net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <asm/uaccess.h>
  30 #include <linux/types.h>
  31 #include <linux/capability.h>
  32 #include <linux/errno.h>
  33 #include <linux/timer.h>
  34 #include <linux/mm.h>
  35 #include <linux/kernel.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/stat.h>
  38 #include <linux/socket.h>
  39 #include <linux/in.h>
  40 #include <linux/inet.h>
  41 #include <linux/netdevice.h>
  42 #include <linux/inetdevice.h>
  43 #include <linux/igmp.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/mroute.h>
  47 #include <linux/init.h>
  48 #include <linux/if_ether.h>
  49 #include <linux/slab.h>
  50 #include <net/net_namespace.h>
  51 #include <net/ip.h>
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54 #include <net/route.h>
  55 #include <net/sock.h>
  56 #include <net/icmp.h>
  57 #include <net/udp.h>
  58 #include <net/raw.h>
  59 #include <linux/notifier.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/netfilter_ipv4.h>
  62 #include <linux/compat.h>
  63 #include <linux/export.h>
  64 #include <net/ip_tunnels.h>
  65 #include <net/checksum.h>
  66 #include <net/netlink.h>
  67 #include <net/fib_rules.h>
  68 #include <linux/netconf.h>
  69
  70 struct ipmr_rule {
  71         struct fib_rule         common;
  72 };
  73
  74 struct ipmr_result {
  75         struct mr_table         *mrt;
  76 };
  77
  78 static inline bool pimsm_enabled(void)
  79 {
  80         return IS_BUILTIN(CONFIG_IP_PIMSM_V1) || IS_BUILTIN(CONFIG_IP_PIMSM_V2);
  81 }
  82
  83 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  84  * Note that the changes are semaphored via rtnl_lock.
  85  */
  86
  87 static DEFINE_RWLOCK(mrt_lock);
  88
  89 /* Multicast router control variables */
  90
  91 /* Special spinlock for queue of unresolved entries */
  92 static DEFINE_SPINLOCK(mfc_unres_lock);
  93
  94 /* We return to original Alan's scheme. Hash table of resolved
  95  * entries is changed only in process context and protected
  96  * with weak lock mrt_lock. Queue of unresolved entries is protected
  97  * with strong spinlock mfc_unres_lock.
  98  *
  99  * In this case data path is free of exclusive locks at all.
 100  */
 101
 102 static struct kmem_cache *mrt_cachep __read_mostly;
 103
 104 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 105 static void ipmr_free_table(struct mr_table *mrt);
 106
 107 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 108                           struct sk_buff *skb, struct mfc_cache *cache,
 109                           int local);
 110 static int ipmr_cache_report(struct mr_table *mrt,
 111                              struct sk_buff *pkt, vifi_t vifi, int assert);
 112 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 113                               struct mfc_cache *c, struct rtmsg *rtm);
 114 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 115                                  int cmd);
 116 static void mroute_clean_tables(struct mr_table *mrt);
 117 static void ipmr_expire_process(unsigned long arg);
 118
 119 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 120 #define ipmr_for_each_table(mrt, net) \
 121         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 122
 123 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 124 {
 125         struct mr_table *mrt;
 126
 127         ipmr_for_each_table(mrt, net) {
 128                 if (mrt->id == id)
 129                         return mrt;
 130         }
 131         return NULL;
 132 }
 133
 134 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 135                            struct mr_table **mrt)
 136 {
 137         int err;
 138         struct ipmr_result res;
 139         struct fib_lookup_arg arg = {
 140                 .result = &res,
 141                 .flags = FIB_LOOKUP_NOREF,
 142         };
 143
 144         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 145                                flowi4_to_flowi(flp4), 0, &arg);
 146         if (err < 0)
 147                 return err;
 148         *mrt = res.mrt;
 149         return 0;
 150 }
 151
 152 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 153                             int flags, struct fib_lookup_arg *arg)
 154 {
 155         struct ipmr_result *res = arg->result;
 156         struct mr_table *mrt;
 157
 158         switch (rule->action) {
 159         case FR_ACT_TO_TBL:
 160                 break;
 161         case FR_ACT_UNREACHABLE:
 162                 return -ENETUNREACH;
 163         case FR_ACT_PROHIBIT:
 164                 return -EACCES;
 165         case FR_ACT_BLACKHOLE:
 166         default:
 167                 return -EINVAL;
 168         }
 169
 170         mrt = ipmr_get_table(rule->fr_net, rule->table);
 171         if (!mrt)
 172                 return -EAGAIN;
 173         res->mrt = mrt;
 174         return 0;
 175 }
 176
 177 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 178 {
 179         return 1;
 180 }
 181
 182 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 183         FRA_GENERIC_POLICY,
 184 };
 185
 186 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 187                                struct fib_rule_hdr *frh, struct nlattr **tb)
 188 {
 189         return 0;
 190 }
 191
 192 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 193                              struct nlattr **tb)
 194 {
 195         return 1;
 196 }
 197
 198 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 199                           struct fib_rule_hdr *frh)
 200 {
 201         frh->dst_len = 0;
 202         frh->src_len = 0;
 203         frh->tos     = 0;
 204         return 0;
 205 }
 206
 207 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 208         .family         = RTNL_FAMILY_IPMR,
 209         .rule_size      = sizeof(struct ipmr_rule),
 210         .addr_size      = sizeof(u32),
 211         .action         = ipmr_rule_action,
 212         .match          = ipmr_rule_match,
 213         .configure      = ipmr_rule_configure,
 214         .compare        = ipmr_rule_compare,
 215         .fill           = ipmr_rule_fill,
 216         .nlgroup        = RTNLGRP_IPV4_RULE,
 217         .policy         = ipmr_rule_policy,
 218         .owner          = THIS_MODULE,
 219 };
 220
 221 static int __net_init ipmr_rules_init(struct net *net)
 222 {
 223         struct fib_rules_ops *ops;
 224         struct mr_table *mrt;
 225         int err;
 226
 227         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 228         if (IS_ERR(ops))
 229                 return PTR_ERR(ops);
 230
 231         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 232
 233         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 234         if (IS_ERR(mrt)) {
 235                 err = PTR_ERR(mrt);
 236                 goto err1;
 237         }
 238
 239         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 240         if (err < 0)
 241                 goto err2;
 242
 243         net->ipv4.mr_rules_ops = ops;
 244         return 0;
 245
 246 err2:
 247         ipmr_free_table(mrt);
 248 err1:
 249         fib_rules_unregister(ops);
 250         return err;
 251 }
 252
 253 static void __net_exit ipmr_rules_exit(struct net *net)
 254 {
 255         struct mr_table *mrt, *next;
 256
 257         rtnl_lock();
 258         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 259                 list_del(&mrt->list);
 260                 ipmr_free_table(mrt);
 261         }
 262         fib_rules_unregister(net->ipv4.mr_rules_ops);
 263         rtnl_unlock();
 264 }
 265 #else
 266 #define ipmr_for_each_table(mrt, net) \
 267         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 268
 269 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 270 {
 271         return net->ipv4.mrt;
 272 }
 273
 274 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 275                            struct mr_table **mrt)
 276 {
 277         *mrt = net->ipv4.mrt;
 278         return 0;
 279 }
 280
 281 static int __net_init ipmr_rules_init(struct net *net)
 282 {
 283         struct mr_table *mrt;
 284
 285         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 286         if (IS_ERR(mrt))
 287                 return PTR_ERR(mrt);
 288         net->ipv4.mrt = mrt;
 289         return 0;
 290 }
 291
 292 static void __net_exit ipmr_rules_exit(struct net *net)
 293 {
 294         rtnl_lock();
 295         ipmr_free_table(net->ipv4.mrt);
 296         net->ipv4.mrt = NULL;
 297         rtnl_unlock();
 298 }
 299 #endif
 300
 301 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 302 {
 303         struct mr_table *mrt;
 304         unsigned int i;
 305
 306         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 307         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 308                 return ERR_PTR(-EINVAL);
 309
 310         mrt = ipmr_get_table(net, id);
 311         if (mrt)
 312                 return mrt;
 313
 314         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 315         if (!mrt)
 316                 return ERR_PTR(-ENOMEM);
 317         write_pnet(&mrt->net, net);
 318         mrt->id = id;
 319
 320         /* Forwarding cache */
 321         for (i = 0; i < MFC_LINES; i++)
 322                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 323
 324         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 325
 326         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 327                     (unsigned long)mrt);
 328
 329         mrt->mroute_reg_vif_num = -1;
 330 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 331         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 332 #endif
 333         return mrt;
 334 }
 335
 336 static void ipmr_free_table(struct mr_table *mrt)
 337 {
 338         del_timer_sync(&mrt->ipmr_expire_timer);
 339         mroute_clean_tables(mrt);
 340         kfree(mrt);
 341 }
 342
 343 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 344
 345 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 346 {
 347         struct net *net = dev_net(dev);
