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  *      @mrt: Table to delete from
 640  *      @vifi: VIF identifier to delete
 641  *      @notify: Set to 1, if the caller is a notifier_call
 642  *      @head: if unregistering the VIF, place it on this queue
 643  */
 644 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 645                       struct list_head *head)
 646 {
 647         struct net *net = read_pnet(&mrt->net);
 648         struct vif_device *v;
 649         struct net_device *dev;
 650         struct in_device *in_dev;
 651
 652         if (vifi < 0 || vifi >= mrt->maxvif)
 653                 return -EADDRNOTAVAIL;
 654
 655         v = &mrt->vif_table[vifi];
 656
 657         if (VIF_EXISTS(mrt, vifi))
 658                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 659                                               mrt->id);
 660
 661         write_lock_bh(&mrt_lock);
 662         dev = v->dev;
 663         v->dev = NULL;
 664
 665         if (!dev) {
 666                 write_unlock_bh(&mrt_lock);
 667                 return -EADDRNOTAVAIL;
 668         }
 669
 670         if (vifi == mrt->mroute_reg_vif_num)
 671                 mrt->mroute_reg_vif_num = -1;
 672
 673         if (vifi + 1 == mrt->maxvif) {
 674                 int tmp;
 675
 676                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 677                         if (VIF_EXISTS(mrt, tmp))
 678                                 break;
 679                 }
 680                 mrt->maxvif = tmp+1;
 681         }
 682
 683         write_unlock_bh(&mrt_lock);
 684
 685         dev_set_allmulti(dev, -1);
 686
 687         in_dev = __in_dev_get_rtnl(dev);
 688         if (in_dev) {
 689                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 690                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 691                                             NETCONFA_MC_FORWARDING,
 692                                             dev->ifindex, &in_dev->cnf);
 693                 ip_rt_multicast_event(in_dev);
 694         }
 695
 696         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 697                 unregister_netdevice_queue(dev, head);
 698
 699         dev_put(dev);
 700         return 0;
 701 }
 702
 703 static void ipmr_cache_free_rcu(struct rcu_head *head)
 704 {
 705         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 706
 707         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 708 }
 709
 710 static void ipmr_cache_free(struct mfc_cache *c)
 711 {
 712         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 713 }
 714
 715 /* Destroy an unresolved cache entry, killing queued skbs
 716  * and reporting error to netlink readers.
 717  */
 718 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 719 {
 720         struct net *net = read_pnet(&mrt->net);
 721         struct sk_buff *skb;
 722         struct nlmsgerr *e;
 723
 724         atomic_dec(&mrt->cache_resolve_queue_len);
 725
 726         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 727                 if (ip_hdr(skb)->version == 0) {
 728                         struct nlmsghdr *nlh = skb_pull(skb,
 729                                                         sizeof(struct iphdr));
 730                         nlh->nlmsg_type = NLMSG_ERROR;
 731                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 732                         skb_trim(skb, nlh->nlmsg_len);
 733                         e = nlmsg_data(nlh);
 734                         e->error = -ETIMEDOUT;
 735                         memset(&e->msg, 0, sizeof(e->msg));
 736
 737                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 738                 } else {
 739                         kfree_skb(skb);
 740                 }
 741         }
 742
 743         ipmr_cache_free(c);
 744 }
 745
 746 /* Timer process for the unresolved queue. */
 747 static void ipmr_expire_process(struct timer_list *t)
 748 {
 749         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 750         struct mr_mfc *c, *next;
 751         unsigned long expires;
 752         unsigned long now;
 753
 754         if (!spin_trylock(&mfc_unres_lock)) {
 755                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 756                 return;
 757         }
 758
 759         if (list_empty(&mrt->mfc_unres_queue))
 760                 goto out;
 761
 762         now = jiffies;
 763         expires = 10*HZ;
 764
 765         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 766                 if (time_after(c->mfc_un.unres.expires, now)) {
 767                         unsigned long interval = c->mfc_un.unres.expires - now;
 768                         if (interval < expires)
 769                                 expires = interval;
 770                         continue;
 771                 }
 772
 773                 list_del(&c->list);
 774                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 775                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 776         }
 777
 778         if (!list_empty(&mrt->mfc_unres_queue))
 779                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 780
 781 out:
 782         spin_unlock(&mfc_unres_lock);
 783 }
 784
 785 /* Fill oifs list. It is called under write locked mrt_lock. */
 786 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 787                                    unsigned char *ttls)
 788 {
 789         int vifi;
 790
 791         cache->mfc_un.res.minvif = MAXVIFS;
 792         cache->mfc_un.res.maxvif = 0;
 793         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 794
 795         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 796                 if (VIF_EXISTS(mrt, vifi) &&
 797                     ttls[vifi] && ttls[vifi] < 255) {
 798                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 799                         if (cache->mfc_un.res.minvif > vifi)
 800                                 cache->mfc_un.res.minvif = vifi;
 801                         if (cache->mfc_un.res.maxvif <= vifi)
 802                                 cache->mfc_un.res.maxvif = vifi + 1;
 803                 }
 804         }
 805         cache->mfc_un.res.lastuse = jiffies;
 806 }
 807
 808 static int vif_add(struct net *net, struct mr_table *mrt,
 809                    struct vifctl *vifc, int mrtsock)
 810 {
 811         struct netdev_phys_item_id ppid = { };
 812         int vifi = vifc->vifc_vifi;
 813         struct vif_device *v = &mrt->vif_table[vifi];
 814         struct net_device *dev;
 815         struct in_device *in_dev;
 816         int err;
 817
 818         /* Is vif busy ? */
 819         if (VIF_EXISTS(mrt, vifi))
 820                 return -EADDRINUSE;
 821
 822         switch (vifc->vifc_flags) {
 823         case VIFF_REGISTER:
 824                 if (!ipmr_pimsm_enabled())
 825                         return -EINVAL;
 826                 /* Special Purpose VIF in PIM
 827                  * All the packets will be sent to the daemon
 828                  */
 829                 if (mrt->mroute_reg_vif_num >= 0)
 830                         return -EADDRINUSE;
 831                 dev = ipmr_reg_vif(net, mrt);
 832                 if (!dev)
 833                         return -ENOBUFS;
 834                 err = dev_set_allmulti(dev, 1);
 835                 if (err) {
 836                         unregister_netdevice(dev);
 837                         dev_put(dev);
 838                         return err;
