net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <[email protected]>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <linux/uaccess.h>
  30 #include <linux/types.h>
  31 #include <linux/cache.h>
  32 #include <linux/capability.h>
  33 #include <linux/errno.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/icmp.h>
  56 #include <net/udp.h>
  57 #include <net/raw.h>
  58 #include <linux/notifier.h>
  59 #include <linux/if_arp.h>
  60 #include <linux/netfilter_ipv4.h>
  61 #include <linux/compat.h>
  62 #include <linux/export.h>
  63 #include <net/ip_tunnels.h>
  64 #include <net/checksum.h>
  65 #include <net/netlink.h>
  66 #include <net/fib_rules.h>
  67 #include <linux/netconf.h>
  68 #include <net/nexthop.h>
  69 #include <net/switchdev.h>
  70
  71 struct ipmr_rule {
  72         struct fib_rule         common;
  73 };
  74
  75 struct ipmr_result {
  76         struct mr_table         *mrt;
  77 };
  78
  79 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  80  * Note that the changes are semaphored via rtnl_lock.
  81  */
  82
  83 static DEFINE_RWLOCK(mrt_lock);
  84
  85 /* Multicast router control variables */
  86
  87 /* Special spinlock for queue of unresolved entries */
  88 static DEFINE_SPINLOCK(mfc_unres_lock);
  89
  90 /* We return to original Alan's scheme. Hash table of resolved
  91  * entries is changed only in process context and protected
  92  * with weak lock mrt_lock. Queue of unresolved entries is protected
  93  * with strong spinlock mfc_unres_lock.
  94  *
  95  * In this case data path is free of exclusive locks at all.
  96  */
  97
  98 static struct kmem_cache *mrt_cachep __ro_after_init;
  99
 100 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 101 static void ipmr_free_table(struct mr_table *mrt);
 102
 103 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 104                           struct net_device *dev, struct sk_buff *skb,
 105                           struct mfc_cache *cache, int local);
 106 static int ipmr_cache_report(struct mr_table *mrt,
 107                              struct sk_buff *pkt, vifi_t vifi, int assert);
 108 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 109                                  int cmd);
 110 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt);
 111 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 112 static void ipmr_expire_process(struct timer_list *t);
 113
 114 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 115 #define ipmr_for_each_table(mrt, net) \
 116         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 117
 118 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 119                                            struct mr_table *mrt)
 120 {
 121         struct mr_table *ret;
 122
 123         if (!mrt)
 124                 ret = list_entry_rcu(net->ipv4.mr_tables.next,
 125                                      struct mr_table, list);
 126         else
 127                 ret = list_entry_rcu(mrt->list.next,
 128                                      struct mr_table, list);
 129
 130         if (&ret->list == &net->ipv4.mr_tables)
 131                 return NULL;
 132         return ret;
 133 }
 134
 135 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 136 {
 137         struct mr_table *mrt;
 138
 139         ipmr_for_each_table(mrt, net) {
 140                 if (mrt->id == id)
 141                         return mrt;
 142         }
 143         return NULL;
 144 }
 145
 146 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 147                            struct mr_table **mrt)
 148 {
 149         int err;
 150         struct ipmr_result res;
 151         struct fib_lookup_arg arg = {
 152                 .result = &res,
 153                 .flags = FIB_LOOKUP_NOREF,
 154         };
 155
 156         /* update flow if oif or iif point to device enslaved to l3mdev */
 157         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 158
 159         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 160                                flowi4_to_flowi(flp4), 0, &arg);
 161         if (err < 0)
 162                 return err;
 163         *mrt = res.mrt;
 164         return 0;
 165 }
 166
 167 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 168                             int flags, struct fib_lookup_arg *arg)
 169 {
 170         struct ipmr_result *res = arg->result;
 171         struct mr_table *mrt;
 172
 173         switch (rule->action) {
 174         case FR_ACT_TO_TBL:
 175                 break;
 176         case FR_ACT_UNREACHABLE:
 177                 return -ENETUNREACH;
 178         case FR_ACT_PROHIBIT:
 179                 return -EACCES;
 180         case FR_ACT_BLACKHOLE:
 181         default:
 182                 return -EINVAL;
 183         }
 184
 185         arg->table = fib_rule_get_table(rule, arg);
 186
 187         mrt = ipmr_get_table(rule->fr_net, arg->table);
 188         if (!mrt)
 189                 return -EAGAIN;
 190         res->mrt = mrt;
 191         return 0;
 192 }
 193
 194 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 195 {
 196         return 1;
 197 }
 198
 199 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 200         FRA_GENERIC_POLICY,
 201 };
 202
 203 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 204                                struct fib_rule_hdr *frh, struct nlattr **tb)
 205 {
 206         return 0;
 207 }
 208
 209 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 210                              struct nlattr **tb)
 211 {
 212         return 1;
 213 }
 214
 215 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 216                           struct fib_rule_hdr *frh)
 217 {
 218         frh->dst_len = 0;
 219         frh->src_len = 0;
 220         frh->tos     = 0;
 221         return 0;
 222 }
 223
 224 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 225         .family         = RTNL_FAMILY_IPMR,
 226         .rule_size      = sizeof(struct ipmr_rule),
 227         .addr_size      = sizeof(u32),
 228         .action         = ipmr_rule_action,
 229         .match          = ipmr_rule_match,
 230         .configure      = ipmr_rule_configure,
 231         .compare        = ipmr_rule_compare,
 232         .fill           = ipmr_rule_fill,
 233         .nlgroup        = RTNLGRP_IPV4_RULE,
 234         .policy         = ipmr_rule_policy,
 235         .owner          = THIS_MODULE,
 236 };
 237
 238 static int __net_init ipmr_rules_init(struct net *net)
 239 {
 240         struct fib_rules_ops *ops;
 241         struct mr_table *mrt;
 242         int err;
 243
 244         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 245         if (IS_ERR(ops))
 246                 return PTR_ERR(ops);
 247
 248         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 249
 250         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 251         if (IS_ERR(mrt)) {
 252                 err = PTR_ERR(mrt);
 253                 goto err1;
 254         }
 255
 256         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 257         if (err < 0)
 258                 goto err2;
 259
 260         net->ipv4.mr_rules_ops = ops;
 261         return 0;
 262
 263 err2:
 264         ipmr_free_table(mrt);
 265 err1:
 266         fib_rules_unregister(ops);
 267         return err;
 268 }
 269
 270 static void __net_exit ipmr_rules_exit(struct net *net)
 271 {
 272         struct mr_table *mrt, *next;
 273
 274         rtnl_lock();
 275         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 276                 list_del(&mrt->list);
 277                 ipmr_free_table(mrt);
 278         }
 279         fib_rules_unregister(net->ipv4.mr_rules_ops);
 280         rtnl_unlock();
 281 }
 282
 283 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 284 {
 285         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR);
 286 }
 287
 288 static unsigned int ipmr_rules_seq_read(struct net *net)
 289 {
 290         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 291 }
 292
 293 bool ipmr_rule_default(const struct fib_rule *rule)
 294 {
 295         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 296 }
 297 EXPORT_SYMBOL(ipmr_rule_default);
 298 #else
 299 #define ipmr_for_each_table(mrt, net) \
 300         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 301
 302 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 303                                            struct mr_table *mrt)
 304 {
 305         if (!mrt)
 306                 return net->ipv4.mrt;
 307         return NULL;
 308 }
 309
 310 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 311 {
 312         return net->ipv4.mrt;
 313 }
 314
 315 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 316                            struct mr_table **mrt)
 317 {
 318         *mrt = net->ipv4.mrt;
 319         return 0;
 320 }
 321
 322 static int __net_init ipmr_rules_init(struct net *net)
 323 {
 324         struct mr_table *mrt;
 325
 326         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 327         if (IS_ERR(mrt))
 328                 return PTR_ERR(mrt);
 329         net->ipv4.mrt = mrt;
 330         return 0;
 331 }
 332
 333 static void __net_exit ipmr_rules_exit(struct net *net)
 334 {
 335         rtnl_lock();
 336         ipmr_free_table(net->ipv4.mrt);
 337         net->ipv4.mrt = NULL;
 338         rtnl_unlock();
 339 }
 340
 341 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 342 {
 343         return 0;
 344 }
 345
 346 static unsigned int ipmr_rules_seq_read(struct net *net)
 347 {
 348         return 0;
 349 }
 350
 351 bool ipmr_rule_default(const struct fib_rule *rule)
 352 {
 353         return true;
 354 }
 355 EXPORT_SYMBOL(ipmr_rule_default);
 356 #endif
 357
 358 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 359                                 const void *ptr)
 360 {
 361         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 362         struct mfc_cache *c = (struct mfc_cache *)ptr;
 363
 364         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 365                cmparg->mfc_origin != c->mfc_origin;
 366 }
 367
 368 static const struct rhashtable_params ipmr_rht_params = {
 369         .head_offset = offsetof(struct mr_mfc, mnode),
 370         .key_offset = offsetof(struct mfc_cache, cmparg),
 371         .key_len = sizeof(struct mfc_cache_cmp_arg),
 372         .nelem_hint = 3,
 373         .locks_mul = 1,
 374         .obj_cmpfn = ipmr_hash_cmp,
 375         .automatic_shrinking = true,
 376 };
 377
 378 static void ipmr_new_table_set(struct mr_table *mrt,
 379                                struct net *net)
 380 {
 381 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 382         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 383 #endif
 384 }
 385
 386 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 387         .mfc_mcastgrp = htonl(INADDR_ANY),
 388         .mfc_origin = htonl(INADDR_ANY),
 389 };
 390
 391 static struct mr_table_ops ipmr_mr_table_ops = {
 392         .rht_params = &ipmr_rht_params,
 393         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 394 };
 395
 396 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 397 {
 398         struct mr_table *mrt;
 399
 400         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 401         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 402                 return ERR_PTR(-EINVAL);
 403
 404         mrt = ipmr_get_table(net, id);
 405         if (mrt)
 406                 return mrt;
 407
 408         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 409                               ipmr_expire_process, ipmr_new_table_set);
 410 }
 411
 412 static void ipmr_free_table(struct mr_table *mrt)
 413 {
 414         del_timer_sync(&mrt->ipmr_expire_timer);
 415         mroute_clean_tables(mrt, true);
 416         rhltable_destroy(&mrt->mfc_hash);
 417         kfree(mrt);
 418 }
 419
 420 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 421
 422 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 423 {
 424         struct net *net = dev_net(dev);
 425
 426         dev_close(dev);
 427
 428         dev = __dev_get_by_name(net, "tunl0");
 429         if (dev) {
 430                 const struct net_device_ops *ops = dev->netdev_ops;
 431                 struct ifreq ifr;
 432                 struct ip_tunnel_parm p;
 433
 434                 memset(&p, 0, sizeof(p));
 435                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 436                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 437                 p.iph.version = 4;