 348
 349         dev_close(dev);
 350
 351         dev = __dev_get_by_name(net, "tunl0");
 352         if (dev) {
 353                 const struct net_device_ops *ops = dev->netdev_ops;
 354                 struct ifreq ifr;
 355                 struct ip_tunnel_parm p;
 356
 357                 memset(&p, 0, sizeof(p));
 358                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 359                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 360                 p.iph.version = 4;
 361                 p.iph.ihl = 5;
 362                 p.iph.protocol = IPPROTO_IPIP;
 363                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 364                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 365
 366                 if (ops->ndo_do_ioctl) {
 367                         mm_segment_t oldfs = get_fs();
 368
 369                         set_fs(KERNEL_DS);
 370                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 371                         set_fs(oldfs);
 372                 }
 373         }
 374 }
 375
 376 /* Initialize ipmr pimreg/tunnel in_device */
 377 static bool ipmr_init_vif_indev(const struct net_device *dev)
 378 {
 379         struct in_device *in_dev;
 380
 381         ASSERT_RTNL();
 382
 383         in_dev = __in_dev_get_rtnl(dev);
 384         if (!in_dev)
 385                 return false;
 386         ipv4_devconf_setall(in_dev);
 387         neigh_parms_data_state_setall(in_dev->arp_parms);
 388         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 389
 390         return true;
 391 }
 392
 393 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 394 {
 395         struct net_device  *dev;
 396
 397         dev = __dev_get_by_name(net, "tunl0");
 398
 399         if (dev) {
 400                 const struct net_device_ops *ops = dev->netdev_ops;
 401                 int err;
 402                 struct ifreq ifr;
 403                 struct ip_tunnel_parm p;
 404
 405                 memset(&p, 0, sizeof(p));
 406                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 407                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 408                 p.iph.version = 4;
 409                 p.iph.ihl = 5;
 410                 p.iph.protocol = IPPROTO_IPIP;
 411                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 412                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 413
 414                 if (ops->ndo_do_ioctl) {
 415                         mm_segment_t oldfs = get_fs();
 416
 417                         set_fs(KERNEL_DS);
 418                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 419                         set_fs(oldfs);
 420                 } else {
 421                         err = -EOPNOTSUPP;
 422                 }
 423                 dev = NULL;
 424
 425                 if (err == 0 &&
 426                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 427                         dev->flags |= IFF_MULTICAST;
 428                         if (!ipmr_init_vif_indev(dev))
 429                                 goto failure;
 430                         if (dev_open(dev))
 431                                 goto failure;
 432                         dev_hold(dev);
 433                 }
 434         }
 435         return dev;
 436
 437 failure:
 438         /* allow the register to be completed before unregistering. */
 439         rtnl_unlock();
 440         rtnl_lock();
 441
 442         unregister_netdevice(dev);
 443         return NULL;
 444 }
 445
 446 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 447 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 448 {
 449         struct net *net = dev_net(dev);
 450         struct mr_table *mrt;
 451         struct flowi4 fl4 = {
 452                 .flowi4_oif     = dev->ifindex,
 453                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 454                 .flowi4_mark    = skb->mark,
 455         };
 456         int err;
 457
 458         err = ipmr_fib_lookup(net, &fl4, &mrt);
 459         if (err < 0) {
 460                 kfree_skb(skb);
 461                 return err;
 462         }
 463
 464         read_lock(&mrt_lock);
 465         dev->stats.tx_bytes += skb->len;
 466         dev->stats.tx_packets++;
 467         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 468         read_unlock(&mrt_lock);
 469         kfree_skb(skb);
 470         return NETDEV_TX_OK;
 471 }
 472
 473 static int reg_vif_get_iflink(const struct net_device *dev)
 474 {
 475         return 0;
 476 }
 477
 478 static const struct net_device_ops reg_vif_netdev_ops = {
 479         .ndo_start_xmit = reg_vif_xmit,
 480         .ndo_get_iflink = reg_vif_get_iflink,
 481 };
 482
 483 static void reg_vif_setup(struct net_device *dev)
 484 {
 485         dev->type               = ARPHRD_PIMREG;
 486         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 487         dev->flags              = IFF_NOARP;
 488         dev->netdev_ops         = &reg_vif_netdev_ops;
 489         dev->destructor         = free_netdev;
 490         dev->features           |= NETIF_F_NETNS_LOCAL;
 491 }
 492
 493 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 494 {
 495         struct net_device *dev;
 496         char name[IFNAMSIZ];
 497
 498         if (mrt->id == RT_TABLE_DEFAULT)
 499                 sprintf(name, "pimreg");
 500         else
 501                 sprintf(name, "pimreg%u", mrt->id);
 502
 503         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 504
 505         if (!dev)
 506                 return NULL;
 507
 508         dev_net_set(dev, net);
 509
 510         if (register_netdevice(dev)) {
 511                 free_netdev(dev);
 512                 return NULL;
 513         }
 514
 515         if (!ipmr_init_vif_indev(dev))
 516                 goto failure;
 517         if (dev_open(dev))
 518                 goto failure;
 519
 520         dev_hold(dev);
 521
 522         return dev;
 523
 524 failure:
 525         /* allow the register to be completed before unregistering. */
 526         rtnl_unlock();
 527         rtnl_lock();
 528
 529         unregister_netdevice(dev);
 530         return NULL;
 531 }
 532
 533 /* called with rcu_read_lock() */
 534 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 535                      unsigned int pimlen)
 536 {
 537         struct net_device *reg_dev = NULL;
 538         struct iphdr *encap;
 539
 540         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 541         /* Check that:
 542          * a. packet is really sent to a multicast group
 543          * b. packet is not a NULL-REGISTER
 544          * c. packet is not truncated
 545          */
 546         if (!ipv4_is_multicast(encap->daddr) ||
 547             encap->tot_len == 0 ||
 548             ntohs(encap->tot_len) + pimlen > skb->len)
 549                 return 1;
 550
 551         read_lock(&mrt_lock);
 552         if (mrt->mroute_reg_vif_num >= 0)
 553                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 554         read_unlock(&mrt_lock);
 555
 556         if (!reg_dev)
 557                 return 1;
 558
 559         skb->mac_header = skb->network_header;
 560         skb_pull(skb, (u8 *)encap - skb->data);
 561         skb_reset_network_header(skb);
 562         skb->protocol = htons(ETH_P_IP);
 563         skb->ip_summed = CHECKSUM_NONE;
 564
 565         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 566
 567         netif_rx(skb);
 568
 569         return NET_RX_SUCCESS;
 570 }
 571 #else
 572 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 573 {
 574         return NULL;
 575 }
 576 #endif
 577
 578 /**
 579  *      vif_delete - Delete a VIF entry
 580  *      @notify: Set to 1, if the caller is a notifier_call
 581  */
 582 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 583                       struct list_head *head)
 584 {
 585         struct vif_device *v;
 586         struct net_device *dev;
 587         struct in_device *in_dev;
 588
 589         if (vifi < 0 || vifi >= mrt->maxvif)
 590                 return -EADDRNOTAVAIL;
 591
 592         v = &mrt->vif_table[vifi];
 593
 594         write_lock_bh(&mrt_lock);
 595         dev = v->dev;
 596         v->dev = NULL;
 597
 598         if (!dev) {
 599                 write_unlock_bh(&mrt_lock);
 600                 return -EADDRNOTAVAIL;
 601         }
 602
 603         if (vifi == mrt->mroute_reg_vif_num)
 604                 mrt->mroute_reg_vif_num = -1;
 605
 606         if (vifi + 1 == mrt->maxvif) {
 607                 int tmp;
 608
 609                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 610                         if (VIF_EXISTS(mrt, tmp))
 611                                 break;
 612                 }
 613                 mrt->maxvif = tmp+1;
 614         }
 615
 616         write_unlock_bh(&mrt_lock);
 617
 618         dev_set_allmulti(dev, -1);
 619
 620         in_dev = __in_dev_get_rtnl(dev);
 621         if (in_dev) {
 622                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 623                 inet_netconf_notify_devconf(dev_net(dev),
 624                                             NETCONFA_MC_FORWARDING,
 625                                             dev->ifindex, &in_dev->cnf);
 626                 ip_rt_multicast_event(in_dev);
 627         }
 628
 629         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 630                 unregister_netdevice_queue(dev, head);
 631
 632         dev_put(dev);
 633         return 0;
 634 }
 635
 636 static void ipmr_cache_free_rcu(struct rcu_head *head)
 637 {
 638         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 639
 640         kmem_cache_free(mrt_cachep, c);
 641 }
 642
 643 static inline void ipmr_cache_free(struct mfc_cache *c)
 644 {
 645         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 646 }
 647
 648 /* Destroy an unresolved cache entry, killing queued skbs
 649  * and reporting error to netlink readers.