 839                 }
 840                 break;
 841         case VIFF_TUNNEL:
 842                 dev = ipmr_new_tunnel(net, vifc);
 843                 if (IS_ERR(dev))
 844                         return PTR_ERR(dev);
 845                 break;
 846         case VIFF_USE_IFINDEX:
 847         case 0:
 848                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 849                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 850                         if (dev && !__in_dev_get_rtnl(dev)) {
 851                                 dev_put(dev);
 852                                 return -EADDRNOTAVAIL;
 853                         }
 854                 } else {
 855                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 856                 }
 857                 if (!dev)
 858                         return -EADDRNOTAVAIL;
 859                 err = dev_set_allmulti(dev, 1);
 860                 if (err) {
 861                         dev_put(dev);
 862                         return err;
 863                 }
 864                 break;
 865         default:
 866                 return -EINVAL;
 867         }
 868
 869         in_dev = __in_dev_get_rtnl(dev);
 870         if (!in_dev) {
 871                 dev_put(dev);
 872                 return -EADDRNOTAVAIL;
 873         }
 874         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 875         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 876                                     dev->ifindex, &in_dev->cnf);
 877         ip_rt_multicast_event(in_dev);
 878
 879         /* Fill in the VIF structures */
 880         vif_device_init(v, dev, vifc->vifc_rate_limit,
 881                         vifc->vifc_threshold,
 882                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 883                         (VIFF_TUNNEL | VIFF_REGISTER));
 884
 885         err = dev_get_port_parent_id(dev, &ppid, true);
 886         if (err == 0) {
 887                 memcpy(v->dev_parent_id.id, ppid.id, ppid.id_len);
 888                 v->dev_parent_id.id_len = ppid.id_len;
 889         } else {
 890                 v->dev_parent_id.id_len = 0;
 891         }
 892
 893         v->local = vifc->vifc_lcl_addr.s_addr;
 894         v->remote = vifc->vifc_rmt_addr.s_addr;
 895
 896         /* And finish update writing critical data */
 897         write_lock_bh(&mrt_lock);
 898         v->dev = dev;
 899         if (v->flags & VIFF_REGISTER)
 900                 mrt->mroute_reg_vif_num = vifi;
 901         if (vifi+1 > mrt->maxvif)
 902                 mrt->maxvif = vifi+1;
 903         write_unlock_bh(&mrt_lock);
 904         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 905         return 0;
 906 }
 907
 908 /* called with rcu_read_lock() */
 909 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 910                                          __be32 origin,
 911                                          __be32 mcastgrp)
 912 {
 913         struct mfc_cache_cmp_arg arg = {
 914                         .mfc_mcastgrp = mcastgrp,
 915                         .mfc_origin = origin
 916         };
 917
 918         return mr_mfc_find(mrt, &arg);
 919 }
 920
 921 /* Look for a (*,G) entry */
 922 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 923                                              __be32 mcastgrp, int vifi)
 924 {
 925         struct mfc_cache_cmp_arg arg = {
 926                         .mfc_mcastgrp = mcastgrp,
 927                         .mfc_origin = htonl(INADDR_ANY)
 928         };
 929
 930         if (mcastgrp == htonl(INADDR_ANY))
 931                 return mr_mfc_find_any_parent(mrt, vifi);
 932         return mr_mfc_find_any(mrt, vifi, &arg);
 933 }
 934
 935 /* Look for a (S,G,iif) entry if parent != -1 */
 936 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 937                                                 __be32 origin, __be32 mcastgrp,
 938                                                 int parent)
 939 {
 940         struct mfc_cache_cmp_arg arg = {
 941                         .mfc_mcastgrp = mcastgrp,
 942                         .mfc_origin = origin,
 943         };
 944
 945         return mr_mfc_find_parent(mrt, &arg, parent);
 946 }
 947
 948 /* Allocate a multicast cache entry */
 949 static struct mfc_cache *ipmr_cache_alloc(void)
 950 {
 951         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 952
 953         if (c) {
 954                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 955                 c->_c.mfc_un.res.minvif = MAXVIFS;
 956                 c->_c.free = ipmr_cache_free_rcu;
 957                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 958         }
 959         return c;
 960 }
 961
 962 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 963 {
 964         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 965
 966         if (c) {
 967                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
 968                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
 969         }
 970         return c;
 971 }
 972
 973 /* A cache entry has gone into a resolved state from queued */
 974 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 975                                struct mfc_cache *uc, struct mfc_cache *c)
 976 {
 977         struct sk_buff *skb;
 978         struct nlmsgerr *e;
 979
 980         /* Play the pending entries through our router */
 981         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
 982                 if (ip_hdr(skb)->version == 0) {
 983                         struct nlmsghdr *nlh = skb_pull(skb,
 984                                                         sizeof(struct iphdr));
 985
 986                         if (mr_fill_mroute(mrt, skb, &c->_c,
 987                                            nlmsg_data(nlh)) > 0) {
 988                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 989                                                  (u8 *)nlh;
 990                         } else {
 991                                 nlh->nlmsg_type = NLMSG_ERROR;
 992                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 993                                 skb_trim(skb, nlh->nlmsg_len);
 994                                 e = nlmsg_data(nlh);
 995                                 e->error = -EMSGSIZE;
 996                                 memset(&e->msg, 0, sizeof(e->msg));
 997                         }
 998
 999                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1000                 } else {
1001                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1002                 }
1003         }
1004 }
1005
1006 /* Bounce a cache query up to mrouted and netlink.
1007  *
1008  * Called under mrt_lock.
1009  */
1010 static int ipmr_cache_report(struct mr_table *mrt,
1011                              struct sk_buff *pkt, vifi_t vifi, int assert)
1012 {
1013         const int ihl = ip_hdrlen(pkt);
1014         struct sock *mroute_sk;
1015         struct igmphdr *igmp;
1016         struct igmpmsg *msg;
1017         struct sk_buff *skb;
1018         int ret;
1019
1020         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1021                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1022         else
1023                 skb = alloc_skb(128, GFP_ATOMIC);
1024
1025         if (!skb)
1026                 return -ENOBUFS;
1027
1028         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1029                 /* Ugly, but we have no choice with this interface.