 438                 p.iph.ihl = 5;
 439                 p.iph.protocol = IPPROTO_IPIP;
 440                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 441                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 442
 443                 if (ops->ndo_do_ioctl) {
 444                         mm_segment_t oldfs = get_fs();
 445
 446                         set_fs(KERNEL_DS);
 447                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 448                         set_fs(oldfs);
 449                 }
 450         }
 451 }
 452
 453 /* Initialize ipmr pimreg/tunnel in_device */
 454 static bool ipmr_init_vif_indev(const struct net_device *dev)
 455 {
 456         struct in_device *in_dev;
 457
 458         ASSERT_RTNL();
 459
 460         in_dev = __in_dev_get_rtnl(dev);
 461         if (!in_dev)
 462                 return false;
 463         ipv4_devconf_setall(in_dev);
 464         neigh_parms_data_state_setall(in_dev->arp_parms);
 465         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 466
 467         return true;
 468 }
 469
 470 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 471 {
 472         struct net_device  *dev;
 473
 474         dev = __dev_get_by_name(net, "tunl0");
 475
 476         if (dev) {
 477                 const struct net_device_ops *ops = dev->netdev_ops;
 478                 int err;
 479                 struct ifreq ifr;
 480                 struct ip_tunnel_parm p;
 481
 482                 memset(&p, 0, sizeof(p));
 483                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 484                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 485                 p.iph.version = 4;
 486                 p.iph.ihl = 5;
 487                 p.iph.protocol = IPPROTO_IPIP;
 488                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 489                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 490
 491                 if (ops->ndo_do_ioctl) {
 492                         mm_segment_t oldfs = get_fs();
 493
 494                         set_fs(KERNEL_DS);
 495                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 496                         set_fs(oldfs);
 497                 } else {
 498                         err = -EOPNOTSUPP;
 499                 }
 500                 dev = NULL;
 501
 502                 if (err == 0 &&
 503                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 504                         dev->flags |= IFF_MULTICAST;
 505                         if (!ipmr_init_vif_indev(dev))
 506                                 goto failure;
 507                         if (dev_open(dev))
 508                                 goto failure;
 509                         dev_hold(dev);
 510                 }
 511         }
 512         return dev;
 513
 514 failure:
 515         unregister_netdevice(dev);
 516         return NULL;
 517 }
 518
 519 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 520 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 521 {
 522         struct net *net = dev_net(dev);
 523         struct mr_table *mrt;
 524         struct flowi4 fl4 = {
 525                 .flowi4_oif     = dev->ifindex,
 526                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 527                 .flowi4_mark    = skb->mark,
 528         };
 529         int err;
 530
 531         err = ipmr_fib_lookup(net, &fl4, &mrt);
 532         if (err < 0) {
 533                 kfree_skb(skb);
 534                 return err;
 535         }
 536
 537         read_lock(&mrt_lock);
 538         dev->stats.tx_bytes += skb->len;
 539         dev->stats.tx_packets++;
 540         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 541         read_unlock(&mrt_lock);
 542         kfree_skb(skb);
 543         return NETDEV_TX_OK;
 544 }
 545
 546 static int reg_vif_get_iflink(const struct net_device *dev)
 547 {
 548         return 0;
 549 }
 550
 551 static const struct net_device_ops reg_vif_netdev_ops = {
 552         .ndo_start_xmit = reg_vif_xmit,
 553         .ndo_get_iflink = reg_vif_get_iflink,
 554 };
 555
 556 static void reg_vif_setup(struct net_device *dev)
 557 {
 558         dev->type               = ARPHRD_PIMREG;
 559         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 560         dev->flags              = IFF_NOARP;
 561         dev->netdev_ops         = &reg_vif_netdev_ops;
 562         dev->needs_free_netdev  = true;
 563         dev->features           |= NETIF_F_NETNS_LOCAL;
 564 }
 565
 566 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 567 {
 568         struct net_device *dev;
 569         char name[IFNAMSIZ];
 570
 571         if (mrt->id == RT_TABLE_DEFAULT)
 572                 sprintf(name, "pimreg");
 573         else
 574                 sprintf(name, "pimreg%u", mrt->id);
 575
 576         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 577
 578         if (!dev)
 579                 return NULL;
 580
 581         dev_net_set(dev, net);
 582
 583         if (register_netdevice(dev)) {
 584                 free_netdev(dev);
 585                 return NULL;
 586         }
 587
 588         if (!ipmr_init_vif_indev(dev))
 589                 goto failure;
 590         if (dev_open(dev))
 591                 goto failure;
 592
 593         dev_hold(dev);
 594
 595         return dev;
 596
 597 failure:
 598         unregister_netdevice(dev);
 599         return NULL;
 600 }
 601
 602 /* called with rcu_read_lock() */
 603 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 604                      unsigned int pimlen)
 605 {
 606         struct net_device *reg_dev = NULL;
 607         struct iphdr *encap;
 608
 609         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 610         /* Check that:
 611          * a. packet is really sent to a multicast group
 612          * b. packet is not a NULL-REGISTER
 613          * c. packet is not truncated
 614          */
 615         if (!ipv4_is_multicast(encap->daddr) ||
 616             encap->tot_len == 0 ||
 617             ntohs(encap->tot_len) + pimlen > skb->len)
 618                 return 1;
 619
 620         read_lock(&mrt_lock);
 621         if (mrt->mroute_reg_vif_num >= 0)
 622                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 623         read_unlock(&mrt_lock);
 624
 625         if (!reg_dev)
 626                 return 1;
 627
 628         skb->mac_header = skb->network_header;
 629         skb_pull(skb, (u8 *)encap - skb->data);
 630         skb_reset_network_header(skb);
 631         skb->protocol = htons(ETH_P_IP);
 632         skb->ip_summed = CHECKSUM_NONE;
 633
 634         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 635
 636         netif_rx(skb);
 637
 638         return NET_RX_SUCCESS;
 639 }
 640 #else
 641 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 642 {
 643         return NULL;
 644 }
 645 #endif
 646
 647 static int call_ipmr_vif_entry_notifier(struct notifier_block *nb,
 648                                         struct net *net,
 649                                         enum fib_event_type event_type,
 650                                         struct vif_device *vif,
 651                                         vifi_t vif_index, u32 tb_id)
 652 {
 653         struct vif_entry_notifier_info info = {
 654                 .info = {
 655                         .family = RTNL_FAMILY_IPMR,
 656                         .net = net,
 657                 },
 658                 .dev = vif->dev,
 659                 .vif_index = vif_index,
 660                 .vif_flags = vif->flags,
 661                 .tb_id = tb_id,
 662         };
 663
 664         return call_fib_notifier(nb, net, event_type, &info.info);
 665 }
 666
 667 static int call_ipmr_vif_entry_notifiers(struct net *net,
 668                                          enum fib_event_type event_type,
 669                                          struct vif_device *vif,
 670                                          vifi_t vif_index, u32 tb_id)
 671 {
 672         struct vif_entry_notifier_info info = {
 673                 .info = {
 674                         .family = RTNL_FAMILY_IPMR,
 675                         .net = net,
 676                 },
 677                 .dev = vif->dev,
 678                 .vif_index = vif_index,
 679                 .vif_flags = vif->flags,
 680                 .tb_id = tb_id,
 681         };
 682
 683         ASSERT_RTNL();
 684         net->ipv4.ipmr_seq++;
 685         return call_fib_notifiers(net, event_type, &info.info);
 686 }
 687
 688 static int call_ipmr_mfc_entry_notifier(struct notifier_block *nb,
 689                                         struct net *net,
 690                                         enum fib_event_type event_type,
 691                                         struct mfc_cache *mfc, u32 tb_id)
 692 {
 693         struct mfc_entry_notifier_info info = {
 694                 .info = {
 695                         .family = RTNL_FAMILY_IPMR,
 696                         .net = net,
 697                 },
 698                 .mfc = mfc,
 699                 .tb_id = tb_id
 700         };
 701
 702         return call_fib_notifier(nb, net, event_type, &info.info);
 703 }
 704
 705 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 706                                          enum fib_event_type event_type,
 707                                          struct mfc_cache *mfc, u32 tb_id)
 708 {
 709         struct mfc_entry_notifier_info info = {
 710                 .info = {
 711                         .family = RTNL_FAMILY_IPMR,
 712                         .net = net,
 713                 },
 714                 .mfc = mfc,
 715                 .tb_id = tb_id
 716         };
 717
 718         ASSERT_RTNL();
 719         net->ipv4.ipmr_seq++;
 720         return call_fib_notifiers(net, event_type, &info.info);
 721 }
 722
 723 /**
 724  *      vif_delete - Delete a VIF entry
 725  *      @notify: Set to 1, if the caller is a notifier_call
 726  */
 727 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 728                       struct list_head *head)
 729 {
 730         struct net *net = read_pnet(&mrt->net);
 731         struct vif_device *v;
 732         struct net_device *dev;
 733         struct in_device *in_dev;
 734
 735         if (vifi < 0 || vifi >= mrt->maxvif)
 736                 return -EADDRNOTAVAIL;
 737
 738         v = &mrt->vif_table[vifi];
 739
 740         if (VIF_EXISTS(mrt, vifi))
 741                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 742                                               mrt->id);
 743
 744         write_lock_bh(&mrt_lock);
 745         dev = v->dev;
 746         v->dev = NULL;
 747
 748         if (!dev) {
 749                 write_unlock_bh(&mrt_lock);
 750                 return -EADDRNOTAVAIL;
 751         }
 752
 753         if (vifi == mrt->mroute_reg_vif_num)
 754                 mrt->mroute_reg_vif_num = -1;
 755
 756         if (vifi + 1 == mrt->maxvif) {
 757                 int tmp;
 758
 759                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 760                         if (VIF_EXISTS(mrt, tmp))
 761                                 break;
 762                 }
 763                 mrt->maxvif = tmp+1;
 764         }
 765
 766         write_unlock_bh(&mrt_lock);
 767
 768         dev_set_allmulti(dev, -1);
 769
 770         in_dev = __in_dev_get_rtnl(dev);
 771         if (in_dev) {
 772                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 773                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 774                                             NETCONFA_MC_FORWARDING,
 775                                             dev->ifindex, &in_dev->cnf);
 776                 ip_rt_multicast_event(in_dev);
 777         }
 778
 779         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 780                 unregister_netdevice_queue(dev, head);
 781
 782         dev_put(dev);
 783         return 0;
 784 }
 785
 786 static void ipmr_cache_free_rcu(struct rcu_head *head)
 787 {
 788         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 789
 790         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 791 }
 792
 793 void ipmr_cache_free(struct mfc_cache *c)
 794 {
 795         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 796 }
 797 EXPORT_SYMBOL(ipmr_cache_free);
 798
 799 /* Destroy an unresolved cache entry, killing queued skbs
 800  * and reporting error to netlink readers.
