net/ipv4/ipmr.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      IP multicast routing support for mrouted 3.6/3.8
   4  *
   5  *              (c) 1995 Alan Cox, <[email protected]>
   6  *        Linux Consultancy and Custom Driver Development
   7  *
   8  *      Fixes:
   9  *      Michael Chastain        :       Incorrect size of copying.
  10  *      Alan Cox                :       Added the cache manager code
  11  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  12  *      Mike McLagan            :       Routing by source
  13  *      Malcolm Beattie         :       Buffer handling fixes.
  14  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  15  *      SVR Anand               :       Fixed several multicast bugs and problems.
  16  *      Alexey Kuznetsov        :       Status, optimisations and more.
  17  *      Brad Parker             :       Better behaviour on mrouted upcall
  18  *                                      overflow.
  19  *      Carlos Picoto           :       PIMv1 Support
  20  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  21  *                                      Relax this requirement to work with older peers.
  22  */
  23
  24 #include <linux/uaccess.h>
  25 #include <linux/types.h>
  26 #include <linux/cache.h>
  27 #include <linux/capability.h>
  28 #include <linux/errno.h>
  29 #include <linux/mm.h>
  30 #include <linux/kernel.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/socket.h>
  34 #include <linux/in.h>
  35 #include <linux/inet.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/inetdevice.h>
  38 #include <linux/igmp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/mroute.h>
  42 #include <linux/init.h>
  43 #include <linux/if_ether.h>
  44 #include <linux/slab.h>
  45 #include <net/net_namespace.h>
  46 #include <net/ip.h>
  47 #include <net/protocol.h>
  48 #include <linux/skbuff.h>
  49 #include <net/route.h>
  50 #include <net/icmp.h>
  51 #include <net/udp.h>
  52 #include <net/raw.h>
  53 #include <linux/notifier.h>
  54 #include <linux/if_arp.h>
  55 #include <linux/netfilter_ipv4.h>
  56 #include <linux/compat.h>
  57 #include <linux/export.h>
  58 #include <linux/rhashtable.h>
  59 #include <net/ip_tunnels.h>
  60 #include <net/checksum.h>
  61 #include <net/netlink.h>
  62 #include <net/fib_rules.h>
  63 #include <linux/netconf.h>
  64 #include <net/rtnh.h>
  65
  66 #include <linux/nospec.h>
  67
  68 struct ipmr_rule {
  69         struct fib_rule         common;
  70 };
  71
  72 struct ipmr_result {
  73         struct mr_table         *mrt;
  74 };
  75
  76 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  77  * Note that the changes are semaphored via rtnl_lock.
  78  */
  79
  80 static DEFINE_RWLOCK(mrt_lock);
  81
  82 /* Multicast router control variables */
  83
  84 /* Special spinlock for queue of unresolved entries */
  85 static DEFINE_SPINLOCK(mfc_unres_lock);
  86
  87 /* We return to original Alan's scheme. Hash table of resolved
  88  * entries is changed only in process context and protected
  89  * with weak lock mrt_lock. Queue of unresolved entries is protected
  90  * with strong spinlock mfc_unres_lock.
  91  *
  92  * In this case data path is free of exclusive locks at all.
  93  */
  94
  95 static struct kmem_cache *mrt_cachep __ro_after_init;
  96
  97 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
  98 static void ipmr_free_table(struct mr_table *mrt);
  99
 100 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 101                           struct net_device *dev, struct sk_buff *skb,
 102                           struct mfc_cache *cache, int local);
 103 static int ipmr_cache_report(struct mr_table *mrt,
 104                              struct sk_buff *pkt, vifi_t vifi, int assert);
 105 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 106                                  int cmd);
 107 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt);
 108 static void mroute_clean_tables(struct mr_table *mrt, int flags);
 109 static void ipmr_expire_process(struct timer_list *t);
 110
 111 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 112 #define ipmr_for_each_table(mrt, net)                                   \
 113         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list,        \
 114                                 lockdep_rtnl_is_held() ||               \
 115                                 list_empty(&net->ipv4.mr_tables))
 116
 117 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 118                                            struct mr_table *mrt)
 119 {
 120         struct mr_table *ret;
 121
 122         if (!mrt)
 123                 ret = list_entry_rcu(net->ipv4.mr_tables.next,
 124                                      struct mr_table, list);
 125         else
 126                 ret = list_entry_rcu(mrt->list.next,
 127                                      struct mr_table, list);
 128
 129         if (&ret->list == &net->ipv4.mr_tables)
 130                 return NULL;
 131         return ret;
 132 }
 133
 134 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 135 {
 136         struct mr_table *mrt;
 137
 138         ipmr_for_each_table(mrt, net) {
 139                 if (mrt->id == id)
 140                         return mrt;
 141         }
 142         return NULL;
 143 }
 144
 145 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 146                            struct mr_table **mrt)
 147 {
 148         int err;
 149         struct ipmr_result res;
 150         struct fib_lookup_arg arg = {
 151                 .result = &res,
 152                 .flags = FIB_LOOKUP_NOREF,
 153         };
 154
 155         /* update flow if oif or iif point to device enslaved to l3mdev */
 156         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 157
 158         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 159                                flowi4_to_flowi(flp4), 0, &arg);
 160         if (err < 0)
 161                 return err;
 162         *mrt = res.mrt;
 163         return 0;
 164 }
 165
 166 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 167                             int flags, struct fib_lookup_arg *arg)
 168 {
 169         struct ipmr_result *res = arg->result;
 170         struct mr_table *mrt;
 171
 172         switch (rule->action) {
 173         case FR_ACT_TO_TBL:
 174                 break;
 175         case FR_ACT_UNREACHABLE:
 176                 return -ENETUNREACH;
 177         case FR_ACT_PROHIBIT:
 178                 return -EACCES;
 179         case FR_ACT_BLACKHOLE:
 180         default:
 181                 return -EINVAL;
 182         }
 183
 184         arg->table = fib_rule_get_table(rule, arg);
 185
 186         mrt = ipmr_get_table(rule->fr_net, arg->table);
 187         if (!mrt)
 188                 return -EAGAIN;
 189         res->mrt = mrt;
 190         return 0;
 191 }
 192
 193 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 194 {
 195         return 1;
 196 }
 197
 198 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 199         FRA_GENERIC_POLICY,
 200 };
 201
 202 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 203                                struct fib_rule_hdr *frh, struct nlattr **tb,
 204                                struct netlink_ext_ack *extack)
 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                            struct netlink_ext_ack *extack)
 285 {
 286         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR, extack);
 287 }
 288
 289 static unsigned int ipmr_rules_seq_read(struct net *net)
 290 {
 291         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 292 }
 293
 294 bool ipmr_rule_default(const struct fib_rule *rule)
 295 {
 296         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 297 }
 298 EXPORT_SYMBOL(ipmr_rule_default);
 299 #else
 300 #define ipmr_for_each_table(mrt, net) \
 301         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 302
 303 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 304                                            struct mr_table *mrt)
 305 {
 306         if (!mrt)
 307                 return net->ipv4.mrt;
 308         return NULL;
 309 }
 310
 311 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 312 {
 313         return net->ipv4.mrt;
 314 }
 315
 316 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 317                            struct mr_table **mrt)
 318 {
 319         *mrt = net->ipv4.mrt;
 320         return 0;
 321 }
 322
 323 static int __net_init ipmr_rules_init(struct net *net)
 324 {
 325         struct mr_table *mrt;
 326
 327         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 328         if (IS_ERR(mrt))
 329                 return PTR_ERR(mrt);
 330         net->ipv4.mrt = mrt;
 331         return 0;
 332 }
 333
 334 static void __net_exit ipmr_rules_exit(struct net *net)
 335 {
 336         rtnl_lock();
 337         ipmr_free_table(net->ipv4.mrt);
 338         net->ipv4.mrt = NULL;
 339         rtnl_unlock();
 340 }
 341
 342 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 343                            struct netlink_ext_ack *extack)
 344 {
 345         return 0;
 346 }
 347
 348 static unsigned int ipmr_rules_seq_read(struct net *net)
 349 {
 350         return 0;
 351 }
 352
 353 bool ipmr_rule_default(const struct fib_rule *rule)
 354 {
 355         return true;
 356 }
 357 EXPORT_SYMBOL(ipmr_rule_default);
 358 #endif
 359
 360 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 361                                 const void *ptr)
 362 {
 363         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 364         struct mfc_cache *c = (struct mfc_cache *)ptr;
 365
 366         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 367                cmparg->mfc_origin != c->mfc_origin;
 368 }
 369
 370 static const struct rhashtable_params ipmr_rht_params = {
 371         .head_offset = offsetof(struct mr_mfc, mnode),
 372         .key_offset = offsetof(struct mfc_cache, cmparg),
 373         .key_len = sizeof(struct mfc_cache_cmp_arg),
 374         .nelem_hint = 3,
 375         .obj_cmpfn = ipmr_hash_cmp,
 376         .automatic_shrinking = true,
 377 };
 378
 379 static void ipmr_new_table_set(struct mr_table *mrt,
 380                                struct net *net)
 381 {
 382 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 383         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 384 #endif
 385 }
 386
 387 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 388         .mfc_mcastgrp = htonl(INADDR_ANY),
 389         .mfc_origin = htonl(INADDR_ANY),
 390 };
 391
 392 static struct mr_table_ops ipmr_mr_table_ops = {
 393         .rht_params = &ipmr_rht_params,
 394         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 395 };
 396
 397 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 398 {
 399         struct mr_table *mrt;
 400
 401         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 402         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 403                 return ERR_PTR(-EINVAL);
 404
 405         mrt = ipmr_get_table(net, id);
 406         if (mrt)
 407                 return mrt;
 408
 409         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 410                               ipmr_expire_process, ipmr_new_table_set);
 411 }
 412
 413 static void ipmr_free_table(struct mr_table *mrt)
 414 {
 415         del_timer_sync(&mrt->ipmr_expire_timer);
 416         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC |
 417                                  MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC);
 418         rhltable_destroy(&mrt->mfc_hash);
 419         kfree(mrt);
 420 }
 421
 422 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 423
 424 /* Initialize ipmr pimreg/tunnel in_device */
 425 static bool ipmr_init_vif_indev(const struct net_device *dev)
 426 {
 427         struct in_device *in_dev;
 428
 429         ASSERT_RTNL();
 430
 431         in_dev = __in_dev_get_rtnl(dev);
 432         if (!in_dev)
 433                 return false;
 434         ipv4_devconf_setall(in_dev);
 435         neigh_parms_data_state_setall(in_dev->arp_parms);
 436         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 437
 438         return true;
 439 }
 440
 441 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 442 {
 443         struct net_device *tunnel_dev, *new_dev;
 444         struct ip_tunnel_parm p = { };
 445         int err;
 446
 447         tunnel_dev = __dev_get_by_name(net, "tunl0");
 448         if (!tunnel_dev)
 449                 goto out;
 450
 451         p.iph.daddr = v->vifc_rmt_addr.s_addr;
 452         p.iph.saddr = v->vifc_lcl_addr.s_addr;
 453         p.iph.version = 4;
 454         p.iph.ihl = 5;
 455         p.iph.protocol = IPPROTO_IPIP;
 456         sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 457
 458         if (!tunnel_dev->netdev_ops->ndo_tunnel_ctl)
 459                 goto out;
 460         err = tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 461                         SIOCADDTUNNEL);
 462         if (err)
 463                 goto out;
 464
 465         new_dev = __dev_get_by_name(net, p.name);
 466         if (!new_dev)
 467                 goto out;
 468
 469         new_dev->flags |= IFF_MULTICAST;
 470         if (!ipmr_init_vif_indev(new_dev))
 471                 goto out_unregister;
 472         if (dev_open(new_dev, NULL))
 473                 goto out_unregister;
 474         dev_hold(new_dev);
 475         err = dev_set_allmulti(new_dev, 1);
 476         if (err) {
 477                 dev_close(new_dev);
 478                 tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 479                                 SIOCDELTUNNEL);
 480                 dev_put(new_dev);
 481                 new_dev = ERR_PTR(err);
 482         }
 483         return new_dev;
 484
 485 out_unregister:
 486         unregister_netdevice(new_dev);
 487 out:
 488         return ERR_PTR(-ENOBUFS);
 489 }
 490
 491 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 492 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 493 {