 650  */
 651 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 652 {
 653         struct net *net = read_pnet(&mrt->net);
 654         struct sk_buff *skb;
 655         struct nlmsgerr *e;
 656
 657         atomic_dec(&mrt->cache_resolve_queue_len);
 658
 659         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 660                 if (ip_hdr(skb)->version == 0) {
 661                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 662                         nlh->nlmsg_type = NLMSG_ERROR;
 663                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 664                         skb_trim(skb, nlh->nlmsg_len);
 665                         e = nlmsg_data(nlh);
 666                         e->error = -ETIMEDOUT;
 667                         memset(&e->msg, 0, sizeof(e->msg));
 668
 669                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 670                 } else {
 671                         kfree_skb(skb);
 672                 }
 673         }
 674
 675         ipmr_cache_free(c);
 676 }
 677
 678 /* Timer process for the unresolved queue. */
 679 static void ipmr_expire_process(unsigned long arg)
 680 {
 681         struct mr_table *mrt = (struct mr_table *)arg;
 682         unsigned long now;
 683         unsigned long expires;
 684         struct mfc_cache *c, *next;
 685
 686         if (!spin_trylock(&mfc_unres_lock)) {
 687                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 688                 return;
 689         }
 690
 691         if (list_empty(&mrt->mfc_unres_queue))
 692                 goto out;
 693
 694         now = jiffies;
 695         expires = 10*HZ;
 696
 697         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 698                 if (time_after(c->mfc_un.unres.expires, now)) {
 699                         unsigned long interval = c->mfc_un.unres.expires - now;
 700                         if (interval < expires)
 701                                 expires = interval;
 702                         continue;
 703                 }
 704
 705                 list_del(&c->list);
 706                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 707                 ipmr_destroy_unres(mrt, c);
 708         }
 709
 710         if (!list_empty(&mrt->mfc_unres_queue))
 711                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 712
 713 out:
 714         spin_unlock(&mfc_unres_lock);
 715 }
 716
 717 /* Fill oifs list. It is called under write locked mrt_lock. */
 718 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 719                                    unsigned char *ttls)
 720 {
 721         int vifi;
 722
 723         cache->mfc_un.res.minvif = MAXVIFS;
 724         cache->mfc_un.res.maxvif = 0;
 725         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 726
 727         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 728                 if (VIF_EXISTS(mrt, vifi) &&
 729                     ttls[vifi] && ttls[vifi] < 255) {
 730                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 731                         if (cache->mfc_un.res.minvif > vifi)
 732                                 cache->mfc_un.res.minvif = vifi;
 733                         if (cache->mfc_un.res.maxvif <= vifi)
 734                                 cache->mfc_un.res.maxvif = vifi + 1;
 735                 }
 736         }
 737 }
 738
 739 static int vif_add(struct net *net, struct mr_table *mrt,
 740                    struct vifctl *vifc, int mrtsock)
 741 {
 742         int vifi = vifc->vifc_vifi;
 743         struct vif_device *v = &mrt->vif_table[vifi];
 744         struct net_device *dev;
 745         struct in_device *in_dev;
 746         int err;
 747
 748         /* Is vif busy ? */
 749         if (VIF_EXISTS(mrt, vifi))
 750                 return -EADDRINUSE;
 751
 752         switch (vifc->vifc_flags) {
 753         case VIFF_REGISTER:
 754                 if (!pimsm_enabled())
 755                         return -EINVAL;
 756                 /* Special Purpose VIF in PIM
 757                  * All the packets will be sent to the daemon
 758                  */
 759                 if (mrt->mroute_reg_vif_num >= 0)
 760                         return -EADDRINUSE;
 761                 dev = ipmr_reg_vif(net, mrt);
 762                 if (!dev)
 763                         return -ENOBUFS;
 764                 err = dev_set_allmulti(dev, 1);
 765                 if (err) {
 766                         unregister_netdevice(dev);
 767                         dev_put(dev);
 768                         return err;
 769                 }
 770                 break;
 771         case VIFF_TUNNEL:
 772                 dev = ipmr_new_tunnel(net, vifc);
 773                 if (!dev)
 774                         return -ENOBUFS;
 775                 err = dev_set_allmulti(dev, 1);
 776                 if (err) {
 777                         ipmr_del_tunnel(dev, vifc);
 778                         dev_put(dev);
 779                         return err;
 780                 }
 781                 break;
 782         case VIFF_USE_IFINDEX:
 783         case 0:
 784                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 785                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 786                         if (dev && !__in_dev_get_rtnl(dev)) {
 787                                 dev_put(dev);
 788                                 return -EADDRNOTAVAIL;
 789                         }
 790                 } else {
 791                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 792                 }
 793                 if (!dev)
 794                         return -EADDRNOTAVAIL;
 795                 err = dev_set_allmulti(dev, 1);
 796                 if (err) {
 797                         dev_put(dev);
 798                         return err;
 799                 }
 800                 break;
 801         default:
 802                 return -EINVAL;
 803         }
 804
 805         in_dev = __in_dev_get_rtnl(dev);
 806         if (!in_dev) {
 807                 dev_put(dev);
 808                 return -EADDRNOTAVAIL;
 809         }
 810         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 811         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
 812                                     &in_dev->cnf);
 813         ip_rt_multicast_event(in_dev);
 814
 815         /* Fill in the VIF structures */
 816
 817         v->rate_limit = vifc->vifc_rate_limit;
 818         v->local = vifc->vifc_lcl_addr.s_addr;
 819         v->remote = vifc->vifc_rmt_addr.s_addr;
 820         v->flags = vifc->vifc_flags;
 821         if (!mrtsock)
 822                 v->flags |= VIFF_STATIC;
 823         v->threshold = vifc->vifc_threshold;
 824         v->bytes_in = 0;
 825         v->bytes_out = 0;
 826         v->pkt_in = 0;
 827         v->pkt_out = 0;
 828         v->link = dev->ifindex;
 829         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 830                 v->link = dev_get_iflink(dev);
 831
 832         /* And finish update writing critical data */
 833         write_lock_bh(&mrt_lock);
 834         v->dev = dev;
 835         if (v->flags & VIFF_REGISTER)
 836                 mrt->mroute_reg_vif_num = vifi;
 837         if (vifi+1 > mrt->maxvif)
 838                 mrt->maxvif = vifi+1;
 839         write_unlock_bh(&mrt_lock);
 840         return 0;
 841 }
 842
 843 /* called with rcu_read_lock() */
 844 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 845                                          __be32 origin,
 846                                          __be32 mcastgrp)
 847 {
 848         int line = MFC_HASH(mcastgrp, origin);
 849         struct mfc_cache *c;
 850
 851         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 852                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 853                         return c;
 854         }
 855         return NULL;
 856 }
 857
 858 /* Look for a (*,*,oif) entry */
 859 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 860                                                     int vifi)
 861 {
 862         int line = MFC_HASH(htonl(INADDR_ANY), htonl(INADDR_ANY));
 863         struct mfc_cache *c;
 864
 865         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 866                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 867                     c->mfc_mcastgrp == htonl(INADDR_ANY) &&
 868                     c->mfc_un.res.ttls[vifi] < 255)
 869                         return c;
 870
 871         return NULL;
 872 }
 873
 874 /* Look for a (*,G) entry */
 875 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 876                                              __be32 mcastgrp, int vifi)
 877 {
 878         int line = MFC_HASH(mcastgrp, htonl(INADDR_ANY));
 879         struct mfc_cache *c, *proxy;
 880
 881         if (mcastgrp == htonl(INADDR_ANY))
 882                 goto skip;
 883
 884         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 885                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 886                     c->mfc_mcastgrp == mcastgrp) {
 887                         if (c->mfc_un.res.ttls[vifi] < 255)
 888                                 return c;
 889
 890                         /* It's ok if the vifi is part of the static tree */
 891                         proxy = ipmr_cache_find_any_parent(mrt,
 892                                                            c->mfc_parent);
 893                         if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 894                                 return c;
 895                 }
 896
 897 skip:
 898         return ipmr_cache_find_any_parent(mrt, vifi);
 899 }
 900
 901 /* Allocate a multicast cache entry */
 902 static struct mfc_cache *ipmr_cache_alloc(void)
 903 {
 904         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 905
 906         if (c)
 907                 c->mfc_un.res.minvif = MAXVIFS;
 908         return c;
 909 }
 910
 911 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 912 {
 913         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 914
 915         if (c) {
 916                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 917                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 918         }
 919         return c;
 920 }
 921
 922 /* A cache entry has gone into a resolved state from queued */
 923 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 924                                struct mfc_cache *uc, struct mfc_cache *c)
 925 {
 926         struct sk_buff *skb;
 927         struct nlmsgerr *e;
 928
 929         /* Play the pending entries through our router */
 930         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 931                 if (ip_hdr(skb)->version == 0) {
 932                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 933
 934                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 935                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 936                                                  (u8 *)nlh;
 937                         } else {
 938                                 nlh->nlmsg_type = NLMSG_ERROR;
 939                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 940                                 skb_trim(skb, nlh->nlmsg_len);
 941                                 e = nlmsg_data(nlh);
 942                                 e->error = -EMSGSIZE;
 943                                 memset(&e->msg, 0, sizeof(e->msg));
 944                         }
 945
 946                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 947                 } else {
 948                         ip_mr_forward(net, mrt, skb, c, 0);
 949                 }
 950         }
 951 }
 952
 953 /* Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 954  * expects the following bizarre scheme.
 955  *
 956  * Called under mrt_lock.
 957  */
 958 static int ipmr_cache_report(struct mr_table *mrt,
 959                              struct sk_buff *pkt, vifi_t vifi, int assert)
 960 {
 961         const int ihl = ip_hdrlen(pkt);
 962         struct sock *mroute_sk;
 963         struct igmphdr *igmp;
 964         struct igmpmsg *msg;
 965         struct sk_buff *skb;
 966         int ret;
 967
 968         if (assert == IGMPMSG_WHOLEPKT)
 969                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 970         else
 971                 skb = alloc_skb(128, GFP_ATOMIC);
 972
 973         if (!skb)
 974                 return -ENOBUFS;
 975
 976         if (assert == IGMPMSG_WHOLEPKT) {
 977                 /* Ugly, but we have no choice with this interface.
 978                  * Duplicate old header, fix ihl, length etc.