 801  */
 802 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 803 {
 804         struct net *net = read_pnet(&mrt->net);
 805         struct sk_buff *skb;
 806         struct nlmsgerr *e;
 807
 808         atomic_dec(&mrt->cache_resolve_queue_len);
 809
 810         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 811                 if (ip_hdr(skb)->version == 0) {
 812                         struct nlmsghdr *nlh = skb_pull(skb,
 813                                                         sizeof(struct iphdr));
 814                         nlh->nlmsg_type = NLMSG_ERROR;
 815                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 816                         skb_trim(skb, nlh->nlmsg_len);
 817                         e = nlmsg_data(nlh);
 818                         e->error = -ETIMEDOUT;
 819                         memset(&e->msg, 0, sizeof(e->msg));
 820
 821                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 822                 } else {
 823                         kfree_skb(skb);
 824                 }
 825         }
 826
 827         ipmr_cache_free(c);
 828 }
 829
 830 /* Timer process for the unresolved queue. */
 831 static void ipmr_expire_process(struct timer_list *t)
 832 {
 833         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 834         struct mr_mfc *c, *next;
 835         unsigned long expires;
 836         unsigned long now;
 837
 838         if (!spin_trylock(&mfc_unres_lock)) {
 839                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 840                 return;
 841         }
 842
 843         if (list_empty(&mrt->mfc_unres_queue))
 844                 goto out;
 845
 846         now = jiffies;
 847         expires = 10*HZ;
 848
 849         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 850                 if (time_after(c->mfc_un.unres.expires, now)) {
 851                         unsigned long interval = c->mfc_un.unres.expires - now;
 852                         if (interval < expires)
 853                                 expires = interval;
 854                         continue;
 855                 }
 856
 857                 list_del(&c->list);
 858                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 859                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 860         }
 861
 862         if (!list_empty(&mrt->mfc_unres_queue))
 863                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 864
 865 out:
 866         spin_unlock(&mfc_unres_lock);
 867 }
 868
 869 /* Fill oifs list. It is called under write locked mrt_lock. */
 870 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 871                                    unsigned char *ttls)
 872 {
 873         int vifi;
 874
 875         cache->mfc_un.res.minvif = MAXVIFS;
 876         cache->mfc_un.res.maxvif = 0;
 877         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 878
 879         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 880                 if (VIF_EXISTS(mrt, vifi) &&
 881                     ttls[vifi] && ttls[vifi] < 255) {
 882                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 883                         if (cache->mfc_un.res.minvif > vifi)
 884                                 cache->mfc_un.res.minvif = vifi;
 885                         if (cache->mfc_un.res.maxvif <= vifi)
 886                                 cache->mfc_un.res.maxvif = vifi + 1;
 887                 }
 888         }
 889         cache->mfc_un.res.lastuse = jiffies;
 890 }
 891
 892 static int vif_add(struct net *net, struct mr_table *mrt,
 893                    struct vifctl *vifc, int mrtsock)
 894 {
 895         int vifi = vifc->vifc_vifi;
 896         struct switchdev_attr attr = {
 897                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
 898         };
 899         struct vif_device *v = &mrt->vif_table[vifi];
 900         struct net_device *dev;
 901         struct in_device *in_dev;
 902         int err;
 903
 904         /* Is vif busy ? */
 905         if (VIF_EXISTS(mrt, vifi))
 906                 return -EADDRINUSE;
 907
 908         switch (vifc->vifc_flags) {
 909         case VIFF_REGISTER:
 910                 if (!ipmr_pimsm_enabled())
 911                         return -EINVAL;
 912                 /* Special Purpose VIF in PIM
 913                  * All the packets will be sent to the daemon
 914                  */
 915                 if (mrt->mroute_reg_vif_num >= 0)
 916                         return -EADDRINUSE;
 917                 dev = ipmr_reg_vif(net, mrt);
 918                 if (!dev)
 919                         return -ENOBUFS;
 920                 err = dev_set_allmulti(dev, 1);
 921                 if (err) {
 922                         unregister_netdevice(dev);
 923                         dev_put(dev);
 924                         return err;
 925                 }
 926                 break;
 927         case VIFF_TUNNEL:
 928                 dev = ipmr_new_tunnel(net, vifc);
 929                 if (!dev)
 930                         return -ENOBUFS;
 931                 err = dev_set_allmulti(dev, 1);
 932                 if (err) {
 933                         ipmr_del_tunnel(dev, vifc);
 934                         dev_put(dev);
 935                         return err;
 936                 }
 937                 break;
 938         case VIFF_USE_IFINDEX:
 939         case 0:
 940                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 941                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 942                         if (dev && !__in_dev_get_rtnl(dev)) {
 943                                 dev_put(dev);
 944                                 return -EADDRNOTAVAIL;
 945                         }
 946                 } else {
 947                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 948                 }
 949                 if (!dev)
 950                         return -EADDRNOTAVAIL;
 951                 err = dev_set_allmulti(dev, 1);
 952                 if (err) {
 953                         dev_put(dev);
 954                         return err;
 955                 }
 956                 break;
 957         default:
 958                 return -EINVAL;
 959         }
 960
 961         in_dev = __in_dev_get_rtnl(dev);
 962         if (!in_dev) {
 963                 dev_put(dev);
 964                 return -EADDRNOTAVAIL;
 965         }
 966         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 967         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 968                                     dev->ifindex, &in_dev->cnf);
 969         ip_rt_multicast_event(in_dev);
 970
 971         /* Fill in the VIF structures */
 972         vif_device_init(v, dev, vifc->vifc_rate_limit,
 973                         vifc->vifc_threshold,
 974                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 975                         (VIFF_TUNNEL | VIFF_REGISTER));
 976
 977         attr.orig_dev = dev;
 978         if (!switchdev_port_attr_get(dev, &attr)) {
 979                 memcpy(v->dev_parent_id.id, attr.u.ppid.id, attr.u.ppid.id_len);
 980                 v->dev_parent_id.id_len = attr.u.ppid.id_len;
 981         } else {
 982                 v->dev_parent_id.id_len = 0;
 983         }
 984
 985         v->local = vifc->vifc_lcl_addr.s_addr;
 986         v->remote = vifc->vifc_rmt_addr.s_addr;
 987
 988         /* And finish update writing critical data */
 989         write_lock_bh(&mrt_lock);
 990         v->dev = dev;
 991         if (v->flags & VIFF_REGISTER)
 992                 mrt->mroute_reg_vif_num = vifi;
 993         if (vifi+1 > mrt->maxvif)
 994                 mrt->maxvif = vifi+1;
 995         write_unlock_bh(&mrt_lock);
 996         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 997         return 0;
 998 }
 999
1000 /* called with rcu_read_lock() */
1001 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
1002                                          __be32 origin,
1003                                          __be32 mcastgrp)
1004 {
1005         struct mfc_cache_cmp_arg arg = {
1006                         .mfc_mcastgrp = mcastgrp,
1007                         .mfc_origin = origin
1008         };
1009
1010         return mr_mfc_find(mrt, &arg);
1011 }
1012
1013 /* Look for a (*,G) entry */
1014 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
1015                                              __be32 mcastgrp, int vifi)
1016 {
1017         struct mfc_cache_cmp_arg arg = {
1018                         .mfc_mcastgrp = mcastgrp,
1019                         .mfc_origin = htonl(INADDR_ANY)
1020         };
1021
1022         if (mcastgrp == htonl(INADDR_ANY))
1023                 return mr_mfc_find_any_parent(mrt, vifi);
1024         return mr_mfc_find_any(mrt, vifi, &arg);
1025 }
1026
1027 /* Look for a (S,G,iif) entry if parent != -1 */
1028 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
1029                                                 __be32 origin, __be32 mcastgrp,
1030                                                 int parent)
1031 {
1032         struct mfc_cache_cmp_arg arg = {
1033                         .mfc_mcastgrp = mcastgrp,
1034                         .mfc_origin = origin,
1035         };
1036
1037         return mr_mfc_find_parent(mrt, &arg, parent);
1038 }
1039
1040 /* Allocate a multicast cache entry */
1041 static struct mfc_cache *ipmr_cache_alloc(void)
1042 {
1043         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
1044
1045         if (c) {
1046                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
1047                 c->_c.mfc_un.res.minvif = MAXVIFS;
1048                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
1049         }
1050         return c;
1051 }
1052
1053 static struct mfc_cache *ipmr_cache_alloc_unres(void)
1054 {
1055         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
1056
1057         if (c) {
1058                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
1059                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
1060         }
1061         return c;
1062 }
1063
1064 /* A cache entry has gone into a resolved state from queued */
1065 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
1066                                struct mfc_cache *uc, struct mfc_cache *c)
1067 {
1068         struct sk_buff *skb;
1069         struct nlmsgerr *e;
1070
1071         /* Play the pending entries through our router */
1072         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
1073                 if (ip_hdr(skb)->version == 0) {
1074                         struct nlmsghdr *nlh = skb_pull(skb,
1075                                                         sizeof(struct iphdr));
1076
1077                         if (mr_fill_mroute(mrt, skb, &c->_c,
1078                                            nlmsg_data(nlh)) > 0) {
1079                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
1080                                                  (u8 *)nlh;
1081                         } else {
1082                                 nlh->nlmsg_type = NLMSG_ERROR;
1083                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
1084                                 skb_trim(skb, nlh->nlmsg_len);
1085                                 e = nlmsg_data(nlh);
1086                                 e->error = -EMSGSIZE;
1087                                 memset(&e->msg, 0, sizeof(e->msg));
1088                         }
1089
1090                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1091                 } else {
1092                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1093                 }
1094         }
1095 }
1096
1097 /* Bounce a cache query up to mrouted and netlink.
1098  *
1099  * Called under mrt_lock.
1100  */
1101 static int ipmr_cache_report(struct mr_table *mrt,
1102                              struct sk_buff *pkt, vifi_t vifi, int assert)
1103 {
1104         const int ihl = ip_hdrlen(pkt);
1105         struct sock *mroute_sk;
1106         struct igmphdr *igmp;
1107         struct igmpmsg *msg;
1108         struct sk_buff *skb;
1109         int ret;
1110
1111         if (assert == IGMPMSG_WHOLEPKT)
1112                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1113         else
1114                 skb = alloc_skb(128, GFP_ATOMIC);
1115
1116         if (!skb)
1117                 return -ENOBUFS;
1118
1119         if (assert == IGMPMSG_WHOLEPKT) {
1120                 /* Ugly, but we have no choice with this interface.
1121                  * Duplicate old header, fix ihl, length etc.
