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 <linux/export.h>
  65 #include <net/ipip.h>
  66 #include <net/checksum.h>
  67 #include <net/netlink.h>
  68 #include <net/fib_rules.h>
  69
  70 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  71 #define CONFIG_IP_PIMSM 1
  72 #endif
  73
  74 struct mr_table {
  75         struct list_head        list;
  76 #ifdef CONFIG_NET_NS
  77         struct net              *net;
  78 #endif
  79         u32                     id;
  80         struct sock __rcu       *mroute_sk;
  81         struct timer_list       ipmr_expire_timer;
  82         struct list_head        mfc_unres_queue;
  83         struct list_head        mfc_cache_array[MFC_LINES];
  84         struct vif_device       vif_table[MAXVIFS];
  85         int                     maxvif;
  86         atomic_t                cache_resolve_queue_len;
  87         int                     mroute_do_assert;
  88         int                     mroute_do_pim;
  89 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
  90         int                     mroute_reg_vif_num;
  91 #endif
  92 };
  93
  94 struct ipmr_rule {
  95         struct fib_rule         common;
  96 };
  97
  98 struct ipmr_result {
  99         struct mr_table         *mrt;
 100 };
 101
 102 /* Big lock, protecting vif table, mrt cache and mroute socket state.
 103  * Note that the changes are semaphored via rtnl_lock.
 104  */
 105
 106 static DEFINE_RWLOCK(mrt_lock);
 107
 108 /*
 109  *      Multicast router control variables
 110  */
 111
 112 #define VIF_EXISTS(_mrt, _idx) ((_mrt)->vif_table[_idx].dev != NULL)
 113
 114 /* Special spinlock for queue of unresolved entries */
 115 static DEFINE_SPINLOCK(mfc_unres_lock);
 116
 117 /* We return to original Alan's scheme. Hash table of resolved
 118  * entries is changed only in process context and protected
 119  * with weak lock mrt_lock. Queue of unresolved entries is protected
 120  * with strong spinlock mfc_unres_lock.
 121  *
 122  * In this case data path is free of exclusive locks at all.
 123  */
 124
 125 static struct kmem_cache *mrt_cachep __read_mostly;
 126
 127 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 128 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
 129                          struct sk_buff *skb, struct mfc_cache *cache,
 130                          int local);
 131 static int ipmr_cache_report(struct mr_table *mrt,
 132                              struct sk_buff *pkt, vifi_t vifi, int assert);
 133 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 134                               struct mfc_cache *c, struct rtmsg *rtm);
 135 static void ipmr_expire_process(unsigned long arg);
 136
 137 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 138 #define ipmr_for_each_table(mrt, net) \
 139         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 140
 141 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 142 {
 143         struct mr_table *mrt;
 144
 145         ipmr_for_each_table(mrt, net) {
 146                 if (mrt->id == id)
 147                         return mrt;
 148         }
 149         return NULL;
 150 }
 151
 152 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 153                            struct mr_table **mrt)
 154 {
 155         struct ipmr_result res;
 156         struct fib_lookup_arg arg = { .result = &res, };
 157         int err;
 158
 159         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 160                                flowi4_to_flowi(flp4), 0, &arg);
 161         if (err < 0)
 162                 return err;
 163         *mrt = res.mrt;
 164         return 0;
 165 }
 166
 167 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 168                             int flags, struct fib_lookup_arg *arg)
 169 {
 170         struct ipmr_result *res = arg->result;
 171         struct mr_table *mrt;
 172
 173         switch (rule->action) {
 174         case FR_ACT_TO_TBL:
 175                 break;
 176         case FR_ACT_UNREACHABLE:
 177                 return -ENETUNREACH;
 178         case FR_ACT_PROHIBIT:
 179                 return -EACCES;
 180         case FR_ACT_BLACKHOLE:
 181         default:
 182                 return -EINVAL;
 183         }
 184
 185         mrt = ipmr_get_table(rule->fr_net, rule->table);
 186         if (mrt == NULL)
 187                 return -EAGAIN;
 188         res->mrt = mrt;
 189         return 0;
 190 }
 191
 192 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 193 {
 194         return 1;
 195 }
 196
 197 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 198         FRA_GENERIC_POLICY,
 199 };
 200
 201 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 202                                struct fib_rule_hdr *frh, struct nlattr **tb)
 203 {
 204         return 0;
 205 }
 206
 207 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 208                              struct nlattr **tb)
 209 {
 210         return 1;
 211 }
 212
 213 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 214                           struct fib_rule_hdr *frh)
 215 {
 216         frh->dst_len = 0;
 217         frh->src_len = 0;
 218         frh->tos     = 0;
 219         return 0;
 220 }
 221
 222 static const struct fib_rules_ops __net_initdata ipmr_rules_ops_template = {
 223         .family         = RTNL_FAMILY_IPMR,
 224         .rule_size      = sizeof(struct ipmr_rule),
 225         .addr_size      = sizeof(u32),
 226         .action         = ipmr_rule_action,
 227         .match          = ipmr_rule_match,
 228         .configure      = ipmr_rule_configure,
 229         .compare        = ipmr_rule_compare,
 230         .default_pref   = fib_default_rule_pref,
 231         .fill           = ipmr_rule_fill,
 232         .nlgroup        = RTNLGRP_IPV4_RULE,
 233         .policy         = ipmr_rule_policy,
 234         .owner          = THIS_MODULE,
 235 };
 236
 237 static int __net_init ipmr_rules_init(struct net *net)
 238 {
 239         struct fib_rules_ops *ops;
 240         struct mr_table *mrt;
 241         int err;
 242
 243         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 244         if (IS_ERR(ops))
 245                 return PTR_ERR(ops);
 246
 247         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 248
 249         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 250         if (mrt == NULL) {
 251                 err = -ENOMEM;
 252                 goto err1;
 253         }
 254
 255         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 256         if (err < 0)
 257                 goto err2;
 258
 259         net->ipv4.mr_rules_ops = ops;
 260         return 0;
 261
 262 err2:
 263         kfree(mrt);
 264 err1:
 265         fib_rules_unregister(ops);
 266         return err;
 267 }
 268
 269 static void __net_exit ipmr_rules_exit(struct net *net)
 270 {
 271         struct mr_table *mrt, *next;
 272
 273         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 274                 list_del(&mrt->list);
 275                 kfree(mrt);
 276         }
 277         fib_rules_unregister(net->ipv4.mr_rules_ops);
 278 }
 279 #else
 280 #define ipmr_for_each_table(mrt, net) \
 281         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 282
 283 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 284 {
 285         return net->ipv4.mrt;
 286 }
 287
 288 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 289                            struct mr_table **mrt)
 290 {
 291         *mrt = net->ipv4.mrt;
 292         return 0;
 293 }
 294
 295 static int __net_init ipmr_rules_init(struct net *net)
 296 {
 297         net->ipv4.mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 298         return net->ipv4.mrt ? 0 : -ENOMEM;
 299 }
 300
 301 static void __net_exit ipmr_rules_exit(struct net *net)
 302 {
 303         kfree(net->ipv4.mrt);
 304 }
 305 #endif
 306
 307 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 308 {
 309         struct mr_table *mrt;
 310         unsigned int i;
 311
 312         mrt = ipmr_get_table(net, id);
 313         if (mrt != NULL)
 314                 return mrt;
 315
 316         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 317         if (mrt == NULL)
 318                 return NULL;
 319         write_pnet(&mrt->net, net);
 320         mrt->id = id;
 321
 322         /* Forwarding cache */
 323         for (i = 0; i < MFC_LINES; i++)
 324                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 325
 326         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 327
 328         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 329                     (unsigned long)mrt);
 330
 331 #ifdef CONFIG_IP_PIMSM
 332         mrt->mroute_reg_vif_num = -1;
 333 #endif
 334 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 335         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 336 #endif
 337         return mrt;
 338 }
 339
 340 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 341
 342 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 343 {
 344         struct net *net = dev_net(dev);
 345
 346         dev_close(dev);
 347
 348         dev = __dev_get_by_name(net, "tunl0");
 349         if (dev) {
 350                 const struct net_device_ops *ops = dev->netdev_ops;
 351                 struct ifreq ifr;
 352                 struct ip_tunnel_parm p;
 353
 354                 memset(&p, 0, sizeof(p));
 355                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 356                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 357                 p.iph.version = 4;
 358                 p.iph.ihl = 5;
 359                 p.iph.protocol = IPPROTO_IPIP;
 360                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 361                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 362
 363                 if (ops->ndo_do_ioctl) {
 364                         mm_segment_t oldfs = get_fs();
 365
 366                         set_fs(KERNEL_DS);
 367                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 368                         set_fs(oldfs);
 369                 }
 370         }
 371 }
 372
 373 static
 374 struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 375 {
 376         struct net_device  *dev;
 377
 378         dev = __dev_get_by_name(net, "tunl0");
 379
 380         if (dev) {
 381                 const struct net_device_ops *ops = dev->netdev_ops;
 382                 int err;
 383                 struct ifreq ifr;
 384                 struct ip_tunnel_parm p;
 385                 struct in_device  *in_dev;
 386
 387                 memset(&p, 0, sizeof(p));
 388                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 389                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 390                 p.iph.version = 4;
 391                 p.iph.ihl = 5;
 392                 p.iph.protocol = IPPROTO_IPIP;
 393                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 394                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 395
 396                 if (ops->ndo_do_ioctl) {
 397                         mm_segment_t oldfs = get_fs();
 398
 399                         set_fs(KERNEL_DS);
 400                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 401                         set_fs(oldfs);
 402                 } else {
 403                         err = -EOPNOTSUPP;
 404                 }
 405                 dev = NULL;
 406
 407                 if (err == 0 &&
 408                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 409                         dev->flags |= IFF_MULTICAST;
 410
 411                         in_dev = __in_dev_get_rtnl(dev);
 412                         if (in_dev == NULL)
 413                                 goto failure;
 414
 415                         ipv4_devconf_setall(in_dev);
 416                         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 417