 979                  * And all this only to mangle msg->im_msgtype and
 980                  * to set msg->im_mbz to "mbz" :-)
 981                  */
 982                 skb_push(skb, sizeof(struct iphdr));
 983                 skb_reset_network_header(skb);
 984                 skb_reset_transport_header(skb);
 985                 msg = (struct igmpmsg *)skb_network_header(skb);
 986                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 987                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 988                 msg->im_mbz = 0;
 989                 msg->im_vif = mrt->mroute_reg_vif_num;
 990                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 991                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 992                                              sizeof(struct iphdr));
 993         } else {
 994                 /* Copy the IP header */
 995                 skb_set_network_header(skb, skb->len);
 996                 skb_put(skb, ihl);
 997                 skb_copy_to_linear_data(skb, pkt->data, ihl);
 998                 /* Flag to the kernel this is a route add */
 999                 ip_hdr(skb)->protocol = 0;
1000                 msg = (struct igmpmsg *)skb_network_header(skb);
1001                 msg->im_vif = vifi;
1002                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1003                 /* Add our header */
1004                 igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
1005                 igmp->type = assert;
1006                 msg->im_msgtype = assert;
1007                 igmp->code = 0;
1008                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1009                 skb->transport_header = skb->network_header;
1010         }
1011
1012         rcu_read_lock();
1013         mroute_sk = rcu_dereference(mrt->mroute_sk);
1014         if (!mroute_sk) {
1015                 rcu_read_unlock();
1016                 kfree_skb(skb);
1017                 return -EINVAL;
1018         }
1019
1020         /* Deliver to mrouted */
1021         ret = sock_queue_rcv_skb(mroute_sk, skb);
1022         rcu_read_unlock();
1023         if (ret < 0) {
1024                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1025                 kfree_skb(skb);
1026         }
1027
1028         return ret;
1029 }
1030
1031 /* Queue a packet for resolution. It gets locked cache entry! */
1032 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1033                                  struct sk_buff *skb)
1034 {
1035         bool found = false;
1036         int err;
1037         struct mfc_cache *c;
1038         const struct iphdr *iph = ip_hdr(skb);
1039
1040         spin_lock_bh(&mfc_unres_lock);
1041         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1042                 if (c->mfc_mcastgrp == iph->daddr &&
1043                     c->mfc_origin == iph->saddr) {
1044                         found = true;
1045                         break;
1046                 }
1047         }
1048
1049         if (!found) {
1050                 /* Create a new entry if allowable */
1051                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1052                     (c = ipmr_cache_alloc_unres()) == NULL) {
1053                         spin_unlock_bh(&mfc_unres_lock);
1054
1055                         kfree_skb(skb);
1056                         return -ENOBUFS;
1057                 }
1058
1059                 /* Fill in the new cache entry */
1060                 c->mfc_parent   = -1;
1061                 c->mfc_origin   = iph->saddr;
1062                 c->mfc_mcastgrp = iph->daddr;
1063
1064                 /* Reflect first query at mrouted. */
1065                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1066                 if (err < 0) {
1067                         /* If the report failed throw the cache entry
1068                            out - Brad Parker
1069                          */
1070                         spin_unlock_bh(&mfc_unres_lock);
1071
1072                         ipmr_cache_free(c);
1073                         kfree_skb(skb);
1074                         return err;
1075                 }
1076
1077                 atomic_inc(&mrt->cache_resolve_queue_len);
1078                 list_add(&c->list, &mrt->mfc_unres_queue);
1079                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1080
1081                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1082                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1083         }
1084
1085         /* See if we can append the packet */
1086         if (c->mfc_un.unres.unresolved.qlen > 3) {
1087                 kfree_skb(skb);
1088                 err = -ENOBUFS;
1089         } else {
1090                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1091                 err = 0;
1092         }
1093
1094         spin_unlock_bh(&mfc_unres_lock);
1095         return err;
1096 }
1097
1098 /* MFC cache manipulation by user space mroute daemon */
1099
1100 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1101 {
1102         int line;
1103         struct mfc_cache *c, *next;
1104
1105         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1106
1107         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1108                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1109                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1110                     (parent == -1 || parent == c->mfc_parent)) {
1111                         list_del_rcu(&c->list);
1112                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1113                         ipmr_cache_free(c);
1114                         return 0;
1115                 }
1116         }
1117         return -ENOENT;
1118 }
1119
1120 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1121                         struct mfcctl *mfc, int mrtsock, int parent)
1122 {
1123         bool found = false;
1124         int line;
1125         struct mfc_cache *uc, *c;
1126
1127         if (mfc->mfcc_parent >= MAXVIFS)
1128                 return -ENFILE;
1129
1130         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1131
1132         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1133                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1134                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1135                     (parent == -1 || parent == c->mfc_parent)) {
1136                         found = true;
1137                         break;
1138                 }
1139         }
1140
1141         if (found) {
1142                 write_lock_bh(&mrt_lock);
1143                 c->mfc_parent = mfc->mfcc_parent;
1144                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1145                 if (!mrtsock)
1146                         c->mfc_flags |= MFC_STATIC;
1147                 write_unlock_bh(&mrt_lock);
1148                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1149                 return 0;
1150         }
1151
1152         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1153             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1154                 return -EINVAL;
1155
1156         c = ipmr_cache_alloc();
1157         if (!c)
1158                 return -ENOMEM;
1159
1160         c->mfc_origin = mfc->mfcc_origin.s_addr;
1161         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1162         c->mfc_parent = mfc->mfcc_parent;
1163         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1164         if (!mrtsock)
1165                 c->mfc_flags |= MFC_STATIC;
1166
1167         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1168
1169         /* Check to see if we resolved a queued list. If so we
1170          * need to send on the frames and tidy up.
1171          */
1172         found = false;
1173         spin_lock_bh(&mfc_unres_lock);
1174         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1175                 if (uc->mfc_origin == c->mfc_origin &&
1176                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1177                         list_del(&uc->list);
1178                         atomic_dec(&mrt->cache_resolve_queue_len);
1179                         found = true;
1180                         break;
1181                 }
1182         }
1183         if (list_empty(&mrt->mfc_unres_queue))
1184                 del_timer(&mrt->ipmr_expire_timer);
1185         spin_unlock_bh(&mfc_unres_lock);
1186
1187         if (found) {
1188                 ipmr_cache_resolve(net, mrt, uc, c);
1189                 ipmr_cache_free(uc);
1190         }
1191         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1192         return 0;
1193 }
1194
1195 /* Close the multicast socket, and clear the vif tables etc */
1196 static void mroute_clean_tables(struct mr_table *mrt)
1197 {
1198         int i;
1199         LIST_HEAD(list);
1200         struct mfc_cache *c, *next;
1201
1202         /* Shut down all active vif entries */
1203         for (i = 0; i < mrt->maxvif; i++) {
1204                 if (!(mrt->vif_table[i].flags & VIFF_STATIC))
1205                         vif_delete(mrt, i, 0, &list);
1206         }
1207         unregister_netdevice_many(&list);
1208
1209         /* Wipe the cache */
1210         for (i = 0; i < MFC_LINES; i++) {
1211                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1212                         if (c->mfc_flags & MFC_STATIC)
1213                                 continue;
1214                         list_del_rcu(&c->list);
1215                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1216                         ipmr_cache_free(c);
1217                 }
1218         }
1219
1220         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1221                 spin_lock_bh(&mfc_unres_lock);
1222                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1223                         list_del(&c->list);
1224                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1225                         ipmr_destroy_unres(mrt, c);
1226                 }
1227                 spin_unlock_bh(&mfc_unres_lock);
1228         }
1229 }
1230
1231 /* called from ip_ra_control(), before an RCU grace period,
1232  * we dont need to call synchronize_rcu() here
1233  */
1234 static void mrtsock_destruct(struct sock *sk)
1235 {
1236         struct net *net = sock_net(sk);
1237         struct mr_table *mrt;
1238
1239         rtnl_lock();
1240         ipmr_for_each_table(mrt, net) {
1241                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1242                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1243                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1244                                                     NETCONFA_IFINDEX_ALL,
1245                                                     net->ipv4.devconf_all);
1246                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1247                         mroute_clean_tables(mrt);
1248                 }
1249         }
1250         rtnl_unlock();
1251 }
1252
1253 /* Socket options and virtual interface manipulation. The whole
1254  * virtual interface system is a complete heap, but unfortunately
1255  * that's how BSD mrouted happens to think. Maybe one day with a proper
1256  * MOSPF/PIM router set up we can clean this up.
1257  */
1258
1259 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1260                          unsigned int optlen)
1261 {
1262         struct net *net = sock_net(sk);
1263         int val, ret = 0, parent = 0;
1264         struct mr_table *mrt;
1265         struct vifctl vif;
1266         struct mfcctl mfc;
1267         u32 uval;
1268
1269         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1270         rtnl_lock();
1271         if (sk->sk_type != SOCK_RAW ||
1272             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1273                 ret = -EOPNOTSUPP;
1274                 goto out_unlock;
1275         }
1276
1277         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1278         if (!mrt) {
1279                 ret = -ENOENT;
1280                 goto out_unlock;
1281         }
1282         if (optname != MRT_INIT) {
1283                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1284                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1285                         ret = -EACCES;
1286                         goto out_unlock;
1287                 }
1288         }
1289
1290         switch (optname) {
1291         case MRT_INIT:
1292                 if (optlen != sizeof(int))
1293                         ret = -EINVAL;
1294                 if (rtnl_dereference(mrt->mroute_sk))
1295                         ret = -EADDRINUSE;
1296                 if (ret)
1297                         break;
1298
1299                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1300                 if (ret == 0) {
1301                         rcu_assign_pointer(mrt->mroute_sk, sk);
1302                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1303                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1304                                                     NETCONFA_IFINDEX_ALL,
1305                                                     net->ipv4.devconf_all);
1306                 }
1307                 break;
1308         case MRT_DONE:
1309                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1310                         ret = -EACCES;
1311                 } else {
1312                         /* We need to unlock here because mrtsock_destruct takes
1313                          * care of rtnl itself and we can't change that due to
1314                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1315                          */
1316                         rtnl_unlock();
1317                         ret = ip_ra_control(sk, 0, NULL);
1318                         goto out;
1319                 }
1320                 break;
1321         case MRT_ADD_VIF:
1322         case MRT_DEL_VIF:
1323                 if (optlen != sizeof(vif)) {
1324                         ret = -EINVAL;
1325                         break;
1326                 }
1327                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1328                         ret = -EFAULT;
1329                         break;
1330                 }
1331                 if (vif.vifc_vifi >= MAXVIFS) {
1332                         ret = -ENFILE;
1333                         break;
1334                 }
1335                 if (optname == MRT_ADD_VIF) {
1336                         ret = vif_add(net, mrt, &vif,
1337                                       sk == rtnl_dereference(mrt->mroute_sk));
1338                 } else {
1339                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1340                 }
1341                 break;
1342         /* Manipulate the forwarding caches. These live
1343          * in a sort of kernel/user symbiosis.