1122                  * And all this only to mangle msg->im_msgtype and
1123                  * to set msg->im_mbz to "mbz" :-)
1124                  */
1125                 skb_push(skb, sizeof(struct iphdr));
1126                 skb_reset_network_header(skb);
1127                 skb_reset_transport_header(skb);
1128                 msg = (struct igmpmsg *)skb_network_header(skb);
1129                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1130                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
1131                 msg->im_mbz = 0;
1132                 msg->im_vif = mrt->mroute_reg_vif_num;
1133                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1134                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1135                                              sizeof(struct iphdr));
1136         } else {
1137                 /* Copy the IP header */
1138                 skb_set_network_header(skb, skb->len);
1139                 skb_put(skb, ihl);
1140                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1141                 /* Flag to the kernel this is a route add */
1142                 ip_hdr(skb)->protocol = 0;
1143                 msg = (struct igmpmsg *)skb_network_header(skb);
1144                 msg->im_vif = vifi;
1145                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1146                 /* Add our header */
1147                 igmp = skb_put(skb, sizeof(struct igmphdr));
1148                 igmp->type = assert;
1149                 msg->im_msgtype = assert;
1150                 igmp->code = 0;
1151                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1152                 skb->transport_header = skb->network_header;
1153         }
1154
1155         rcu_read_lock();
1156         mroute_sk = rcu_dereference(mrt->mroute_sk);
1157         if (!mroute_sk) {
1158                 rcu_read_unlock();
1159                 kfree_skb(skb);
1160                 return -EINVAL;
1161         }
1162
1163         igmpmsg_netlink_event(mrt, skb);
1164
1165         /* Deliver to mrouted */
1166         ret = sock_queue_rcv_skb(mroute_sk, skb);
1167         rcu_read_unlock();
1168         if (ret < 0) {
1169                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1170                 kfree_skb(skb);
1171         }
1172
1173         return ret;
1174 }
1175
1176 /* Queue a packet for resolution. It gets locked cache entry! */
1177 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1178                                  struct sk_buff *skb, struct net_device *dev)
1179 {
1180         const struct iphdr *iph = ip_hdr(skb);
1181         struct mfc_cache *c;
1182         bool found = false;
1183         int err;
1184
1185         spin_lock_bh(&mfc_unres_lock);
1186         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1187                 if (c->mfc_mcastgrp == iph->daddr &&
1188                     c->mfc_origin == iph->saddr) {
1189                         found = true;
1190                         break;
1191                 }
1192         }
1193
1194         if (!found) {
1195                 /* Create a new entry if allowable */
1196                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1197                     (c = ipmr_cache_alloc_unres()) == NULL) {
1198                         spin_unlock_bh(&mfc_unres_lock);
1199
1200                         kfree_skb(skb);
1201                         return -ENOBUFS;
1202                 }
1203
1204                 /* Fill in the new cache entry */
1205                 c->_c.mfc_parent = -1;
1206                 c->mfc_origin   = iph->saddr;
1207                 c->mfc_mcastgrp = iph->daddr;
1208
1209                 /* Reflect first query at mrouted. */
1210                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1211
1212                 if (err < 0) {
1213                         /* If the report failed throw the cache entry
1214                            out - Brad Parker
1215                          */
1216                         spin_unlock_bh(&mfc_unres_lock);
1217
1218                         ipmr_cache_free(c);
1219                         kfree_skb(skb);
1220                         return err;
1221                 }
1222
1223                 atomic_inc(&mrt->cache_resolve_queue_len);
1224                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1225                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1226
1227                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1228                         mod_timer(&mrt->ipmr_expire_timer,
1229                                   c->_c.mfc_un.unres.expires);
1230         }
1231
1232         /* See if we can append the packet */
1233         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1234                 kfree_skb(skb);
1235                 err = -ENOBUFS;
1236         } else {
1237                 if (dev) {
1238                         skb->dev = dev;
1239                         skb->skb_iif = dev->ifindex;
1240                 }
1241                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1242                 err = 0;
1243         }
1244
1245         spin_unlock_bh(&mfc_unres_lock);
1246         return err;
1247 }
1248
1249 /* MFC cache manipulation by user space mroute daemon */
1250
1251 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1252 {
1253         struct net *net = read_pnet(&mrt->net);
1254         struct mfc_cache *c;
1255
1256         /* The entries are added/deleted only under RTNL */
1257         rcu_read_lock();
1258         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1259                                    mfc->mfcc_mcastgrp.s_addr, parent);
1260         rcu_read_unlock();
1261         if (!c)
1262                 return -ENOENT;
1263         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1264         list_del_rcu(&c->_c.list);
1265         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1266         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1267         ipmr_cache_put(c);
1268
1269         return 0;
1270 }
1271
1272 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1273                         struct mfcctl *mfc, int mrtsock, int parent)
1274 {
1275         struct mfc_cache *uc, *c;
1276         struct mr_mfc *_uc;
1277         bool found;
1278         int ret;
1279
1280         if (mfc->mfcc_parent >= MAXVIFS)
1281                 return -ENFILE;
1282
1283         /* The entries are added/deleted only under RTNL */
1284         rcu_read_lock();
1285         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1286                                    mfc->mfcc_mcastgrp.s_addr, parent);
1287         rcu_read_unlock();
1288         if (c) {
1289                 write_lock_bh(&mrt_lock);
1290                 c->_c.mfc_parent = mfc->mfcc_parent;
1291                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1292                 if (!mrtsock)
1293                         c->_c.mfc_flags |= MFC_STATIC;
1294                 write_unlock_bh(&mrt_lock);
1295                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1296                                               mrt->id);
1297                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1298                 return 0;
1299         }
1300
1301         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1302             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1303                 return -EINVAL;
1304
1305         c = ipmr_cache_alloc();
1306         if (!c)
1307                 return -ENOMEM;
1308
1309         c->mfc_origin = mfc->mfcc_origin.s_addr;
1310         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1311         c->_c.mfc_parent = mfc->mfcc_parent;
1312         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1313         if (!mrtsock)
1314                 c->_c.mfc_flags |= MFC_STATIC;
1315
1316         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1317                                   ipmr_rht_params);
1318         if (ret) {
1319                 pr_err("ipmr: rhtable insert error %d\n", ret);
1320                 ipmr_cache_free(c);
1321                 return ret;
1322         }
1323         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1324         /* Check to see if we resolved a queued list. If so we
1325          * need to send on the frames and tidy up.
1326          */
1327         found = false;
1328         spin_lock_bh(&mfc_unres_lock);
1329         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1330                 uc = (struct mfc_cache *)_uc;
1331                 if (uc->mfc_origin == c->mfc_origin &&
1332                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1333                         list_del(&_uc->list);
1334                         atomic_dec(&mrt->cache_resolve_queue_len);
1335                         found = true;
1336                         break;
1337                 }
1338         }
1339         if (list_empty(&mrt->mfc_unres_queue))
1340                 del_timer(&mrt->ipmr_expire_timer);
1341         spin_unlock_bh(&mfc_unres_lock);
1342
1343         if (found) {
1344                 ipmr_cache_resolve(net, mrt, uc, c);
1345                 ipmr_cache_free(uc);
1346         }
1347         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1348         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1349         return 0;
1350 }
1351
1352 /* Close the multicast socket, and clear the vif tables etc */
1353 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1354 {
1355         struct net *net = read_pnet(&mrt->net);
1356         struct mr_mfc *c, *tmp;
1357         struct mfc_cache *cache;
1358         LIST_HEAD(list);
1359         int i;
1360
1361         /* Shut down all active vif entries */
1362         for (i = 0; i < mrt->maxvif; i++) {
1363                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1364                         continue;
1365                 vif_delete(mrt, i, 0, &list);
1366         }
1367         unregister_netdevice_many(&list);
1368
1369         /* Wipe the cache */
1370         list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1371                 if (!all && (c->mfc_flags & MFC_STATIC))
1372                         continue;
1373                 rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1374                 list_del_rcu(&c->list);
1375                 cache = (struct mfc_cache *)c;
1376                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1377                                               mrt->id);
1378                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1379                 ipmr_cache_put(cache);
1380         }
1381
1382         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1383                 spin_lock_bh(&mfc_unres_lock);
1384                 list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1385                         list_del(&c->list);
1386                         cache = (struct mfc_cache *)c;
1387                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1388                         ipmr_destroy_unres(mrt, cache);
1389                 }
1390                 spin_unlock_bh(&mfc_unres_lock);
1391         }
1392 }
1393
1394 /* called from ip_ra_control(), before an RCU grace period,
1395  * we dont need to call synchronize_rcu() here
1396  */
1397 static void mrtsock_destruct(struct sock *sk)
1398 {
1399         struct net *net = sock_net(sk);
1400         struct mr_table *mrt;
1401
1402         ASSERT_RTNL();
1403         ipmr_for_each_table(mrt, net) {
1404                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1405                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1406                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1407                                                     NETCONFA_MC_FORWARDING,
1408                                                     NETCONFA_IFINDEX_ALL,
1409                                                     net->ipv4.devconf_all);
1410                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1411                         mroute_clean_tables(mrt, false);
1412                 }
1413         }
1414 }
1415
1416 /* Socket options and virtual interface manipulation. The whole
1417  * virtual interface system is a complete heap, but unfortunately
1418  * that's how BSD mrouted happens to think. Maybe one day with a proper
1419  * MOSPF/PIM router set up we can clean this up.
1420  */
1421
1422 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1423                          unsigned int optlen)
1424 {
1425         struct net *net = sock_net(sk);
1426         int val, ret = 0, parent = 0;
1427         struct mr_table *mrt;
1428         struct vifctl vif;
1429         struct mfcctl mfc;
1430         u32 uval;
1431
1432         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1433         rtnl_lock();
1434         if (sk->sk_type != SOCK_RAW ||
1435             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1436                 ret = -EOPNOTSUPP;
1437                 goto out_unlock;
1438         }
1439
1440         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1441         if (!mrt) {
1442                 ret = -ENOENT;
1443                 goto out_unlock;
1444         }
1445         if (optname != MRT_INIT) {
1446                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1447                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1448                         ret = -EACCES;
1449                         goto out_unlock;
1450                 }
1451         }
1452
1453         switch (optname) {
1454         case MRT_INIT:
1455                 if (optlen != sizeof(int)) {
1456                         ret = -EINVAL;
1457                         break;
1458                 }
1459                 if (rtnl_dereference(mrt->mroute_sk)) {
1460                         ret = -EADDRINUSE;
1461                         break;
1462                 }
1463
1464                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1465                 if (ret == 0) {
1466                         rcu_assign_pointer(mrt->mroute_sk, sk);
1467                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1468                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1469                                                     NETCONFA_MC_FORWARDING,
1470                                                     NETCONFA_IFINDEX_ALL,
1471                                                     net->ipv4.devconf_all);
1472                 }
1473                 break;
1474         case MRT_DONE:
1475                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1476                         ret = -EACCES;
1477                 } else {
1478                         ret = ip_ra_control(sk, 0, NULL);
1479                         goto out_unlock;
1480                 }
1481                 break;
1482         case MRT_ADD_VIF:
1483         case MRT_DEL_VIF:
1484                 if (optlen != sizeof(vif)) {
1485                         ret = -EINVAL;
1486                         break;
1487                 }
1488                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1489                         ret = -EFAULT;
1490                         break;
1491                 }
1492                 if (vif.vifc_vifi >= MAXVIFS) {
1493                         ret = -ENFILE;
1494                         break;
1495                 }
1496                 if (optname == MRT_ADD_VIF) {
1497                         ret = vif_add(net, mrt, &vif,
1498                                       sk == rtnl_dereference(mrt->mroute_sk));
1499                 } else {
1500                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1501                 }
1502                 break;
1503         /* Manipulate the forwarding caches. These live
1504          * in a sort of kernel/user symbiosis.