 418                         if (dev_open(dev))
 419                                 goto failure;
 420                         dev_hold(dev);
 421                 }
 422         }
 423         return dev;
 424
 425 failure:
 426         /* allow the register to be completed before unregistering. */
 427         rtnl_unlock();
 428         rtnl_lock();
 429
 430         unregister_netdevice(dev);
 431         return NULL;
 432 }
 433
 434 #ifdef CONFIG_IP_PIMSM
 435
 436 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 437 {
 438         struct net *net = dev_net(dev);
 439         struct mr_table *mrt;
 440         struct flowi4 fl4 = {
 441                 .flowi4_oif     = dev->ifindex,
 442                 .flowi4_iif     = skb->skb_iif,
 443                 .flowi4_mark    = skb->mark,
 444         };
 445         int err;
 446
 447         err = ipmr_fib_lookup(net, &fl4, &mrt);
 448         if (err < 0) {
 449                 kfree_skb(skb);
 450                 return err;
 451         }
 452
 453         read_lock(&mrt_lock);
 454         dev->stats.tx_bytes += skb->len;
 455         dev->stats.tx_packets++;
 456         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 457         read_unlock(&mrt_lock);
 458         kfree_skb(skb);
 459         return NETDEV_TX_OK;
 460 }
 461
 462 static const struct net_device_ops reg_vif_netdev_ops = {
 463         .ndo_start_xmit = reg_vif_xmit,
 464 };
 465
 466 static void reg_vif_setup(struct net_device *dev)
 467 {
 468         dev->type               = ARPHRD_PIMREG;
 469         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 470         dev->flags              = IFF_NOARP;
 471         dev->netdev_ops         = &reg_vif_netdev_ops,
 472         dev->destructor         = free_netdev;
 473         dev->features           |= NETIF_F_NETNS_LOCAL;
 474 }
 475
 476 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 477 {
 478         struct net_device *dev;
 479         struct in_device *in_dev;
 480         char name[IFNAMSIZ];
 481
 482         if (mrt->id == RT_TABLE_DEFAULT)
 483                 sprintf(name, "pimreg");
 484         else
 485                 sprintf(name, "pimreg%u", mrt->id);
 486
 487         dev = alloc_netdev(0, name, reg_vif_setup);
 488
 489         if (dev == NULL)
 490                 return NULL;
 491
 492         dev_net_set(dev, net);
 493
 494         if (register_netdevice(dev)) {
 495                 free_netdev(dev);
 496                 return NULL;
 497         }
 498         dev->iflink = 0;
 499
 500         rcu_read_lock();
 501         in_dev = __in_dev_get_rcu(dev);
 502         if (!in_dev) {
 503                 rcu_read_unlock();
 504                 goto failure;
 505         }
 506
 507         ipv4_devconf_setall(in_dev);
 508         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 509         rcu_read_unlock();
 510
 511         if (dev_open(dev))
 512                 goto failure;
 513
 514         dev_hold(dev);
 515
 516         return dev;
 517
 518 failure:
 519         /* allow the register to be completed before unregistering. */
 520         rtnl_unlock();
 521         rtnl_lock();
 522
 523         unregister_netdevice(dev);
 524         return NULL;
 525 }
 526 #endif
 527
 528 /*
 529  *      Delete a VIF entry
 530  *      @notify: Set to 1, if the caller is a notifier_call
 531  */
 532
 533 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 534                       struct list_head *head)
 535 {
 536         struct vif_device *v;
 537         struct net_device *dev;
 538         struct in_device *in_dev;
 539
 540         if (vifi < 0 || vifi >= mrt->maxvif)
 541                 return -EADDRNOTAVAIL;
 542
 543         v = &mrt->vif_table[vifi];
 544
 545         write_lock_bh(&mrt_lock);
 546         dev = v->dev;
 547         v->dev = NULL;
 548
 549         if (!dev) {
 550                 write_unlock_bh(&mrt_lock);
 551                 return -EADDRNOTAVAIL;
 552         }
 553
 554 #ifdef CONFIG_IP_PIMSM
 555         if (vifi == mrt->mroute_reg_vif_num)
 556                 mrt->mroute_reg_vif_num = -1;
 557 #endif
 558
 559         if (vifi + 1 == mrt->maxvif) {
 560                 int tmp;
 561
 562                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 563                         if (VIF_EXISTS(mrt, tmp))
 564                                 break;
 565                 }
 566                 mrt->maxvif = tmp+1;
 567         }
 568
 569         write_unlock_bh(&mrt_lock);
 570
 571         dev_set_allmulti(dev, -1);
 572
 573         in_dev = __in_dev_get_rtnl(dev);
 574         if (in_dev) {
 575                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 576                 ip_rt_multicast_event(in_dev);
 577         }
 578
 579         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 580                 unregister_netdevice_queue(dev, head);
 581
 582         dev_put(dev);
 583         return 0;
 584 }
 585
 586 static void ipmr_cache_free_rcu(struct rcu_head *head)
 587 {
 588         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 589
 590         kmem_cache_free(mrt_cachep, c);
 591 }
 592
 593 static inline void ipmr_cache_free(struct mfc_cache *c)
 594 {
 595         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 596 }
 597
 598 /* Destroy an unresolved cache entry, killing queued skbs
 599  * and reporting error to netlink readers.
 600  */
 601
 602 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 603 {
 604         struct net *net = read_pnet(&mrt->net);
 605         struct sk_buff *skb;
 606         struct nlmsgerr *e;
 607
 608         atomic_dec(&mrt->cache_resolve_queue_len);
 609
 610         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 611                 if (ip_hdr(skb)->version == 0) {
 612                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 613                         nlh->nlmsg_type = NLMSG_ERROR;
 614                         nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 615                         skb_trim(skb, nlh->nlmsg_len);
 616                         e = NLMSG_DATA(nlh);
 617                         e->error = -ETIMEDOUT;
 618                         memset(&e->msg, 0, sizeof(e->msg));
 619
 620                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
 621                 } else {
 622                         kfree_skb(skb);
 623                 }
 624         }
 625
 626         ipmr_cache_free(c);
 627 }
 628
 629
 630 /* Timer process for the unresolved queue. */
 631
 632 static void ipmr_expire_process(unsigned long arg)
 633 {
 634         struct mr_table *mrt = (struct mr_table *)arg;
 635         unsigned long now;
 636         unsigned long expires;
 637         struct mfc_cache *c, *next;
 638
 639         if (!spin_trylock(&mfc_unres_lock)) {
 640                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 641                 return;
 642         }
 643
 644         if (list_empty(&mrt->mfc_unres_queue))
 645                 goto out;
 646
 647         now = jiffies;
 648         expires = 10*HZ;
 649
 650         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 651                 if (time_after(c->mfc_un.unres.expires, now)) {
 652                         unsigned long interval = c->mfc_un.unres.expires - now;
 653                         if (interval < expires)
 654                                 expires = interval;
 655                         continue;
 656                 }
 657
 658                 list_del(&c->list);
 659                 ipmr_destroy_unres(mrt, c);
 660         }
 661
 662         if (!list_empty(&mrt->mfc_unres_queue))
 663                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 664
 665 out:
 666         spin_unlock(&mfc_unres_lock);
 667 }
 668
 669 /* Fill oifs list. It is called under write locked mrt_lock. */
 670
 671 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 672                                    unsigned char *ttls)
 673 {
 674         int vifi;
 675
 676         cache->mfc_un.res.minvif = MAXVIFS;
 677         cache->mfc_un.res.maxvif = 0;
 678         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 679
 680         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 681                 if (VIF_EXISTS(mrt, vifi) &&
 682                     ttls[vifi] && ttls[vifi] < 255) {
 683                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 684                         if (cache->mfc_un.res.minvif > vifi)
 685                                 cache->mfc_un.res.minvif = vifi;
 686                         if (cache->mfc_un.res.maxvif <= vifi)
 687                                 cache->mfc_un.res.maxvif = vifi + 1;
 688                 }
 689         }
 690 }
 691
 692 static int vif_add(struct net *net, struct mr_table *mrt,
 693                    struct vifctl *vifc, int mrtsock)
 694 {
 695         int vifi = vifc->vifc_vifi;
 696         struct vif_device *v = &mrt->vif_table[vifi];
 697         struct net_device *dev;
 698         struct in_device *in_dev;
 699         int err;
 700
 701         /* Is vif busy ? */
 702         if (VIF_EXISTS(mrt, vifi))
 703                 return -EADDRINUSE;
 704
 705         switch (vifc->vifc_flags) {
 706 #ifdef CONFIG_IP_PIMSM
 707         case VIFF_REGISTER:
 708                 /*
 709                  * Special Purpose VIF in PIM
 710                  * All the packets will be sent to the daemon
 711                  */
 712                 if (mrt->mroute_reg_vif_num >= 0)
 713                         return -EADDRINUSE;
 714                 dev = ipmr_reg_vif(net, mrt);
 715                 if (!dev)
 716                         return -ENOBUFS;
 717                 err = dev_set_allmulti(dev, 1);
 718                 if (err) {
 719                         unregister_netdevice(dev);
 720                         dev_put(dev);
 721                         return err;
 722                 }
 723                 break;
 724 #endif
 725         case VIFF_TUNNEL:
 726                 dev = ipmr_new_tunnel(net, vifc);
 727                 if (!dev)
 728                         return -ENOBUFS;
 729                 err = dev_set_allmulti(dev, 1);
 730                 if (err) {
 731                         ipmr_del_tunnel(dev, vifc);
 732                         dev_put(dev);
 733                         return err;
 734                 }
 735                 break;
 736
 737         case VIFF_USE_IFINDEX:
 738         case 0:
 739                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 740                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 741                         if (dev && __in_dev_get_rtnl(dev) == NULL) {
 742                                 dev_put(dev);
 743                                 return -EADDRNOTAVAIL;
 744                         }
 745                 } else {
 746                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 747                 }
 748                 if (!dev)
 749                         return -EADDRNOTAVAIL;
 750                 err = dev_set_allmulti(dev, 1);
 751                 if (err) {
 752                         dev_put(dev);
 753                         return err;
 754                 }
 755                 break;
 756         default:
 757                 return -EINVAL;
 758         }
 759
 760         in_dev = __in_dev_get_rtnl(dev);
 761         if (!in_dev) {
 762                 dev_put(dev);
 763                 return -EADDRNOTAVAIL;
 764         }
 765         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 766         ip_rt_multicast_event(in_dev);
 767
 768         /* Fill in the VIF structures */
 769
 770         v->rate_limit = vifc->vifc_rate_limit;
 771         v->local = vifc->vifc_lcl_addr.s_addr;
 772         v->remote = vifc->vifc_rmt_addr.s_addr;
 773         v->flags = vifc->vifc_flags;
 774         if (!mrtsock)
 775                 v->flags |= VIFF_STATIC;
 776         v->threshold = vifc->vifc_threshold;