1344          */
1345         case MRT_ADD_MFC:
1346         case MRT_DEL_MFC:
1347                 parent = -1;
1348         case MRT_ADD_MFC_PROXY:
1349         case MRT_DEL_MFC_PROXY:
1350                 if (optlen != sizeof(mfc)) {
1351                         ret = -EINVAL;
1352                         break;
1353                 }
1354                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1355                         ret = -EFAULT;
1356                         break;
1357                 }
1358                 if (parent == 0)
1359                         parent = mfc.mfcc_parent;
1360                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1361                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1362                 else
1363                         ret = ipmr_mfc_add(net, mrt, &mfc,
1364                                            sk == rtnl_dereference(mrt->mroute_sk),
1365                                            parent);
1366                 break;
1367         /* Control PIM assert. */
1368         case MRT_ASSERT:
1369                 if (optlen != sizeof(val)) {
1370                         ret = -EINVAL;
1371                         break;
1372                 }
1373                 if (get_user(val, (int __user *)optval)) {
1374                         ret = -EFAULT;
1375                         break;
1376                 }
1377                 mrt->mroute_do_assert = val;
1378                 break;
1379         case MRT_PIM:
1380                 if (!pimsm_enabled()) {
1381                         ret = -ENOPROTOOPT;
1382                         break;
1383                 }
1384                 if (optlen != sizeof(val)) {
1385                         ret = -EINVAL;
1386                         break;
1387                 }
1388                 if (get_user(val, (int __user *)optval)) {
1389                         ret = -EFAULT;
1390                         break;
1391                 }
1392
1393                 val = !!val;
1394                 if (val != mrt->mroute_do_pim) {
1395                         mrt->mroute_do_pim = val;
1396                         mrt->mroute_do_assert = val;
1397                 }
1398                 break;
1399         case MRT_TABLE:
1400                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1401                         ret = -ENOPROTOOPT;
1402                         break;
1403                 }
1404                 if (optlen != sizeof(uval)) {
1405                         ret = -EINVAL;
1406                         break;
1407                 }
1408                 if (get_user(uval, (u32 __user *)optval)) {
1409                         ret = -EFAULT;
1410                         break;
1411                 }
1412
1413                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1414                         ret = -EBUSY;
1415                 } else {
1416                         mrt = ipmr_new_table(net, uval);
1417                         if (IS_ERR(mrt))
1418                                 ret = PTR_ERR(mrt);
1419                         else
1420                                 raw_sk(sk)->ipmr_table = uval;
1421                 }
1422                 break;
1423         /* Spurious command, or MRT_VERSION which you cannot set. */
1424         default:
1425                 ret = -ENOPROTOOPT;
1426         }
1427 out_unlock:
1428         rtnl_unlock();
1429 out:
1430         return ret;
1431 }
1432
1433 /* Getsock opt support for the multicast routing system. */
1434 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1435 {
1436         int olr;
1437         int val;
1438         struct net *net = sock_net(sk);
1439         struct mr_table *mrt;
1440
1441         if (sk->sk_type != SOCK_RAW ||
1442             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1443                 return -EOPNOTSUPP;
1444
1445         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1446         if (!mrt)
1447                 return -ENOENT;
1448
1449         switch (optname) {
1450         case MRT_VERSION:
1451                 val = 0x0305;
1452                 break;
1453         case MRT_PIM:
1454                 if (!pimsm_enabled())
1455                         return -ENOPROTOOPT;
1456                 val = mrt->mroute_do_pim;
1457                 break;
1458         case MRT_ASSERT:
1459                 val = mrt->mroute_do_assert;
1460                 break;
1461         default:
1462                 return -ENOPROTOOPT;
1463         }
1464
1465         if (get_user(olr, optlen))
1466                 return -EFAULT;
1467         olr = min_t(unsigned int, olr, sizeof(int));
1468         if (olr < 0)
1469                 return -EINVAL;
1470         if (put_user(olr, optlen))
1471                 return -EFAULT;
1472         if (copy_to_user(optval, &val, olr))
1473                 return -EFAULT;
1474         return 0;
1475 }
1476
1477 /* The IP multicast ioctl support routines. */
1478 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1479 {
1480         struct sioc_sg_req sr;
1481         struct sioc_vif_req vr;
1482         struct vif_device *vif;
1483         struct mfc_cache *c;
1484         struct net *net = sock_net(sk);
1485         struct mr_table *mrt;
1486
1487         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1488         if (!mrt)
1489                 return -ENOENT;
1490
1491         switch (cmd) {
1492         case SIOCGETVIFCNT:
1493                 if (copy_from_user(&vr, arg, sizeof(vr)))
1494                         return -EFAULT;
1495                 if (vr.vifi >= mrt->maxvif)
1496                         return -EINVAL;
1497                 read_lock(&mrt_lock);
1498                 vif = &mrt->vif_table[vr.vifi];
1499                 if (VIF_EXISTS(mrt, vr.vifi)) {
1500                         vr.icount = vif->pkt_in;
1501                         vr.ocount = vif->pkt_out;
1502                         vr.ibytes = vif->bytes_in;
1503                         vr.obytes = vif->bytes_out;
1504                         read_unlock(&mrt_lock);
1505
1506                         if (copy_to_user(arg, &vr, sizeof(vr)))
1507                                 return -EFAULT;
1508                         return 0;
1509                 }
1510                 read_unlock(&mrt_lock);
1511                 return -EADDRNOTAVAIL;
1512         case SIOCGETSGCNT:
1513                 if (copy_from_user(&sr, arg, sizeof(sr)))
1514                         return -EFAULT;
1515
1516                 rcu_read_lock();
1517                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1518                 if (c) {
1519                         sr.pktcnt = c->mfc_un.res.pkt;
1520                         sr.bytecnt = c->mfc_un.res.bytes;
1521                         sr.wrong_if = c->mfc_un.res.wrong_if;
1522                         rcu_read_unlock();
1523
1524                         if (copy_to_user(arg, &sr, sizeof(sr)))
1525                                 return -EFAULT;
1526                         return 0;
1527                 }
1528                 rcu_read_unlock();
1529                 return -EADDRNOTAVAIL;
1530         default:
1531                 return -ENOIOCTLCMD;
1532         }
1533 }
1534
1535 #ifdef CONFIG_COMPAT
1536 struct compat_sioc_sg_req {
1537         struct in_addr src;
1538         struct in_addr grp;
1539         compat_ulong_t pktcnt;
1540         compat_ulong_t bytecnt;
1541         compat_ulong_t wrong_if;
1542 };
1543
1544 struct compat_sioc_vif_req {
1545         vifi_t  vifi;           /* Which iface */
1546         compat_ulong_t icount;
1547         compat_ulong_t ocount;
1548         compat_ulong_t ibytes;
1549         compat_ulong_t obytes;
1550 };
1551
1552 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1553 {
1554         struct compat_sioc_sg_req sr;
1555         struct compat_sioc_vif_req vr;
1556         struct vif_device *vif;
1557         struct mfc_cache *c;
1558         struct net *net = sock_net(sk);
1559         struct mr_table *mrt;
1560
1561         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1562         if (!mrt)
1563                 return -ENOENT;
1564
1565         switch (cmd) {
1566         case SIOCGETVIFCNT:
1567                 if (copy_from_user(&vr, arg, sizeof(vr)))
1568                         return -EFAULT;
1569                 if (vr.vifi >= mrt->maxvif)
1570                         return -EINVAL;
1571                 read_lock(&mrt_lock);
1572                 vif = &mrt->vif_table[vr.vifi];
1573                 if (VIF_EXISTS(mrt, vr.vifi)) {
1574                         vr.icount = vif->pkt_in;
1575                         vr.ocount = vif->pkt_out;
1576                         vr.ibytes = vif->bytes_in;
1577                         vr.obytes = vif->bytes_out;
1578                         read_unlock(&mrt_lock);
1579
1580                         if (copy_to_user(arg, &vr, sizeof(vr)))
1581                                 return -EFAULT;
1582                         return 0;
1583                 }
1584                 read_unlock(&mrt_lock);
1585                 return -EADDRNOTAVAIL;
1586         case SIOCGETSGCNT:
1587                 if (copy_from_user(&sr, arg, sizeof(sr)))
1588                         return -EFAULT;
1589
1590                 rcu_read_lock();
1591                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1592                 if (c) {
1593                         sr.pktcnt = c->mfc_un.res.pkt;
1594                         sr.bytecnt = c->mfc_un.res.bytes;
1595                         sr.wrong_if = c->mfc_un.res.wrong_if;
1596                         rcu_read_unlock();
1597
1598                         if (copy_to_user(arg, &sr, sizeof(sr)))
1599                                 return -EFAULT;
1600                         return 0;
1601                 }
1602                 rcu_read_unlock();
1603                 return -EADDRNOTAVAIL;
1604         default:
1605                 return -ENOIOCTLCMD;
1606         }
1607 }
1608 #endif
1609
1610 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1611 {
1612         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1613         struct net *net = dev_net(dev);
1614         struct mr_table *mrt;
1615         struct vif_device *v;
1616         int ct;
1617
1618         if (event != NETDEV_UNREGISTER)
1619                 return NOTIFY_DONE;
1620
1621         ipmr_for_each_table(mrt, net) {
1622                 v = &mrt->vif_table[0];
1623                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1624                         if (v->dev == dev)
1625                                 vif_delete(mrt, ct, 1, NULL);
1626                 }
1627         }
1628         return NOTIFY_DONE;
1629 }
1630
1631 static struct notifier_block ip_mr_notifier = {
1632         .notifier_call = ipmr_device_event,
1633 };
1634
1635 /* Encapsulate a packet by attaching a valid IPIP header to it.
1636  * This avoids tunnel drivers and other mess and gives us the speed so
1637  * important for multicast video.
1638  */
1639 static void ip_encap(struct net *net, struct sk_buff *skb,
1640                      __be32 saddr, __be32 daddr)
1641 {
1642         struct iphdr *iph;
1643         const struct iphdr *old_iph = ip_hdr(skb);
1644
1645         skb_push(skb, sizeof(struct iphdr));
1646         skb->transport_header = skb->network_header;
1647         skb_reset_network_header(skb);
1648         iph = ip_hdr(skb);
1649
1650         iph->version    =       4;
1651         iph->tos        =       old_iph->tos;
1652         iph->ttl        =       old_iph->ttl;
1653         iph->frag_off   =       0;
1654         iph->daddr      =       daddr;
1655         iph->saddr      =       saddr;
1656         iph->protocol   =       IPPROTO_IPIP;
1657         iph->ihl        =       5;
1658         iph->tot_len    =       htons(skb->len);
1659         ip_select_ident(net, skb, NULL);
1660         ip_send_check(iph);
1661
1662         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1663         nf_reset(skb);
1664 }
1665
1666 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1667                                       struct sk_buff *skb)
1668 {
1669         struct ip_options *opt = &(IPCB(skb)->opt);
1670
1671         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1672         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1673
1674         if (unlikely(opt->optlen))
1675                 ip_forward_options(skb);
1676
1677         return dst_output(net, sk, skb);
1678 }
1679
1680 /* Processing handlers for ipmr_forward */
1681
1682 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1683                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1684 {
1685         const struct iphdr *iph = ip_hdr(skb);
1686         struct vif_device *vif = &mrt->vif_table[vifi];
1687         struct net_device *dev;
1688         struct rtable *rt;
1689         struct flowi4 fl4;
1690         int    encap = 0;
1691
1692         if (!vif->dev)
1693                 goto out_free;
1694
1695         if (vif->flags & VIFF_REGISTER) {
1696                 vif->pkt_out++;
1697                 vif->bytes_out += skb->len;
1698                 vif->dev->stats.tx_bytes += skb->len;
1699                 vif->dev->stats.tx_packets++;
1700                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1701                 goto out_free;
1702         }
1703
1704         if (vif->flags & VIFF_TUNNEL) {
1705                 rt = ip_route_output_ports(net, &fl4, NULL,
1706                                            vif->remote, vif->local,
1707                                            0, 0,
1708                                            IPPROTO_IPIP,
1709                                            RT_TOS(iph->tos), vif->link);
1710                 if (IS_ERR(rt))
1711                         goto out_free;
1712                 encap = sizeof(struct iphdr);
1713         } else {
1714                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1715                                            0, 0,
1716                                            IPPROTO_IPIP,
1717                                            RT_TOS(iph->tos), vif->link);
1718                 if (IS_ERR(rt))
1719                         goto out_free;
1720         }
1721
1722         dev = rt->dst.dev;
1723
1724         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1725                 /* Do not fragment multicasts. Alas, IPv4 does not
1726                  * allow to send ICMP, so that packets will disappear
1727                  * to blackhole.