1505          */
1506         case MRT_ADD_MFC:
1507         case MRT_DEL_MFC:
1508                 parent = -1;
1509                 /* fall through */
1510         case MRT_ADD_MFC_PROXY:
1511         case MRT_DEL_MFC_PROXY:
1512                 if (optlen != sizeof(mfc)) {
1513                         ret = -EINVAL;
1514                         break;
1515                 }
1516                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1517                         ret = -EFAULT;
1518                         break;
1519                 }
1520                 if (parent == 0)
1521                         parent = mfc.mfcc_parent;
1522                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1523                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1524                 else
1525                         ret = ipmr_mfc_add(net, mrt, &mfc,
1526                                            sk == rtnl_dereference(mrt->mroute_sk),
1527                                            parent);
1528                 break;
1529         /* Control PIM assert. */
1530         case MRT_ASSERT:
1531                 if (optlen != sizeof(val)) {
1532                         ret = -EINVAL;
1533                         break;
1534                 }
1535                 if (get_user(val, (int __user *)optval)) {
1536                         ret = -EFAULT;
1537                         break;
1538                 }
1539                 mrt->mroute_do_assert = val;
1540                 break;
1541         case MRT_PIM:
1542                 if (!ipmr_pimsm_enabled()) {
1543                         ret = -ENOPROTOOPT;
1544                         break;
1545                 }
1546                 if (optlen != sizeof(val)) {
1547                         ret = -EINVAL;
1548                         break;
1549                 }
1550                 if (get_user(val, (int __user *)optval)) {
1551                         ret = -EFAULT;
1552                         break;
1553                 }
1554
1555                 val = !!val;
1556                 if (val != mrt->mroute_do_pim) {
1557                         mrt->mroute_do_pim = val;
1558                         mrt->mroute_do_assert = val;
1559                 }
1560                 break;
1561         case MRT_TABLE:
1562                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1563                         ret = -ENOPROTOOPT;
1564                         break;
1565                 }
1566                 if (optlen != sizeof(uval)) {
1567                         ret = -EINVAL;
1568                         break;
1569                 }
1570                 if (get_user(uval, (u32 __user *)optval)) {
1571                         ret = -EFAULT;
1572                         break;
1573                 }
1574
1575                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1576                         ret = -EBUSY;
1577                 } else {
1578                         mrt = ipmr_new_table(net, uval);
1579                         if (IS_ERR(mrt))
1580                                 ret = PTR_ERR(mrt);
1581                         else
1582                                 raw_sk(sk)->ipmr_table = uval;
1583                 }
1584                 break;
1585         /* Spurious command, or MRT_VERSION which you cannot set. */
1586         default:
1587                 ret = -ENOPROTOOPT;
1588         }
1589 out_unlock:
1590         rtnl_unlock();
1591         return ret;
1592 }
1593
1594 /* Getsock opt support for the multicast routing system. */
1595 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1596 {
1597         int olr;
1598         int val;
1599         struct net *net = sock_net(sk);
1600         struct mr_table *mrt;
1601
1602         if (sk->sk_type != SOCK_RAW ||
1603             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1604                 return -EOPNOTSUPP;
1605
1606         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1607         if (!mrt)
1608                 return -ENOENT;
1609
1610         switch (optname) {
1611         case MRT_VERSION:
1612                 val = 0x0305;
1613                 break;
1614         case MRT_PIM:
1615                 if (!ipmr_pimsm_enabled())
1616                         return -ENOPROTOOPT;
1617                 val = mrt->mroute_do_pim;
1618                 break;
1619         case MRT_ASSERT:
1620                 val = mrt->mroute_do_assert;
1621                 break;
1622         default:
1623                 return -ENOPROTOOPT;
1624         }
1625
1626         if (get_user(olr, optlen))
1627                 return -EFAULT;
1628         olr = min_t(unsigned int, olr, sizeof(int));
1629         if (olr < 0)
1630                 return -EINVAL;
1631         if (put_user(olr, optlen))
1632                 return -EFAULT;
1633         if (copy_to_user(optval, &val, olr))
1634                 return -EFAULT;
1635         return 0;
1636 }
1637
1638 /* The IP multicast ioctl support routines. */
1639 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1640 {
1641         struct sioc_sg_req sr;
1642         struct sioc_vif_req vr;
1643         struct vif_device *vif;
1644         struct mfc_cache *c;
1645         struct net *net = sock_net(sk);
1646         struct mr_table *mrt;
1647
1648         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1649         if (!mrt)
1650                 return -ENOENT;
1651
1652         switch (cmd) {
1653         case SIOCGETVIFCNT:
1654                 if (copy_from_user(&vr, arg, sizeof(vr)))
1655                         return -EFAULT;
1656                 if (vr.vifi >= mrt->maxvif)
1657                         return -EINVAL;
1658                 read_lock(&mrt_lock);
1659                 vif = &mrt->vif_table[vr.vifi];
1660                 if (VIF_EXISTS(mrt, vr.vifi)) {
1661                         vr.icount = vif->pkt_in;
1662                         vr.ocount = vif->pkt_out;
1663                         vr.ibytes = vif->bytes_in;
1664                         vr.obytes = vif->bytes_out;
1665                         read_unlock(&mrt_lock);
1666
1667                         if (copy_to_user(arg, &vr, sizeof(vr)))
1668                                 return -EFAULT;
1669                         return 0;
1670                 }
1671                 read_unlock(&mrt_lock);
1672                 return -EADDRNOTAVAIL;
1673         case SIOCGETSGCNT:
1674                 if (copy_from_user(&sr, arg, sizeof(sr)))
1675                         return -EFAULT;
1676
1677                 rcu_read_lock();
1678                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1679                 if (c) {
1680                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1681                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1682                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1683                         rcu_read_unlock();
1684
1685                         if (copy_to_user(arg, &sr, sizeof(sr)))
1686                                 return -EFAULT;
1687                         return 0;
1688                 }
1689                 rcu_read_unlock();
1690                 return -EADDRNOTAVAIL;
1691         default:
1692                 return -ENOIOCTLCMD;
1693         }
1694 }
1695
1696 #ifdef CONFIG_COMPAT
1697 struct compat_sioc_sg_req {
1698         struct in_addr src;
1699         struct in_addr grp;
1700         compat_ulong_t pktcnt;
1701         compat_ulong_t bytecnt;
1702         compat_ulong_t wrong_if;
1703 };
1704
1705 struct compat_sioc_vif_req {
1706         vifi_t  vifi;           /* Which iface */
1707         compat_ulong_t icount;
1708         compat_ulong_t ocount;
1709         compat_ulong_t ibytes;
1710         compat_ulong_t obytes;
1711 };
1712
1713 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1714 {
1715         struct compat_sioc_sg_req sr;
1716         struct compat_sioc_vif_req vr;
1717         struct vif_device *vif;
1718         struct mfc_cache *c;
1719         struct net *net = sock_net(sk);
1720         struct mr_table *mrt;
1721
1722         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1723         if (!mrt)
1724                 return -ENOENT;
1725
1726         switch (cmd) {
1727         case SIOCGETVIFCNT:
1728                 if (copy_from_user(&vr, arg, sizeof(vr)))
1729                         return -EFAULT;
1730                 if (vr.vifi >= mrt->maxvif)
1731                         return -EINVAL;
1732                 read_lock(&mrt_lock);
1733                 vif = &mrt->vif_table[vr.vifi];
1734                 if (VIF_EXISTS(mrt, vr.vifi)) {
1735                         vr.icount = vif->pkt_in;
1736                         vr.ocount = vif->pkt_out;
1737                         vr.ibytes = vif->bytes_in;
1738                         vr.obytes = vif->bytes_out;
1739                         read_unlock(&mrt_lock);
1740
1741                         if (copy_to_user(arg, &vr, sizeof(vr)))
1742                                 return -EFAULT;
1743                         return 0;
1744                 }
1745                 read_unlock(&mrt_lock);
1746                 return -EADDRNOTAVAIL;
1747         case SIOCGETSGCNT:
1748                 if (copy_from_user(&sr, arg, sizeof(sr)))
1749                         return -EFAULT;
1750
1751                 rcu_read_lock();
1752                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1753                 if (c) {
1754                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1755                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1756                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1757                         rcu_read_unlock();
1758
1759                         if (copy_to_user(arg, &sr, sizeof(sr)))
1760                                 return -EFAULT;
1761                         return 0;
1762                 }
1763                 rcu_read_unlock();
1764                 return -EADDRNOTAVAIL;
1765         default:
1766                 return -ENOIOCTLCMD;
1767         }
1768 }
1769 #endif
1770
1771 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1772 {
1773         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1774         struct net *net = dev_net(dev);
1775         struct mr_table *mrt;
1776         struct vif_device *v;
1777         int ct;
1778
1779         if (event != NETDEV_UNREGISTER)
1780                 return NOTIFY_DONE;
1781
1782         ipmr_for_each_table(mrt, net) {
1783                 v = &mrt->vif_table[0];
1784                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1785                         if (v->dev == dev)
1786                                 vif_delete(mrt, ct, 1, NULL);
1787                 }
1788         }
1789         return NOTIFY_DONE;
1790 }
1791
1792 static struct notifier_block ip_mr_notifier = {
1793         .notifier_call = ipmr_device_event,
1794 };
1795
1796 /* Encapsulate a packet by attaching a valid IPIP header to it.
1797  * This avoids tunnel drivers and other mess and gives us the speed so
1798  * important for multicast video.
1799  */
1800 static void ip_encap(struct net *net, struct sk_buff *skb,
1801                      __be32 saddr, __be32 daddr)
1802 {
1803         struct iphdr *iph;
1804         const struct iphdr *old_iph = ip_hdr(skb);
1805
1806         skb_push(skb, sizeof(struct iphdr));
1807         skb->transport_header = skb->network_header;
1808         skb_reset_network_header(skb);
1809         iph = ip_hdr(skb);
1810
1811         iph->version    =       4;
1812         iph->tos        =       old_iph->tos;
1813         iph->ttl        =       old_iph->ttl;
1814         iph->frag_off   =       0;
1815         iph->daddr      =       daddr;
1816         iph->saddr      =       saddr;
1817         iph->protocol   =       IPPROTO_IPIP;
1818         iph->ihl        =       5;
1819         iph->tot_len    =       htons(skb->len);
1820         ip_select_ident(net, skb, NULL);
1821         ip_send_check(iph);
1822
1823         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1824         nf_reset(skb);
1825 }
1826
1827 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1828                                       struct sk_buff *skb)
1829 {
1830         struct ip_options *opt = &(IPCB(skb)->opt);
1831
1832         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1833         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1834
1835         if (unlikely(opt->optlen))
1836                 ip_forward_options(skb);
1837
1838         return dst_output(net, sk, skb);
1839 }
1840
1841 #ifdef CONFIG_NET_SWITCHDEV
1842 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1843                                    int in_vifi, int out_vifi)
1844 {
1845         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1846         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1847
1848         if (!skb->offload_mr_fwd_mark)
1849                 return false;
1850         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1851                 return false;
1852         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1853                                         &in_vif->dev_parent_id);
1854 }
1855 #else
1856 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1857                                    int in_vifi, int out_vifi)
1858 {
1859         return false;
1860 }
1861 #endif
1862
1863 /* Processing handlers for ipmr_forward */
1864
1865 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1866                             int in_vifi, struct sk_buff *skb,
1867                             struct mfc_cache *c, int vifi)
1868 {
1869         const struct iphdr *iph = ip_hdr(skb);
1870         struct vif_device *vif = &mrt->vif_table[vifi];
1871         struct net_device *dev;
1872         struct rtable *rt;
1873         struct flowi4 fl4;
1874         int    encap = 0;
1875
1876         if (!vif->dev)
1877                 goto out_free;
1878
1879         if (vif->flags & VIFF_REGISTER) {
1880                 vif->pkt_out++;
1881                 vif->bytes_out += skb->len;
1882                 vif->dev->stats.tx_bytes += skb->len;
1883                 vif->dev->stats.tx_packets++;
1884                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1885                 goto out_free;
1886         }
1887
1888         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1889                 goto out_free;
1890
1891         if (vif->flags & VIFF_TUNNEL) {
1892                 rt = ip_route_output_ports(net, &fl4, NULL,
1893                                            vif->remote, vif->local,
1894                                            0, 0,
1895                                            IPPROTO_IPIP,
1896                                            RT_TOS(iph->tos), vif->link);
1897                 if (IS_ERR(rt))
1898                         goto out_free;
1899                 encap = sizeof(struct iphdr);
1900         } else {
1901                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1902                                            0, 0,
1903                                            IPPROTO_IPIP,
1904                                            RT_TOS(iph->tos), vif->link);
1905                 if (IS_ERR(rt))
1906                         goto out_free;
1907         }
1908
1909         dev = rt->dst.dev;
1910
1911         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1912                 /* Do not fragment multicasts. Alas, IPv4 does not
1913                  * allow to send ICMP, so that packets will disappear
1914                  * to blackhole.
1915                  */
1916                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1917                 ip_rt_put(rt);
1918                 goto out_free;
1919         }
1920
1921         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1922
1923         if (skb_cow(skb, encap)) {
1924                 ip_rt_put(rt);
1925                 goto out_free;
1926         }
1927
1928         vif->pkt_out++;
1929         vif->bytes_out += skb->len;
1930
1931         skb_dst_drop(skb);
1932         skb_dst_set(skb, &rt->dst);
1933         ip_decrease_ttl(ip_hdr(skb));
1934
1935         /* FIXME: forward and output firewalls used to be called here.
1936          * What do we do with netfilter? -- RR
1937          */
1938         if (vif->flags & VIFF_TUNNEL) {
1939                 ip_encap(net, skb, vif->local, vif->remote);
1940                 /* FIXME: extra output firewall step used to be here. --RR */
1941                 vif->dev->stats.tx_packets++;
1942                 vif->dev->stats.tx_bytes += skb->len;
1943         }
1944
1945         IPCB(skb)->flags |= IPSKB_FORWARDED;
1946
1947         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1948          * not only before forwarding, but after forwarding on all output
1949          * interfaces. It is clear, if mrouter runs a multicasting
1950          * program, it should receive packets not depending to what interface
1951          * program is joined.
