr_init: can't add PIM protocol\n");
2529                 err = -EAGAIN;
2530                 goto add_proto_fail;
2531         }
2532 #endif
2533         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE, NULL, ipmr_rtm_dumproute);
2534         return 0;
2535
2536 #ifdef CONFIG_IP_PIMSM_V2
2537 add_proto_fail:
2538         unregister_netdevice_notifier(&ip_mr_notifier);
2539 #endif
2540 reg_notif_fail:
2541         unregister_pernet_subsys(&ipmr_net_ops);
2542 reg_pernet_fail:
2543         kmem_cache_destroy(mrt_cachep);
2544         return err;
2545 }