1728                  */
1729                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1730                 ip_rt_put(rt);
1731                 goto out_free;
1732         }
1733
1734         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1735
1736         if (skb_cow(skb, encap)) {
1737                 ip_rt_put(rt);
1738                 goto out_free;
1739         }
1740
1741         vif->pkt_out++;
1742         vif->bytes_out += skb->len;
1743
1744         skb_dst_drop(skb);
1745         skb_dst_set(skb, &rt->dst);
1746         ip_decrease_ttl(ip_hdr(skb));
1747
1748         /* FIXME: forward and output firewalls used to be called here.
1749          * What do we do with netfilter? -- RR
1750          */
1751         if (vif->flags & VIFF_TUNNEL) {
1752                 ip_encap(net, skb, vif->local, vif->remote);
1753                 /* FIXME: extra output firewall step used to be here. --RR */
1754                 vif->dev->stats.tx_packets++;
1755                 vif->dev->stats.tx_bytes += skb->len;
1756         }
1757
1758         IPCB(skb)->flags |= IPSKB_FORWARDED;
1759
1760         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1761          * not only before forwarding, but after forwarding on all output
1762          * interfaces. It is clear, if mrouter runs a multicasting
1763          * program, it should receive packets not depending to what interface
1764          * program is joined.
1765          * If we will not make it, the program will have to join on all
1766          * interfaces. On the other hand, multihoming host (or router, but
1767          * not mrouter) cannot join to more than one interface - it will
1768          * result in receiving multiple packets.
1769          */
1770         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1771                 net, NULL, skb, skb->dev, dev,
1772                 ipmr_forward_finish);
1773         return;
1774
1775 out_free:
1776         kfree_skb(skb);
1777 }
1778
1779 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1780 {
1781         int ct;
1782
1783         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1784                 if (mrt->vif_table[ct].dev == dev)
1785                         break;
1786         }
1787         return ct;
1788 }
1789
1790 /* "local" means that we should preserve one skb (for local delivery) */
1791 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1792                           struct sk_buff *skb, struct mfc_cache *cache,
1793                           int local)
1794 {
1795         int psend = -1;
1796         int vif, ct;
1797         int true_vifi = ipmr_find_vif(mrt, skb->dev);
1798
1799         vif = cache->mfc_parent;
1800         cache->mfc_un.res.pkt++;
1801         cache->mfc_un.res.bytes += skb->len;
1802
1803         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1804                 struct mfc_cache *cache_proxy;
1805
1806                 /* For an (*,G) entry, we only check that the incomming
1807                  * interface is part of the static tree.
1808                  */
1809                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1810                 if (cache_proxy &&
1811                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1812                         goto forward;
1813         }
1814
1815         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1816         if (mrt->vif_table[vif].dev != skb->dev) {
1817                 if (rt_is_output_route(skb_rtable(skb))) {
1818                         /* It is our own packet, looped back.
1819                          * Very complicated situation...
1820                          *
1821                          * The best workaround until routing daemons will be
1822                          * fixed is not to redistribute packet, if it was
1823                          * send through wrong interface. It means, that
1824                          * multicast applications WILL NOT work for
1825                          * (S,G), which have default multicast route pointing
1826                          * to wrong oif. In any case, it is not a good
1827                          * idea to use multicasting applications on router.
1828                          */
1829                         goto dont_forward;
1830                 }
1831
1832                 cache->mfc_un.res.wrong_if++;
1833
1834                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1835                     /* pimsm uses asserts, when switching from RPT to SPT,
1836                      * so that we cannot check that packet arrived on an oif.
1837                      * It is bad, but otherwise we would need to move pretty
1838                      * large chunk of pimd to kernel. Ough... --ANK
1839                      */
1840                     (mrt->mroute_do_pim ||
1841                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1842                     time_after(jiffies,
1843                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1844                         cache->mfc_un.res.last_assert = jiffies;
1845                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1846                 }
1847                 goto dont_forward;
1848         }
1849
1850 forward:
1851         mrt->vif_table[vif].pkt_in++;
1852         mrt->vif_table[vif].bytes_in += skb->len;
1853
1854         /* Forward the frame */
1855         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1856             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1857                 if (true_vifi >= 0 &&
1858                     true_vifi != cache->mfc_parent &&
1859                     ip_hdr(skb)->ttl >
1860                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1861                         /* It's an (*,*) entry and the packet is not coming from
1862                          * the upstream: forward the packet to the upstream
1863                          * only.
1864                          */
1865                         psend = cache->mfc_parent;
1866                         goto last_forward;
1867                 }
1868                 goto dont_forward;
1869         }
1870         for (ct = cache->mfc_un.res.maxvif - 1;
1871              ct >= cache->mfc_un.res.minvif; ct--) {
1872                 /* For (*,G) entry, don't forward to the incoming interface */
1873                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1874                      ct != true_vifi) &&
1875                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1876                         if (psend != -1) {
1877                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1878
1879                                 if (skb2)
1880                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1881                                                         psend);
1882                         }
1883                         psend = ct;
1884                 }
1885         }
1886 last_forward:
1887         if (psend != -1) {
1888                 if (local) {
1889                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1890
1891                         if (skb2)
1892                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1893                 } else {
1894                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1895                         return;
1896                 }
1897         }
1898
1899 dont_forward:
1900         if (!local)
1901                 kfree_skb(skb);
1902 }
1903
1904 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1905 {
1906         struct rtable *rt = skb_rtable(skb);
1907         struct iphdr *iph = ip_hdr(skb);
1908         struct flowi4 fl4 = {
1909                 .daddr = iph->daddr,
1910                 .saddr = iph->saddr,
1911                 .flowi4_tos = RT_TOS(iph->tos),
1912                 .flowi4_oif = (rt_is_output_route(rt) ?
1913                                skb->dev->ifindex : 0),
1914                 .flowi4_iif = (rt_is_output_route(rt) ?
1915                                LOOPBACK_IFINDEX :
1916                                skb->dev->ifindex),
1917                 .flowi4_mark = skb->mark,
1918         };
1919         struct mr_table *mrt;
1920         int err;
1921
1922         err = ipmr_fib_lookup(net, &fl4, &mrt);
1923         if (err)
1924                 return ERR_PTR(err);
1925         return mrt;
1926 }
1927
1928 /* Multicast packets for forwarding arrive here
1929  * Called with rcu_read_lock();
1930  */
1931 int ip_mr_input(struct sk_buff *skb)
1932 {
1933         struct mfc_cache *cache;
1934         struct net *net = dev_net(skb->dev);
1935         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1936         struct mr_table *mrt;
1937
1938         /* Packet is looped back after forward, it should not be
1939          * forwarded second time, but still can be delivered locally.
1940          */
1941         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1942                 goto dont_forward;
1943
1944         mrt = ipmr_rt_fib_lookup(net, skb);
1945         if (IS_ERR(mrt)) {
1946                 kfree_skb(skb);
1947                 return PTR_ERR(mrt);
1948         }
1949         if (!local) {
1950                 if (IPCB(skb)->opt.router_alert) {
1951                         if (ip_call_ra_chain(skb))
1952                                 return 0;
1953                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1954                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1955                          * Cisco IOS <= 11.2(8)) do not put router alert
1956                          * option to IGMP packets destined to routable
1957                          * groups. It is very bad, because it means
1958                          * that we can forward NO IGMP messages.