 777         v->bytes_in = 0;
 778         v->bytes_out = 0;
 779         v->pkt_in = 0;
 780         v->pkt_out = 0;
 781         v->link = dev->ifindex;
 782         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 783                 v->link = dev->iflink;
 784
 785         /* And finish update writing critical data */
 786         write_lock_bh(&mrt_lock);
 787         v->dev = dev;
 788 #ifdef CONFIG_IP_PIMSM
 789         if (v->flags & VIFF_REGISTER)
 790                 mrt->mroute_reg_vif_num = vifi;
 791 #endif
 792         if (vifi+1 > mrt->maxvif)
 793                 mrt->maxvif = vifi+1;
 794         write_unlock_bh(&mrt_lock);
 795         return 0;
 796 }
 797
 798 /* called with rcu_read_lock() */
 799 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 800                                          __be32 origin,
 801                                          __be32 mcastgrp)
 802 {
 803         int line = MFC_HASH(mcastgrp, origin);
 804         struct mfc_cache *c;
 805
 806         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 807                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 808                         return c;
 809         }
 810         return NULL;
 811 }
 812
 813 /*
 814  *      Allocate a multicast cache entry
 815  */
 816 static struct mfc_cache *ipmr_cache_alloc(void)
 817 {
 818         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 819
 820         if (c)
 821                 c->mfc_un.res.minvif = MAXVIFS;
 822         return c;
 823 }
 824
 825 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 826 {
 827         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 828
 829         if (c) {
 830                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 831                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 832         }
 833         return c;
 834 }
 835
 836 /*
 837  *      A cache entry has gone into a resolved state from queued
 838  */
 839
 840 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 841                                struct mfc_cache *uc, struct mfc_cache *c)
 842 {
 843         struct sk_buff *skb;
 844         struct nlmsgerr *e;
 845
 846         /* Play the pending entries through our router */
 847
 848         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 849                 if (ip_hdr(skb)->version == 0) {
 850                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 851
 852                         if (__ipmr_fill_mroute(mrt, skb, c, NLMSG_DATA(nlh)) > 0) {
 853                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 854                                                  (u8 *)nlh;
 855                         } else {
 856                                 nlh->nlmsg_type = NLMSG_ERROR;
 857                                 nlh->nlmsg_len = NLMSG_LENGTH(sizeof(struct nlmsgerr));
 858                                 skb_trim(skb, nlh->nlmsg_len);
 859                                 e = NLMSG_DATA(nlh);
 860                                 e->error = -EMSGSIZE;
 861                                 memset(&e->msg, 0, sizeof(e->msg));
 862                         }
 863
 864                         rtnl_unicast(skb, net, NETLINK_CB(skb).pid);
 865                 } else {
 866                         ip_mr_forward(net, mrt, skb, c, 0);
 867                 }
 868         }
 869 }
 870
 871 /*
 872  *      Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 873  *      expects the following bizarre scheme.
 874  *
 875  *      Called under mrt_lock.
 876  */
 877
 878 static int ipmr_cache_report(struct mr_table *mrt,
 879                              struct sk_buff *pkt, vifi_t vifi, int assert)
 880 {
 881         struct sk_buff *skb;
 882         const int ihl = ip_hdrlen(pkt);
 883         struct igmphdr *igmp;
 884         struct igmpmsg *msg;
 885         struct sock *mroute_sk;
 886         int ret;
 887
 888 #ifdef CONFIG_IP_PIMSM
 889         if (assert == IGMPMSG_WHOLEPKT)
 890                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 891         else
 892 #endif
 893                 skb = alloc_skb(128, GFP_ATOMIC);
 894
 895         if (!skb)
 896                 return -ENOBUFS;
 897
 898 #ifdef CONFIG_IP_PIMSM
 899         if (assert == IGMPMSG_WHOLEPKT) {
 900                 /* Ugly, but we have no choice with this interface.
 901                  * Duplicate old header, fix ihl, length etc.
 902                  * And all this only to mangle msg->im_msgtype and
 903                  * to set msg->im_mbz to "mbz" :-)
 904                  */
 905                 skb_push(skb, sizeof(struct iphdr));
 906                 skb_reset_network_header(skb);
 907                 skb_reset_transport_header(skb);
 908                 msg = (struct igmpmsg *)skb_network_header(skb);
 909                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 910                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 911                 msg->im_mbz = 0;
 912                 msg->im_vif = mrt->mroute_reg_vif_num;
 913                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 914                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 915                                              sizeof(struct iphdr));
 916         } else
 917 #endif
 918         {
 919
 920         /* Copy the IP header */
 921
 922         skb->network_header = skb->tail;
 923         skb_put(skb, ihl);
 924         skb_copy_to_linear_data(skb, pkt->data, ihl);
 925         ip_hdr(skb)->protocol = 0;      /* Flag to the kernel this is a route add */
 926         msg = (struct igmpmsg *)skb_network_header(skb);
 927         msg->im_vif = vifi;
 928         skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 929
 930         /* Add our header */
 931
 932         igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 933         igmp->type      =
 934         msg->im_msgtype = assert;
 935         igmp->code      = 0;
 936         ip_hdr(skb)->tot_len = htons(skb->len);         /* Fix the length */
 937         skb->transport_header = skb->network_header;
 938         }
 939
 940         rcu_read_lock();
 941         mroute_sk = rcu_dereference(mrt->mroute_sk);
 942         if (mroute_sk == NULL) {
 943                 rcu_read_unlock();
 944                 kfree_skb(skb);
 945                 return -EINVAL;
 946         }
 947
 948         /* Deliver to mrouted */
 949
 950         ret = sock_queue_rcv_skb(mroute_sk, skb);
 951         rcu_read_unlock();
 952         if (ret < 0) {
 953                 if (net_ratelimit())
 954                         printk(KERN_WARNING "mroute: pending queue full, dropping entries.\n");
 955                 kfree_skb(skb);
 956         }
 957
 958         return ret;
 959 }
 960
 961 /*
 962  *      Queue a packet for resolution. It gets locked cache entry!
 963  */
 964
 965 static int
 966 ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi, struct sk_buff *skb)
 967 {
 968         bool found = false;
 969         int err;
 970         struct mfc_cache *c;
 971         const struct iphdr *iph = ip_hdr(skb);
 972
 973         spin_lock_bh(&mfc_unres_lock);
 974         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
 975                 if (c->mfc_mcastgrp == iph->daddr &&
 976                     c->mfc_origin == iph->saddr) {
 977                         found = true;
 978                         break;
 979                 }
 980         }
 981
 982         if (!found) {
 983                 /* Create a new entry if allowable */
 984
 985                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
 986                     (c = ipmr_cache_alloc_unres()) == NULL) {
 987                         spin_unlock_bh(&mfc_unres_lock);
 988
 989                         kfree_skb(skb);
 990                         return -ENOBUFS;
 991                 }
 992
 993                 /* Fill in the new cache entry */
 994
 995                 c->mfc_parent   = -1;
 996                 c->mfc_origin   = iph->saddr;
 997                 c->mfc_mcastgrp = iph->daddr;
 998
 999                 /* Reflect first query at mrouted. */
1000
1001                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1002                 if (err < 0) {
1003                         /* If the report failed throw the cache entry
1004                            out - Brad Parker
1005                          */
1006                         spin_unlock_bh(&mfc_unres_lock);
1007
1008                         ipmr_cache_free(c);
1009                         kfree_skb(skb);
1010                         return err;
1011                 }
1012
1013                 atomic_inc(&mrt->cache_resolve_queue_len);
1014                 list_add(&c->list, &mrt->mfc_unres_queue);
1015
1016                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1017                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1018         }
1019
1020         /* See if we can append the packet */
1021
1022         if (c->mfc_un.unres.unresolved.qlen > 3) {
1023                 kfree_skb(skb);
1024                 err = -ENOBUFS;
1025         } else {
1026                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1027                 err = 0;
1028         }
1029
1030         spin_unlock_bh(&mfc_unres_lock);
1031         return err;
1032 }
1033
1034 /*
1035  *      MFC cache manipulation by user space mroute daemon
1036  */
1037
1038 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc)
1039 {
1040         int line;
1041         struct mfc_cache *c, *next;
1042
1043         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1044
1045         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1046                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1047                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1048                         list_del_rcu(&c->list);
1049
1050                         ipmr_cache_free(c);
1051                         return 0;
1052                 }
1053         }
1054         return -ENOENT;
1055 }
1056
1057 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1058                         struct mfcctl *mfc, int mrtsock)
1059 {
1060         bool found = false;
1061         int line;
1062         struct mfc_cache *uc, *c;
1063
1064         if (mfc->mfcc_parent >= MAXVIFS)
1065                 return -ENFILE;
1066
1067         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1068
1069         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1070                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1071                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr) {
1072                         found = true;
1073                         break;
1074                 }
1075         }
1076
1077         if (found) {
1078                 write_lock_bh(&mrt_lock);
1079                 c->mfc_parent = mfc->mfcc_parent;
1080                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1081                 if (!mrtsock)
1082                         c->mfc_flags |= MFC_STATIC;
1083                 write_unlock_bh(&mrt_lock);
1084                 return 0;
1085         }
1086
1087         if (!ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1088                 return -EINVAL;
1089
1090         c = ipmr_cache_alloc();
1091         if (c == NULL)
1092                 return -ENOMEM;
1093
1094         c->mfc_origin = mfc->mfcc_origin.s_addr;
1095         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1096         c->mfc_parent = mfc->mfcc_parent;
1097         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1098         if (!mrtsock)
1099                 c->mfc_flags |= MFC_STATIC;
1100
1101         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1102
1103         /*
1104          *      Check to see if we resolved a queued list. If so we
1105          *      need to send on the frames and tidy up.