 494         struct net *net = dev_net(dev);
 495         struct mr_table *mrt;
 496         struct flowi4 fl4 = {
 497                 .flowi4_oif     = dev->ifindex,
 498                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 499                 .flowi4_mark    = skb->mark,
 500         };
 501         int err;
 502
 503         err = ipmr_fib_lookup(net, &fl4, &mrt);
 504         if (err < 0) {
 505                 kfree_skb(skb);
 506                 return err;
 507         }
 508
 509         read_lock(&mrt_lock);
 510         dev->stats.tx_bytes += skb->len;
 511         dev->stats.tx_packets++;
 512         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 513         read_unlock(&mrt_lock);
 514         kfree_skb(skb);
 515         return NETDEV_TX_OK;
 516 }
 517
 518 static int reg_vif_get_iflink(const struct net_device *dev)
 519 {
 520         return 0;
 521 }
 522
 523 static const struct net_device_ops reg_vif_netdev_ops = {
 524         .ndo_start_xmit = reg_vif_xmit,
 525         .ndo_get_iflink = reg_vif_get_iflink,
 526 };
 527
 528 static void reg_vif_setup(struct net_device *dev)
 529 {
 530         dev->type               = ARPHRD_PIMREG;
 531         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 532         dev->flags              = IFF_NOARP;
 533         dev->netdev_ops         = &reg_vif_netdev_ops;
 534         dev->needs_free_netdev  = true;
 535         dev->features           |= NETIF_F_NETNS_LOCAL;
 536 }
 537
 538 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 539 {
 540         struct net_device *dev;
 541         char name[IFNAMSIZ];
 542
 543         if (mrt->id == RT_TABLE_DEFAULT)
 544                 sprintf(name, "pimreg");
 545         else
 546                 sprintf(name, "pimreg%u", mrt->id);
 547
 548         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 549
 550         if (!dev)
 551                 return NULL;
 552
 553         dev_net_set(dev, net);
 554
 555         if (register_netdevice(dev)) {
 556                 free_netdev(dev);
 557                 return NULL;
 558         }
 559
 560         if (!ipmr_init_vif_indev(dev))
 561                 goto failure;
 562         if (dev_open(dev, NULL))
 563                 goto failure;
 564
 565         dev_hold(dev);
 566
 567         return dev;
 568
 569 failure:
 570         unregister_netdevice(dev);
 571         return NULL;
 572 }
 573
 574 /* called with rcu_read_lock() */
 575 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 576                      unsigned int pimlen)
 577 {
 578         struct net_device *reg_dev = NULL;
 579         struct iphdr *encap;
 580
 581         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 582         /* Check that:
 583          * a. packet is really sent to a multicast group
 584          * b. packet is not a NULL-REGISTER
 585          * c. packet is not truncated
 586          */
 587         if (!ipv4_is_multicast(encap->daddr) ||
 588             encap->tot_len == 0 ||
 589             ntohs(encap->tot_len) + pimlen > skb->len)
 590                 return 1;
 591
 592         read_lock(&mrt_lock);
 593         if (mrt->mroute_reg_vif_num >= 0)
 594                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 595         read_unlock(&mrt_lock);
 596
 597         if (!reg_dev)
 598                 return 1;
 599
 600         skb->mac_header = skb->network_header;
 601         skb_pull(skb, (u8 *)encap - skb->data);
 602         skb_reset_network_header(skb);
 603         skb->protocol = htons(ETH_P_IP);
 604         skb->ip_summed = CHECKSUM_NONE;
 605
 606         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 607
 608         netif_rx(skb);
 609
 610         return NET_RX_SUCCESS;
 611 }
 612 #else
 613 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 614 {
 615         return NULL;
 616 }
 617 #endif
 618
 619 static int call_ipmr_vif_entry_notifiers(struct net *net,
 620                                          enum fib_event_type event_type,
 621                                          struct vif_device *vif,
 622                                          vifi_t vif_index, u32 tb_id)
 623 {
 624         return mr_call_vif_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 625                                      vif, vif_index, tb_id,
 626                                      &net->ipv4.ipmr_seq);
 627 }
 628
 629 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 630                                          enum fib_event_type event_type,
 631                                          struct mfc_cache *mfc, u32 tb_id)
 632 {
 633         return mr_call_mfc_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 634                                      &mfc->_c, tb_id, &net->ipv4.ipmr_seq);
 635 }
 636
 637 /**
 638  *      vif_delete - Delete a VIF entry
 639  *      @notify: Set to 1, if the caller is a notifier_call
 640  */
 641 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 642                       struct list_head *head)
 643 {
 644         struct net *net = read_pnet(&mrt->net);
 645         struct vif_device *v;
 646         struct net_device *dev;
 647         struct in_device *in_dev;
 648
 649         if (vifi < 0 || vifi >= mrt->maxvif)
 650                 return -EADDRNOTAVAIL;
 651
 652         v = &mrt->vif_table[vifi];
 653
 654         if (VIF_EXISTS(mrt, vifi))
 655                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 656                                               mrt->id);
 657
 658         write_lock_bh(&mrt_lock);
 659         dev = v->dev;
 660         v->dev = NULL;
 661
 662         if (!dev) {
 663                 write_unlock_bh(&mrt_lock);
 664                 return -EADDRNOTAVAIL;
 665         }
 666
 667         if (vifi == mrt->mroute_reg_vif_num)
 668                 mrt->mroute_reg_vif_num = -1;
 669
 670         if (vifi + 1 == mrt->maxvif) {
 671                 int tmp;
 672
 673                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 674                         if (VIF_EXISTS(mrt, tmp))
 675                                 break;
 676                 }
 677                 mrt->maxvif = tmp+1;
 678         }
 679
 680         write_unlock_bh(&mrt_lock);
 681
 682         dev_set_allmulti(dev, -1);
 683
 684         in_dev = __in_dev_get_rtnl(dev);
 685         if (in_dev) {
 686                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 687                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 688                                             NETCONFA_MC_FORWARDING,
 689                                             dev->ifindex, &in_dev->cnf);
 690                 ip_rt_multicast_event(in_dev);
 691         }
 692
 693         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 694                 unregister_netdevice_queue(dev, head);
 695
 696         dev_put(dev);
 697         return 0;
 698 }
 699
 700 static void ipmr_cache_free_rcu(struct rcu_head *head)
 701 {
 702         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 703
 704         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 705 }
 706
 707 static void ipmr_cache_free(struct mfc_cache *c)
 708 {
 709         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 710 }
 711
 712 /* Destroy an unresolved cache entry, killing queued skbs
 713  * and reporting error to netlink readers.
 714  */
 715 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 716 {
 717         struct net *net = read_pnet(&mrt->net);
 718         struct sk_buff *skb;
 719         struct nlmsgerr *e;
 720
 721         atomic_dec(&mrt->cache_resolve_queue_len);
 722
 723         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 724                 if (ip_hdr(skb)->version == 0) {
 725                         struct nlmsghdr *nlh = skb_pull(skb,
 726                                                         sizeof(struct iphdr));
 727                         nlh->nlmsg_type = NLMSG_ERROR;
 728                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 729                         skb_trim(skb, nlh->nlmsg_len);
 730                         e = nlmsg_data(nlh);
 731                         e->error = -ETIMEDOUT;
 732                         memset(&e->msg, 0, sizeof(e->msg));
 733
 734                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 735                 } else {
 736                         kfree_skb(skb);
 737                 }
 738         }
 739
 740         ipmr_cache_free(c);
 741 }
 742
 743 /* Timer process for the unresolved queue. */
 744 static void ipmr_expire_process(struct timer_list *t)
 745 {
 746         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 747         struct mr_mfc *c, *next;
 748         unsigned long expires;
 749         unsigned long now;
 750
 751         if (!spin_trylock(&mfc_unres_lock)) {
 752                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 753                 return;
 754         }
 755
 756         if (list_empty(&mrt->mfc_unres_queue))
 757                 goto out;
 758
 759         now = jiffies;
 760         expires = 10*HZ;
 761
 762         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 763                 if (time_after(c->mfc_un.unres.expires, now)) {
 764                         unsigned long interval = c->mfc_un.unres.expires - now;
 765                         if (interval < expires)
 766                                 expires = interval;
 767                         continue;
 768                 }
 769
 770                 list_del(&c->list);
 771                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 772                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 773         }
 774
 775         if (!list_empty(&mrt->mfc_unres_queue))
 776                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 777
 778 out:
 779         spin_unlock(&mfc_unres_lock);
 780 }
 781
 782 /* Fill oifs list. It is called under write locked mrt_lock. */
 783 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 784                                    unsigned char *ttls)
 785 {
 786         int vifi;
 787
 788         cache->mfc_un.res.minvif = MAXVIFS;
 789         cache->mfc_un.res.maxvif = 0;
 790         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 791
 792         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 793                 if (VIF_EXISTS(mrt, vifi) &&
 794                     ttls[vifi] && ttls[vifi] < 255) {
 795                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 796                         if (cache->mfc_un.res.minvif > vifi)
 797                                 cache->mfc_un.res.minvif = vifi;
 798                         if (cache->mfc_un.res.maxvif <= vifi)
 799                                 cache->mfc_un.res.maxvif = vifi + 1;
 800                 }
 801         }
 802         cache->mfc_un.res.lastuse = jiffies;
 803 }
 804
 805 static int vif_add(struct net *net, struct mr_table *mrt,
 806                    struct vifctl *vifc, int mrtsock)
 807 {
 808         struct netdev_phys_item_id ppid = { };
 809         int vifi = vifc->vifc_vifi;
 810         struct vif_device *v = &mrt->vif_table[vifi];
 811         struct net_device *dev;
 812         struct in_device *in_dev;
 813         int err;
 814
 815         /* Is vif busy ? */
 816         if (VIF_EXISTS(mrt, vifi))
 817                 return -EADDRINUSE;
 818
 819         switch (vifc->vifc_flags) {
 820         case VIFF_REGISTER:
 821                 if (!ipmr_pimsm_enabled())
 822                         return -EINVAL;
 823                 /* Special Purpose VIF in PIM
 824                  * All the packets will be sent to the daemon
 825                  */
 826                 if (mrt->mroute_reg_vif_num >= 0)
 827                         return -EADDRINUSE;
 828                 dev = ipmr_reg_vif(net, mrt);
 829                 if (!dev)
 830                         return -ENOBUFS;
 831                 err = dev_set_allmulti(dev, 1);
 832                 if (err) {
 833                         unregister_netdevice(dev);
 834                         dev_put(dev);
 835                         return err;
 836                 }
 837                 break;
 838         case VIFF_TUNNEL:
 839                 dev = ipmr_new_tunnel(net, vifc);
 840                 if (IS_ERR(dev))
 841                         return PTR_ERR(dev);
 842                 break;
 843         case VIFF_USE_IFINDEX:
 844         case 0:
 845                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 846                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 847                         if (dev && !__in_dev_get_rtnl(dev)) {
 848                                 dev_put(dev);
 849                                 return -EADDRNOTAVAIL;
 850                         }
 851                 } else {
 852                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 853                 }
 854                 if (!dev)
 855                         return -EADDRNOTAVAIL;
 856                 err = dev_set_allmulti(dev, 1);
 857                 if (err) {
 858                         dev_put(dev);
 859                         return err;
 860                 }
 861                 break;
 862         default:
 863                 return -EINVAL;
 864         }
 865
 866         in_dev = __in_dev_get_rtnl(dev);
 867         if (!in_dev) {
 868                 dev_put(dev);
 869                 return -EADDRNOTAVAIL;
 870         }
 871         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 872         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 873                                     dev->ifindex, &in_dev->cnf);
 874         ip_rt_multicast_event(in_dev);
 875
 876         /* Fill in the VIF structures */
 877         vif_device_init(v, dev, vifc->vifc_rate_limit,
 878                         vifc->vifc_threshold,
 879                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 880                         (VIFF_TUNNEL | VIFF_REGISTER));
 881
 882         err = dev_get_port_parent_id(dev, &ppid, true);
 883         if (err == 0) {