1030                  * Duplicate old header, fix ihl, length etc.
1031                  * And all this only to mangle msg->im_msgtype and
1032                  * to set msg->im_mbz to "mbz" :-)
1033                  */
1034                 skb_push(skb, sizeof(struct iphdr));
1035                 skb_reset_network_header(skb);
1036                 skb_reset_transport_header(skb);
1037                 msg = (struct igmpmsg *)skb_network_header(skb);
1038                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1039                 msg->im_msgtype = assert;
1040                 msg->im_mbz = 0;
1041                 if (assert == IGMPMSG_WRVIFWHOLE) {
1042                         msg->im_vif = vifi;
1043                         msg->im_vif_hi = vifi >> 8;
1044                 } else {
1045                         msg->im_vif = mrt->mroute_reg_vif_num;
1046                         msg->im_vif_hi = mrt->mroute_reg_vif_num >> 8;
1047                 }
1048                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1049                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1050                                              sizeof(struct iphdr));
1051         } else {
1052                 /* Copy the IP header */
1053                 skb_set_network_header(skb, skb->len);
1054                 skb_put(skb, ihl);
1055                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1056                 /* Flag to the kernel this is a route add */
1057                 ip_hdr(skb)->protocol = 0;
1058                 msg = (struct igmpmsg *)skb_network_header(skb);
1059                 msg->im_vif = vifi;
1060                 msg->im_vif_hi = vifi >> 8;
1061                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1062                 /* Add our header */
1063                 igmp = skb_put(skb, sizeof(struct igmphdr));
1064                 igmp->type = assert;
1065                 msg->im_msgtype = assert;
1066                 igmp->code = 0;
1067                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1068                 skb->transport_header = skb->network_header;
1069         }
1070
1071         rcu_read_lock();
1072         mroute_sk = rcu_dereference(mrt->mroute_sk);
1073         if (!mroute_sk) {
1074                 rcu_read_unlock();
1075                 kfree_skb(skb);
1076                 return -EINVAL;
1077         }
1078
1079         igmpmsg_netlink_event(mrt, skb);
1080
1081         /* Deliver to mrouted */
1082         ret = sock_queue_rcv_skb(mroute_sk, skb);
1083         rcu_read_unlock();
1084         if (ret < 0) {
1085                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1086                 kfree_skb(skb);
1087         }
1088
1089         return ret;
1090 }
1091
1092 /* Queue a packet for resolution. It gets locked cache entry! */
1093 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1094                                  struct sk_buff *skb, struct net_device *dev)
1095 {
1096         const struct iphdr *iph = ip_hdr(skb);
1097         struct mfc_cache *c;
1098         bool found = false;
1099         int err;
1100
1101         spin_lock_bh(&mfc_unres_lock);
1102         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1103                 if (c->mfc_mcastgrp == iph->daddr &&
1104                     c->mfc_origin == iph->saddr) {
1105                         found = true;
1106                         break;
1107                 }
1108         }
1109
1110         if (!found) {
1111                 /* Create a new entry if allowable */
1112                 c = ipmr_cache_alloc_unres();
1113                 if (!c) {
1114                         spin_unlock_bh(&mfc_unres_lock);
1115
1116                         kfree_skb(skb);
1117                         return -ENOBUFS;
1118                 }
1119
1120                 /* Fill in the new cache entry */
1121                 c->_c.mfc_parent = -1;
1122                 c->mfc_origin   = iph->saddr;
1123                 c->mfc_mcastgrp = iph->daddr;
1124
1125                 /* Reflect first query at mrouted. */
1126                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1127
1128                 if (err < 0) {
1129                         /* If the report failed throw the cache entry
1130                            out - Brad Parker
1131                          */
1132                         spin_unlock_bh(&mfc_unres_lock);
1133
1134                         ipmr_cache_free(c);
1135                         kfree_skb(skb);
1136                         return err;
1137                 }
1138
1139                 atomic_inc(&mrt->cache_resolve_queue_len);
1140                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1141                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1142
1143                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1144                         mod_timer(&mrt->ipmr_expire_timer,
1145                                   c->_c.mfc_un.unres.expires);
1146         }
1147
1148         /* See if we can append the packet */
1149         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1150                 kfree_skb(skb);
1151                 err = -ENOBUFS;
1152         } else {
1153                 if (dev) {
1154                         skb->dev = dev;
1155                         skb->skb_iif = dev->ifindex;
1156                 }
1157                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1158                 err = 0;
1159         }
1160
1161         spin_unlock_bh(&mfc_unres_lock);
1162         return err;
1163 }
1164
1165 /* MFC cache manipulation by user space mroute daemon */
1166
1167 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1168 {
1169         struct net *net = read_pnet(&mrt->net);
1170         struct mfc_cache *c;
1171
1172         /* The entries are added/deleted only under RTNL */
1173         rcu_read_lock();
1174         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1175                                    mfc->mfcc_mcastgrp.s_addr, parent);
1176         rcu_read_unlock();
1177         if (!c)
1178                 return -ENOENT;
1179         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1180         list_del_rcu(&c->_c.list);
1181         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1182         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1183         mr_cache_put(&c->_c);
1184
1185         return 0;
1186 }
1187
1188 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1189                         struct mfcctl *mfc, int mrtsock, int parent)
1190 {
1191         struct mfc_cache *uc, *c;
1192         struct mr_mfc *_uc;
1193         bool found;
1194         int ret;
1195
1196         if (mfc->mfcc_parent >= MAXVIFS)
1197                 return -ENFILE;
1198
1199         /* The entries are added/deleted only under RTNL */
1200         rcu_read_lock();
1201         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1202                                    mfc->mfcc_mcastgrp.s_addr, parent);
1203         rcu_read_unlock();
1204         if (c) {
1205                 write_lock_bh(&mrt_lock);
1206                 c->_c.mfc_parent = mfc->mfcc_parent;
1207                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1208                 if (!mrtsock)
1209                         c->_c.mfc_flags |= MFC_STATIC;
1210                 write_unlock_bh(&mrt_lock);
1211                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1212                                               mrt->id);
1213                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1214                 return 0;
1215         }
1216
1217         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1218             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1219                 return -EINVAL;
1220
1221         c = ipmr_cache_alloc();
1222         if (!c)
1223                 return -ENOMEM;
1224
1225         c->mfc_origin = mfc->mfcc_origin.s_addr;
1226         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1227         c->_c.mfc_parent = mfc->mfcc_parent;
1228         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1229         if (!mrtsock)
1230                 c->_c.mfc_flags |= MFC_STATIC;
1231
1232         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1233                                   ipmr_rht_params);
1234         if (ret) {
1235                 pr_err("ipmr: rhtable insert error %d\n", ret);
1236                 ipmr_cache_free(c);
1237                 return ret;
1238         }
1239         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1240         /* Check to see if we resolved a queued list. If so we
1241          * need to send on the frames and tidy up.