1106          */
1107         found = false;
1108         spin_lock_bh(&mfc_unres_lock);
1109         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1110                 if (uc->mfc_origin == c->mfc_origin &&
1111                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1112                         list_del(&uc->list);
1113                         atomic_dec(&mrt->cache_resolve_queue_len);
1114                         found = true;
1115                         break;
1116                 }
1117         }
1118         if (list_empty(&mrt->mfc_unres_queue))
1119                 del_timer(&mrt->ipmr_expire_timer);
1120         spin_unlock_bh(&mfc_unres_lock);
1121
1122         if (found) {
1123                 ipmr_cache_resolve(net, mrt, uc, c);
1124                 ipmr_cache_free(uc);
1125         }
1126         return 0;
1127 }
1128
1129 /*
1130  *      Close the multicast socket, and clear the vif tables etc
1131  */
1132
1133 static void mroute_clean_tables(struct mr_table *mrt)
1134 {
1135         int i;
1136         LIST_HEAD(list);
1137         struct mfc_cache *c, *next;
1138
1139         /* Shut down all active vif entries */
1140
1141         for (i = 0; i < mrt->maxvif; i++) {
1142                 if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1143                         vif_delete(mrt, i, 0, &list);
1144         }
1145         unregister_netdevice_many(&list);
1146
1147         /* Wipe the cache */
1148
1149         for (i = 0; i < MFC_LINES; i++) {
1150                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1151                         if (c->mfc_flags & MFC_STATIC)
1152                                 continue;
1153                         list_del_rcu(&c->list);
1154                         ipmr_cache_free(c);
1155                 }
1156         }
1157
1158         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1159                 spin_lock_bh(&mfc_unres_lock);
1160                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1161                         list_del(&c->list);
1162                         ipmr_destroy_unres(mrt, c);
1163                 }
1164                 spin_unlock_bh(&mfc_unres_lock);
1165         }
1166 }
1167
1168 /* called from ip_ra_control(), before an RCU grace period,
1169  * we dont need to call synchronize_rcu() here
1170  */
1171 static void mrtsock_destruct(struct sock *sk)
1172 {
1173         struct net *net = sock_net(sk);
1174         struct mr_table *mrt;
1175
1176         rtnl_lock();
1177         ipmr_for_each_table(mrt, net) {
1178                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1179                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1180                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1181                         mroute_clean_tables(mrt);
1182                 }
1183         }
1184         rtnl_unlock();
1185 }
1186
1187 /*
1188  *      Socket options and virtual interface manipulation. The whole
1189  *      virtual interface system is a complete heap, but unfortunately
1190  *      that's how BSD mrouted happens to think. Maybe one day with a proper
1191  *      MOSPF/PIM router set up we can clean this up.
1192  */
1193
1194 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval, unsigned int optlen)
1195 {
1196         int ret;
1197         struct vifctl vif;
1198         struct mfcctl mfc;
1199         struct net *net = sock_net(sk);
1200         struct mr_table *mrt;
1201
1202         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1203         if (mrt == NULL)
1204                 return -ENOENT;
1205
1206         if (optname != MRT_INIT) {
1207                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1208                     !capable(CAP_NET_ADMIN))
1209                         return -EACCES;
1210         }
1211
1212         switch (optname) {
1213         case MRT_INIT:
1214                 if (sk->sk_type != SOCK_RAW ||
1215                     inet_sk(sk)->inet_num != IPPROTO_IGMP)
1216                         return -EOPNOTSUPP;
1217                 if (optlen != sizeof(int))
1218                         return -ENOPROTOOPT;
1219
1220                 rtnl_lock();
1221                 if (rtnl_dereference(mrt->mroute_sk)) {
1222                         rtnl_unlock();
1223                         return -EADDRINUSE;
1224                 }
1225
1226                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1227                 if (ret == 0) {
1228                         RCU_INIT_POINTER(mrt->mroute_sk, sk);
1229                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1230                 }
1231                 rtnl_unlock();
1232                 return ret;
1233         case MRT_DONE:
1234                 if (sk != rcu_access_pointer(mrt->mroute_sk))
1235                         return -EACCES;
1236                 return ip_ra_control(sk, 0, NULL);
1237         case MRT_ADD_VIF:
1238         case MRT_DEL_VIF:
1239                 if (optlen != sizeof(vif))
1240                         return -EINVAL;
1241                 if (copy_from_user(&vif, optval, sizeof(vif)))
1242                         return -EFAULT;
1243                 if (vif.vifc_vifi >= MAXVIFS)
1244                         return -ENFILE;
1245                 rtnl_lock();
1246                 if (optname == MRT_ADD_VIF) {
1247                         ret = vif_add(net, mrt, &vif,
1248                                       sk == rtnl_dereference(mrt->mroute_sk));
1249                 } else {
1250                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1251                 }
1252                 rtnl_unlock();
1253                 return ret;
1254
1255                 /*
1256                  *      Manipulate the forwarding caches. These live
1257                  *      in a sort of kernel/user symbiosis.
1258                  */
1259         case MRT_ADD_MFC:
1260         case MRT_DEL_MFC:
1261                 if (optlen != sizeof(mfc))
1262                         return -EINVAL;
1263                 if (copy_from_user(&mfc, optval, sizeof(mfc)))
1264                         return -EFAULT;
1265                 rtnl_lock();
1266                 if (optname == MRT_DEL_MFC)
1267                         ret = ipmr_mfc_delete(mrt, &mfc);
1268                 else
1269                         ret = ipmr_mfc_add(net, mrt, &mfc,
1270                                            sk == rtnl_dereference(mrt->mroute_sk));
1271                 rtnl_unlock();
1272                 return ret;
1273                 /*
1274                  *      Control PIM assert.
1275                  */
1276         case MRT_ASSERT:
1277         {
1278                 int v;
1279                 if (get_user(v, (int __user *)optval))
1280                         return -EFAULT;
1281                 mrt->mroute_do_assert = (v) ? 1 : 0;
1282                 return 0;
1283         }
1284 #ifdef CONFIG_IP_PIMSM
1285         case MRT_PIM:
1286         {
1287                 int v;
1288
1289                 if (get_user(v, (int __user *)optval))
1290                         return -EFAULT;
1291                 v = (v) ? 1 : 0;
1292
1293                 rtnl_lock();
1294                 ret = 0;
1295                 if (v != mrt->mroute_do_pim) {
1296                         mrt->mroute_do_pim = v;
1297                         mrt->mroute_do_assert = v;
1298                 }
1299                 rtnl_unlock();
1300                 return ret;
1301         }
1302 #endif
1303 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
1304         case MRT_TABLE:
1305         {
1306                 u32 v;
1307
1308                 if (optlen != sizeof(u32))
1309                         return -EINVAL;
1310                 if (get_user(v, (u32 __user *)optval))
1311                         return -EFAULT;
1312
1313                 rtnl_lock();
1314                 ret = 0;
1315                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1316                         ret = -EBUSY;
1317                 } else {
1318                         if (!ipmr_new_table(net, v))
1319                                 ret = -ENOMEM;
1320                         raw_sk(sk)->ipmr_table = v;
1321                 }
1322                 rtnl_unlock();
1323                 return ret;
1324         }
1325 #endif
1326         /*
1327          *      Spurious command, or MRT_VERSION which you cannot
1328          *      set.
1329          */
1330         default:
1331                 return -ENOPROTOOPT;
1332         }
1333 }
1334
1335 /*
1336  *      Getsock opt support for the multicast routing system.
1337  */
1338
1339 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1340 {
1341         int olr;
1342         int val;
1343         struct net *net = sock_net(sk);
1344         struct mr_table *mrt;
1345
1346         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1347         if (mrt == NULL)
1348                 return -ENOENT;
1349
1350         if (optname != MRT_VERSION &&
1351 #ifdef CONFIG_IP_PIMSM
1352            optname != MRT_PIM &&
1353 #endif
1354            optname != MRT_ASSERT)
1355                 return -ENOPROTOOPT;
1356
1357         if (get_user(olr, optlen))
1358                 return -EFAULT;
1359
1360         olr = min_t(unsigned int, olr, sizeof(int));
1361         if (olr < 0)
1362                 return -EINVAL;
1363
1364         if (put_user(olr, optlen))
1365                 return -EFAULT;
1366         if (optname == MRT_VERSION)
1367                 val = 0x0305;
1368 #ifdef CONFIG_IP_PIMSM
1369         else if (optname == MRT_PIM)
1370                 val = mrt->mroute_do_pim;
1371 #endif
1372         else
1373                 val = mrt->mroute_do_assert;
1374         if (copy_to_user(optval, &val, olr))
1375                 return -EFAULT;
1376         return 0;
1377 }
1378
1379 /*
1380  *      The IP multicast ioctl support routines.