1952          * If we will not make it, the program will have to join on all
1953          * interfaces. On the other hand, multihoming host (or router, but
1954          * not mrouter) cannot join to more than one interface - it will
1955          * result in receiving multiple packets.
1956          */
1957         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1958                 net, NULL, skb, skb->dev, dev,
1959                 ipmr_forward_finish);
1960         return;
1961
1962 out_free:
1963         kfree_skb(skb);
1964 }
1965
1966 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1967 {
1968         int ct;
1969
1970         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1971                 if (mrt->vif_table[ct].dev == dev)
1972                         break;
1973         }
1974         return ct;
1975 }
1976
1977 /* "local" means that we should preserve one skb (for local delivery) */
1978 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1979                           struct net_device *dev, struct sk_buff *skb,
1980                           struct mfc_cache *c, int local)
1981 {
1982         int true_vifi = ipmr_find_vif(mrt, dev);
1983         int psend = -1;
1984         int vif, ct;
1985
1986         vif = c->_c.mfc_parent;
1987         c->_c.mfc_un.res.pkt++;
1988         c->_c.mfc_un.res.bytes += skb->len;
1989         c->_c.mfc_un.res.lastuse = jiffies;
1990
1991         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1992                 struct mfc_cache *cache_proxy;
1993
1994                 /* For an (*,G) entry, we only check that the incomming
1995                  * interface is part of the static tree.
1996                  */
1997                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1998                 if (cache_proxy &&
1999                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
2000                         goto forward;
2001         }
2002
2003         /* Wrong interface: drop packet and (maybe) send PIM assert. */
2004         if (mrt->vif_table[vif].dev != dev) {
2005                 if (rt_is_output_route(skb_rtable(skb))) {
2006                         /* It is our own packet, looped back.
2007                          * Very complicated situation...
2008                          *
2009                          * The best workaround until routing daemons will be
2010                          * fixed is not to redistribute packet, if it was
2011                          * send through wrong interface. It means, that
2012                          * multicast applications WILL NOT work for
2013                          * (S,G), which have default multicast route pointing
2014                          * to wrong oif. In any case, it is not a good
2015                          * idea to use multicasting applications on router.
2016                          */
2017                         goto dont_forward;
2018                 }
2019
2020                 c->_c.mfc_un.res.wrong_if++;
2021
2022                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
2023                     /* pimsm uses asserts, when switching from RPT to SPT,
2024                      * so that we cannot check that packet arrived on an oif.
2025                      * It is bad, but otherwise we would need to move pretty
2026                      * large chunk of pimd to kernel. Ough... --ANK
2027                      */
2028                     (mrt->mroute_do_pim ||
2029                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
2030                     time_after(jiffies,
2031                                c->_c.mfc_un.res.last_assert +
2032                                MFC_ASSERT_THRESH)) {
2033                         c->_c.mfc_un.res.last_assert = jiffies;
2034                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
2035                 }
2036                 goto dont_forward;
2037         }
2038
2039 forward:
2040         mrt->vif_table[vif].pkt_in++;
2041         mrt->vif_table[vif].bytes_in += skb->len;
2042
2043         /* Forward the frame */
2044         if (c->mfc_origin == htonl(INADDR_ANY) &&
2045             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
2046                 if (true_vifi >= 0 &&
2047                     true_vifi != c->_c.mfc_parent &&
2048                     ip_hdr(skb)->ttl >
2049                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2050                         /* It's an (*,*) entry and the packet is not coming from
2051                          * the upstream: forward the packet to the upstream
2052                          * only.
2053                          */
2054                         psend = c->_c.mfc_parent;
2055                         goto last_forward;
2056                 }
2057                 goto dont_forward;
2058         }
2059         for (ct = c->_c.mfc_un.res.maxvif - 1;
2060              ct >= c->_c.mfc_un.res.minvif; ct--) {
2061                 /* For (*,G) entry, don't forward to the incoming interface */
2062                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2063                      ct != true_vifi) &&
2064                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2065                         if (psend != -1) {
2066                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2067
2068                                 if (skb2)
2069                                         ipmr_queue_xmit(net, mrt, true_vifi,
2070                                                         skb2, c, psend);
2071                         }
2072                         psend = ct;
2073                 }
2074         }
2075 last_forward:
2076         if (psend != -1) {
2077                 if (local) {
2078                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2079
2080                         if (skb2)
2081                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2082                                                 c, psend);
2083                 } else {
2084                         ipmr_queue_xmit(net, mrt, true_vifi, skb, c, psend);
2085                         return;
2086                 }
2087         }
2088
2089 dont_forward:
2090         if (!local)
2091                 kfree_skb(skb);
2092 }
2093
2094 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2095 {
2096         struct rtable *rt = skb_rtable(skb);
2097         struct iphdr *iph = ip_hdr(skb);
2098         struct flowi4 fl4 = {
2099                 .daddr = iph->daddr,
2100                 .saddr = iph->saddr,
2101                 .flowi4_tos = RT_TOS(iph->tos),
2102                 .flowi4_oif = (rt_is_output_route(rt) ?
2103                                skb->dev->ifindex : 0),
2104                 .flowi4_iif = (rt_is_output_route(rt) ?
2105                                LOOPBACK_IFINDEX :
2106                                skb->dev->ifindex),
2107                 .flowi4_mark = skb->mark,
2108         };
2109         struct mr_table *mrt;
2110         int err;
2111
2112         err = ipmr_fib_lookup(net, &fl4, &mrt);
2113         if (err)
2114                 return ERR_PTR(err);
2115         return mrt;
2116 }
2117
2118 /* Multicast packets for forwarding arrive here
2119  * Called with rcu_read_lock();
2120  */
2121 int ip_mr_input(struct sk_buff *skb)
2122 {
2123         struct mfc_cache *cache;
2124         struct net *net = dev_net(skb->dev);
2125         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2126         struct mr_table *mrt;
2127         struct net_device *dev;
2128
2129         /* skb->dev passed in is the loX master dev for vrfs.
2130          * As there are no vifs associated with loopback devices,
2131          * get the proper interface that does have a vif associated with it.
2132          */
2133         dev = skb->dev;
2134         if (netif_is_l3_master(skb->dev)) {
2135                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2136                 if (!dev) {
2137                         kfree_skb(skb);
2138                         return -ENODEV;
2139                 }
2140         }
2141
2142         /* Packet is looped back after forward, it should not be
2143          * forwarded second time, but still can be delivered locally.
2144          */
2145         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2146                 goto dont_forward;
2147
2148         mrt = ipmr_rt_fib_lookup(net, skb);
2149         if (IS_ERR(mrt)) {
2150                 kfree_skb(skb);
2151                 return PTR_ERR(mrt);
2152         }
2153         if (!local) {
2154                 if (IPCB(skb)->opt.router_alert) {
2155                         if (ip_call_ra_chain(skb))
2156                                 return 0;
2157                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2158                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2159                          * Cisco IOS <= 11.2(8)) do not put router alert
2160                          * option to IGMP packets destined to routable
2161                          * groups. It is very bad, because it means
2162                          * that we can forward NO IGMP messages.
2163                          */
2164                         struct sock *mroute_sk;
2165
2166                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2167                         if (mroute_sk) {
2168                                 nf_reset(skb);
2169                                 raw_rcv(mroute_sk, skb);
2170                                 return 0;
2171                         }
2172                     }
2173         }
2174
2175         /* already under rcu_read_lock() */
2176         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2177         if (!cache) {
2178                 int vif = ipmr_find_vif(mrt, dev);
2179
2180                 if (vif >= 0)
2181                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2182                                                     vif);
2183         }
2184
2185         /* No usable cache entry */
2186         if (!cache) {
2187                 int vif;
2188
2189                 if (local) {
2190                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2191                         ip_local_deliver(skb);
2192                         if (!skb2)
2193                                 return -ENOBUFS;
2194                         skb = skb2;
2195                 }
2196
2197                 read_lock(&mrt_lock);
2198                 vif = ipmr_find_vif(mrt, dev);
2199                 if (vif >= 0) {
2200                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2201                         read_unlock(&mrt_lock);
2202
2203                         return err2;
2204                 }
2205                 read_unlock(&mrt_lock);
2206                 kfree_skb(skb);
2207                 return -ENODEV;
2208         }
2209
2210         read_lock(&mrt_lock);
2211         ip_mr_forward(net, mrt, dev, skb, cache, local);
2212         read_unlock(&mrt_lock);
2213
2214         if (local)
2215                 return ip_local_deliver(skb);
2216
2217         return 0;
2218
2219 dont_forward:
2220         if (local)
2221                 return ip_local_deliver(skb);
2222         kfree_skb(skb);
2223         return 0;
2224 }
2225
2226 #ifdef CONFIG_IP_PIMSM_V1
2227 /* Handle IGMP messages of PIMv1 */
2228 int pim_rcv_v1(struct sk_buff *skb)
2229 {
2230         struct igmphdr *pim;
2231         struct net *net = dev_net(skb->dev);
2232         struct mr_table *mrt;
2233
2234         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2235                 goto drop;
2236
2237         pim = igmp_hdr(skb);
2238
2239         mrt = ipmr_rt_fib_lookup(net, skb);
2240         if (IS_ERR(mrt))
2241                 goto drop;
2242         if (!mrt->mroute_do_pim ||
2243             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2244                 goto drop;
2245
2246         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2247 drop:
2248                 kfree_skb(skb);
2249         }
2250         return 0;
2251 }
2252 #endif
2253
2254 #ifdef CONFIG_IP_PIMSM_V2
2255 static int pim_rcv(struct sk_buff *skb)
2256 {
2257         struct pimreghdr *pim;
2258         struct net *net = dev_net(skb->dev);
2259         struct mr_table *mrt;
2260
2261         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2262                 goto drop;
2263
2264         pim = (struct pimreghdr *)skb_transport_header(skb);
2265         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2266             (pim->flags & PIM_NULL_REGISTER) ||
2267             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2268              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2269                 goto drop;
2270
2271         mrt = ipmr_rt_fib_lookup(net, skb);
2272         if (IS_ERR(mrt))
2273                 goto drop;
2274         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2275 drop:
2276                 kfree_skb(skb);
2277         }
2278         return 0;
2279 }
2280 #endif
2281
2282 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2283                    __be32 saddr, __be32 daddr,
2284                    struct rtmsg *rtm, u32 portid)
2285 {
2286         struct mfc_cache *cache;
2287         struct mr_table *mrt;
2288         int err;
2289
2290         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2291         if (!mrt)
2292                 return -ENOENT;
2293
2294         rcu_read_lock();
2295         cache = ipmr_cache_find(mrt, saddr, daddr);
2296         if (!cache && skb->dev) {
2297                 int vif = ipmr_find_vif(mrt, skb->dev);
2298
2299                 if (vif >= 0)
2300                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2301         }
2302         if (!cache) {
2303                 struct sk_buff *skb2;
2304                 struct iphdr *iph;
2305                 struct net_device *dev;
2306                 int vif = -1;
2307
2308                 dev = skb->dev;
2309                 read_lock(&mrt_lock);
2310                 if (dev)
2311                         vif = ipmr_find_vif(mrt, dev);
2312                 if (vif < 0) {
2313                         read_unlock(&mrt_lock);
2314                         rcu_read_unlock();
2315                         return -ENODEV;
2316                 }
2317                 skb2 = skb_clone(skb, GFP_ATOMIC);
2318                 if (!skb2) {
2319                         read_unlock(&mrt_lock);
2320                         rcu_read_unlock();
2321                         return -ENOMEM;
2322                 }
2323
2324                 NETLINK_CB(skb2).portid = portid;
2325                 skb_push(skb2, sizeof(struct iphdr));
2326                 skb_reset_network_header(skb2);
2327                 iph = ip_hdr(skb2);
2328                 iph->ihl = sizeof(struct iphdr) >> 2;
2329                 iph->saddr = saddr;
2330                 iph->daddr = daddr;
2331                 iph->version = 0;
2332                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2333                 read_unlock(&mrt_lock);
2334                 rcu_read_unlock();
2335                 return err;
2336         }
2337
2338         read_lock(&mrt_lock);
2339         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2340         read_unlock(&mrt_lock);
2341         rcu_read_unlock();
2342         return err;
2343 }
2344
2345 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2346                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2347                             int flags)
2348 {
2349         struct nlmsghdr *nlh;
2350         struct rtmsg *rtm;
2351         int err;
2352
2353         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2354         if (!nlh)
2355                 return -EMSGSIZE;
2356
2357         rtm = nlmsg_data(nlh);
2358         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2359         rtm->rtm_dst_len  = 32;
2360         rtm->rtm_src_len  = 32;
2361         rtm->rtm_tos      = 0;
2362         rtm->rtm_table    = mrt->id;
2363         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2364                 goto nla_put_failure;
2365         rtm->rtm_type     = RTN_MULTICAST;
2366         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2367         if (c->_c.mfc_flags & MFC_STATIC)
2368                 rtm->rtm_protocol = RTPROT_STATIC;
2369         else
2370                 rtm->rtm_protocol = RTPROT_MROUTED;
2371         rtm->rtm_flags    = 0;
2372
2373         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2374             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2375                 goto nla_put_failure;
2376         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2377         /* do not break the dump if cache is unresolved */
2378         if (err < 0 && err != -ENOENT)
2379                 goto nla_put_failure;
2380
2381         nlmsg_end(skb, nlh);
2382         return 0;
2383
2384 nla_put_failure:
2385         nlmsg_cancel(skb, nlh);
2386         return -EMSGSIZE;
2387 }
2388
2389 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2390                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2391                              int flags)