 884                 memcpy(v->dev_parent_id.id, ppid.id, ppid.id_len);
 885                 v->dev_parent_id.id_len = ppid.id_len;
 886         } else {
 887                 v->dev_parent_id.id_len = 0;
 888         }
 889
 890         v->local = vifc->vifc_lcl_addr.s_addr;
 891         v->remote = vifc->vifc_rmt_addr.s_addr;
 892
 893         /* And finish update writing critical data */
 894         write_lock_bh(&mrt_lock);
 895         v->dev = dev;
 896         if (v->flags & VIFF_REGISTER)
 897                 mrt->mroute_reg_vif_num = vifi;
 898         if (vifi+1 > mrt->maxvif)
 899                 mrt->maxvif = vifi+1;
 900         write_unlock_bh(&mrt_lock);
 901         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 902         return 0;
 903 }
 904
 905 /* called with rcu_read_lock() */
 906 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 907                                          __be32 origin,
 908                                          __be32 mcastgrp)
 909 {
 910         struct mfc_cache_cmp_arg arg = {
 911                         .mfc_mcastgrp = mcastgrp,
 912                         .mfc_origin = origin
 913         };
 914
 915         return mr_mfc_find(mrt, &arg);
 916 }
 917
 918 /* Look for a (*,G) entry */
 919 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 920                                              __be32 mcastgrp, int vifi)
 921 {
 922         struct mfc_cache_cmp_arg arg = {
 923                         .mfc_mcastgrp = mcastgrp,
 924                         .mfc_origin = htonl(INADDR_ANY)
 925         };
 926
 927         if (mcastgrp == htonl(INADDR_ANY))
 928                 return mr_mfc_find_any_parent(mrt, vifi);
 929         return mr_mfc_find_any(mrt, vifi, &arg);
 930 }
 931
 932 /* Look for a (S,G,iif) entry if parent != -1 */
 933 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 934                                                 __be32 origin, __be32 mcastgrp,
 935                                                 int parent)
 936 {
 937         struct mfc_cache_cmp_arg arg = {
 938                         .mfc_mcastgrp = mcastgrp,
 939                         .mfc_origin = origin,
 940         };
 941
 942         return mr_mfc_find_parent(mrt, &arg, parent);
 943 }
 944
 945 /* Allocate a multicast cache entry */
 946 static struct mfc_cache *ipmr_cache_alloc(void)
 947 {
 948         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 949
 950         if (c) {
 951                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 952                 c->_c.mfc_un.res.minvif = MAXVIFS;
 953                 c->_c.free = ipmr_cache_free_rcu;
 954                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 955         }
 956         return c;
 957 }
 958
 959 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 960 {
 961         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 962
 963         if (c) {
 964                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
 965                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
 966         }
 967         return c;
 968 }
 969
 970 /* A cache entry has gone into a resolved state from queued */
 971 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 972                                struct mfc_cache *uc, struct mfc_cache *c)
 973 {
 974         struct sk_buff *skb;
 975         struct nlmsgerr *e;
 976
 977         /* Play the pending entries through our router */
 978         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
 979                 if (ip_hdr(skb)->version == 0) {
 980                         struct nlmsghdr *nlh = skb_pull(skb,
 981                                                         sizeof(struct iphdr));
 982
 983                         if (mr_fill_mroute(mrt, skb, &c->_c,
 984                                            nlmsg_data(nlh)) > 0) {
 985                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 986                                                  (u8 *)nlh;
 987                         } else {
 988                                 nlh->nlmsg_type = NLMSG_ERROR;
 989                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 990                                 skb_trim(skb, nlh->nlmsg_len);
 991                                 e = nlmsg_data(nlh);
 992                                 e->error = -EMSGSIZE;
 993                                 memset(&e->msg, 0, sizeof(e->msg));
 994                         }
 995
 996                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 997                 } else {
 998                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
 999                 }
1000         }
1001 }
1002
1003 /* Bounce a cache query up to mrouted and netlink.
1004  *
1005  * Called under mrt_lock.
1006  */
1007 static int ipmr_cache_report(struct mr_table *mrt,
1008                              struct sk_buff *pkt, vifi_t vifi, int assert)
1009 {
1010         const int ihl = ip_hdrlen(pkt);
1011         struct sock *mroute_sk;
1012         struct igmphdr *igmp;
1013         struct igmpmsg *msg;
1014         struct sk_buff *skb;
1015         int ret;
1016
1017         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1018                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1019         else
1020                 skb = alloc_skb(128, GFP_ATOMIC);
1021
1022         if (!skb)
1023                 return -ENOBUFS;
1024
1025         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1026                 /* Ugly, but we have no choice with this interface.
1027                  * Duplicate old header, fix ihl, length etc.
1028                  * And all this only to mangle msg->im_msgtype and
1029                  * to set msg->im_mbz to "mbz" :-)
1030                  */
1031                 skb_push(skb, sizeof(struct iphdr));
1032                 skb_reset_network_header(skb);
1033                 skb_reset_transport_header(skb);
1034                 msg = (struct igmpmsg *)skb_network_header(skb);
1035                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1036                 msg->im_msgtype = assert;
1037                 msg->im_mbz = 0;
1038                 if (assert == IGMPMSG_WRVIFWHOLE)
1039                         msg->im_vif = vifi;
1040                 else
1041                         msg->im_vif = mrt->mroute_reg_vif_num;
1042                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1043                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1044                                              sizeof(struct iphdr));
1045         } else {
1046                 /* Copy the IP header */
1047                 skb_set_network_header(skb, skb->len);
1048                 skb_put(skb, ihl);
1049                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1050                 /* Flag to the kernel this is a route add */
1051                 ip_hdr(skb)->protocol = 0;
1052                 msg = (struct igmpmsg *)skb_network_header(skb);
1053                 msg->im_vif = vifi;
1054                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1055                 /* Add our header */
1056                 igmp = skb_put(skb, sizeof(struct igmphdr));
1057                 igmp->type = assert;
1058                 msg->im_msgtype = assert;
1059                 igmp->code = 0;
1060                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1061                 skb->transport_header = skb->network_header;
1062         }
1063
1064         rcu_read_lock();
1065         mroute_sk = rcu_dereference(mrt->mroute_sk);
1066         if (!mroute_sk) {
1067                 rcu_read_unlock();
1068                 kfree_skb(skb);
1069                 return -EINVAL;
1070         }
1071
1072         igmpmsg_netlink_event(mrt, skb);
1073
1074         /* Deliver to mrouted */
1075         ret = sock_queue_rcv_skb(mroute_sk, skb);
1076         rcu_read_unlock();
1077         if (ret < 0) {
1078                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1079                 kfree_skb(skb);
1080         }
1081
1082         return ret;
1083 }
1084
1085 /* Queue a packet for resolution. It gets locked cache entry! */
1086 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1087                                  struct sk_buff *skb, struct net_device *dev)
1088 {
1089         const struct iphdr *iph = ip_hdr(skb);
1090         struct mfc_cache *c;
1091         bool found = false;
1092         int err;
1093
1094         spin_lock_bh(&mfc_unres_lock);
1095         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1096                 if (c->mfc_mcastgrp == iph->daddr &&
1097                     c->mfc_origin == iph->saddr) {
1098                         found = true;
1099                         break;
1100                 }
1101         }
1102
1103         if (!found) {
1104                 /* Create a new entry if allowable */
1105                 c = ipmr_cache_alloc_unres();
1106                 if (!c) {
1107                         spin_unlock_bh(&mfc_unres_lock);
1108
1109                         kfree_skb(skb);
1110                         return -ENOBUFS;
1111                 }
1112
1113                 /* Fill in the new cache entry */
1114                 c->_c.mfc_parent = -1;
1115                 c->mfc_origin   = iph->saddr;
1116                 c->mfc_mcastgrp = iph->daddr;
1117
1118                 /* Reflect first query at mrouted. */
1119                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1120
1121                 if (err < 0) {
1122                         /* If the report failed throw the cache entry
1123                            out - Brad Parker
1124                          */
1125                         spin_unlock_bh(&mfc_unres_lock);
1126
1127                         ipmr_cache_free(c);
1128                         kfree_skb(skb);
1129                         return err;
1130                 }
1131
1132                 atomic_inc(&mrt->cache_resolve_queue_len);
1133                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1134                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1135
1136                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1137                         mod_timer(&mrt->ipmr_expire_timer,
1138                                   c->_c.mfc_un.unres.expires);
1139         }
1140
1141         /* See if we can append the packet */
1142         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1143                 kfree_skb(skb);
1144                 err = -ENOBUFS;
1145         } else {
1146                 if (dev) {
1147                         skb->dev = dev;
1148                         skb->skb_iif = dev->ifindex;
1149                 }
1150                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1151                 err = 0;
1152         }
1153
1154         spin_unlock_bh(&mfc_unres_lock);
1155         return err;
1156 }
1157
1158 /* MFC cache manipulation by user space mroute daemon */
1159
1160 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1161 {
1162         struct net *net = read_pnet(&mrt->net);
1163         struct mfc_cache *c;
1164
1165         /* The entries are added/deleted only under RTNL */
1166         rcu_read_lock();
1167         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1168                                    mfc->mfcc_mcastgrp.s_addr, parent);
1169         rcu_read_unlock();
1170         if (!c)
1171                 return -ENOENT;
1172         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1173         list_del_rcu(&c->_c.list);
1174         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1175         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1176         mr_cache_put(&c->_c);
1177
1178         return 0;
1179 }
1180
1181 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1182                         struct mfcctl *mfc, int mrtsock, int parent)
1183 {
1184         struct mfc_cache *uc, *c;
1185         struct mr_mfc *_uc;
1186         bool found;
1187         int ret;
1188
1189         if (mfc->mfcc_parent >= MAXVIFS)
1190                 return -ENFILE;
1191
1192         /* The entries are added/deleted only under RTNL */
1193         rcu_read_lock();
1194         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1195                                    mfc->mfcc_mcastgrp.s_addr, parent);
1196         rcu_read_unlock();
1197         if (c) {
1198                 write_lock_bh(&mrt_lock);
1199                 c->_c.mfc_parent = mfc->mfcc_parent;
1200                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1201                 if (!mrtsock)
1202                         c->_c.mfc_flags |= MFC_STATIC;
1203                 write_unlock_bh(&mrt_lock);
1204                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1205                                               mrt->id);
1206                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1207                 return 0;
1208         }
1209
1210         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1211             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1212                 return -EINVAL;
1213
1214         c = ipmr_cache_alloc();
1215         if (!c)
1216                 return -ENOMEM;
1217
1218         c->mfc_origin = mfc->mfcc_origin.s_addr;
1219         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1220         c->_c.mfc_parent = mfc->mfcc_parent;
1221         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1222         if (!mrtsock)
1223                 c->_c.mfc_flags |= MFC_STATIC;
1224
1225         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1226                                   ipmr_rht_params);
1227         if (ret) {
1228                 pr_err("ipmr: rhtable insert error %d\n", ret);
1229                 ipmr_cache_free(c);
1230                 return ret;
1231         }
1232         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1233         /* Check to see if we resolved a queued list. If so we
1234          * need to send on the frames and tidy up.