1381  */
1382
1383 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1384 {
1385         struct sioc_sg_req sr;
1386         struct sioc_vif_req vr;
1387         struct vif_device *vif;
1388         struct mfc_cache *c;
1389         struct net *net = sock_net(sk);
1390         struct mr_table *mrt;
1391
1392         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1393         if (mrt == NULL)
1394                 return -ENOENT;
1395
1396         switch (cmd) {
1397         case SIOCGETVIFCNT:
1398                 if (copy_from_user(&vr, arg, sizeof(vr)))
1399                         return -EFAULT;
1400                 if (vr.vifi >= mrt->maxvif)
1401                         return -EINVAL;
1402                 read_lock(&mrt_lock);
1403                 vif = &mrt->vif_table[vr.vifi];
1404                 if (VIF_EXISTS(mrt, vr.vifi)) {
1405                         vr.icount = vif->pkt_in;
1406                         vr.ocount = vif->pkt_out;
1407                         vr.ibytes = vif->bytes_in;
1408                         vr.obytes = vif->bytes_out;
1409                         read_unlock(&mrt_lock);
1410
1411                         if (copy_to_user(arg, &vr, sizeof(vr)))
1412                                 return -EFAULT;
1413                         return 0;
1414                 }
1415                 read_unlock(&mrt_lock);
1416                 return -EADDRNOTAVAIL;
1417         case SIOCGETSGCNT:
1418                 if (copy_from_user(&sr, arg, sizeof(sr)))
1419                         return -EFAULT;
1420
1421                 rcu_read_lock();
1422                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1423                 if (c) {
1424                         sr.pktcnt = c->mfc_un.res.pkt;
1425                         sr.bytecnt = c->mfc_un.res.bytes;
1426                         sr.wrong_if = c->mfc_un.res.wrong_if;
1427                         rcu_read_unlock();
1428
1429                         if (copy_to_user(arg, &sr, sizeof(sr)))
1430                                 return -EFAULT;
1431                         return 0;
1432                 }
1433                 rcu_read_unlock();
1434                 return -EADDRNOTAVAIL;
1435         default:
1436                 return -ENOIOCTLCMD;
1437         }
1438 }
1439
1440 #ifdef CONFIG_COMPAT
1441 struct compat_sioc_sg_req {
1442         struct in_addr src;
1443         struct in_addr grp;
1444         compat_ulong_t pktcnt;
1445         compat_ulong_t bytecnt;
1446         compat_ulong_t wrong_if;
1447 };
1448
1449 struct compat_sioc_vif_req {
1450         vifi_t  vifi;           /* Which iface */
1451         compat_ulong_t icount;
1452         compat_ulong_t ocount;
1453         compat_ulong_t ibytes;
1454         compat_ulong_t obytes;
1455 };
1456
1457 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1458 {
1459         struct compat_sioc_sg_req sr;
1460         struct compat_sioc_vif_req vr;
1461         struct vif_device *vif;
1462         struct mfc_cache *c;
1463         struct net *net = sock_net(sk);
1464         struct mr_table *mrt;
1465
1466         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1467         if (mrt == NULL)
1468                 return -ENOENT;
1469
1470         switch (cmd) {
1471         case SIOCGETVIFCNT:
1472                 if (copy_from_user(&vr, arg, sizeof(vr)))
1473                         return -EFAULT;
1474                 if (vr.vifi >= mrt->maxvif)
1475                         return -EINVAL;
1476                 read_lock(&mrt_lock);
1477                 vif = &mrt->vif_table[vr.vifi];
1478                 if (VIF_EXISTS(mrt, vr.vifi)) {
1479                         vr.icount = vif->pkt_in;
1480                         vr.ocount = vif->pkt_out;
1481                         vr.ibytes = vif->bytes_in;
1482                         vr.obytes = vif->bytes_out;
1483                         read_unlock(&mrt_lock);
1484
1485                         if (copy_to_user(arg, &vr, sizeof(vr)))
1486                                 return -EFAULT;
1487                         return 0;
1488                 }
1489                 read_unlock(&mrt_lock);
1490                 return -EADDRNOTAVAIL;
1491         case SIOCGETSGCNT:
1492                 if (copy_from_user(&sr, arg, sizeof(sr)))
1493                         return -EFAULT;
1494
1495                 rcu_read_lock();
1496                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1497                 if (c) {
1498                         sr.pktcnt = c->mfc_un.res.pkt;
1499                         sr.bytecnt = c->mfc_un.res.bytes;
1500                         sr.wrong_if = c->mfc_un.res.wrong_if;
1501                         rcu_read_unlock();
1502
1503                         if (copy_to_user(arg, &sr, sizeof(sr)))
1504                                 return -EFAULT;
1505                         return 0;
1506                 }
1507                 rcu_read_unlock();
1508                 return -EADDRNOTAVAIL;
1509         default:
1510                 return -ENOIOCTLCMD;
1511         }
1512 }
1513 #endif
1514
1515
1516 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1517 {
1518         struct net_device *dev = ptr;
1519         struct net *net = dev_net(dev);
1520         struct mr_table *mrt;
1521         struct vif_device *v;
1522         int ct;
1523         LIST_HEAD(list);
1524
1525         if (event != NETDEV_UNREGISTER)
1526                 return NOTIFY_DONE;
1527
1528         ipmr_for_each_table(mrt, net) {
1529                 v = &mrt->vif_table[0];
1530                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1531                         if (v->dev == dev)
1532                                 vif_delete(mrt, ct, 1, &list);
1533                 }
1534         }
1535         unregister_netdevice_many(&list);
1536         return NOTIFY_DONE;
1537 }
1538
1539
1540 static struct notifier_block ip_mr_notifier = {
1541         .notifier_call = ipmr_device_event,
1542 };
1543
1544 /*
1545  *      Encapsulate a packet by attaching a valid IPIP header to it.
1546  *      This avoids tunnel drivers and other mess and gives us the speed so
1547  *      important for multicast video.
1548  */
1549
1550 static void ip_encap(struct sk_buff *skb, __be32 saddr, __be32 daddr)
1551 {
1552         struct iphdr *iph;
1553         const struct iphdr *old_iph = ip_hdr(skb);
1554
1555         skb_push(skb, sizeof(struct iphdr));
1556         skb->transport_header = skb->network_header;
1557         skb_reset_network_header(skb);
1558         iph = ip_hdr(skb);
1559
1560         iph->version    =       4;
1561         iph->tos        =       old_iph->tos;
1562         iph->ttl        =       old_iph->ttl;
1563         iph->frag_off   =       0;
1564         iph->daddr      =       daddr;
1565         iph->saddr      =       saddr;
1566         iph->protocol   =       IPPROTO_IPIP;
1567         iph->ihl        =       5;
1568         iph->tot_len    =       htons(skb->len);
1569         ip_select_ident(iph, skb_dst(skb), NULL);
1570         ip_send_check(iph);
1571
1572         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1573         nf_reset(skb);
1574 }
1575
1576 static inline int ipmr_forward_finish(struct sk_buff *skb)
1577 {
1578         struct ip_options *opt = &(IPCB(skb)->opt);
1579
1580         IP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), IPSTATS_MIB_OUTFORWDATAGRAMS);
1581
1582         if (unlikely(opt->optlen))
1583                 ip_forward_options(skb);
1584
1585         return dst_output(skb);
1586 }
1587
1588 /*
1589  *      Processing handlers for ipmr_forward
1590  */
1591
1592 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1593                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1594 {
1595         const struct iphdr *iph = ip_hdr(skb);
1596         struct vif_device *vif = &mrt->vif_table[vifi];
1597         struct net_device *dev;
1598         struct rtable *rt;
1599         struct flowi4 fl4;
1600         int    encap = 0;
1601
1602         if (vif->dev == NULL)
1603                 goto out_free;
1604
1605 #ifdef CONFIG_IP_PIMSM
1606         if (vif->flags & VIFF_REGISTER) {
1607                 vif->pkt_out++;
1608                 vif->bytes_out += skb->len;
1609                 vif->dev->stats.tx_bytes += skb->len;
1610                 vif->dev->stats.tx_packets++;
1611                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1612                 goto out_free;
1613         }
1614 #endif
1615
1616         if (vif->flags & VIFF_TUNNEL) {
1617                 rt = ip_route_output_ports(net, &fl4, NULL,
1618                                            vif->remote, vif->local,
1619                                            0, 0,
1620                                            IPPROTO_IPIP,
1621                                            RT_TOS(iph->tos), vif->link);
1622                 if (IS_ERR(rt))
1623                         goto out_free;
1624                 encap = sizeof(struct iphdr);
1625         } else {
1626                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1627                                            0, 0,
1628                                            IPPROTO_IPIP,
1629                                            RT_TOS(iph->tos), vif->link);
1630                 if (IS_ERR(rt))
1631                         goto out_free;
1632         }
1633
1634         dev = rt->dst.dev;
1635
1636         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1637                 /* Do not fragment multicasts. Alas, IPv4 does not
1638                  * allow to send ICMP, so that packets will disappear
1639                  * to blackhole.
1640                  */
1641
1642                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
1643                 ip_rt_put(rt);
1644                 goto out_free;
1645         }
1646
1647         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1648
1649         if (skb_cow(skb, encap)) {
1650                 ip_rt_put(rt);
1651                 goto out_free;
1652         }
1653
1654         vif->pkt_out++;
1655         vif->bytes_out += skb->len;
1656
1657         skb_dst_drop(skb);
1658         skb_dst_set(skb, &rt->dst);
1659         ip_decrease_ttl(ip_hdr(skb));
1660
1661         /* FIXME: forward and output firewalls used to be called here.