1242          */
1243         found = false;
1244         spin_lock_bh(&mfc_unres_lock);
1245         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1246                 uc = (struct mfc_cache *)_uc;
1247                 if (uc->mfc_origin == c->mfc_origin &&
1248                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1249                         list_del(&_uc->list);
1250                         atomic_dec(&mrt->cache_resolve_queue_len);
1251                         found = true;
1252                         break;
1253                 }
1254         }
1255         if (list_empty(&mrt->mfc_unres_queue))
1256                 del_timer(&mrt->ipmr_expire_timer);
1257         spin_unlock_bh(&mfc_unres_lock);
1258
1259         if (found) {
1260                 ipmr_cache_resolve(net, mrt, uc, c);
1261                 ipmr_cache_free(uc);
1262         }
1263         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1264         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1265         return 0;
1266 }
1267
1268 /* Close the multicast socket, and clear the vif tables etc */
1269 static void mroute_clean_tables(struct mr_table *mrt, int flags)
1270 {
1271         struct net *net = read_pnet(&mrt->net);
1272         struct mr_mfc *c, *tmp;
1273         struct mfc_cache *cache;
1274         LIST_HEAD(list);
1275         int i;
1276
1277         /* Shut down all active vif entries */
1278         if (flags & (MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC)) {
1279                 for (i = 0; i < mrt->maxvif; i++) {
1280                         if (((mrt->vif_table[i].flags & VIFF_STATIC) &&
1281                              !(flags & MRT_FLUSH_VIFS_STATIC)) ||
1282                             (!(mrt->vif_table[i].flags & VIFF_STATIC) && !(flags & MRT_FLUSH_VIFS)))
1283                                 continue;
1284                         vif_delete(mrt, i, 0, &list);
1285                 }
1286                 unregister_netdevice_many(&list);
1287         }
1288
1289         /* Wipe the cache */
1290         if (flags & (MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC)) {
1291                 list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1292                         if (((c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC_STATIC)) ||
1293                             (!(c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC)))
1294                                 continue;
1295                         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1296                         list_del_rcu(&c->list);
1297                         cache = (struct mfc_cache *)c;
1298                         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1299                                                       mrt->id);
1300                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1301                         mr_cache_put(c);
1302                 }
1303         }
1304
1305         if (flags & MRT_FLUSH_MFC) {
1306                 if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1307                         spin_lock_bh(&mfc_unres_lock);
1308                         list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1309                                 list_del(&c->list);
1310                                 cache = (struct mfc_cache *)c;
1311                                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1312                                 ipmr_destroy_unres(mrt, cache);
1313                         }
1314                         spin_unlock_bh(&mfc_unres_lock);
1315                 }
1316         }
1317 }
1318
1319 /* called from ip_ra_control(), before an RCU grace period,
1320  * we dont need to call synchronize_rcu() here
1321  */
1322 static void mrtsock_destruct(struct sock *sk)
1323 {
1324         struct net *net = sock_net(sk);
1325         struct mr_table *mrt;
1326
1327         rtnl_lock();
1328         ipmr_for_each_table(mrt, net) {
1329                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1330                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1331                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1332                                                     NETCONFA_MC_FORWARDING,
1333                                                     NETCONFA_IFINDEX_ALL,
1334                                                     net->ipv4.devconf_all);
1335                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1336                         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_MFC);
1337                 }
1338         }
1339         rtnl_unlock();
1340 }
1341
1342 /* Socket options and virtual interface manipulation. The whole
1343  * virtual interface system is a complete heap, but unfortunately
1344  * that's how BSD mrouted happens to think. Maybe one day with a proper
1345  * MOSPF/PIM router set up we can clean this up.
1346  */
1347
1348 int ip_mroute_setsockopt(struct sock *sk, int optname, sockptr_t optval,
1349                          unsigned int optlen)
1350 {
1351         struct net *net = sock_net(sk);
1352         int val, ret = 0, parent = 0;
1353         struct mr_table *mrt;
1354         struct vifctl vif;
1355         struct mfcctl mfc;
1356         bool do_wrvifwhole;
1357         u32 uval;
1358
1359         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1360         rtnl_lock();
1361         if (sk->sk_type != SOCK_RAW ||
1362             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1363                 ret = -EOPNOTSUPP;
1364                 goto out_unlock;
1365         }
1366
1367         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1368         if (!mrt) {
1369                 ret = -ENOENT;
1370                 goto out_unlock;
1371         }
1372         if (optname != MRT_INIT) {
1373                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1374                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1375                         ret = -EACCES;
1376                         goto out_unlock;
1377                 }
1378         }
1379
1380         switch (optname) {
1381         case MRT_INIT:
1382                 if (optlen != sizeof(int)) {
1383                         ret = -EINVAL;
1384                         break;
1385                 }
1386                 if (rtnl_dereference(mrt->mroute_sk)) {
1387                         ret = -EADDRINUSE;
1388                         break;
1389                 }
1390
1391                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1392                 if (ret == 0) {
1393                         rcu_assign_pointer(mrt->mroute_sk, sk);
1394                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1395                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1396                                                     NETCONFA_MC_FORWARDING,
1397                                                     NETCONFA_IFINDEX_ALL,
1398                                                     net->ipv4.devconf_all);
1399                 }
1400                 break;
1401         case MRT_DONE:
1402                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1403                         ret = -EACCES;
1404                 } else {
1405                         /* We need to unlock here because mrtsock_destruct takes
1406                          * care of rtnl itself and we can't change that due to
1407                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1408                          */
1409                         rtnl_unlock();
1410                         ret = ip_ra_control(sk, 0, NULL);
1411                         goto out;
1412                 }
1413                 break;
1414         case MRT_ADD_VIF:
1415         case MRT_DEL_VIF:
1416                 if (optlen != sizeof(vif)) {
1417                         ret = -EINVAL;
1418                         break;
1419                 }
1420                 if (copy_from_sockptr(&vif, optval, sizeof(vif))) {
1421                         ret = -EFAULT;
1422                         break;
1423                 }
1424                 if (vif.vifc_vifi >= MAXVIFS) {
1425                         ret = -ENFILE;
1426                         break;
1427                 }
1428                 if (optname == MRT_ADD_VIF) {
1429                         ret = vif_add(net, mrt, &vif,
1430                                       sk == rtnl_dereference(mrt->mroute_sk));
1431                 } else {
1432                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1433                 }
1434                 break;
1435         /* Manipulate the forwarding caches. These live
1436          * in a sort of kernel/user symbiosis.