2392 {
2393         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2394                                 cmd, flags);
2395 }
2396
2397 static size_t mroute_msgsize(bool unresolved, int maxvif)
2398 {
2399         size_t len =
2400                 NLMSG_ALIGN(sizeof(struct rtmsg))
2401                 + nla_total_size(4)     /* RTA_TABLE */
2402                 + nla_total_size(4)     /* RTA_SRC */
2403                 + nla_total_size(4)     /* RTA_DST */
2404                 ;
2405
2406         if (!unresolved)
2407                 len = len
2408                       + nla_total_size(4)       /* RTA_IIF */
2409                       + nla_total_size(0)       /* RTA_MULTIPATH */
2410                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2411                                                 /* RTA_MFC_STATS */
2412                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2413                 ;
2414
2415         return len;
2416 }
2417
2418 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2419                                  int cmd)
2420 {
2421         struct net *net = read_pnet(&mrt->net);
2422         struct sk_buff *skb;
2423         int err = -ENOBUFS;
2424
2425         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2426                                        mrt->maxvif),
2427                         GFP_ATOMIC);
2428         if (!skb)
2429                 goto errout;
2430
2431         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2432         if (err < 0)
2433                 goto errout;
2434
2435         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2436         return;
2437
2438 errout:
2439         kfree_skb(skb);
2440         if (err < 0)
2441                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2442 }
2443
2444 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2445 {
2446         size_t len =
2447                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2448                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2449                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2450                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2451                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2452                                         /* IPMRA_CREPORT_PKT */
2453                 + nla_total_size(payloadlen)
2454                 ;
2455
2456         return len;
2457 }
2458
2459 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2460 {
2461         struct net *net = read_pnet(&mrt->net);
2462         struct nlmsghdr *nlh;
2463         struct rtgenmsg *rtgenm;
2464         struct igmpmsg *msg;
2465         struct sk_buff *skb;
2466         struct nlattr *nla;
2467         int payloadlen;
2468
2469         payloadlen = pkt->len - sizeof(struct igmpmsg);
2470         msg = (struct igmpmsg *)skb_network_header(pkt);
2471
2472         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2473         if (!skb)
2474                 goto errout;
2475
2476         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2477                         sizeof(struct rtgenmsg), 0);
2478         if (!nlh)
2479                 goto errout;
2480         rtgenm = nlmsg_data(nlh);
2481         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2482         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2483             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif) ||
2484             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2485                             msg->im_src.s_addr) ||
2486             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2487                             msg->im_dst.s_addr))
2488                 goto nla_put_failure;
2489
2490         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2491         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2492                                   nla_data(nla), payloadlen))
2493                 goto nla_put_failure;
2494
2495         nlmsg_end(skb, nlh);
2496
2497         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2498         return;
2499
2500 nla_put_failure:
2501         nlmsg_cancel(skb, nlh);
2502 errout:
2503         kfree_skb(skb);
2504         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2505 }
2506
2507 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2508                              struct netlink_ext_ack *extack)
2509 {
2510         struct net *net = sock_net(in_skb->sk);
2511         struct nlattr *tb[RTA_MAX + 1];
2512         struct sk_buff *skb = NULL;
2513         struct mfc_cache *cache;
2514         struct mr_table *mrt;
2515         struct rtmsg *rtm;
2516         __be32 src, grp;
2517         u32 tableid;
2518         int err;
2519
2520         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2521                           rtm_ipv4_policy, extack);
2522         if (err < 0)
2523                 goto errout;
2524
2525         rtm = nlmsg_data(nlh);
2526
2527         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2528         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2529         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2530
2531         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2532         if (!mrt) {
2533                 err = -ENOENT;
2534                 goto errout_free;
2535         }
2536
2537         /* entries are added/deleted only under RTNL */
2538         rcu_read_lock();
2539         cache = ipmr_cache_find(mrt, src, grp);
2540         rcu_read_unlock();
2541         if (!cache) {
2542                 err = -ENOENT;
2543                 goto errout_free;
2544         }
2545
2546         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2547         if (!skb) {
2548                 err = -ENOBUFS;
2549                 goto errout_free;
2550         }
2551
2552         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2553                                nlh->nlmsg_seq, cache,
2554                                RTM_NEWROUTE, 0);
2555         if (err < 0)
2556                 goto errout_free;
2557
2558         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2559
2560 errout:
2561         return err;
2562
2563 errout_free:
2564         kfree_skb(skb);
2565         goto errout;
2566 }
2567
2568 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2569 {
2570         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2571                                 _ipmr_fill_mroute, &mfc_unres_lock);
2572 }
2573
2574 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2575         [RTA_SRC]       = { .type = NLA_U32 },
2576         [RTA_DST]       = { .type = NLA_U32 },
2577         [RTA_IIF]       = { .type = NLA_U32 },
2578         [RTA_TABLE]     = { .type = NLA_U32 },
2579         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2580 };
2581
2582 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2583 {
2584         switch (rtm_protocol) {
2585         case RTPROT_STATIC:
2586         case RTPROT_MROUTED:
2587                 return true;
2588         }
2589         return false;
2590 }
2591
2592 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2593 {
2594         struct rtnexthop *rtnh = nla_data(nla);
2595         int remaining = nla_len(nla), vifi = 0;
2596
2597         while (rtnh_ok(rtnh, remaining)) {
2598                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2599                 if (++vifi == MAXVIFS)
2600                         break;
2601                 rtnh = rtnh_next(rtnh, &remaining);
2602         }
2603
2604         return remaining > 0 ? -EINVAL : vifi;
2605 }
2606
2607 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2608 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2609                             struct mfcctl *mfcc, int *mrtsock,
2610                             struct mr_table **mrtret,
2611                             struct netlink_ext_ack *extack)
2612 {
2613         struct net_device *dev = NULL;
2614         u32 tblid = RT_TABLE_DEFAULT;
2615         struct mr_table *mrt;
2616         struct nlattr *attr;
2617         struct rtmsg *rtm;
2618         int ret, rem;
2619
2620         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy,
2621                              extack);
2622         if (ret < 0)
2623                 goto out;
2624         rtm = nlmsg_data(nlh);
2625
2626         ret = -EINVAL;
2627         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2628             rtm->rtm_type != RTN_MULTICAST ||
2629             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2630             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2631                 goto out;
2632
2633         memset(mfcc, 0, sizeof(*mfcc));
2634         mfcc->mfcc_parent = -1;
2635         ret = 0;
2636         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2637                 switch (nla_type(attr)) {
2638                 case RTA_SRC:
2639                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2640                         break;
2641                 case RTA_DST:
2642                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2643                         break;
2644                 case RTA_IIF:
2645                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2646                         if (!dev) {
2647                                 ret = -ENODEV;
2648                                 goto out;
2649                         }
2650                         break;
2651                 case RTA_MULTIPATH:
2652                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2653                                 ret = -EINVAL;
2654                                 goto out;
2655                         }
2656                         break;
2657                 case RTA_PREFSRC:
2658                         ret = 1;
2659                         break;
2660                 case RTA_TABLE:
2661                         tblid = nla_get_u32(attr);
2662                         break;
2663                 }
2664         }
2665         mrt = ipmr_get_table(net, tblid);
2666         if (!mrt) {
2667                 ret = -ENOENT;
2668                 goto out;
2669         }
2670         *mrtret = mrt;
2671         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2672         if (dev)
2673                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2674
2675 out:
2676         return ret;
2677 }
2678
2679 /* takes care of both newroute and delroute */
2680 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2681                           struct netlink_ext_ack *extack)
2682 {
2683         struct net *net = sock_net(skb->sk);
2684         int ret, mrtsock, parent;
2685         struct mr_table *tbl;
2686         struct mfcctl mfcc;
2687
2688         mrtsock = 0;
2689         tbl = NULL;
2690         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2691         if (ret < 0)
2692                 return ret;
2693
2694         parent = ret ? mfcc.mfcc_parent : -1;
2695         if (nlh->nlmsg_type == RTM_NEWROUTE)
2696                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2697         else
2698                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2699 }
2700
2701 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2702 {
2703         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2704
2705         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2706             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2707             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2708                         mrt->mroute_reg_vif_num) ||
2709             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2710                        mrt->mroute_do_assert) ||
2711             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim))
2712                 return false;
2713
2714         return true;
2715 }
2716
2717 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2718 {
2719         struct nlattr *vif_nest;
2720         struct vif_device *vif;
2721
2722         /* if the VIF doesn't exist just continue */
2723         if (!VIF_EXISTS(mrt, vifid))
2724                 return true;
2725
2726         vif = &mrt->vif_table[vifid];
2727         vif_nest = nla_nest_start(skb, IPMRA_VIF);
2728         if (!vif_nest)
2729                 return false;
2730         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2731             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2732             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2733             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2734                               IPMRA_VIFA_PAD) ||
2735             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2736                               IPMRA_VIFA_PAD) ||
2737             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2738                               IPMRA_VIFA_PAD) ||
2739             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2740                               IPMRA_VIFA_PAD) ||
2741             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2742             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2743                 nla_nest_cancel(skb, vif_nest);
2744                 return false;
2745         }
2746         nla_nest_end(skb, vif_nest);
2747
2748         return true;
2749 }
2750
2751 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2752 {
2753         struct net *net = sock_net(skb->sk);
2754         struct nlmsghdr *nlh = NULL;
2755         unsigned int t = 0, s_t;
2756         unsigned int e = 0, s_e;
2757         struct mr_table *mrt;
2758
2759         s_t = cb->args[0];
2760         s_e = cb->args[1];
2761
2762         ipmr_for_each_table(mrt, net) {
2763                 struct nlattr *vifs, *af;
2764                 struct ifinfomsg *hdr;
2765                 u32 i;
2766
2767                 if (t < s_t)
2768                         goto skip_table;
2769                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2770                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2771                                 sizeof(*hdr), NLM_F_MULTI);
2772                 if (!nlh)
2773                         break;
2774
2775                 hdr = nlmsg_data(nlh);
2776                 memset(hdr, 0, sizeof(*hdr));
2777                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2778
2779                 af = nla_nest_start(skb, IFLA_AF_SPEC);
2780                 if (!af) {
2781                         nlmsg_cancel(skb, nlh);
2782                         goto out;
2783                 }
2784
2785                 if (!ipmr_fill_table(mrt, skb)) {
2786                         nlmsg_cancel(skb, nlh);
2787                         goto out;
2788                 }
2789
2790                 vifs = nla_nest_start(skb, IPMRA_TABLE_VIFS);
2791                 if (!vifs) {
2792                         nla_nest_end(skb, af);
2793                         nlmsg_end(skb, nlh);
2794                         goto out;
2795                 }
2796                 for (i = 0; i < mrt->maxvif; i++) {
2797                         if (e < s_e)
2798                                 goto skip_entry;
2799                         if (!ipmr_fill_vif(mrt, i, skb)) {
2800                                 nla_nest_end(skb, vifs);
2801                                 nla_nest_end(skb, af);
2802                                 nlmsg_end(skb, nlh);
2803                                 goto out;
2804                         }
2805 skip_entry:
2806                         e++;
2807                 }
2808                 s_e = 0;
2809                 e = 0;
2810                 nla_nest_end(skb, vifs);
2811                 nla_nest_end(skb, af);
2812                 nlmsg_end(skb, nlh);
2813 skip_table:
2814                 t++;
2815         }
2816
2817 out:
2818         cb->args[1] = e;
2819         cb->args[0] = t;
2820
2821         return skb->len;
2822 }
2823
2824 #ifdef CONFIG_PROC_FS
2825 /* The /proc interfaces to multicast routing :
2826  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2827  */
2828
2829 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2830         __acquires(mrt_lock)
2831 {
2832         struct mr_vif_iter *iter = seq->private;
2833         struct net *net = seq_file_net(seq);
2834         struct mr_table *mrt;
2835
2836         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2837         if (!mrt)
2838                 return ERR_PTR(-ENOENT);
2839
2840         iter->mrt = mrt;
2841
2842         read_lock(&mrt_lock);
2843         return mr_vif_seq_start(seq, pos);
2844 }
2845
2846 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2847         __releases(mrt_lock)
2848 {
2849         read_unlock(&mrt_lock);
2850 }
2851
2852 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2853 {
2854         struct mr_vif_iter *iter = seq->private;
2855         struct mr_table *mrt = iter->mrt;
2856
2857         if (v == SEQ_START_TOKEN) {
2858                 seq_puts(seq,
2859                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2860         } else {
2861                 const struct vif_device *vif = v;
2862                 const char *name =  vif->dev ?