1959                          */
1960                         struct sock *mroute_sk;
1961
1962                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1963                         if (mroute_sk) {
1964                                 nf_reset(skb);
1965                                 raw_rcv(mroute_sk, skb);
1966                                 return 0;
1967                         }
1968                     }
1969         }
1970
1971         /* already under rcu_read_lock() */
1972         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1973         if (!cache) {
1974                 int vif = ipmr_find_vif(mrt, skb->dev);
1975
1976                 if (vif >= 0)
1977                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
1978                                                     vif);
1979         }
1980
1981         /* No usable cache entry */
1982         if (!cache) {
1983                 int vif;
1984
1985                 if (local) {
1986                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1987                         ip_local_deliver(skb);
1988                         if (!skb2)
1989                                 return -ENOBUFS;
1990                         skb = skb2;
1991                 }
1992
1993                 read_lock(&mrt_lock);
1994                 vif = ipmr_find_vif(mrt, skb->dev);
1995                 if (vif >= 0) {
1996                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1997                         read_unlock(&mrt_lock);
1998
1999                         return err2;
2000                 }
2001                 read_unlock(&mrt_lock);
2002                 kfree_skb(skb);
2003                 return -ENODEV;
2004         }
2005
2006         read_lock(&mrt_lock);
2007         ip_mr_forward(net, mrt, skb, cache, local);
2008         read_unlock(&mrt_lock);
2009
2010         if (local)
2011                 return ip_local_deliver(skb);
2012
2013         return 0;
2014
2015 dont_forward:
2016         if (local)
2017                 return ip_local_deliver(skb);
2018         kfree_skb(skb);
2019         return 0;
2020 }
2021
2022 #ifdef CONFIG_IP_PIMSM_V1
2023 /* Handle IGMP messages of PIMv1 */
2024 int pim_rcv_v1(struct sk_buff *skb)
2025 {
2026         struct igmphdr *pim;
2027         struct net *net = dev_net(skb->dev);
2028         struct mr_table *mrt;
2029
2030         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2031                 goto drop;
2032
2033         pim = igmp_hdr(skb);
2034
2035         mrt = ipmr_rt_fib_lookup(net, skb);
2036         if (IS_ERR(mrt))
2037                 goto drop;
2038         if (!mrt->mroute_do_pim ||
2039             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2040                 goto drop;
2041
2042         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2043 drop:
2044                 kfree_skb(skb);
2045         }
2046         return 0;
2047 }
2048 #endif
2049
2050 #ifdef CONFIG_IP_PIMSM_V2
2051 static int pim_rcv(struct sk_buff *skb)
2052 {
2053         struct pimreghdr *pim;
2054         struct net *net = dev_net(skb->dev);
2055         struct mr_table *mrt;
2056
2057         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2058                 goto drop;
2059
2060         pim = (struct pimreghdr *)skb_transport_header(skb);
2061         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2062             (pim->flags & PIM_NULL_REGISTER) ||
2063             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2064              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2065                 goto drop;
2066
2067         mrt = ipmr_rt_fib_lookup(net, skb);
2068         if (IS_ERR(mrt))
2069                 goto drop;
2070         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2071 drop:
2072                 kfree_skb(skb);
2073         }
2074         return 0;
2075 }
2076 #endif
2077
2078 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2079                               struct mfc_cache *c, struct rtmsg *rtm)
2080 {
2081         int ct;
2082         struct rtnexthop *nhp;
2083         struct nlattr *mp_attr;
2084         struct rta_mfc_stats mfcs;
2085
2086         /* If cache is unresolved, don't try to parse IIF and OIF */
2087         if (c->mfc_parent >= MAXVIFS)
2088                 return -ENOENT;
2089
2090         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2091             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2092                 return -EMSGSIZE;
2093
2094         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2095                 return -EMSGSIZE;
2096
2097         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2098                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2099                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2100                                 nla_nest_cancel(skb, mp_attr);
2101                                 return -EMSGSIZE;
2102                         }
2103
2104                         nhp->rtnh_flags = 0;
2105                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2106                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2107                         nhp->rtnh_len = sizeof(*nhp);
2108                 }
2109         }
2110
2111         nla_nest_end(skb, mp_attr);
2112
2113         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2114         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2115         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2116         if (nla_put(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs) < 0)
2117                 return -EMSGSIZE;
2118
2119         rtm->rtm_type = RTN_MULTICAST;
2120         return 1;
2121 }
2122
2123 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2124                    __be32 saddr, __be32 daddr,
2125                    struct rtmsg *rtm, int nowait)
2126 {
2127         struct mfc_cache *cache;
2128         struct mr_table *mrt;
2129         int err;
2130
2131         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2132         if (!mrt)
2133                 return -ENOENT;
2134
2135         rcu_read_lock();
2136         cache = ipmr_cache_find(mrt, saddr, daddr);
2137         if (!cache && skb->dev) {
2138                 int vif = ipmr_find_vif(mrt, skb->dev);
2139
2140                 if (vif >= 0)
2141                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2142         }
2143         if (!cache) {
2144                 struct sk_buff *skb2;
2145                 struct iphdr *iph;
2146                 struct net_device *dev;
2147                 int vif = -1;
2148
2149                 if (nowait) {
2150                         rcu_read_unlock();
2151                         return -EAGAIN;
2152                 }
2153
2154                 dev = skb->dev;
2155                 read_lock(&mrt_lock);
2156                 if (dev)
2157                         vif = ipmr_find_vif(mrt, dev);
2158                 if (vif < 0) {
2159                         read_unlock(&mrt_lock);
2160                         rcu_read_unlock();
2161                         return -ENODEV;
2162                 }
2163                 skb2 = skb_clone(skb, GFP_ATOMIC);
2164                 if (!skb2) {
2165                         read_unlock(&mrt_lock);
2166                         rcu_read_unlock();
2167                         return -ENOMEM;
2168                 }
2169
2170                 skb_push(skb2, sizeof(struct iphdr));
2171                 skb_reset_network_header(skb2);
2172                 iph = ip_hdr(skb2);
2173                 iph->ihl = sizeof(struct iphdr) >> 2;
2174                 iph->saddr = saddr;
2175                 iph->daddr = daddr;
2176                 iph->version = 0;
2177                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2178                 read_unlock(&mrt_lock);
2179                 rcu_read_unlock();
2180                 return err;
2181         }
2182
2183         read_lock(&mrt_lock);
2184         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2185         read_unlock(&mrt_lock);
2186         rcu_read_unlock();
2187         return err;
2188 }
2189
2190 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2191                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2192                             int flags)
2193 {
2194         struct nlmsghdr *nlh;
2195         struct rtmsg *rtm;
2196         int err;
2197
2198         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2199         if (!nlh)
2200                 return -EMSGSIZE;
2201
2202         rtm = nlmsg_data(nlh);
2203         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2204         rtm->rtm_dst_len  = 32;
2205         rtm->rtm_src_len  = 32;
2206         rtm->rtm_tos      = 0;
2207         rtm->rtm_table    = mrt->id;
2208         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2209                 goto nla_put_failure;
2210         rtm->rtm_type     = RTN_MULTICAST;
2211         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2212         if (c->mfc_flags & MFC_STATIC)
2213                 rtm->rtm_protocol = RTPROT_STATIC;
2214         else
2215                 rtm->rtm_protocol = RTPROT_MROUTED;
2216         rtm->rtm_flags    = 0;
2217
2218         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2219             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2220                 goto nla_put_failure;
2221         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2222         /* do not break the dump if cache is unresolved */
2223         if (err < 0 && err != -ENOENT)
2224                 goto nla_put_failure;
2225
2226         nlmsg_end(skb, nlh);
2227         return 0;
2228
2229 nla_put_failure:
2230         nlmsg_cancel(skb, nlh);
2231         return -EMSGSIZE;
2232 }
2233
2234 static size_t mroute_msgsize(bool unresolved, int maxvif)
2235 {
2236         size_t len =
2237                 NLMSG_ALIGN(sizeof(struct rtmsg))
2238                 + nla_total_size(4)     /* RTA_TABLE */
2239                 + nla_total_size(4)     /* RTA_SRC */
2240                 + nla_total_size(4)     /* RTA_DST */
2241                 ;
2242
2243         if (!unresolved)
2244                 len = len
2245                       + nla_total_size(4)       /* RTA_IIF */
2246                       + nla_total_size(0)       /* RTA_MULTIPATH */
2247                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2248                                                 /* RTA_MFC_STATS */
2249                       + nla_total_size(sizeof(struct rta_mfc_stats))
2250                 ;
2251
2252         return len;
2253 }
2254
2255 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2256                                  int cmd)
2257 {
2258         struct net *net = read_pnet(&mrt->net);
2259         struct sk_buff *skb;
2260         int err = -ENOBUFS;
2261
2262         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2263                         GFP_ATOMIC);
2264         if (!skb)
2265                 goto errout;
2266
2267         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2268         if (err < 0)
2269                 goto errout;
2270
2271         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2272         return;
2273
2274 errout:
2275         kfree_skb(skb);
2276         if (err < 0)
2277                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2278 }
2279
2280 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2281 {
2282         struct net *net = sock_net(skb->sk);
2283         struct mr_table *mrt;
2284         struct mfc_cache *mfc;
2285         unsigned int t = 0, s_t;
2286         unsigned int h = 0, s_h;
2287         unsigned int e = 0, s_e;
2288
2289         s_t = cb->args[0];
2290         s_h = cb->args[1];
2291         s_e = cb->args[2];
2292
2293         rcu_read_lock();
2294         ipmr_for_each_table(mrt, net) {
2295                 if (t < s_t)
2296                         goto next_table;
2297                 if (t > s_t)
2298                         s_h = 0;
2299                 for (h = s_h; h < MFC_LINES; h++) {
2300                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2301                                 if (e < s_e)
2302                                         goto next_entry;
2303                                 if (ipmr_fill_mroute(mrt, skb,
2304                                                      NETLINK_CB(cb->skb).portid,
2305                                                      cb->nlh->nlmsg_seq,
2306                                                      mfc, RTM_NEWROUTE,
2307                                                      NLM_F_MULTI) < 0)
2308                                         goto done;
2309 next_entry:
2310                                 e++;
2311                         }
2312                         e = s_e = 0;
2313                 }
2314                 spin_lock_bh(&mfc_unres_lock);
2315                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2316                         if (e < s_e)
2317                                 goto next_entry2;
2318                         if (ipmr_fill_mroute(mrt, skb,
2319                                              NETLINK_CB(cb->skb).portid,