1437          */
1438         case MRT_ADD_MFC:
1439         case MRT_DEL_MFC:
1440                 parent = -1;
1441                 fallthrough;
1442         case MRT_ADD_MFC_PROXY:
1443         case MRT_DEL_MFC_PROXY:
1444                 if (optlen != sizeof(mfc)) {
1445                         ret = -EINVAL;
1446                         break;
1447                 }
1448                 if (copy_from_sockptr(&mfc, optval, sizeof(mfc))) {
1449                         ret = -EFAULT;
1450                         break;
1451                 }
1452                 if (parent == 0)
1453                         parent = mfc.mfcc_parent;
1454                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1455                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1456                 else
1457                         ret = ipmr_mfc_add(net, mrt, &mfc,
1458                                            sk == rtnl_dereference(mrt->mroute_sk),
1459                                            parent);
1460                 break;
1461         case MRT_FLUSH:
1462                 if (optlen != sizeof(val)) {
1463                         ret = -EINVAL;
1464                         break;
1465                 }
1466                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1467                         ret = -EFAULT;
1468                         break;
1469                 }
1470                 mroute_clean_tables(mrt, val);
1471                 break;
1472         /* Control PIM assert. */
1473         case MRT_ASSERT:
1474                 if (optlen != sizeof(val)) {
1475                         ret = -EINVAL;
1476                         break;
1477                 }
1478                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1479                         ret = -EFAULT;
1480                         break;
1481                 }
1482                 mrt->mroute_do_assert = val;
1483                 break;
1484         case MRT_PIM:
1485                 if (!ipmr_pimsm_enabled()) {
1486                         ret = -ENOPROTOOPT;
1487                         break;
1488                 }
1489                 if (optlen != sizeof(val)) {
1490                         ret = -EINVAL;
1491                         break;
1492                 }
1493                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1494                         ret = -EFAULT;
1495                         break;
1496                 }
1497
1498                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1499                 val = !!val;
1500                 if (val != mrt->mroute_do_pim) {
1501                         mrt->mroute_do_pim = val;
1502                         mrt->mroute_do_assert = val;
1503                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1504                 }
1505                 break;
1506         case MRT_TABLE:
1507                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1508                         ret = -ENOPROTOOPT;
1509                         break;
1510                 }
1511                 if (optlen != sizeof(uval)) {
1512                         ret = -EINVAL;
1513                         break;
1514                 }
1515                 if (copy_from_sockptr(&uval, optval, sizeof(uval))) {
1516                         ret = -EFAULT;
1517                         break;
1518                 }
1519
1520                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1521                         ret = -EBUSY;
1522                 } else {
1523                         mrt = ipmr_new_table(net, uval);
1524                         if (IS_ERR(mrt))
1525                                 ret = PTR_ERR(mrt);
1526                         else
1527                                 raw_sk(sk)->ipmr_table = uval;
1528                 }
1529                 break;
1530         /* Spurious command, or MRT_VERSION which you cannot set. */
1531         default:
1532                 ret = -ENOPROTOOPT;
1533         }
1534 out_unlock:
1535         rtnl_unlock();
1536 out:
1537         return ret;
1538 }
1539
1540 /* Getsock opt support for the multicast routing system. */
1541 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1542 {
1543         int olr;
1544         int val;
1545         struct net *net = sock_net(sk);
1546         struct mr_table *mrt;
1547
1548         if (sk->sk_type != SOCK_RAW ||
1549             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1550                 return -EOPNOTSUPP;
1551
1552         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1553         if (!mrt)
1554                 return -ENOENT;
1555
1556         switch (optname) {
1557         case MRT_VERSION:
1558                 val = 0x0305;
1559                 break;
1560         case MRT_PIM:
1561                 if (!ipmr_pimsm_enabled())
1562                         return -ENOPROTOOPT;
1563                 val = mrt->mroute_do_pim;
1564                 break;
1565         case MRT_ASSERT:
1566                 val = mrt->mroute_do_assert;
1567                 break;
1568         default:
1569                 return -ENOPROTOOPT;
1570         }
1571
1572         if (get_user(olr, optlen))
1573                 return -EFAULT;
1574         olr = min_t(unsigned int, olr, sizeof(int));
1575         if (olr < 0)
1576                 return -EINVAL;
1577         if (put_user(olr, optlen))
1578                 return -EFAULT;
1579         if (copy_to_user(optval, &val, olr))
1580                 return -EFAULT;
1581         return 0;
1582 }
1583
1584 /* The IP multicast ioctl support routines. */
1585 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1586 {
1587         struct sioc_sg_req sr;
1588         struct sioc_vif_req vr;
1589         struct vif_device *vif;
1590         struct mfc_cache *c;
1591         struct net *net = sock_net(sk);
1592         struct mr_table *mrt;
1593