1235          */
1236         found = false;
1237         spin_lock_bh(&mfc_unres_lock);
1238         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1239                 uc = (struct mfc_cache *)_uc;
1240                 if (uc->mfc_origin == c->mfc_origin &&
1241                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1242                         list_del(&_uc->list);
1243                         atomic_dec(&mrt->cache_resolve_queue_len);
1244                         found = true;
1245                         break;
1246                 }
1247         }
1248         if (list_empty(&mrt->mfc_unres_queue))
1249                 del_timer(&mrt->ipmr_expire_timer);
1250         spin_unlock_bh(&mfc_unres_lock);
1251
1252         if (found) {
1253                 ipmr_cache_resolve(net, mrt, uc, c);
1254                 ipmr_cache_free(uc);
1255         }
1256         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1257         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1258         return 0;
1259 }
1260
1261 /* Close the multicast socket, and clear the vif tables etc */
1262 static void mroute_clean_tables(struct mr_table *mrt, int flags)
1263 {
1264         struct net *net = read_pnet(&mrt->net);
1265         struct mr_mfc *c, *tmp;
1266         struct mfc_cache *cache;
1267         LIST_HEAD(list);
1268         int i;
1269
1270         /* Shut down all active vif entries */
1271         if (flags & (MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC)) {
1272                 for (i = 0; i < mrt->maxvif; i++) {
1273                         if (((mrt->vif_table[i].flags & VIFF_STATIC) &&
1274                              !(flags & MRT_FLUSH_VIFS_STATIC)) ||
1275                             (!(mrt->vif_table[i].flags & VIFF_STATIC) && !(flags & MRT_FLUSH_VIFS)))
1276                                 continue;
1277                         vif_delete(mrt, i, 0, &list);
1278                 }
1279                 unregister_netdevice_many(&list);
1280         }
1281
1282         /* Wipe the cache */
1283         if (flags & (MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC)) {
1284                 list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1285                         if (((c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC_STATIC)) ||
1286                             (!(c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC)))
1287                                 continue;
1288                         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1289                         list_del_rcu(&c->list);
1290                         cache = (struct mfc_cache *)c;
1291                         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1292                                                       mrt->id);
1293                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1294                         mr_cache_put(c);
1295                 }
1296         }
1297
1298         if (flags & MRT_FLUSH_MFC) {
1299                 if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1300                         spin_lock_bh(&mfc_unres_lock);
1301                         list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1302                                 list_del(&c->list);
1303                                 cache = (struct mfc_cache *)c;
1304                                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1305                                 ipmr_destroy_unres(mrt, cache);
1306                         }
1307                         spin_unlock_bh(&mfc_unres_lock);
1308                 }
1309         }
1310 }
1311
1312 /* called from ip_ra_control(), before an RCU grace period,
1313  * we dont need to call synchronize_rcu() here
1314  */
1315 static void mrtsock_destruct(struct sock *sk)
1316 {
1317         struct net *net = sock_net(sk);
1318         struct mr_table *mrt;
1319
1320         rtnl_lock();
1321         ipmr_for_each_table(mrt, net) {
1322                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1323                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1324                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1325                                                     NETCONFA_MC_FORWARDING,
1326                                                     NETCONFA_IFINDEX_ALL,
1327                                                     net->ipv4.devconf_all);
1328                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1329                         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_MFC);
1330                 }
1331         }
1332         rtnl_unlock();
1333 }
1334
1335 /* Socket options and virtual interface manipulation. The whole
1336  * virtual interface system is a complete heap, but unfortunately
1337  * that's how BSD mrouted happens to think. Maybe one day with a proper
1338  * MOSPF/PIM router set up we can clean this up.
1339  */
1340
1341 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1342                          unsigned int optlen)
1343 {
1344         struct net *net = sock_net(sk);
1345         int val, ret = 0, parent = 0;
1346         struct mr_table *mrt;
1347         struct vifctl vif;
1348         struct mfcctl mfc;
1349         bool do_wrvifwhole;
1350         u32 uval;
1351
1352         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1353         rtnl_lock();
1354         if (sk->sk_type != SOCK_RAW ||
1355             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1356                 ret = -EOPNOTSUPP;
1357                 goto out_unlock;
1358         }
1359
1360         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1361         if (!mrt) {
1362                 ret = -ENOENT;
1363                 goto out_unlock;
1364         }
1365         if (optname != MRT_INIT) {
1366                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1367                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1368                         ret = -EACCES;
1369                         goto out_unlock;
1370                 }
1371         }
1372
1373         switch (optname) {
1374         case MRT_INIT:
1375                 if (optlen != sizeof(int)) {
1376                         ret = -EINVAL;
1377                         break;
1378                 }
1379                 if (rtnl_dereference(mrt->mroute_sk)) {
1380                         ret = -EADDRINUSE;
1381                         break;
1382                 }
1383
1384                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1385                 if (ret == 0) {
1386                         rcu_assign_pointer(mrt->mroute_sk, sk);
1387                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1388                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1389                                                     NETCONFA_MC_FORWARDING,
1390                                                     NETCONFA_IFINDEX_ALL,
1391                                                     net->ipv4.devconf_all);
1392                 }
1393                 break;
1394         case MRT_DONE:
1395                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1396                         ret = -EACCES;
1397                 } else {
1398                         /* We need to unlock here because mrtsock_destruct takes
1399                          * care of rtnl itself and we can't change that due to
1400                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1401                          */
1402                         rtnl_unlock();
1403                         ret = ip_ra_control(sk, 0, NULL);
1404                         goto out;
1405                 }
1406                 break;
1407         case MRT_ADD_VIF:
1408         case MRT_DEL_VIF:
1409                 if (optlen != sizeof(vif)) {
1410                         ret = -EINVAL;
1411                         break;
1412                 }
1413                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1414                         ret = -EFAULT;
1415                         break;
1416                 }
1417                 if (vif.vifc_vifi >= MAXVIFS) {
1418                         ret = -ENFILE;
1419                         break;
1420                 }
1421                 if (optname == MRT_ADD_VIF) {
1422                         ret = vif_add(net, mrt, &vif,
1423                                       sk == rtnl_dereference(mrt->mroute_sk));
1424                 } else {
1425                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1426                 }
1427                 break;
1428         /* Manipulate the forwarding caches. These live
1429          * in a sort of kernel/user symbiosis.
1430          */
1431         case MRT_ADD_MFC:
1432         case MRT_DEL_MFC:
1433                 parent = -1;
1434                 fallthrough;
1435         case MRT_ADD_MFC_PROXY:
1436         case MRT_DEL_MFC_PROXY:
1437                 if (optlen != sizeof(mfc)) {
1438                         ret = -EINVAL;
1439                         break;
1440                 }
1441                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1442                         ret = -EFAULT;
1443                         break;
1444                 }
1445                 if (parent == 0)
1446                         parent = mfc.mfcc_parent;
1447                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1448                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1449                 else
1450                         ret = ipmr_mfc_add(net, mrt, &mfc,
1451                                            sk == rtnl_dereference(mrt->mroute_sk),
1452                                            parent);
1453                 break;
1454         case MRT_FLUSH:
1455                 if (optlen != sizeof(val)) {
1456                         ret = -EINVAL;
1457                         break;
1458                 }
1459                 if (get_user(val, (int __user *)optval)) {
1460                         ret = -EFAULT;
1461                         break;
1462                 }
1463                 mroute_clean_tables(mrt, val);
1464                 break;
1465         /* Control PIM assert. */
1466         case MRT_ASSERT:
1467                 if (optlen != sizeof(val)) {
1468                         ret = -EINVAL;
1469                         break;
1470                 }
1471                 if (get_user(val, (int __user *)optval)) {
1472                         ret = -EFAULT;
1473                         break;
1474                 }
1475                 mrt->mroute_do_assert = val;
1476                 break;
1477         case MRT_PIM:
1478                 if (!ipmr_pimsm_enabled()) {
1479                         ret = -ENOPROTOOPT;
1480                         break;
1481                 }
1482                 if (optlen != sizeof(val)) {
1483                         ret = -EINVAL;
1484                         break;
1485                 }
1486                 if (get_user(val, (int __user *)optval)) {
1487                         ret = -EFAULT;
1488                         break;
1489                 }
1490
1491                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1492                 val = !!val;
1493                 if (val != mrt->mroute_do_pim) {
1494                         mrt->mroute_do_pim = val;
1495                         mrt->mroute_do_assert = val;
1496                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1497                 }
1498                 break;
1499         case MRT_TABLE:
1500                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1501                         ret = -ENOPROTOOPT;
1502                         break;
1503                 }
1504                 if (optlen != sizeof(uval)) {
1505                         ret = -EINVAL;
1506                         break;
1507                 }
1508                 if (get_user(uval, (u32 __user *)optval)) {
1509                         ret = -EFAULT;
1510                         break;
1511                 }
1512
1513                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1514                         ret = -EBUSY;
1515                 } else {
1516                         mrt = ipmr_new_table(net, uval);
1517                         if (IS_ERR(mrt))
1518                                 ret = PTR_ERR(mrt);
1519                         else
1520                                 raw_sk(sk)->ipmr_table = uval;
1521                 }
1522                 break;
1523         /* Spurious command, or MRT_VERSION which you cannot set. */
1524         default:
1525                 ret = -ENOPROTOOPT;
1526         }
1527 out_unlock:
1528         rtnl_unlock();
1529 out:
1530         return ret;
1531 }
1532
1533 /* Getsock opt support for the multicast routing system. */
1534 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1535 {
1536         int olr;
1537         int val;
1538         struct net *net = sock_net(sk);
1539         struct mr_table *mrt;
1540
1541         if (sk->sk_type != SOCK_RAW ||
1542             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1543                 return -EOPNOTSUPP;
1544
1545         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1546         if (!mrt)
1547                 return -ENOENT;
1548
1549         switch (optname) {
1550         case MRT_VERSION:
1551                 val = 0x0305;
1552                 break;
1553         case MRT_PIM:
1554                 if (!ipmr_pimsm_enabled())
1555                         return -ENOPROTOOPT;
1556                 val = mrt->mroute_do_pim;
1557                 break;
1558         case MRT_ASSERT:
1559                 val = mrt->mroute_do_assert;
1560                 break;
1561         default:
1562                 return -ENOPROTOOPT;
1563         }
1564
1565         if (get_user(olr, optlen))
1566                 return -EFAULT;
1567         olr = min_t(unsigned int, olr, sizeof(int));
1568         if (olr < 0)
1569                 return -EINVAL;