2320                                              cb->nlh->nlmsg_seq,
2321                                              mfc, RTM_NEWROUTE,
2322                                              NLM_F_MULTI) < 0) {
2323                                 spin_unlock_bh(&mfc_unres_lock);
2324                                 goto done;
2325                         }
2326 next_entry2:
2327                         e++;
2328                 }
2329                 spin_unlock_bh(&mfc_unres_lock);
2330                 e = s_e = 0;
2331                 s_h = 0;
2332 next_table:
2333                 t++;
2334         }
2335 done:
2336         rcu_read_unlock();
2337
2338         cb->args[2] = e;
2339         cb->args[1] = h;
2340         cb->args[0] = t;
2341
2342         return skb->len;
2343 }
2344
2345 #ifdef CONFIG_PROC_FS
2346 /* The /proc interfaces to multicast routing :
2347  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2348  */
2349 struct ipmr_vif_iter {
2350         struct seq_net_private p;
2351         struct mr_table *mrt;
2352         int ct;
2353 };
2354
2355 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2356                                            struct ipmr_vif_iter *iter,
2357                                            loff_t pos)
2358 {
2359         struct mr_table *mrt = iter->mrt;
2360
2361         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2362                 if (!VIF_EXISTS(mrt, iter->ct))
2363                         continue;
2364                 if (pos-- == 0)
2365                         return &mrt->vif_table[iter->ct];
2366         }
2367         return NULL;
2368 }
2369
2370 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2371         __acquires(mrt_lock)
2372 {
2373         struct ipmr_vif_iter *iter = seq->private;
2374         struct net *net = seq_file_net(seq);
2375         struct mr_table *mrt;
2376
2377         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2378         if (!mrt)
2379                 return ERR_PTR(-ENOENT);
2380
2381         iter->mrt = mrt;
2382
2383         read_lock(&mrt_lock);
2384         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2385                 : SEQ_START_TOKEN;
2386 }
2387
2388 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2389 {
2390         struct ipmr_vif_iter *iter = seq->private;
2391         struct net *net = seq_file_net(seq);
2392         struct mr_table *mrt = iter->mrt;
2393
2394         ++*pos;
2395         if (v == SEQ_START_TOKEN)
2396                 return ipmr_vif_seq_idx(net, iter, 0);
2397
2398         while (++iter->ct < mrt->maxvif) {
2399                 if (!VIF_EXISTS(mrt, iter->ct))
2400                         continue;
2401                 return &mrt->vif_table[iter->ct];
2402         }
2403         return NULL;
2404 }
2405
2406 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2407         __releases(mrt_lock)
2408 {
2409         read_unlock(&mrt_lock);
2410 }
2411
2412 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2413 {
2414         struct ipmr_vif_iter *iter = seq->private;
2415         struct mr_table *mrt = iter->mrt;
2416
2417         if (v == SEQ_START_TOKEN) {
2418                 seq_puts(seq,
2419                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2420         } else {
2421                 const struct vif_device *vif = v;
2422                 const char *name =  vif->dev ? vif->dev->name : "none";
2423
2424                 seq_printf(seq,
2425                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2426                            vif - mrt->vif_table,
2427                            name, vif->bytes_in, vif->pkt_in,
2428                            vif->bytes_out, vif->pkt_out,
2429                            vif->flags, vif->local, vif->remote);
2430         }
2431         return 0;
2432 }
2433
2434 static const struct seq_operations ipmr_vif_seq_ops = {
2435         .start = ipmr_vif_seq_start,
2436         .next  = ipmr_vif_seq_next,
2437         .stop  = ipmr_vif_seq_stop,
2438         .show  = ipmr_vif_seq_show,
2439 };
2440
2441 static int ipmr_vif_open(struct inode *inode, struct file *file)
2442 {
2443         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2444                             sizeof(struct ipmr_vif_iter));
2445 }
2446
2447 static const struct file_operations ipmr_vif_fops = {
2448         .owner   = THIS_MODULE,
2449         .open    = ipmr_vif_open,
2450         .read    = seq_read,
2451         .llseek  = seq_lseek,
2452         .release = seq_release_net,
2453 };
2454
2455 struct ipmr_mfc_iter {
2456         struct seq_net_private p;
2457         struct mr_table *mrt;
2458         struct list_head *cache;
2459         int ct;
2460 };
2461
2462
2463 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2464                                           struct ipmr_mfc_iter *it, loff_t pos)
2465 {
2466         struct mr_table *mrt = it->mrt;
2467         struct mfc_cache *mfc;
2468
2469         rcu_read_lock();
2470         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2471                 it->cache = &mrt->mfc_cache_array[it->ct];
2472                 list_for_each_entry_rcu(mfc, it->cache, list)
2473                         if (pos-- == 0)
2474                                 return mfc;
2475         }
2476         rcu_read_unlock();
2477
2478         spin_lock_bh(&mfc_unres_lock);
2479         it->cache = &mrt->mfc_unres_queue;
2480         list_for_each_entry(mfc, it->cache, list)
2481                 if (pos-- == 0)
2482                         return mfc;
2483         spin_unlock_bh(&mfc_unres_lock);
2484
2485         it->cache = NULL;
2486         return NULL;
2487 }
2488
2489
2490 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2491 {
2492         struct ipmr_mfc_iter *it = seq->private;
2493         struct net *net = seq_file_net(seq);
2494         struct mr_table *mrt;
2495
2496         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2497         if (!mrt)
2498                 return ERR_PTR(-ENOENT);
2499
2500         it->mrt = mrt;
2501         it->cache = NULL;
2502         it->ct = 0;
2503         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2504                 : SEQ_START_TOKEN;
2505 }
2506
2507 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2508 {
2509         struct mfc_cache *mfc = v;
2510         struct ipmr_mfc_iter *it = seq->private;
2511         struct net *net = seq_file_net(seq);
2512         struct mr_table *mrt = it->mrt;
2513
2514         ++*pos;
2515
2516         if (v == SEQ_START_TOKEN)
2517                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2518
2519         if (mfc->list.next != it->cache)
2520                 return list_entry(mfc->list.next, struct mfc_cache, list);
2521
2522         if (it->cache == &mrt->mfc_unres_queue)
2523                 goto end_of_list;
2524
2525         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2526
2527         while (++it->ct < MFC_LINES) {
2528                 it->cache = &mrt->mfc_cache_array[it->ct];
2529                 if (list_empty(it->cache))
2530                         continue;
2531                 return list_first_entry(it->cache, struct mfc_cache, list);
2532         }
2533
2534         /* exhausted cache_array, show unresolved */
2535         rcu_read_unlock();
2536         it->cache = &mrt->mfc_unres_queue;
2537         it->ct = 0;
2538
2539         spin_lock_bh(&mfc_unres_lock);
2540         if (!list_empty(it->cache))
2541                 return list_first_entry(it->cache, struct mfc_cache, list);
2542
2543 end_of_list:
2544         spin_unlock_bh(&mfc_unres_lock);
2545         it->cache = NULL;
2546
2547         return NULL;
2548 }
2549
2550 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2551 {
2552         struct ipmr_mfc_iter *it = seq->private;
2553         struct mr_table *mrt = it->mrt;
2554
2555         if (it->cache == &mrt->mfc_unres_queue)
2556                 spin_unlock_bh(&mfc_unres_lock);
2557         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2558                 rcu_read_unlock();
2559 }
2560
2561 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2562 {
2563         int n;
2564
2565         if (v == SEQ_START_TOKEN) {
2566                 seq_puts(seq,
2567                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2568         } else {
2569                 const struct mfc_cache *mfc = v;
2570                 const struct ipmr_mfc_iter *it = seq->private;
2571                 const struct mr_table *mrt = it->mrt;
2572
2573                 seq_printf(seq, "%08X %08X %-3hd",
2574                            (__force u32) mfc->mfc_mcastgrp,
2575                            (__force u32) mfc->mfc_origin,
2576                            mfc->mfc_parent);
2577
2578                 if (it->cache != &mrt->mfc_unres_queue) {
2579                         seq_printf(seq, " %8lu %8lu %8lu",
2580                                    mfc->mfc_un.res.pkt,
2581                                    mfc->mfc_un.res.bytes,
2582                                    mfc->mfc_un.res.wrong_if);
2583                         for (n = mfc->mfc_un.res.minvif;
2584                              n < mfc->mfc_un.res.maxvif; n++) {
2585                                 if (VIF_EXISTS(mrt, n) &&
2586                                     mfc->mfc_un.res.ttls[n] < 255)
2587                                         seq_printf(seq,
2588                                            " %2d:%-3d",
2589                                            n, mfc->mfc_un.res.ttls[n]);
2590                         }
2591                 } else {
2592                         /* unresolved mfc_caches don't contain
2593                          * pkt, bytes and wrong_if values
2594                          */
2595                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2596                 }
2597                 seq_putc(seq, '\n');
2598         }
2599         return 0;
2600 }
2601
2602 static const struct seq_operations ipmr_mfc_seq_ops = {
2603         .start = ipmr_mfc_seq_start,
2604         .next  = ipmr_mfc_seq_next,
2605         .stop  = ipmr_mfc_seq_stop,
2606         .show  = ipmr_mfc_seq_show,
2607 };
2608
2609 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2610 {
2611         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2612                             sizeof(struct ipmr_mfc_iter));
2613 }
2614
2615 static const struct file_operations ipmr_mfc_fops = {
2616         .owner   = THIS_MODULE,
2617         .open    = ipmr_mfc_open,
2618         .read    = seq_read,
2619         .llseek  = seq_lseek,
2620         .release = seq_release_net,
2621 };
2622 #endif
2623
2624 #ifdef CONFIG_IP_PIMSM_V2
2625 static const struct net_protocol pim_protocol = {
2626         .handler        =       pim_rcv,
2627         .netns_ok       =       1,
2628 };
2629 #endif
2630
2631 /* Setup for IP multicast routing */
2632 static int __net_init ipmr_net_init(struct net *net)
2633 {
2634         int err;
2635
2636         err = ipmr_rules_init(net);
2637         if (err < 0)
2638                 goto fail;
2639
2640 #ifdef CONFIG_PROC_FS
2641         err = -ENOMEM;
2642         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
2643                 goto proc_vif_fail;
2644         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
2645                 goto proc_cache_fail;
2646 #endif
2647         return 0;
2648
2649 #ifdef CONFIG_PROC_FS
2650 proc_cache_fail:
2651         remove_proc_entry("ip_mr_vif", net->proc_net);
2652 proc_vif_fail:
2653         ipmr_rules_exit(net);
2654 #endif
2655 fail:
2656         return err;
2657 }
2658
2659 static void __net_exit ipmr_net_exit(struct net *net)
2660 {
2661 #ifdef CONFIG_PROC_FS
2662         remove_proc_entry("ip_mr_cache", net->proc_net);
2663         remove_proc_entry("ip_mr_vif", net->proc_net);
2664 #endif
2665         ipmr_rules_exit(net);
2666 }
2667
2668 static struct pernet_operations ipmr_net_ops = {
2669         .init = ipmr_net_init,
2670         .exit = ipmr_net_exit,
2671 };
2672
2673 int __init ip_mr_init(void)
2674 {
2675         int err;
2676
2677         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2678                                        sizeof(struct mfc_cache),
2679                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2680                                        NULL);
2681
2682         err = register_pernet_subsys(&ipmr_net_ops);
2683         if (err)
2684                 goto reg_pernet_fail;
2685
2686         err = register_netdevice_notifier(&ip_mr_notifier);
2687         if (err)
2688                 goto reg_notif_fail;
2689 #ifdef CONFIG_IP_PIMSM_V2
2690         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2691                 pr_err("%s: can't add PIM protocol\n", __func__);
2692                 err = -EAGAIN;
2693                 goto add_proto_fail;
2694         }
2695 #endif
2696         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2697                       NULL, ipmr_rtm_dumproute, NULL);
2698         return 0;
2699
2700 #ifdef CONFIG_IP_PIMSM_V2
2701 add_proto_fail:
2702         unregister_netdevice_notifier(&ip_mr_notifier);
2703 #endif
2704 reg_notif_fail:
2705         unregister_pernet_subsys(&ipmr_net_ops);
2706 reg_pernet_fail:
2707         kmem_cache_destroy(mrt_cachep);
2708         return err;
2709 }