1594         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1595         if (!mrt)
1596                 return -ENOENT;
1597
1598         switch (cmd) {
1599         case SIOCGETVIFCNT:
1600                 if (copy_from_user(&vr, arg, sizeof(vr)))
1601                         return -EFAULT;
1602                 if (vr.vifi >= mrt->maxvif)
1603                         return -EINVAL;
1604                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1605                 read_lock(&mrt_lock);
1606                 vif = &mrt->vif_table[vr.vifi];
1607                 if (VIF_EXISTS(mrt, vr.vifi)) {
1608                         vr.icount = vif->pkt_in;
1609                         vr.ocount = vif->pkt_out;
1610                         vr.ibytes = vif->bytes_in;
1611                         vr.obytes = vif->bytes_out;
1612                         read_unlock(&mrt_lock);
1613
1614                         if (copy_to_user(arg, &vr, sizeof(vr)))
1615                                 return -EFAULT;
1616                         return 0;
1617                 }
1618                 read_unlock(&mrt_lock);
1619                 return -EADDRNOTAVAIL;
1620         case SIOCGETSGCNT:
1621                 if (copy_from_user(&sr, arg, sizeof(sr)))
1622                         return -EFAULT;
1623
1624                 rcu_read_lock();
1625                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1626                 if (c) {
1627                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1628                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1629                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1630                         rcu_read_unlock();
1631
1632                         if (copy_to_user(arg, &sr, sizeof(sr)))
1633                                 return -EFAULT;
1634                         return 0;
1635                 }
1636                 rcu_read_unlock();
1637                 return -EADDRNOTAVAIL;
1638         default:
1639                 return -ENOIOCTLCMD;
1640         }
1641 }
1642
1643 #ifdef CONFIG_COMPAT
1644 struct compat_sioc_sg_req {
1645         struct in_addr src;
1646         struct in_addr grp;
1647         compat_ulong_t pktcnt;
1648         compat_ulong_t bytecnt;
1649         compat_ulong_t wrong_if;
1650 };
1651
1652 struct compat_sioc_vif_req {
1653         vifi_t  vifi;           /* Which iface */
1654         compat_ulong_t icount;
1655         compat_ulong_t ocount;
1656         compat_ulong_t ibytes;
1657         compat_ulong_t obytes;
1658 };
1659
1660 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1661 {
1662         struct compat_sioc_sg_req sr;
1663         struct compat_sioc_vif_req vr;
1664         struct vif_device *vif;
1665         struct mfc_cache *c;
1666         struct net *net = sock_net(sk);
1667         struct mr_table *mrt;
1668
1669         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1670         if (!mrt)
1671                 return -ENOENT;
1672
1673         switch (cmd) {
1674         case SIOCGETVIFCNT:
1675                 if (copy_from_user(&vr, arg, sizeof(vr)))
1676                         return -EFAULT;
1677                 if (vr.vifi >= mrt->maxvif)
1678                         return -EINVAL;
1679                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1680                 read_lock(&mrt_lock);
1681                 vif = &mrt->vif_table[vr.vifi];
1682                 if (VIF_EXISTS(mrt, vr.vifi)) {
1683                         vr.icount = vif->pkt_in;
1684                         vr.ocount = vif->pkt_out;
1685                         vr.ibytes = vif->bytes_in;
1686                         vr.obytes = vif->bytes_out;
1687                         read_unlock(&mrt_lock);
1688
1689                         if (copy_to_user(arg, &vr, sizeof(vr)))
1690                                 return -EFAULT;
1691                         return 0;
1692                 }
1693                 read_unlock(&mrt_lock);
1694                 return -EADDRNOTAVAIL;
1695         case SIOCGETSGCNT:
1696                 if (copy_from_user(&sr, arg, sizeof(sr)))
1697                         return -EFAULT;
1698
1699                 rcu_read_lock();
1700                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1701                 if (c) {
1702                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1703                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1704                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1705                         rcu_read_unlock();
1706
1707                         if (copy_to_user(arg, &sr, sizeof(sr)))
1708                                 return -EFAULT;
1709                         return 0;
1710                 }
1711                 rcu_read_unlock();
1712                 return -EADDRNOTAVAIL;
1713         default:
1714                 return -ENOIOCTLCMD;
1715         }
1716 }
1717 #endif
1718
1719 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1720 {
1721         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1722         struct net *net = dev_net(dev);
1723         struct mr_table *mrt;
1724         struct vif_device *v;
1725         int ct;
1726
1727         if (event != NETDEV_UNREGISTER)
1728                 return NOTIFY_DONE;
1729
1730         ipmr_for_each_table(mrt, net) {
1731                 v = &mrt->vif_table[0];
1732                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1733                         if (v->dev == dev)
1734                                 vif_delete(mrt, ct, 1, NULL);
1735                 }
1736         }
1737         return NOTIFY_DONE;
1738 }
1739
1740 static struct notifier_block ip_mr_notifier = {
1741         .notifier_call = ipmr_device_event,
1742 };
1743
1744 /* Encapsulate a packet by attaching a valid IPIP header to it.
1745  * This avoids tunnel drivers and other mess and gives us the speed so
1746  * important for multicast video.