1662          * What do we do with netfilter? -- RR
1663          */
1664         if (vif->flags & VIFF_TUNNEL) {
1665                 ip_encap(skb, vif->local, vif->remote);
1666                 /* FIXME: extra output firewall step used to be here. --RR */
1667                 vif->dev->stats.tx_packets++;
1668                 vif->dev->stats.tx_bytes += skb->len;
1669         }
1670
1671         IPCB(skb)->flags |= IPSKB_FORWARDED;
1672
1673         /*
1674          * RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1675          * not only before forwarding, but after forwarding on all output
1676          * interfaces. It is clear, if mrouter runs a multicasting
1677          * program, it should receive packets not depending to what interface
1678          * program is joined.
1679          * If we will not make it, the program will have to join on all
1680          * interfaces. On the other hand, multihoming host (or router, but
1681          * not mrouter) cannot join to more than one interface - it will
1682          * result in receiving multiple packets.
1683          */
1684         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD, skb, skb->dev, dev,
1685                 ipmr_forward_finish);
1686         return;
1687
1688 out_free:
1689         kfree_skb(skb);
1690 }
1691
1692 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1693 {
1694         int ct;
1695
1696         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1697                 if (mrt->vif_table[ct].dev == dev)
1698                         break;
1699         }
1700         return ct;
1701 }
1702
1703 /* "local" means that we should preserve one skb (for local delivery) */
1704
1705 static int ip_mr_forward(struct net *net, struct mr_table *mrt,
1706                          struct sk_buff *skb, struct mfc_cache *cache,
1707                          int local)
1708 {
1709         int psend = -1;
1710         int vif, ct;
1711
1712         vif = cache->mfc_parent;
1713         cache->mfc_un.res.pkt++;
1714         cache->mfc_un.res.bytes += skb->len;
1715
1716         /*
1717          * Wrong interface: drop packet and (maybe) send PIM assert.
1718          */
1719         if (mrt->vif_table[vif].dev != skb->dev) {
1720                 int true_vifi;
1721
1722                 if (rt_is_output_route(skb_rtable(skb))) {
1723                         /* It is our own packet, looped back.
1724                          * Very complicated situation...
1725                          *
1726                          * The best workaround until routing daemons will be
1727                          * fixed is not to redistribute packet, if it was
1728                          * send through wrong interface. It means, that
1729                          * multicast applications WILL NOT work for
1730                          * (S,G), which have default multicast route pointing
1731                          * to wrong oif. In any case, it is not a good
1732                          * idea to use multicasting applications on router.
1733                          */
1734                         goto dont_forward;
1735                 }
1736
1737                 cache->mfc_un.res.wrong_if++;
1738                 true_vifi = ipmr_find_vif(mrt, skb->dev);
1739
1740                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1741                     /* pimsm uses asserts, when switching from RPT to SPT,
1742                      * so that we cannot check that packet arrived on an oif.
1743                      * It is bad, but otherwise we would need to move pretty
1744                      * large chunk of pimd to kernel. Ough... --ANK
1745                      */
1746                     (mrt->mroute_do_pim ||
1747                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1748                     time_after(jiffies,
1749                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1750                         cache->mfc_un.res.last_assert = jiffies;
1751                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1752                 }
1753                 goto dont_forward;
1754         }
1755
1756         mrt->vif_table[vif].pkt_in++;
1757         mrt->vif_table[vif].bytes_in += skb->len;
1758
1759         /*
1760          *      Forward the frame
1761          */
1762         for (ct = cache->mfc_un.res.maxvif - 1;
1763              ct >= cache->mfc_un.res.minvif; ct--) {
1764                 if (ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1765                         if (psend != -1) {
1766                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1767
1768                                 if (skb2)
1769                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1770                                                         psend);
1771                         }
1772                         psend = ct;
1773                 }
1774         }
1775         if (psend != -1) {
1776                 if (local) {
1777                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1778
1779                         if (skb2)
1780                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1781                 } else {
1782                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1783                         return 0;
1784                 }
1785         }
1786
1787 dont_forward:
1788         if (!local)
1789                 kfree_skb(skb);
1790         return 0;
1791 }
1792
1793 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1794 {
1795         struct rtable *rt = skb_rtable(skb);
1796         struct iphdr *iph = ip_hdr(skb);
1797         struct flowi4 fl4 = {
1798                 .daddr = iph->daddr,
1799                 .saddr = iph->saddr,
1800                 .flowi4_tos = RT_TOS(iph->tos),
1801                 .flowi4_oif = rt->rt_oif,
1802                 .flowi4_iif = rt->rt_iif,
1803                 .flowi4_mark = rt->rt_mark,
1804         };
1805         struct mr_table *mrt;
1806         int err;
1807
1808         err = ipmr_fib_lookup(net, &fl4, &mrt);
1809         if (err)
1810                 return ERR_PTR(err);
1811         return mrt;
1812 }
1813
1814 /*
1815  *      Multicast packets for forwarding arrive here
1816  *      Called with rcu_read_lock();
1817  */
1818
1819 int ip_mr_input(struct sk_buff *skb)
1820 {
1821         struct mfc_cache *cache;
1822         struct net *net = dev_net(skb->dev);
1823         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1824         struct mr_table *mrt;
1825
1826         /* Packet is looped back after forward, it should not be
1827          * forwarded second time, but still can be delivered locally.
1828          */
1829         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1830                 goto dont_forward;
1831
1832         mrt = ipmr_rt_fib_lookup(net, skb);
1833         if (IS_ERR(mrt)) {
1834                 kfree_skb(skb);
1835                 return PTR_ERR(mrt);
1836         }
1837         if (!local) {
1838                 if (IPCB(skb)->opt.router_alert) {
1839                         if (ip_call_ra_chain(skb))
1840                                 return 0;
1841                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1842                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1843                          * Cisco IOS <= 11.2(8)) do not put router alert
1844                          * option to IGMP packets destined to routable
1845                          * groups. It is very bad, because it means
1846                          * that we can forward NO IGMP messages.
1847                          */
1848                         struct sock *mroute_sk;
1849
1850                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1851                         if (mroute_sk) {
1852                                 nf_reset(skb);
1853                                 raw_rcv(mroute_sk, skb);
1854                                 return 0;
1855                         }
1856                     }
1857         }
1858
1859         /* already under rcu_read_lock() */
1860         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1861
1862         /*
1863          *      No usable cache entry
1864          */
1865         if (cache == NULL) {
1866                 int vif;
1867
1868                 if (local) {
1869                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1870                         ip_local_deliver(skb);
1871                         if (skb2 == NULL)
1872                                 return -ENOBUFS;
1873                         skb = skb2;
1874                 }
1875
1876                 read_lock(&mrt_lock);
1877                 vif = ipmr_find_vif(mrt, skb->dev);
1878                 if (vif >= 0) {
1879                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1880                         read_unlock(&mrt_lock);
1881
1882                         return err2;
1883                 }
1884                 read_unlock(&mrt_lock);
1885                 kfree_skb(skb);
1886                 return -ENODEV;
1887         }
1888
1889         read_lock(&mrt_lock);
1890         ip_mr_forward(net, mrt, skb, cache, local);
1891         read_unlock(&mrt_lock);
1892
1893         if (local)
1894                 return ip_local_deliver(skb);
1895
1896         return 0;
1897
1898 dont_forward:
1899         if (local)
1900                 return ip_local_deliver(skb);
1901         kfree_skb(skb);
1902         return 0;
1903 }
1904
1905 #ifdef CONFIG_IP_PIMSM
1906 /* called with rcu_read_lock() */
1907 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
1908                      unsigned int pimlen)
1909 {
1910         struct net_device *reg_dev = NULL;
1911         struct iphdr *encap;
1912
1913         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
1914         /*
1915          * Check that:
1916          * a. packet is really sent to a multicast group
1917          * b. packet is not a NULL-REGISTER
1918          * c. packet is not truncated
1919          */
1920         if (!ipv4_is_multicast(encap->daddr) ||
1921             encap->tot_len == 0 ||
1922             ntohs(encap->tot_len) + pimlen > skb->len)
1923                 return 1;
1924
1925         read_lock(&mrt_lock);
1926         if (mrt->mroute_reg_vif_num >= 0)
1927                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
1928         read_unlock(&mrt_lock);
1929
1930         if (reg_dev == NULL)
1931                 return 1;
1932
1933         skb->mac_header = skb->network_header;
1934         skb_pull(skb, (u8 *)encap - skb->data);
1935         skb_reset_network_header(skb);
1936         skb->protocol = htons(ETH_P_IP);
1937         skb->ip_summed = CHECKSUM_NONE;