1570         if (put_user(olr, optlen))
1571                 return -EFAULT;
1572         if (copy_to_user(optval, &val, olr))
1573                 return -EFAULT;
1574         return 0;
1575 }
1576
1577 /* The IP multicast ioctl support routines. */
1578 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1579 {
1580         struct sioc_sg_req sr;
1581         struct sioc_vif_req vr;
1582         struct vif_device *vif;
1583         struct mfc_cache *c;
1584         struct net *net = sock_net(sk);
1585         struct mr_table *mrt;
1586
1587         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1588         if (!mrt)
1589                 return -ENOENT;
1590
1591         switch (cmd) {
1592         case SIOCGETVIFCNT:
1593                 if (copy_from_user(&vr, arg, sizeof(vr)))
1594                         return -EFAULT;
1595                 if (vr.vifi >= mrt->maxvif)
1596                         return -EINVAL;
1597                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1598                 read_lock(&mrt_lock);
1599                 vif = &mrt->vif_table[vr.vifi];
1600                 if (VIF_EXISTS(mrt, vr.vifi)) {
1601                         vr.icount = vif->pkt_in;
1602                         vr.ocount = vif->pkt_out;
1603                         vr.ibytes = vif->bytes_in;
1604                         vr.obytes = vif->bytes_out;
1605                         read_unlock(&mrt_lock);
1606
1607                         if (copy_to_user(arg, &vr, sizeof(vr)))
1608                                 return -EFAULT;
1609                         return 0;
1610                 }
1611                 read_unlock(&mrt_lock);
1612                 return -EADDRNOTAVAIL;
1613         case SIOCGETSGCNT:
1614                 if (copy_from_user(&sr, arg, sizeof(sr)))
1615                         return -EFAULT;
1616
1617                 rcu_read_lock();
1618                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1619                 if (c) {
1620                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1621                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1622                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1623                         rcu_read_unlock();
1624
1625                         if (copy_to_user(arg, &sr, sizeof(sr)))
1626                                 return -EFAULT;
1627                         return 0;
1628                 }
1629                 rcu_read_unlock();
1630                 return -EADDRNOTAVAIL;
1631         default:
1632                 return -ENOIOCTLCMD;
1633         }
1634 }
1635
1636 #ifdef CONFIG_COMPAT
1637 struct compat_sioc_sg_req {
1638         struct in_addr src;
1639         struct in_addr grp;
1640         compat_ulong_t pktcnt;
1641         compat_ulong_t bytecnt;
1642         compat_ulong_t wrong_if;
1643 };
1644
1645 struct compat_sioc_vif_req {
1646         vifi_t  vifi;           /* Which iface */
1647         compat_ulong_t icount;
1648         compat_ulong_t ocount;
1649         compat_ulong_t ibytes;
1650         compat_ulong_t obytes;
1651 };
1652
1653 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1654 {
1655         struct compat_sioc_sg_req sr;
1656         struct compat_sioc_vif_req vr;
1657         struct vif_device *vif;
1658         struct mfc_cache *c;
1659         struct net *net = sock_net(sk);
1660         struct mr_table *mrt;
1661
1662         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1663         if (!mrt)
1664                 return -ENOENT;
1665
1666         switch (cmd) {
1667         case SIOCGETVIFCNT:
1668                 if (copy_from_user(&vr, arg, sizeof(vr)))
1669                         return -EFAULT;
1670                 if (vr.vifi >= mrt->maxvif)
1671                         return -EINVAL;
1672                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1673                 read_lock(&mrt_lock);
1674                 vif = &mrt->vif_table[vr.vifi];
1675                 if (VIF_EXISTS(mrt, vr.vifi)) {
1676                         vr.icount = vif->pkt_in;
1677                         vr.ocount = vif->pkt_out;
1678                         vr.ibytes = vif->bytes_in;
1679                         vr.obytes = vif->bytes_out;
1680                         read_unlock(&mrt_lock);
1681
1682                         if (copy_to_user(arg, &vr, sizeof(vr)))
1683                                 return -EFAULT;
1684                         return 0;
1685                 }
1686                 read_unlock(&mrt_lock);
1687                 return -EADDRNOTAVAIL;
1688         case SIOCGETSGCNT:
1689                 if (copy_from_user(&sr, arg, sizeof(sr)))
1690                         return -EFAULT;
1691
1692                 rcu_read_lock();
1693                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1694                 if (c) {
1695                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1696                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1697                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1698                         rcu_read_unlock();
1699
1700                         if (copy_to_user(arg, &sr, sizeof(sr)))
1701                                 return -EFAULT;
1702                         return 0;
1703                 }
1704                 rcu_read_unlock();
1705                 return -EADDRNOTAVAIL;
1706         default:
1707                 return -ENOIOCTLCMD;
1708         }
1709 }
1710 #endif
1711
1712 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1713 {
1714         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1715         struct net *net = dev_net(dev);
1716         struct mr_table *mrt;
1717         struct vif_device *v;
1718         int ct;
1719
1720         if (event != NETDEV_UNREGISTER)
1721                 return NOTIFY_DONE;
1722
1723         ipmr_for_each_table(mrt, net) {
1724                 v = &mrt->vif_table[0];
1725                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1726                         if (v->dev == dev)
1727                                 vif_delete(mrt, ct, 1, NULL);
1728                 }
1729         }
1730         return NOTIFY_DONE;
1731 }
1732
1733 static struct notifier_block ip_mr_notifier = {
1734         .notifier_call = ipmr_device_event,
1735 };
1736
1737 /* Encapsulate a packet by attaching a valid IPIP header to it.
1738  * This avoids tunnel drivers and other mess and gives us the speed so
1739  * important for multicast video.
1740  */
1741 static void ip_encap(struct net *net, struct sk_buff *skb,
1742                      __be32 saddr, __be32 daddr)
1743 {
1744         struct iphdr *iph;
1745         const struct iphdr *old_iph = ip_hdr(skb);
1746
1747         skb_push(skb, sizeof(struct iphdr));
1748         skb->transport_header = skb->network_header;
1749         skb_reset_network_header(skb);
1750         iph = ip_hdr(skb);
1751
1752         iph->version    =       4;
1753         iph->tos        =       old_iph->tos;
1754         iph->ttl        =       old_iph->ttl;
1755         iph->frag_off   =       0;
1756         iph->daddr      =       daddr;
1757         iph->saddr      =       saddr;
1758         iph->protocol   =       IPPROTO_IPIP;
1759         iph->ihl        =       5;
1760         iph->tot_len    =       htons(skb->len);
1761         ip_select_ident(net, skb, NULL);
1762         ip_send_check(iph);
1763
1764         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1765         nf_reset_ct(skb);
1766 }
1767
1768 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1769                                       struct sk_buff *skb)
1770 {
1771         struct ip_options *opt = &(IPCB(skb)->opt);
1772
1773         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1774         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1775
1776         if (unlikely(opt->optlen))
1777                 ip_forward_options(skb);
1778
1779         return dst_output(net, sk, skb);
1780 }
1781
1782 #ifdef CONFIG_NET_SWITCHDEV
1783 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1784                                    int in_vifi, int out_vifi)
1785 {
1786         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1787         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1788
1789         if (!skb->offload_l3_fwd_mark)
1790                 return false;
1791         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1792                 return false;
1793         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1794                                         &in_vif->dev_parent_id);
1795 }
1796 #else
1797 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1798                                    int in_vifi, int out_vifi)
1799 {
1800         return false;
1801 }
1802 #endif
1803
1804 /* Processing handlers for ipmr_forward */
1805
1806 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1807                             int in_vifi, struct sk_buff *skb, int vifi)
1808 {
1809         const struct iphdr *iph = ip_hdr(skb);
1810         struct vif_device *vif = &mrt->vif_table[vifi];
1811         struct net_device *dev;
1812         struct rtable *rt;
1813         struct flowi4 fl4;
1814         int    encap = 0;
1815
1816         if (!vif->dev)
1817                 goto out_free;
1818
1819         if (vif->flags & VIFF_REGISTER) {
1820                 vif->pkt_out++;
1821                 vif->bytes_out += skb->len;
1822                 vif->dev->stats.tx_bytes += skb->len;
1823                 vif->dev->stats.tx_packets++;
1824                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1825                 goto out_free;
1826         }
1827
1828         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1829                 goto out_free;
1830
1831         if (vif->flags & VIFF_TUNNEL) {
1832                 rt = ip_route_output_ports(net, &fl4, NULL,
1833                                            vif->remote, vif->local,
1834                                            0, 0,
1835                                            IPPROTO_IPIP,
1836                                            RT_TOS(iph->tos), vif->link);
1837                 if (IS_ERR(rt))
1838                         goto out_free;
1839                 encap = sizeof(struct iphdr);
1840         } else {
1841                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1842                                            0, 0,
1843                                            IPPROTO_IPIP,
1844                                            RT_TOS(iph->tos), vif->link);
1845                 if (IS_ERR(rt))
1846                         goto out_free;
1847         }
1848
1849         dev = rt->dst.dev;
1850
1851         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1852                 /* Do not fragment multicasts. Alas, IPv4 does not
1853                  * allow to send ICMP, so that packets will disappear
1854                  * to blackhole.
1855                  */
1856                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1857                 ip_rt_put(rt);
1858                 goto out_free;
1859         }
1860
1861         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1862
1863         if (skb_cow(skb, encap)) {
1864                 ip_rt_put(rt);
1865                 goto out_free;
1866         }
1867
1868         vif->pkt_out++;
1869         vif->bytes_out += skb->len;
1870
1871         skb_dst_drop(skb);
1872         skb_dst_set(skb, &rt->dst);
1873         ip_decrease_ttl(ip_hdr(skb));
1874
1875         /* FIXME: forward and output firewalls used to be called here.
1876          * What do we do with netfilter? -- RR
1877          */
1878         if (vif->flags & VIFF_TUNNEL) {
1879                 ip_encap(net, skb, vif->local, vif->remote);
1880                 /* FIXME: extra output firewall step used to be here. --RR */
1881                 vif->dev->stats.tx_packets++;
1882                 vif->dev->stats.tx_bytes += skb->len;
1883         }
1884
1885         IPCB(skb)->flags |= IPSKB_FORWARDED;
1886
1887         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1888          * not only before forwarding, but after forwarding on all output
1889          * interfaces. It is clear, if mrouter runs a multicasting
1890          * program, it should receive packets not depending to what interface
1891          * program is joined.
1892          * If we will not make it, the program will have to join on all
1893          * interfaces. On the other hand, multihoming host (or router, but
1894          * not mrouter) cannot join to more than one interface - it will
1895          * result in receiving multiple packets.
1896          */
1897         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1898                 net, NULL, skb, skb->dev, dev,
1899                 ipmr_forward_finish);
1900         return;
1901
1902 out_free:
1903         kfree_skb(skb);
1904 }
1905
1906 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1907 {
1908         int ct;
1909
1910         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1911                 if (mrt->vif_table[ct].dev == dev)
1912                         break;
1913         }
1914         return ct;
1915 }
1916
1917 /* "local" means that we should preserve one skb (for local delivery) */
1918 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1919                           struct net_device *dev, struct sk_buff *skb,
1920                           struct mfc_cache *c, int local)
1921 {
1922         int true_vifi = ipmr_find_vif(mrt, dev);
1923         int psend = -1;
1924         int vif, ct;
1925
1926         vif = c->_c.mfc_parent;
1927         c->_c.mfc_un.res.pkt++;
1928         c->_c.mfc_un.res.bytes += skb->len;
1929         c->_c.mfc_un.res.lastuse = jiffies;
1930
1931         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1932                 struct mfc_cache *cache_proxy;
1933
1934                 /* For an (*,G) entry, we only check that the incomming
1935                  * interface is part of the static tree.