1747  */
1748 static void ip_encap(struct net *net, struct sk_buff *skb,
1749                      __be32 saddr, __be32 daddr)
1750 {
1751         struct iphdr *iph;
1752         const struct iphdr *old_iph = ip_hdr(skb);
1753
1754         skb_push(skb, sizeof(struct iphdr));
1755         skb->transport_header = skb->network_header;
1756         skb_reset_network_header(skb);
1757         iph = ip_hdr(skb);
1758
1759         iph->version    =       4;
1760         iph->tos        =       old_iph->tos;
1761         iph->ttl        =       old_iph->ttl;
1762         iph->frag_off   =       0;
1763         iph->daddr      =       daddr;
1764         iph->saddr      =       saddr;
1765         iph->protocol   =       IPPROTO_IPIP;
1766         iph->ihl        =       5;
1767         iph->tot_len    =       htons(skb->len);
1768         ip_select_ident(net, skb, NULL);
1769         ip_send_check(iph);
1770
1771         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1772         nf_reset_ct(skb);
1773 }
1774
1775 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1776                                       struct sk_buff *skb)
1777 {
1778         struct ip_options *opt = &(IPCB(skb)->opt);
1779
1780         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1781         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1782
1783         if (unlikely(opt->optlen))
1784                 ip_forward_options(skb);
1785
1786         return dst_output(net, sk, skb);
1787 }
1788
1789 #ifdef CONFIG_NET_SWITCHDEV
1790 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1791                                    int in_vifi, int out_vifi)
1792 {
1793         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1794         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1795
1796         if (!skb->offload_l3_fwd_mark)
1797                 return false;
1798         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1799                 return false;
1800         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1801                                         &in_vif->dev_parent_id);
1802 }
1803 #else
1804 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1805                                    int in_vifi, int out_vifi)
1806 {
1807         return false;
1808 }
1809 #endif
1810
1811 /* Processing handlers for ipmr_forward */
1812
1813 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1814                             int in_vifi, struct sk_buff *skb, int vifi)
1815 {
1816         const struct iphdr *iph = ip_hdr(skb);
1817         struct vif_device *vif = &mrt->vif_table[vifi];
1818         struct net_device *dev;
1819         struct rtable *rt;
1820         struct flowi4 fl4;
1821         int    encap = 0;
1822
1823         if (!vif->dev)
1824                 goto out_free;
1825
1826         if (vif->flags & VIFF_REGISTER) {
1827                 vif->pkt_out++;
1828                 vif->bytes_out += skb->len;
1829                 vif->dev->stats.tx_bytes += skb->len;
1830                 vif->dev->stats.tx_packets++;
1831                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1832                 goto out_free;
1833         }
1834
1835         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1836                 goto out_free;
1837
1838         if (vif->flags & VIFF_TUNNEL) {
1839                 rt = ip_route_output_ports(net, &fl4, NULL,
1840                                            vif->remote, vif->local,
1841                                            0, 0,
1842                                            IPPROTO_IPIP,
1843                                            RT_TOS(iph->tos), vif->link);
1844                 if (IS_ERR(rt))
1845                         goto out_free;
1846                 encap = sizeof(struct iphdr);
1847         } else {
1848                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1849                                            0, 0,
1850                                            IPPROTO_IPIP,
1851                                            RT_TOS(iph->tos), vif->link);
1852                 if (IS_ERR(rt))
1853                         goto out_free;
1854         }
1855
1856         dev = rt->dst.dev;
1857
1858         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1859                 /* Do not fragment multicasts. Alas, IPv4 does not
1860                  * allow to send ICMP, so that packets will disappear
1861                  * to blackhole.
1862                  */
1863                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1864                 ip_rt_put(rt);
1865                 goto out_free;
1866         }
1867
1868         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1869
1870         if (skb_cow(skb, encap)) {
1871                 ip_rt_put(rt);
1872                 goto out_free;
1873         }
1874
1875         vif->pkt_out++;
1876         vif->bytes_out += skb->len;
1877
1878         skb_dst_drop(skb);
1879         skb_dst_set(skb, &rt->dst);
1880         ip_decrease_ttl(ip_hdr(skb));
1881
1882         /* FIXME: forward and output firewalls used to be called here.
1883          * What do we do with netfilter? -- RR
1884          */
1885         if (vif->flags & VIFF_TUNNEL) {
1886                 ip_encap(net, skb, vif->local, vif->remote);
1887                 /* FIXME: extra output firewall step used to be here. --RR */
1888                 vif->dev->stats.tx_packets++;
1889                 vif->dev->stats.tx_bytes += skb->len;
1890         }
1891
1892         IPCB(skb)->flags |= IPSKB_FORWARDED;
1893
1894         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1895          * not only before forwarding, but after forwarding on all output
1896          * interfaces. It is clear, if mrouter runs a multicasting
1897          * program, it should receive packets not depending to what interface
1898          * program is joined.
1899          * If we will not make it, the program will have to join on all
1900          * interfaces. On the other hand, multihoming host (or router, but
1901          * not mrouter) cannot join to more than one interface - it will
1902          * result in receiving multiple packets.
1903          */
1904         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1905                 net, NULL, skb, skb->dev, dev,
1906                 ipmr_forward_finish);
1907         return;
1908
1909 out_free:
1910         kfree_skb(skb);
1911 }
1912
1913 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1914 {
1915         int ct;
1916
1917         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1918                 if (mrt->vif_table[ct].dev == dev)
1919                         break;
1920         }
1921         return ct;
1922 }
1923
1924 /* "local" means that we should preserve one skb (for local delivery) */
1925 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1926                           struct net_device *dev, struct sk_buff *skb,
1927                           struct mfc_cache *c, int local)
1928 {
1929         int true_vifi = ipmr_find_vif(mrt, dev);
1930         int psend = -1;
1931         int vif, ct;
1932
1933         vif = c->_c.mfc_parent;
1934         c->_c.mfc_un.res.pkt++;
1935         c->_c.mfc_un.res.bytes += skb->len;
1936         c->_c.mfc_un.res.lastuse = jiffies;
1937
1938         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1939                 struct mfc_cache *cache_proxy;
1940
1941                 /* For an (*,G) entry, we only check that the incomming
1942                  * interface is part of the static tree.
1943                  */
1944                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1945                 if (cache_proxy &&
1946                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1947                         goto forward;
1948         }
1949
1950         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1951         if (mrt->vif_table[vif].dev != dev) {
1952                 if (rt_is_output_route(skb_rtable(skb))) {
1953                         /* It is our own packet, looped back.
1954                          * Very complicated situation...
1955                          *
1956                          * The best workaround until routing daemons will be
1957                          * fixed is not to redistribute packet, if it was
1958                          * send through wrong interface. It means, that
1959                          * multicast applications WILL NOT work for
1960                          * (S,G), which have default multicast route pointing
1961                          * to wrong oif. In any case, it is not a good
1962                          * idea to use multicasting applications on router.
1963                          */
1964                         goto dont_forward;
1965                 }
1966
1967                 c->_c.mfc_un.res.wrong_if++;
1968
1969                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1970                     /* pimsm uses asserts, when switching from RPT to SPT,
1971                      * so that we cannot check that packet arrived on an oif.