1938         skb->pkt_type = PACKET_HOST;
1939
1940         skb_tunnel_rx(skb, reg_dev);
1941
1942         netif_rx(skb);
1943
1944         return NET_RX_SUCCESS;
1945 }
1946 #endif
1947
1948 #ifdef CONFIG_IP_PIMSM_V1
1949 /*
1950  * Handle IGMP messages of PIMv1
1951  */
1952
1953 int pim_rcv_v1(struct sk_buff *skb)
1954 {
1955         struct igmphdr *pim;
1956         struct net *net = dev_net(skb->dev);
1957         struct mr_table *mrt;
1958
1959         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1960                 goto drop;
1961
1962         pim = igmp_hdr(skb);
1963
1964         mrt = ipmr_rt_fib_lookup(net, skb);
1965         if (IS_ERR(mrt))
1966                 goto drop;
1967         if (!mrt->mroute_do_pim ||
1968             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
1969                 goto drop;
1970
1971         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
1972 drop:
1973                 kfree_skb(skb);
1974         }
1975         return 0;
1976 }
1977 #endif
1978
1979 #ifdef CONFIG_IP_PIMSM_V2
1980 static int pim_rcv(struct sk_buff *skb)
1981 {
1982         struct pimreghdr *pim;
1983         struct net *net = dev_net(skb->dev);
1984         struct mr_table *mrt;
1985
1986         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
1987                 goto drop;
1988
1989         pim = (struct pimreghdr *)skb_transport_header(skb);
1990         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
1991             (pim->flags & PIM_NULL_REGISTER) ||
1992             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
1993              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
1994                 goto drop;
1995
1996         mrt = ipmr_rt_fib_lookup(net, skb);
1997         if (IS_ERR(mrt))
1998                 goto drop;
1999         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2000 drop:
2001                 kfree_skb(skb);
2002         }
2003         return 0;
2004 }
2005 #endif
2006
2007 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2008                               struct mfc_cache *c, struct rtmsg *rtm)
2009 {
2010         int ct;
2011         struct rtnexthop *nhp;
2012         u8 *b = skb_tail_pointer(skb);
2013         struct rtattr *mp_head;
2014
2015         /* If cache is unresolved, don't try to parse IIF and OIF */
2016         if (c->mfc_parent >= MAXVIFS)
2017                 return -ENOENT;
2018
2019         if (VIF_EXISTS(mrt, c->mfc_parent))
2020                 RTA_PUT(skb, RTA_IIF, 4, &mrt->vif_table[c->mfc_parent].dev->ifindex);
2021
2022         mp_head = (struct rtattr *)skb_put(skb, RTA_LENGTH(0));
2023
2024         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2025                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2026                         if (skb_tailroom(skb) < RTA_ALIGN(RTA_ALIGN(sizeof(*nhp)) + 4))
2027                                 goto rtattr_failure;
2028                         nhp = (struct rtnexthop *)skb_put(skb, RTA_ALIGN(sizeof(*nhp)));
2029                         nhp->rtnh_flags = 0;
2030                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2031                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2032                         nhp->rtnh_len = sizeof(*nhp);
2033                 }
2034         }
2035         mp_head->rta_type = RTA_MULTIPATH;
2036         mp_head->rta_len = skb_tail_pointer(skb) - (u8 *)mp_head;
2037         rtm->rtm_type = RTN_MULTICAST;
2038         return 1;
2039
2040 rtattr_failure:
2041         nlmsg_trim(skb, b);
2042         return -EMSGSIZE;
2043 }
2044
2045 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2046                    __be32 saddr, __be32 daddr,
2047                    struct rtmsg *rtm, int nowait)
2048 {
2049         struct mfc_cache *cache;
2050         struct mr_table *mrt;
2051         int err;
2052
2053         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2054         if (mrt == NULL)
2055                 return -ENOENT;
2056
2057         rcu_read_lock();
2058         cache = ipmr_cache_find(mrt, saddr, daddr);
2059
2060         if (cache == NULL) {
2061                 struct sk_buff *skb2;
2062                 struct iphdr *iph;
2063                 struct net_device *dev;
2064                 int vif = -1;
2065
2066                 if (nowait) {
2067                         rcu_read_unlock();
2068                         return -EAGAIN;
2069                 }
2070
2071                 dev = skb->dev;
2072                 read_lock(&mrt_lock);
2073                 if (dev)
2074                         vif = ipmr_find_vif(mrt, dev);
2075                 if (vif < 0) {
2076                         read_unlock(&mrt_lock);
2077                         rcu_read_unlock();
2078                         return -ENODEV;
2079                 }
2080                 skb2 = skb_clone(skb, GFP_ATOMIC);
2081                 if (!skb2) {
2082                         read_unlock(&mrt_lock);
2083                         rcu_read_unlock();
2084                         return -ENOMEM;
2085                 }
2086
2087                 skb_push(skb2, sizeof(struct iphdr));
2088                 skb_reset_network_header(skb2);
2089                 iph = ip_hdr(skb2);
2090                 iph->ihl = sizeof(struct iphdr) >> 2;
2091                 iph->saddr = saddr;
2092                 iph->daddr = daddr;
2093                 iph->version = 0;
2094                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2095                 read_unlock(&mrt_lock);
2096                 rcu_read_unlock();
2097                 return err;
2098         }
2099
2100         read_lock(&mrt_lock);
2101         if (!nowait && (rtm->rtm_flags & RTM_F_NOTIFY))
2102                 cache->mfc_flags |= MFC_NOTIFY;
2103         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2104         read_unlock(&mrt_lock);
2105         rcu_read_unlock();
2106         return err;
2107 }
2108
2109 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2110                             u32 pid, u32 seq, struct mfc_cache *c)
2111 {
2112         struct nlmsghdr *nlh;
2113         struct rtmsg *rtm;
2114
2115         nlh = nlmsg_put(skb, pid, seq, RTM_NEWROUTE, sizeof(*rtm), NLM_F_MULTI);
2116         if (nlh == NULL)
2117                 return -EMSGSIZE;
2118
2119         rtm = nlmsg_data(nlh);
2120         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2121         rtm->rtm_dst_len  = 32;
2122         rtm->rtm_src_len  = 32;
2123         rtm->rtm_tos      = 0;
2124         rtm->rtm_table    = mrt->id;
2125         NLA_PUT_U32(skb, RTA_TABLE, mrt->id);
2126         rtm->rtm_type     = RTN_MULTICAST;
2127         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2128         rtm->rtm_protocol = RTPROT_UNSPEC;
2129         rtm->rtm_flags    = 0;
2130
2131         NLA_PUT_BE32(skb, RTA_SRC, c->mfc_origin);
2132         NLA_PUT_BE32(skb, RTA_DST, c->mfc_mcastgrp);
2133
2134         if (__ipmr_fill_mroute(mrt, skb, c, rtm) < 0)
2135                 goto nla_put_failure;
2136
2137         return nlmsg_end(skb, nlh);
2138
2139 nla_put_failure:
2140         nlmsg_cancel(skb, nlh);
2141         return -EMSGSIZE;
2142 }
2143
2144 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2145 {
2146         struct net *net = sock_net(skb->sk);
2147         struct mr_table *mrt;
2148         struct mfc_cache *mfc;
2149         unsigned int t = 0, s_t;
2150         unsigned int h = 0, s_h;
2151         unsigned int e = 0, s_e;
2152
2153         s_t = cb->args[0];
2154         s_h = cb->args[1];
2155         s_e = cb->args[2];
2156
2157         rcu_read_lock();
2158         ipmr_for_each_table(mrt, net) {
2159                 if (t < s_t)
2160                         goto next_table;
2161                 if (t > s_t)
2162                         s_h = 0;
2163                 for (h = s_h; h < MFC_LINES; h++) {
2164                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2165                                 if (e < s_e)
2166                                         goto next_entry;
2167                                 if (ipmr_fill_mroute(mrt, skb,
2168                                                      NETLINK_CB(cb->skb).pid,
2169                                                      cb->nlh->nlmsg_seq,
2170                                                      mfc) < 0)
2171                                         goto done;
2172 next_entry:
2173                                 e++;
2174                         }
2175                         e = s_e = 0;
2176                 }
2177                 s_h = 0;
2178 next_table:
2179                 t++;
2180         }
2181 done:
2182         rcu_read_unlock();
2183
2184         cb->args[2] = e;
2185         cb->args[1] = h;
2186         cb->args[0] = t;
2187
2188         return skb->len;
2189 }
2190
2191 #ifdef CONFIG_PROC_FS
2192 /*
2193  *      The /proc interfaces to multicast routing :
2194  *      /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2195  */
2196 struct ipmr_vif_iter {
2197         struct seq_net_private p;
2198         struct mr_table *mrt;
2199         int ct;
2200 };
2201
2202 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2203                                            struct ipmr_vif_iter *iter,
2204                                            loff_t pos)
2205 {
2206         struct mr_table *mrt = iter->mrt;
2207
2208         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2209                 if (!VIF_EXISTS(mrt, iter->ct))
2210                         continue;
2211                 if (pos-- == 0)
2212                         return &mrt->vif_table[iter->ct];
2213         }
2214         return NULL;
2215 }
2216
2217 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2218         __acquires(mrt_lock)
2219 {
2220         struct ipmr_vif_iter *iter = seq->private;
2221         struct net *net = seq_file_net(seq);
2222         struct mr_table *mrt;
2223
2224         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2225         if (mrt == NULL)
2226                 return ERR_PTR(-ENOENT);
2227
2228         iter->mrt = mrt;
2229
2230         read_lock(&mrt_lock);
2231         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2232                 : SEQ_START_TOKEN;
2233 }
2234
2235 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2236 {
2237         struct ipmr_vif_iter *iter = seq->private;
2238         struct net *net = seq_file_net(seq);
2239         struct mr_table *mrt = iter->mrt;
2240
2241         ++*pos;
2242         if (v == SEQ_START_TOKEN)
2243                 return ipmr_vif_seq_idx(net, iter, 0);
2244
2245         while (++iter->ct < mrt->maxvif) {
2246                 if (!VIF_EXISTS(mrt, iter->ct))
2247                         continue;