2863                                     vif->dev->name : "none";
2864
2865                 seq_printf(seq,
2866                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2867                            vif - mrt->vif_table,
2868                            name, vif->bytes_in, vif->pkt_in,
2869                            vif->bytes_out, vif->pkt_out,
2870                            vif->flags, vif->local, vif->remote);
2871         }
2872         return 0;
2873 }
2874
2875 static const struct seq_operations ipmr_vif_seq_ops = {
2876         .start = ipmr_vif_seq_start,
2877         .next  = mr_vif_seq_next,
2878         .stop  = ipmr_vif_seq_stop,
2879         .show  = ipmr_vif_seq_show,
2880 };
2881
2882 static int ipmr_vif_open(struct inode *inode, struct file *file)
2883 {
2884         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2885                             sizeof(struct mr_vif_iter));
2886 }
2887
2888 static const struct file_operations ipmr_vif_fops = {
2889         .open    = ipmr_vif_open,
2890         .read    = seq_read,
2891         .llseek  = seq_lseek,
2892         .release = seq_release_net,
2893 };
2894
2895 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2896 {
2897         struct net *net = seq_file_net(seq);
2898         struct mr_table *mrt;
2899
2900         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2901         if (!mrt)
2902                 return ERR_PTR(-ENOENT);
2903
2904         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2905 }
2906
2907 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2908 {
2909         int n;
2910
2911         if (v == SEQ_START_TOKEN) {
2912                 seq_puts(seq,
2913                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2914         } else {
2915                 const struct mfc_cache *mfc = v;
2916                 const struct mr_mfc_iter *it = seq->private;
2917                 const struct mr_table *mrt = it->mrt;
2918
2919                 seq_printf(seq, "%08X %08X %-3hd",
2920                            (__force u32) mfc->mfc_mcastgrp,
2921                            (__force u32) mfc->mfc_origin,
2922                            mfc->_c.mfc_parent);
2923
2924                 if (it->cache != &mrt->mfc_unres_queue) {
2925                         seq_printf(seq, " %8lu %8lu %8lu",
2926                                    mfc->_c.mfc_un.res.pkt,
2927                                    mfc->_c.mfc_un.res.bytes,
2928                                    mfc->_c.mfc_un.res.wrong_if);
2929                         for (n = mfc->_c.mfc_un.res.minvif;
2930                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2931                                 if (VIF_EXISTS(mrt, n) &&
2932                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2933                                         seq_printf(seq,
2934                                            " %2d:%-3d",
2935                                            n, mfc->_c.mfc_un.res.ttls[n]);
2936                         }
2937                 } else {
2938                         /* unresolved mfc_caches don't contain
2939                          * pkt, bytes and wrong_if values
2940                          */
2941                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2942                 }
2943                 seq_putc(seq, '\n');
2944         }
2945         return 0;
2946 }
2947
2948 static const struct seq_operations ipmr_mfc_seq_ops = {
2949         .start = ipmr_mfc_seq_start,
2950         .next  = mr_mfc_seq_next,
2951         .stop  = mr_mfc_seq_stop,
2952         .show  = ipmr_mfc_seq_show,
2953 };
2954
2955 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2956 {
2957         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2958                             sizeof(struct mr_mfc_iter));
2959 }
2960
2961 static const struct file_operations ipmr_mfc_fops = {
2962         .open    = ipmr_mfc_open,
2963         .read    = seq_read,
2964         .llseek  = seq_lseek,
2965         .release = seq_release_net,
2966 };
2967 #endif
2968
2969 #ifdef CONFIG_IP_PIMSM_V2
2970 static const struct net_protocol pim_protocol = {
2971         .handler        =       pim_rcv,
2972         .netns_ok       =       1,
2973 };
2974 #endif
2975
2976 static unsigned int ipmr_seq_read(struct net *net)
2977 {
2978         ASSERT_RTNL();
2979
2980         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
2981 }
2982
2983 static int ipmr_dump(struct net *net, struct notifier_block *nb)
2984 {
2985         struct mr_table *mrt;
2986         int err;
2987
2988         err = ipmr_rules_dump(net, nb);
2989         if (err)
2990                 return err;
2991
2992         ipmr_for_each_table(mrt, net) {
2993                 struct vif_device *v = &mrt->vif_table[0];
2994                 struct mr_mfc *mfc;
2995                 int vifi;
2996
2997                 /* Notifiy on table VIF entries */
2998                 read_lock(&mrt_lock);
2999                 for (vifi = 0; vifi < mrt->maxvif; vifi++, v++) {
3000                         if (!v->dev)
3001                                 continue;
3002
3003                         call_ipmr_vif_entry_notifier(nb, net, FIB_EVENT_VIF_ADD,
3004                                                      v, vifi, mrt->id);
3005                 }
3006                 read_unlock(&mrt_lock);
3007
3008                 /* Notify on table MFC entries */
3009                 list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list)
3010                         call_ipmr_mfc_entry_notifier(nb, net,
3011                                                      FIB_EVENT_ENTRY_ADD,
3012                                                      (struct mfc_cache *)mfc,
3013                                                      mrt->id);
3014         }
3015
3016         return 0;
3017 }
3018
3019 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3020         .family         = RTNL_FAMILY_IPMR,
3021         .fib_seq_read   = ipmr_seq_read,
3022         .fib_dump       = ipmr_dump,
3023         .owner          = THIS_MODULE,
3024 };
3025
3026 static int __net_init ipmr_notifier_init(struct net *net)
3027 {
3028         struct fib_notifier_ops *ops;
3029
3030         net->ipv4.ipmr_seq = 0;
3031
3032         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3033         if (IS_ERR(ops))
3034                 return PTR_ERR(ops);
3035         net->ipv4.ipmr_notifier_ops = ops;
3036
3037         return 0;
3038 }
3039
3040 static void __net_exit ipmr_notifier_exit(struct net *net)
3041 {
3042         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3043         net->ipv4.ipmr_notifier_ops = NULL;
3044 }
3045
3046 /* Setup for IP multicast routing */
3047 static int __net_init ipmr_net_init(struct net *net)
3048 {
3049         int err;
3050
3051         err = ipmr_notifier_init(net);
3052         if (err)
3053                 goto ipmr_notifier_fail;
3054
3055         err = ipmr_rules_init(net);
3056         if (err < 0)
3057                 goto ipmr_rules_fail;
3058
3059 #ifdef CONFIG_PROC_FS
3060         err = -ENOMEM;
3061         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
3062                 goto proc_vif_fail;
3063         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
3064                 goto proc_cache_fail;
3065 #endif
3066         return 0;
3067
3068 #ifdef CONFIG_PROC_FS
3069 proc_cache_fail:
3070         remove_proc_entry("ip_mr_vif", net->proc_net);
3071 proc_vif_fail:
3072         ipmr_rules_exit(net);
3073 #endif
3074 ipmr_rules_fail:
3075         ipmr_notifier_exit(net);
3076 ipmr_notifier_fail:
3077         return err;
3078 }
3079
3080 static void __net_exit ipmr_net_exit(struct net *net)
3081 {
3082 #ifdef CONFIG_PROC_FS
3083         remove_proc_entry("ip_mr_cache", net->proc_net);
3084         remove_proc_entry("ip_mr_vif", net->proc_net);
3085 #endif
3086         ipmr_notifier_exit(net);
3087         ipmr_rules_exit(net);
3088 }
3089
3090 static struct pernet_operations ipmr_net_ops = {
3091         .init = ipmr_net_init,
3092         .exit = ipmr_net_exit,
3093         .async = true,
3094 };
3095
3096 int __init ip_mr_init(void)
3097 {
3098         int err;
3099
3100         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3101                                        sizeof(struct mfc_cache),
3102                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3103                                        NULL);
3104
3105         err = register_pernet_subsys(&ipmr_net_ops);
3106         if (err)
3107                 goto reg_pernet_fail;
3108
3109         err = register_netdevice_notifier(&ip_mr_notifier);
3110         if (err)
3111                 goto reg_notif_fail;
3112 #ifdef CONFIG_IP_PIMSM_V2
3113         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3114                 pr_err("%s: can't add PIM protocol\n", __func__);
3115                 err = -EAGAIN;
3116                 goto add_proto_fail;
3117         }
3118 #endif
3119         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3120                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3121         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3122                       ipmr_rtm_route, NULL, 0);
3123         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3124                       ipmr_rtm_route, NULL, 0);
3125
3126         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3127                       NULL, ipmr_rtm_dumplink, 0);
3128         return 0;
3129
3130 #ifdef CONFIG_IP_PIMSM_V2
3131 add_proto_fail:
3132         unregister_netdevice_notifier(&ip_mr_notifier);
3133 #endif
3134 reg_notif_fail:
3135         unregister_pernet_subsys(&ipmr_net_ops);
3136 reg_pernet_fail:
3137         kmem_cache_destroy(mrt_cachep);
3138         return err;
3139 }