1936                  */
1937                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1938                 if (cache_proxy &&
1939                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1940                         goto forward;
1941         }
1942
1943         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1944         if (mrt->vif_table[vif].dev != dev) {
1945                 if (rt_is_output_route(skb_rtable(skb))) {
1946                         /* It is our own packet, looped back.
1947                          * Very complicated situation...
1948                          *
1949                          * The best workaround until routing daemons will be
1950                          * fixed is not to redistribute packet, if it was
1951                          * send through wrong interface. It means, that
1952                          * multicast applications WILL NOT work for
1953                          * (S,G), which have default multicast route pointing
1954                          * to wrong oif. In any case, it is not a good
1955                          * idea to use multicasting applications on router.
1956                          */
1957                         goto dont_forward;
1958                 }
1959
1960                 c->_c.mfc_un.res.wrong_if++;
1961
1962                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1963                     /* pimsm uses asserts, when switching from RPT to SPT,
1964                      * so that we cannot check that packet arrived on an oif.
1965                      * It is bad, but otherwise we would need to move pretty
1966                      * large chunk of pimd to kernel. Ough... --ANK
1967                      */
1968                     (mrt->mroute_do_pim ||
1969                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1970                     time_after(jiffies,
1971                                c->_c.mfc_un.res.last_assert +
1972                                MFC_ASSERT_THRESH)) {
1973                         c->_c.mfc_un.res.last_assert = jiffies;
1974                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1975                         if (mrt->mroute_do_wrvifwhole)
1976                                 ipmr_cache_report(mrt, skb, true_vifi,
1977                                                   IGMPMSG_WRVIFWHOLE);
1978                 }
1979                 goto dont_forward;
1980         }
1981
1982 forward:
1983         mrt->vif_table[vif].pkt_in++;
1984         mrt->vif_table[vif].bytes_in += skb->len;
1985
1986         /* Forward the frame */
1987         if (c->mfc_origin == htonl(INADDR_ANY) &&
1988             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
1989                 if (true_vifi >= 0 &&
1990                     true_vifi != c->_c.mfc_parent &&
1991                     ip_hdr(skb)->ttl >
1992                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
1993                         /* It's an (*,*) entry and the packet is not coming from
1994                          * the upstream: forward the packet to the upstream
1995                          * only.
1996                          */
1997                         psend = c->_c.mfc_parent;
1998                         goto last_forward;
1999                 }
2000                 goto dont_forward;
2001         }
2002         for (ct = c->_c.mfc_un.res.maxvif - 1;
2003              ct >= c->_c.mfc_un.res.minvif; ct--) {
2004                 /* For (*,G) entry, don't forward to the incoming interface */
2005                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2006                      ct != true_vifi) &&
2007                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2008                         if (psend != -1) {
2009                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2010
2011                                 if (skb2)
2012                                         ipmr_queue_xmit(net, mrt, true_vifi,
2013                                                         skb2, psend);
2014                         }
2015                         psend = ct;
2016                 }
2017         }
2018 last_forward:
2019         if (psend != -1) {
2020                 if (local) {
2021                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2022
2023                         if (skb2)
2024                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2025                                                 psend);
2026                 } else {
2027                         ipmr_queue_xmit(net, mrt, true_vifi, skb, psend);
2028                         return;
2029                 }
2030         }
2031
2032 dont_forward:
2033         if (!local)
2034                 kfree_skb(skb);
2035 }
2036
2037 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2038 {
2039         struct rtable *rt = skb_rtable(skb);
2040         struct iphdr *iph = ip_hdr(skb);
2041         struct flowi4 fl4 = {
2042                 .daddr = iph->daddr,
2043                 .saddr = iph->saddr,
2044                 .flowi4_tos = RT_TOS(iph->tos),
2045                 .flowi4_oif = (rt_is_output_route(rt) ?
2046                                skb->dev->ifindex : 0),
2047                 .flowi4_iif = (rt_is_output_route(rt) ?
2048                                LOOPBACK_IFINDEX :
2049                                skb->dev->ifindex),
2050                 .flowi4_mark = skb->mark,
2051         };
2052         struct mr_table *mrt;
2053         int err;
2054
2055         err = ipmr_fib_lookup(net, &fl4, &mrt);
2056         if (err)
2057                 return ERR_PTR(err);
2058         return mrt;
2059 }
2060
2061 /* Multicast packets for forwarding arrive here
2062  * Called with rcu_read_lock();
2063  */
2064 int ip_mr_input(struct sk_buff *skb)
2065 {
2066         struct mfc_cache *cache;
2067         struct net *net = dev_net(skb->dev);
2068         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2069         struct mr_table *mrt;
2070         struct net_device *dev;
2071
2072         /* skb->dev passed in is the loX master dev for vrfs.
2073          * As there are no vifs associated with loopback devices,
2074          * get the proper interface that does have a vif associated with it.
2075          */
2076         dev = skb->dev;
2077         if (netif_is_l3_master(skb->dev)) {
2078                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2079                 if (!dev) {
2080                         kfree_skb(skb);
2081                         return -ENODEV;
2082                 }
2083         }
2084
2085         /* Packet is looped back after forward, it should not be
2086          * forwarded second time, but still can be delivered locally.
2087          */
2088         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2089                 goto dont_forward;
2090
2091         mrt = ipmr_rt_fib_lookup(net, skb);
2092         if (IS_ERR(mrt)) {
2093                 kfree_skb(skb);
2094                 return PTR_ERR(mrt);
2095         }
2096         if (!local) {
2097                 if (IPCB(skb)->opt.router_alert) {
2098                         if (ip_call_ra_chain(skb))
2099                                 return 0;
2100                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2101                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2102                          * Cisco IOS <= 11.2(8)) do not put router alert
2103                          * option to IGMP packets destined to routable
2104                          * groups. It is very bad, because it means
2105                          * that we can forward NO IGMP messages.
2106                          */
2107                         struct sock *mroute_sk;
2108
2109                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2110                         if (mroute_sk) {
2111                                 nf_reset_ct(skb);
2112                                 raw_rcv(mroute_sk, skb);
2113                                 return 0;
2114                         }
2115                     }
2116         }
2117
2118         /* already under rcu_read_lock() */
2119         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2120         if (!cache) {
2121                 int vif = ipmr_find_vif(mrt, dev);
2122
2123                 if (vif >= 0)
2124                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2125                                                     vif);
2126         }
2127
2128         /* No usable cache entry */
2129         if (!cache) {
2130                 int vif;
2131
2132                 if (local) {
2133                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2134                         ip_local_deliver(skb);
2135                         if (!skb2)
2136                                 return -ENOBUFS;
2137                         skb = skb2;
2138                 }
2139
2140                 read_lock(&mrt_lock);
2141                 vif = ipmr_find_vif(mrt, dev);
2142                 if (vif >= 0) {
2143                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2144                         read_unlock(&mrt_lock);
2145
2146                         return err2;
2147                 }
2148                 read_unlock(&mrt_lock);
2149                 kfree_skb(skb);
2150                 return -ENODEV;
2151         }
2152
2153         read_lock(&mrt_lock);
2154         ip_mr_forward(net, mrt, dev, skb, cache, local);
2155         read_unlock(&mrt_lock);
2156
2157         if (local)
2158                 return ip_local_deliver(skb);
2159
2160         return 0;
2161
2162 dont_forward:
2163         if (local)
2164                 return ip_local_deliver(skb);
2165         kfree_skb(skb);
2166         return 0;
2167 }
2168
2169 #ifdef CONFIG_IP_PIMSM_V1
2170 /* Handle IGMP messages of PIMv1 */
2171 int pim_rcv_v1(struct sk_buff *skb)
2172 {
2173         struct igmphdr *pim;
2174         struct net *net = dev_net(skb->dev);
2175         struct mr_table *mrt;
2176
2177         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2178                 goto drop;
2179
2180         pim = igmp_hdr(skb);
2181
2182         mrt = ipmr_rt_fib_lookup(net, skb);
2183         if (IS_ERR(mrt))
2184                 goto drop;
2185         if (!mrt->mroute_do_pim ||
2186             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2187                 goto drop;
2188
2189         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2190 drop:
2191                 kfree_skb(skb);
2192         }
2193         return 0;
2194 }
2195 #endif
2196
2197 #ifdef CONFIG_IP_PIMSM_V2
2198 static int pim_rcv(struct sk_buff *skb)
2199 {
2200         struct pimreghdr *pim;
2201         struct net *net = dev_net(skb->dev);
2202         struct mr_table *mrt;
2203
2204         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2205                 goto drop;
2206
2207         pim = (struct pimreghdr *)skb_transport_header(skb);
2208         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2209             (pim->flags & PIM_NULL_REGISTER) ||
2210             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2211              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2212                 goto drop;
2213
2214         mrt = ipmr_rt_fib_lookup(net, skb);
2215         if (IS_ERR(mrt))
2216                 goto drop;
2217         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2218 drop:
2219                 kfree_skb(skb);
2220         }
2221         return 0;
2222 }
2223 #endif
2224
2225 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2226                    __be32 saddr, __be32 daddr,
2227                    struct rtmsg *rtm, u32 portid)
2228 {
2229         struct mfc_cache *cache;
2230         struct mr_table *mrt;
2231         int err;
2232
2233         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2234         if (!mrt)
2235                 return -ENOENT;
2236
2237         rcu_read_lock();
2238         cache = ipmr_cache_find(mrt, saddr, daddr);
2239         if (!cache && skb->dev) {
2240                 int vif = ipmr_find_vif(mrt, skb->dev);
2241
2242                 if (vif >= 0)
2243                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2244         }
2245         if (!cache) {
2246                 struct sk_buff *skb2;
2247                 struct iphdr *iph;
2248                 struct net_device *dev;
2249                 int vif = -1;
2250
2251                 dev = skb->dev;
2252                 read_lock(&mrt_lock);
2253                 if (dev)
2254                         vif = ipmr_find_vif(mrt, dev);
2255                 if (vif < 0) {
2256                         read_unlock(&mrt_lock);
2257                         rcu_read_unlock();
2258                         return -ENODEV;
2259                 }
2260
2261                 skb2 = skb_realloc_headroom(skb, sizeof(struct iphdr));
2262                 if (!skb2) {
2263                         read_unlock(&mrt_lock);
2264                         rcu_read_unlock();
2265                         return -ENOMEM;
2266                 }
2267
2268                 NETLINK_CB(skb2).portid = portid;
2269                 skb_push(skb2, sizeof(struct iphdr));
2270                 skb_reset_network_header(skb2);
2271                 iph = ip_hdr(skb2);
2272                 iph->ihl = sizeof(struct iphdr) >> 2;
2273                 iph->saddr = saddr;
2274                 iph->daddr = daddr;
2275                 iph->version = 0;
2276                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2277                 read_unlock(&mrt_lock);
2278                 rcu_read_unlock();
2279                 return err;
2280         }
2281
2282         read_lock(&mrt_lock);
2283         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2284         read_unlock(&mrt_lock);
2285         rcu_read_unlock();
2286         return err;
2287 }
2288
2289 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2290                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2291                             int flags)
2292 {
2293         struct nlmsghdr *nlh;
2294         struct rtmsg *rtm;
2295         int err;
2296
2297         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2298         if (!nlh)
2299                 return -EMSGSIZE;
2300
2301         rtm = nlmsg_data(nlh);
2302         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2303         rtm->rtm_dst_len  = 32;
2304         rtm->rtm_src_len  = 32;
2305         rtm->rtm_tos      = 0;
2306         rtm->rtm_table    = mrt->id;
2307         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2308                 goto nla_put_failure;