2248                 return &mrt->vif_table[iter->ct];
2249         }
2250         return NULL;
2251 }
2252
2253 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2254         __releases(mrt_lock)
2255 {
2256         read_unlock(&mrt_lock);
2257 }
2258
2259 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2260 {
2261         struct ipmr_vif_iter *iter = seq->private;
2262         struct mr_table *mrt = iter->mrt;
2263
2264         if (v == SEQ_START_TOKEN) {
2265                 seq_puts(seq,
2266                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2267         } else {
2268                 const struct vif_device *vif = v;
2269                 const char *name =  vif->dev ? vif->dev->name : "none";
2270
2271                 seq_printf(seq,
2272                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2273                            vif - mrt->vif_table,
2274                            name, vif->bytes_in, vif->pkt_in,
2275                            vif->bytes_out, vif->pkt_out,
2276                            vif->flags, vif->local, vif->remote);
2277         }
2278         return 0;
2279 }
2280
2281 static const struct seq_operations ipmr_vif_seq_ops = {
2282         .start = ipmr_vif_seq_start,
2283         .next  = ipmr_vif_seq_next,
2284         .stop  = ipmr_vif_seq_stop,
2285         .show  = ipmr_vif_seq_show,
2286 };
2287
2288 static int ipmr_vif_open(struct inode *inode, struct file *file)
2289 {
2290         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2291                             sizeof(struct ipmr_vif_iter));
2292 }
2293
2294 static const struct file_operations ipmr_vif_fops = {
2295         .owner   = THIS_MODULE,
2296         .open    = ipmr_vif_open,
2297         .read    = seq_read,
2298         .llseek  = seq_lseek,
2299         .release = seq_release_net,
2300 };
2301
2302 struct ipmr_mfc_iter {
2303         struct seq_net_private p;
2304         struct mr_table *mrt;
2305         struct list_head *cache;
2306         int ct;
2307 };
2308
2309
2310 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2311                                           struct ipmr_mfc_iter *it, loff_t pos)
2312 {
2313         struct mr_table *mrt = it->mrt;
2314         struct mfc_cache *mfc;
2315
2316         rcu_read_lock();
2317         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2318                 it->cache = &mrt->mfc_cache_array[it->ct];
2319                 list_for_each_entry_rcu(mfc, it->cache, list)
2320                         if (pos-- == 0)
2321                                 return mfc;
2322         }
2323         rcu_read_unlock();
2324
2325         spin_lock_bh(&mfc_unres_lock);
2326         it->cache = &mrt->mfc_unres_queue;
2327         list_for_each_entry(mfc, it->cache, list)
2328                 if (pos-- == 0)
2329                         return mfc;
2330         spin_unlock_bh(&mfc_unres_lock);
2331
2332         it->cache = NULL;
2333         return NULL;
2334 }
2335
2336
2337 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2338 {
2339         struct ipmr_mfc_iter *it = seq->private;
2340         struct net *net = seq_file_net(seq);
2341         struct mr_table *mrt;
2342
2343         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2344         if (mrt == NULL)
2345                 return ERR_PTR(-ENOENT);
2346
2347         it->mrt = mrt;
2348         it->cache = NULL;
2349         it->ct = 0;
2350         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2351                 : SEQ_START_TOKEN;
2352 }
2353
2354 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2355 {
2356         struct mfc_cache *mfc = v;
2357         struct ipmr_mfc_iter *it = seq->private;
2358         struct net *net = seq_file_net(seq);
2359         struct mr_table *mrt = it->mrt;
2360
2361         ++*pos;
2362
2363         if (v == SEQ_START_TOKEN)
2364                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2365
2366         if (mfc->list.next != it->cache)
2367                 return list_entry(mfc->list.next, struct mfc_cache, list);
2368
2369         if (it->cache == &mrt->mfc_unres_queue)
2370                 goto end_of_list;
2371
2372         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2373
2374         while (++it->ct < MFC_LINES) {
2375                 it->cache = &mrt->mfc_cache_array[it->ct];
2376                 if (list_empty(it->cache))
2377                         continue;
2378                 return list_first_entry(it->cache, struct mfc_cache, list);
2379         }
2380
2381         /* exhausted cache_array, show unresolved */
2382         rcu_read_unlock();
2383         it->cache = &mrt->mfc_unres_queue;
2384         it->ct = 0;
2385
2386         spin_lock_bh(&mfc_unres_lock);
2387         if (!list_empty(it->cache))
2388                 return list_first_entry(it->cache, struct mfc_cache, list);
2389
2390 end_of_list:
2391         spin_unlock_bh(&mfc_unres_lock);
2392         it->cache = NULL;
2393
2394         return NULL;
2395 }
2396
2397 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2398 {
2399         struct ipmr_mfc_iter *it = seq->private;
2400         struct mr_table *mrt = it->mrt;
2401
2402         if (it->cache == &mrt->mfc_unres_queue)
2403                 spin_unlock_bh(&mfc_unres_lock);
2404         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2405                 rcu_read_unlock();
2406 }
2407
2408 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2409 {
2410         int n;
2411
2412         if (v == SEQ_START_TOKEN) {
2413                 seq_puts(seq,
2414                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2415         } else {
2416                 const struct mfc_cache *mfc = v;
2417                 const struct ipmr_mfc_iter *it = seq->private;
2418                 const struct mr_table *mrt = it->mrt;
2419
2420                 seq_printf(seq, "%08X %08X %-3hd",
2421                            (__force u32) mfc->mfc_mcastgrp,
2422                            (__force u32) mfc->mfc_origin,
2423                            mfc->mfc_parent);
2424
2425                 if (it->cache != &mrt->mfc_unres_queue) {
2426                         seq_printf(seq, " %8lu %8lu %8lu",
2427                                    mfc->mfc_un.res.pkt,
2428                                    mfc->mfc_un.res.bytes,
2429                                    mfc->mfc_un.res.wrong_if);
2430                         for (n = mfc->mfc_un.res.minvif;
2431                              n < mfc->mfc_un.res.maxvif; n++) {
2432                                 if (VIF_EXISTS(mrt, n) &&
2433                                     mfc->mfc_un.res.ttls[n] < 255)
2434                                         seq_printf(seq,
2435                                            " %2d:%-3d",
2436                                            n, mfc->mfc_un.res.ttls[n]);
2437                         }
2438                 } else {
2439                         /* unresolved mfc_caches don't contain
2440                          * pkt, bytes and wrong_if values
2441                          */
2442                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2443                 }
2444                 seq_putc(seq, '\n');
2445         }
2446         return 0;
2447 }
2448
2449 static const struct seq_operations ipmr_mfc_seq_ops = {
2450         .start = ipmr_mfc_seq_start,
2451         .next  = ipmr_mfc_seq_next,
2452         .stop  = ipmr_mfc_seq_stop,
2453         .show  = ipmr_mfc_seq_show,
2454 };
2455
2456 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2457 {
2458         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2459                             sizeof(struct ipmr_mfc_iter));
2460 }
2461
2462 static const struct file_operations ipmr_mfc_fops = {
2463         .owner   = THIS_MODULE,
2464         .open    = ipmr_mfc_open,
2465         .read    = seq_read,
2466         .llseek  = seq_lseek,
2467         .release = seq_release_net,
2468 };
2469 #endif
2470
2471 #ifdef CONFIG_IP_PIMSM_V2
2472 static const struct net_protocol pim_protocol = {
2473         .handler        =       pim_rcv,
2474         .netns_ok       =       1,
2475 };
2476 #endif
2477
2478
2479 /*
2480  *      Setup for IP multicast routing
2481  */
2482 static int __net_init ipmr_net_init(struct net *net)
2483 {
2484         int err;
2485
2486         err = ipmr_rules_init(net);
2487         if (err < 0)
2488                 goto fail;
2489
2490 #ifdef CONFIG_PROC_FS
2491         err = -ENOMEM;
2492         if (!proc_net_fops_create(net, "ip_mr_vif", 0, &ipmr_vif_fops))
2493                 goto proc_vif_fail;
2494         if (!proc_net_fops_create(net, "ip_mr_cache", 0, &ipmr_mfc_fops))
2495                 goto proc_cache_fail;
2496 #endif
2497         return 0;
2498
2499 #ifdef CONFIG_PROC_FS
2500 proc_cache_fail:
2501         proc_net_remove(net, "ip_mr_vif");
2502 proc_vif_fail:
2503         ipmr_rules_exit(net);
2504 #endif
2505 fail:
2506         return err;
2507 }
2508
2509 static void __net_exit ipmr_net_exit(struct net *net)
2510 {
2511 #ifdef CONFIG_PROC_FS
2512         proc_net_remove(net, "ip_mr_cache");
2513         proc_net_remove(net, "ip_mr_vif");
2514 #endif
2515         ipmr_rules_exit(net);
2516 }
2517
2518 static struct pernet_operations ipmr_net_ops = {
2519         .init = ipmr_net_init,
2520         .exit = ipmr_net_exit,
2521 };
2522
2523 int __init ip_mr_init(void)
2524 {
2525         int err;
2526
2527         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2528                                        sizeof(struct mfc_cache),
2529                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2530                                        NULL);
2531         if (!mrt_cachep)
2532                 return -ENOMEM;
2533
2534         err = register_pernet_subsys(&ipmr_net_ops);
2535         if (err)
2536                 goto reg_pernet_fail;
2537
2538         err = register_netdevice_notifier(&ip_mr_notifier);
2539         if (err)
2540                 goto reg_notif_fail;
2541 #ifdef CONFIG_IP_PIMSM_V2
2542         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2543                 printk(KERN_ERR "ip_mr_init: can't add PIM protocol\n");
2544                 err = -EAGAIN;
2545                 goto add_proto_fail;
2546         }
2547 #endif
2548         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2549                       NULL, ipmr_rtm_dumproute, NULL);
2550         return 0;
2551
2552 #ifdef CONFIG_IP_PIMSM_V2
2553 add_proto_fail:
2554         unregister_netdevice_notifier(&ip_mr_notifier);
2555 #endif
2556 reg_notif_fail:
2557         unregister_pernet_subsys(&ipmr_net_ops);
2558 reg_pernet_fail:
2559         kmem_cache_destroy(mrt_cachep);
2560         return err;
2561 }