2309         rtm->rtm_type     = RTN_MULTICAST;
2310         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2311         if (c->_c.mfc_flags & MFC_STATIC)
2312                 rtm->rtm_protocol = RTPROT_STATIC;
2313         else
2314                 rtm->rtm_protocol = RTPROT_MROUTED;
2315         rtm->rtm_flags    = 0;
2316
2317         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2318             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2319                 goto nla_put_failure;
2320         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2321         /* do not break the dump if cache is unresolved */
2322         if (err < 0 && err != -ENOENT)
2323                 goto nla_put_failure;
2324
2325         nlmsg_end(skb, nlh);
2326         return 0;
2327
2328 nla_put_failure:
2329         nlmsg_cancel(skb, nlh);
2330         return -EMSGSIZE;
2331 }
2332
2333 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2334                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2335                              int flags)
2336 {
2337         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2338                                 cmd, flags);
2339 }
2340
2341 static size_t mroute_msgsize(bool unresolved, int maxvif)
2342 {
2343         size_t len =
2344                 NLMSG_ALIGN(sizeof(struct rtmsg))
2345                 + nla_total_size(4)     /* RTA_TABLE */
2346                 + nla_total_size(4)     /* RTA_SRC */
2347                 + nla_total_size(4)     /* RTA_DST */
2348                 ;
2349
2350         if (!unresolved)
2351                 len = len
2352                       + nla_total_size(4)       /* RTA_IIF */
2353                       + nla_total_size(0)       /* RTA_MULTIPATH */
2354                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2355                                                 /* RTA_MFC_STATS */
2356                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2357                 ;
2358
2359         return len;
2360 }
2361
2362 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2363                                  int cmd)
2364 {
2365         struct net *net = read_pnet(&mrt->net);
2366         struct sk_buff *skb;
2367         int err = -ENOBUFS;
2368
2369         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2370                                        mrt->maxvif),
2371                         GFP_ATOMIC);
2372         if (!skb)
2373                 goto errout;
2374
2375         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2376         if (err < 0)
2377                 goto errout;
2378
2379         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2380         return;
2381
2382 errout:
2383         kfree_skb(skb);
2384         if (err < 0)
2385                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2386 }
2387
2388 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2389 {
2390         size_t len =
2391                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2392                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2393                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2394                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2395                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2396                                         /* IPMRA_CREPORT_PKT */
2397                 + nla_total_size(payloadlen)
2398                 ;
2399
2400         return len;
2401 }
2402
2403 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2404 {
2405         struct net *net = read_pnet(&mrt->net);
2406         struct nlmsghdr *nlh;
2407         struct rtgenmsg *rtgenm;
2408         struct igmpmsg *msg;
2409         struct sk_buff *skb;
2410         struct nlattr *nla;
2411         int payloadlen;
2412
2413         payloadlen = pkt->len - sizeof(struct igmpmsg);
2414         msg = (struct igmpmsg *)skb_network_header(pkt);
2415
2416         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2417         if (!skb)
2418                 goto errout;
2419
2420         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2421                         sizeof(struct rtgenmsg), 0);
2422         if (!nlh)
2423                 goto errout;
2424         rtgenm = nlmsg_data(nlh);
2425         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2426         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2427             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif) ||
2428             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2429                             msg->im_src.s_addr) ||
2430             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2431                             msg->im_dst.s_addr))
2432                 goto nla_put_failure;
2433
2434         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2435         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2436                                   nla_data(nla), payloadlen))
2437                 goto nla_put_failure;
2438
2439         nlmsg_end(skb, nlh);
2440
2441         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2442         return;
2443
2444 nla_put_failure:
2445         nlmsg_cancel(skb, nlh);
2446 errout:
2447         kfree_skb(skb);
2448         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2449 }
2450
2451 static int ipmr_rtm_valid_getroute_req(struct sk_buff *skb,
2452                                        const struct nlmsghdr *nlh,
2453                                        struct nlattr **tb,
2454                                        struct netlink_ext_ack *extack)
2455 {
2456         struct rtmsg *rtm;
2457         int i, err;
2458
2459         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2460                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for multicast route get request");
2461                 return -EINVAL;
2462         }
2463
2464         if (!netlink_strict_get_check(skb))
2465                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2466                                               rtm_ipv4_policy, extack);
2467
2468         rtm = nlmsg_data(nlh);
2469         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2470             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2471             rtm->rtm_tos || rtm->rtm_table || rtm->rtm_protocol ||
2472             rtm->rtm_scope || rtm->rtm_type || rtm->rtm_flags) {
2473                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for multicast route get request");
2474                 return -EINVAL;
2475         }
2476
2477         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2478                                             rtm_ipv4_policy, extack);
2479         if (err)
2480                 return err;
2481
2482         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2483             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2484                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2485                 return -EINVAL;
2486         }
2487
2488         for (i = 0; i <= RTA_MAX; i++) {
2489                 if (!tb[i])
2490                         continue;
2491
2492                 switch (i) {
2493                 case RTA_SRC:
2494                 case RTA_DST:
2495                 case RTA_TABLE:
2496                         break;
2497                 default:
2498                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in multicast route get request");
2499                         return -EINVAL;
2500                 }
2501         }
2502
2503         return 0;
2504 }
2505
2506 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2507                              struct netlink_ext_ack *extack)
2508 {
2509         struct net *net = sock_net(in_skb->sk);
2510         struct nlattr *tb[RTA_MAX + 1];
2511         struct sk_buff *skb = NULL;
2512         struct mfc_cache *cache;
2513         struct mr_table *mrt;
2514         __be32 src, grp;
2515         u32 tableid;
2516         int err;
2517
2518         err = ipmr_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2519         if (err < 0)
2520                 goto errout;
2521
2522         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2523         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2524         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2525
2526         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2527         if (!mrt) {
2528                 err = -ENOENT;
2529                 goto errout_free;
2530         }
2531
2532         /* entries are added/deleted only under RTNL */
2533         rcu_read_lock();
2534         cache = ipmr_cache_find(mrt, src, grp);
2535         rcu_read_unlock();
2536         if (!cache) {
2537                 err = -ENOENT;
2538                 goto errout_free;
2539         }
2540
2541         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2542         if (!skb) {
2543                 err = -ENOBUFS;
2544                 goto errout_free;
2545         }
2546
2547         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2548                                nlh->nlmsg_seq, cache,
2549                                RTM_NEWROUTE, 0);
2550         if (err < 0)
2551                 goto errout_free;
2552
2553         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2554
2555 errout:
2556         return err;
2557
2558 errout_free:
2559         kfree_skb(skb);
2560         goto errout;
2561 }
2562
2563 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2564 {
2565         struct fib_dump_filter filter = {};
2566         int err;
2567
2568         if (cb->strict_check) {
2569                 err = ip_valid_fib_dump_req(sock_net(skb->sk), cb->nlh,
2570                                             &filter, cb);
2571                 if (err < 0)
2572                         return err;
2573         }
2574
2575         if (filter.table_id) {
2576                 struct mr_table *mrt;
2577
2578                 mrt = ipmr_get_table(sock_net(skb->sk), filter.table_id);
2579                 if (!mrt) {
2580                         if (filter.dump_all_families)
2581                                 return skb->len;
2582
2583                         NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
2584                         return -ENOENT;
2585                 }
2586                 err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
2587                                     &mfc_unres_lock, &filter);
2588                 return skb->len ? : err;
2589         }
2590
2591         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2592                                 _ipmr_fill_mroute, &mfc_unres_lock, &filter);
2593 }
2594
2595 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2596         [RTA_SRC]       = { .type = NLA_U32 },
2597         [RTA_DST]       = { .type = NLA_U32 },
2598         [RTA_IIF]       = { .type = NLA_U32 },
2599         [RTA_TABLE]     = { .type = NLA_U32 },
2600         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2601 };
2602
2603 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2604 {
2605         switch (rtm_protocol) {
2606         case RTPROT_STATIC:
2607         case RTPROT_MROUTED:
2608                 return true;
2609         }
2610         return false;
2611 }
2612
2613 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2614 {
2615         struct rtnexthop *rtnh = nla_data(nla);
2616         int remaining = nla_len(nla), vifi = 0;
2617
2618         while (rtnh_ok(rtnh, remaining)) {
2619                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2620                 if (++vifi == MAXVIFS)
2621                         break;
2622                 rtnh = rtnh_next(rtnh, &remaining);
2623         }
2624
2625         return remaining > 0 ? -EINVAL : vifi;
2626 }
2627
2628 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2629 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2630                             struct mfcctl *mfcc, int *mrtsock,
2631                             struct mr_table **mrtret,
2632                             struct netlink_ext_ack *extack)
2633 {
2634         struct net_device *dev = NULL;
2635         u32 tblid = RT_TABLE_DEFAULT;
2636         struct mr_table *mrt;
2637         struct nlattr *attr;
2638         struct rtmsg *rtm;
2639         int ret, rem;
2640
2641         ret = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
2642                                         rtm_ipmr_policy, extack);
2643         if (ret < 0)
2644                 goto out;
2645         rtm = nlmsg_data(nlh);
2646
2647         ret = -EINVAL;
2648         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2649             rtm->rtm_type != RTN_MULTICAST ||
2650             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2651             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2652                 goto out;
2653
2654         memset(mfcc, 0, sizeof(*mfcc));
2655         mfcc->mfcc_parent = -1;
2656         ret = 0;
2657         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2658                 switch (nla_type(attr)) {
2659                 case RTA_SRC:
2660                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2661                         break;
2662                 case RTA_DST:
2663                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2664                         break;
2665                 case RTA_IIF:
2666                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2667                         if (!dev) {
2668                                 ret = -ENODEV;
2669                                 goto out;
2670                         }
2671                         break;
2672                 case RTA_MULTIPATH:
2673                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2674                                 ret = -EINVAL;
2675                                 goto out;
2676                         }
2677                         break;
2678                 case RTA_PREFSRC:
2679                         ret = 1;
2680                         break;
2681                 case RTA_TABLE:
2682                         tblid = nla_get_u32(attr);
2683                         break;
2684                 }
2685         }
2686         mrt = ipmr_get_table(net, tblid);
2687         if (!mrt) {
2688                 ret = -ENOENT;
2689                 goto out;
2690         }
2691         *mrtret = mrt;
2692         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2693         if (dev)
2694                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2695
2696 out:
2697         return ret;
2698 }
2699
2700 /* takes care of both newroute and delroute */
2701 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2702                           struct netlink_ext_ack *extack)
2703 {
2704         struct net *net = sock_net(skb->sk);
2705         int ret, mrtsock, parent;
2706         struct mr_table *tbl;
2707         struct mfcctl mfcc;
2708
2709         mrtsock = 0;
2710         tbl = NULL;
2711         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2712         if (ret < 0)
2713                 return ret;
2714
2715         parent = ret ? mfcc.mfcc_parent : -1;
2716         if (nlh->nlmsg_type == RTM_NEWROUTE)
2717                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2718         else
2719                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2720 }
2721
2722 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2723 {
2724         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2725
2726         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2727             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2728             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2729                         mrt->mroute_reg_vif_num) ||
2730             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2731                        mrt->mroute_do_assert) ||
2732             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2733             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2734                        mrt->mroute_do_wrvifwhole))
2735                 return false;
2736
2737         return true;
2738 }
2739
2740 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2741 {
2742         struct nlattr *vif_nest;
2743         struct vif_device *vif;
2744
2745         /* if the VIF doesn't exist just continue */
2746         if (!VIF_EXISTS(mrt, vifid))
2747                 return true;
2748
2749         vif = &mrt->vif_table[vifid];
2750         vif_nest = nla_nest_start_noflag(skb, IPMRA_VIF);
2751         if (!vif_nest)
2752                 return false;
2753         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2754             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2755             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2756             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2757                               IPMRA_VIFA_PAD) ||
2758             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2759                               IPMRA_VIFA_PAD) ||
2760             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2761                               IPMRA_VIFA_PAD) ||
2762             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2763                               IPMRA_VIFA_PAD) ||
2764             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2765             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2766                 nla_nest_cancel(skb, vif_nest);
2767                 return false;
2768         }
2769         nla_nest_end(skb, vif_nest);
2770
2771         return true;
2772 }
2773
2774 static int ipmr_valid_dumplink(const struct nlmsghdr *nlh,
2775                                struct netlink_ext_ack *extack)
2776 {
2777         struct ifinfomsg *ifm;
2778
2779         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ifm))) {
2780                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for ipmr link dump");
2781                 return -EINVAL;
2782         }
2783
2784         if (nlmsg_attrlen(nlh, sizeof(*ifm))) {
2785                 NL_SET_ERR_MSG(extack, "Invalid data after header in ipmr link dump");
2786                 return -EINVAL;
2787         }
2788
2789         ifm = nlmsg_data(nlh);
2790         if (ifm->__ifi_pad || ifm->ifi_type || ifm->ifi_flags ||
2791             ifm->ifi_change || ifm->ifi_index) {
2792                 NL_SET_ERR_MSG(extack, "Invalid values in header for ipmr link dump request");
2793                 return -EINVAL;
2794         }
2795
2796         return 0;
2797 }
2798
2799 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2800 {
2801         struct net *net = sock_net(skb->sk);
2802         struct nlmsghdr *nlh = NULL;
2803         unsigned int t = 0, s_t;
2804         unsigned int e = 0, s_e;
2805         struct mr_table *mrt;
2806
2807         if (cb->strict_check) {
2808                 int err = ipmr_valid_dumplink(cb->nlh, cb->extack);
2809
2810                 if (err < 0)
2811                         return err;
2812         }
2813
2814         s_t = cb->args[0];
2815         s_e = cb->args[1];
2816
2817         ipmr_for_each_table(mrt, net) {
2818                 struct nlattr *vifs, *af;
2819                 struct ifinfomsg *hdr;
2820                 u32 i;
2821
2822                 if (t < s_t)
2823                         goto skip_table;
2824                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2825                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2826                                 sizeof(*hdr), NLM_F_MULTI);
2827                 if (!nlh)
2828                         break;
2829
2830                 hdr = nlmsg_data(nlh);
2831                 memset(hdr, 0, sizeof(*hdr));
2832                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2833
2834                 af = nla_nest_start_noflag(skb, IFLA_AF_SPEC);
2835                 if (!af) {
2836                         nlmsg_cancel(skb, nlh);
2837                         goto out;
2838                 }
2839
2840                 if (!ipmr_fill_table(mrt, skb)) {
2841                         nlmsg_cancel(skb, nlh);
2842                         goto out;
2843                 }
2844
2845                 vifs = nla_nest_start_noflag(skb, IPMRA_TABLE_VIFS);
2846                 if (!vifs) {
2847                         nla_nest_end(skb, af);
2848                         nlmsg_end(skb, nlh);
2849                         goto out;
2850                 }
2851                 for (i = 0; i < mrt->maxvif; i++) {
2852                         if (e < s_e)
2853                                 goto skip_entry;
2854                         if (!ipmr_fill_vif(mrt, i, skb)) {
2855                                 nla_nest_end(skb, vifs);
2856                                 nla_nest_end(skb, af);
2857                                 nlmsg_end(skb, nlh);
2858                                 goto out;
2859                         }
2860 skip_entry:
2861                         e++;
2862                 }
2863                 s_e = 0;
2864                 e = 0;
2865                 nla_nest_end(skb, vifs);
2866                 nla_nest_end(skb, af);
2867                 nlmsg_end(skb, nlh);
2868 skip_table:
2869                 t++;
2870         }
2871
2872 out:
2873         cb->args[1] = e;
2874         cb->args[0] = t;
2875
2876         return skb->len;
2877 }
2878
2879 #ifdef CONFIG_PROC_FS
2880 /* The /proc interfaces to multicast routing :
2881  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2882  */
2883
2884 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2885         __acquires(mrt_lock)
2886 {
2887         struct mr_vif_iter *iter = seq->private;
2888         struct net *net = seq_file_net(seq);
2889         struct mr_table *mrt;
2890
2891         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2892         if (!mrt)
2893                 return ERR_PTR(-ENOENT);
2894
2895         iter->mrt = mrt;
2896
2897         read_lock(&mrt_lock);
2898         return mr_vif_seq_start(seq, pos);
2899 }
2900
2901 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2902         __releases(mrt_lock)
2903 {
2904         read_unlock(&mrt_lock);
2905 }
2906
2907 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2908 {
2909         struct mr_vif_iter *iter = seq->private;
2910         struct mr_table *mrt = iter->mrt;
2911
2912         if (v == SEQ_START_TOKEN) {
2913                 seq_puts(seq,
2914                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2915         } else {
2916                 const struct vif_device *vif = v;
2917                 const char *name =  vif->dev ?
2918                                     vif->dev->name : "none";
2919
2920                 seq_printf(seq,
2921                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2922                            vif - mrt->vif_table,
2923                            name, vif->bytes_in, vif->pkt_in,
2924                            vif->bytes_out, vif->pkt_out,
2925                            vif->flags, vif->local, vif->remote);
2926         }
2927         return 0;
2928 }
2929
2930 static const struct seq_operations ipmr_vif_seq_ops = {
2931         .start = ipmr_vif_seq_start,
2932         .next  = mr_vif_seq_next,
2933         .stop  = ipmr_vif_seq_stop,
2934         .show  = ipmr_vif_seq_show,
2935 };
2936
2937 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2938 {
2939         struct net *net = seq_file_net(seq);
2940         struct mr_table *mrt;
2941
2942         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2943         if (!mrt)
2944                 return ERR_PTR(-ENOENT);
2945
2946         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2947 }
2948
2949 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2950 {
2951         int n;
2952
2953         if (v == SEQ_START_TOKEN) {
2954                 seq_puts(seq,
2955                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2956         } else {
2957                 const struct mfc_cache *mfc = v;
2958                 const struct mr_mfc_iter *it = seq->private;
2959                 const struct mr_table *mrt = it->mrt;
2960
2961                 seq_printf(seq, "%08X %08X %-3hd",
2962                            (__force u32) mfc->mfc_mcastgrp,
2963                            (__force u32) mfc->mfc_origin,
2964                            mfc->_c.mfc_parent);
2965
2966                 if (it->cache != &mrt->mfc_unres_queue) {
2967                         seq_printf(seq, " %8lu %8lu %8lu",
2968                                    mfc->_c.mfc_un.res.pkt,
2969                                    mfc->_c.mfc_un.res.bytes,
2970                                    mfc->_c.mfc_un.res.wrong_if);
2971                         for (n = mfc->_c.mfc_un.res.minvif;
2972                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2973                                 if (VIF_EXISTS(mrt, n) &&
2974                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2975                                         seq_printf(seq,
2976                                            " %2d:%-3d",
2977                                            n, mfc->_c.mfc_un.res.ttls[n]);
2978                         }
2979                 } else {
2980                         /* unresolved mfc_caches don't contain
2981                          * pkt, bytes and wrong_if values
2982                          */
2983                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2984                 }
2985                 seq_putc(seq, '\n');
2986         }
2987         return 0;
2988 }
2989
2990 static const struct seq_operations ipmr_mfc_seq_ops = {
2991         .start = ipmr_mfc_seq_start,
2992         .next  = mr_mfc_seq_next,
2993         .stop  = mr_mfc_seq_stop,
2994         .show  = ipmr_mfc_seq_show,
2995 };
2996 #endif
2997
2998 #ifdef CONFIG_IP_PIMSM_V2
2999 static const struct net_protocol pim_protocol = {
3000         .handler        =       pim_rcv,
3001         .netns_ok       =       1,
3002 };
3003 #endif
3004
3005 static unsigned int ipmr_seq_read(struct net *net)
3006 {
3007         ASSERT_RTNL();
3008
3009         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3010 }
3011
3012 static int ipmr_dump(struct net *net, struct notifier_block *nb,
3013                      struct netlink_ext_ack *extack)
3014 {
3015         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
3016                        ipmr_mr_table_iter, &mrt_lock, extack);
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_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
3062                         sizeof(struct mr_vif_iter)))
3063                 goto proc_vif_fail;
3064         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
3065                         sizeof(struct mr_mfc_iter)))
3066                 goto proc_cache_fail;
3067 #endif
3068         return 0;
3069
3070 #ifdef CONFIG_PROC_FS
3071 proc_cache_fail:
3072         remove_proc_entry("ip_mr_vif", net->proc_net);
3073 proc_vif_fail:
3074         ipmr_rules_exit(net);
3075 #endif
3076 ipmr_rules_fail:
3077         ipmr_notifier_exit(net);
3078 ipmr_notifier_fail:
3079         return err;
3080 }
3081
3082 static void __net_exit ipmr_net_exit(struct net *net)
3083 {
3084 #ifdef CONFIG_PROC_FS
3085         remove_proc_entry("ip_mr_cache", net->proc_net);
3086         remove_proc_entry("ip_mr_vif", net->proc_net);
3087 #endif
3088         ipmr_notifier_exit(net);
3089         ipmr_rules_exit(net);
3090 }
3091
3092 static struct pernet_operations ipmr_net_ops = {
3093         .init = ipmr_net_init,
3094         .exit = ipmr_net_exit,
3095 };
3096
3097 int __init ip_mr_init(void)
3098 {
3099         int err;
3100
3101         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3102                                        sizeof(struct mfc_cache),
3103                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3104                                        NULL);
3105
3106         err = register_pernet_subsys(&ipmr_net_ops);
3107         if (err)
3108                 goto reg_pernet_fail;
3109
3110         err = register_netdevice_notifier(&ip_mr_notifier);
3111         if (err)
3112                 goto reg_notif_fail;
3113 #ifdef CONFIG_IP_PIMSM_V2
3114         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3115                 pr_err("%s: can't add PIM protocol\n", __func__);
3116                 err = -EAGAIN;
3117                 goto add_proto_fail;
3118         }
3119 #endif
3120         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3121                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3122         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3123                       ipmr_rtm_route, NULL, 0);
3124         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3125                       ipmr_rtm_route, NULL, 0);
3126
3127         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3128                       NULL, ipmr_rtm_dumplink, 0);
3129         return 0;
3130
3131 #ifdef CONFIG_IP_PIMSM_V2
3132 add_proto_fail:
3133         unregister_netdevice_notifier(&ip_mr_notifier);
3134 #endif
3135 reg_notif_fail:
3136         unregister_pernet_subsys(&ipmr_net_ops);
3137 reg_pernet_fail:
3138         kmem_cache_destroy(mrt_cachep);
3139         return err;
3140 }