1972                      * It is bad, but otherwise we would need to move pretty
1973                      * large chunk of pimd to kernel. Ough... --ANK
1974                      */
1975                     (mrt->mroute_do_pim ||
1976                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1977                     time_after(jiffies,
1978                                c->_c.mfc_un.res.last_assert +
1979                                MFC_ASSERT_THRESH)) {
1980                         c->_c.mfc_un.res.last_assert = jiffies;
1981                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1982                         if (mrt->mroute_do_wrvifwhole)
1983                                 ipmr_cache_report(mrt, skb, true_vifi,
1984                                                   IGMPMSG_WRVIFWHOLE);
1985                 }
1986                 goto dont_forward;
1987         }
1988
1989 forward:
1990         mrt->vif_table[vif].pkt_in++;
1991         mrt->vif_table[vif].bytes_in += skb->len;
1992
1993         /* Forward the frame */
1994         if (c->mfc_origin == htonl(INADDR_ANY) &&
1995             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
1996                 if (true_vifi >= 0 &&
1997                     true_vifi != c->_c.mfc_parent &&
1998                     ip_hdr(skb)->ttl >
1999                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2000                         /* It's an (*,*) entry and the packet is not coming from
2001                          * the upstream: forward the packet to the upstream
2002                          * only.
2003                          */
2004                         psend = c->_c.mfc_parent;
2005                         goto last_forward;
2006                 }
2007                 goto dont_forward;
2008         }
2009         for (ct = c->_c.mfc_un.res.maxvif - 1;
2010              ct >= c->_c.mfc_un.res.minvif; ct--) {
2011                 /* For (*,G) entry, don't forward to the incoming interface */
2012                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2013                      ct != true_vifi) &&
2014                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2015                         if (psend != -1) {
2016                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2017
2018                                 if (skb2)
2019                                         ipmr_queue_xmit(net, mrt, true_vifi,
2020                                                         skb2, psend);
2021                         }
2022                         psend = ct;
2023                 }
2024         }
2025 last_forward:
2026         if (psend != -1) {
2027                 if (local) {
2028                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2029
2030                         if (skb2)
2031                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2032                                                 psend);
2033                 } else {
2034                         ipmr_queue_xmit(net, mrt, true_vifi, skb, psend);
2035                         return;
2036                 }
2037         }
2038
2039 dont_forward:
2040         if (!local)
2041                 kfree_skb(skb);
2042 }
2043
2044 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2045 {
2046         struct rtable *rt = skb_rtable(skb);
2047         struct iphdr *iph = ip_hdr(skb);
2048         struct flowi4 fl4 = {
2049                 .daddr = iph->daddr,
2050                 .saddr = iph->saddr,
2051                 .flowi4_tos = RT_TOS(iph->tos),
2052                 .flowi4_oif = (rt_is_output_route(rt) ?
2053                                skb->dev->ifindex : 0),
2054                 .flowi4_iif = (rt_is_output_route(rt) ?
2055                                LOOPBACK_IFINDEX :
2056                                skb->dev->ifindex),
2057                 .flowi4_mark = skb->mark,
2058         };
2059         struct mr_table *mrt;
2060         int err;
2061
2062         err = ipmr_fib_lookup(net, &fl4, &mrt);
2063         if (err)
2064                 return ERR_PTR(err);
2065         return mrt;
2066 }
2067
2068 /* Multicast packets for forwarding arrive here
2069  * Called with rcu_read_lock();
2070  */
2071 int ip_mr_input(struct sk_buff *skb)
2072 {
2073         struct mfc_cache *cache;
2074         struct net *net = dev_net(skb->dev);
2075         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2076         struct mr_table *mrt;
2077         struct net_device *dev;
2078
2079         /* skb->dev passed in is the loX master dev for vrfs.
2080          * As there are no vifs associated with loopback devices,
2081          * get the proper interface that does have a vif associated with it.
2082          */
2083         dev = skb->dev;
2084         if (netif_is_l3_master(skb->dev)) {
2085                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2086                 if (!dev) {
2087                         kfree_skb(skb);
2088                         return -ENODEV;
2089                 }
2090         }
2091
2092         /* Packet is looped back after forward, it should not be
2093          * forwarded second time, but still can be delivered locally.
2094          */
2095         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2096                 goto dont_forward;
2097
2098         mrt = ipmr_rt_fib_lookup(net, skb);
2099         if (IS_ERR(mrt)) {
2100                 kfree_skb(skb);
2101                 return PTR_ERR(mrt);
2102         }
2103         if (!local) {
2104                 if (IPCB(skb)->opt.router_alert) {
2105                         if (ip_call_ra_chain(skb))
2106                                 return 0;
2107                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2108                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2109                          * Cisco IOS <= 11.2(8)) do not put router alert
2110                          * option to IGMP packets destined to routable
2111                          * groups. It is very bad, because it means
2112                          * that we can forward NO IGMP messages.
2113                          */
2114                         struct sock *mroute_sk;
2115
2116                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2117                         if (mroute_sk) {
2118                                 nf_reset_ct(skb);
2119                                 raw_rcv(mroute_sk, skb);
2120                                 return 0;
2121                         }
2122                     }
2123         }
2124
2125         /* already under rcu_read_lock() */
2126         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2127         if (!cache) {
2128                 int vif = ipmr_find_vif(mrt, dev);
2129
2130                 if (vif >= 0)
2131                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2132                                                     vif);
2133         }
2134
2135         /* No usable cache entry */
2136         if (!cache) {
2137                 int vif;
2138
2139                 if (local) {
2140                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2141                         ip_local_deliver(skb);
2142                         if (!skb2)
2143                                 return -ENOBUFS;
2144                         skb = skb2;
2145                 }
2146
2147                 read_lock(&mrt_lock);
2148                 vif = ipmr_find_vif(mrt, dev);
2149                 if (vif >= 0) {
2150                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2151                         read_unlock(&mrt_lock);
2152
2153                         return err2;
2154                 }
2155                 read_unlock(&mrt_lock);
2156                 kfree_skb(skb);
2157                 return -ENODEV;
2158         }
2159
2160         read_lock(&mrt_lock);
2161         ip_mr_forward(net, mrt, dev, skb, cache, local);
2162         read_unlock(&mrt_lock);
2163
2164         if (local)
2165                 return ip_local_deliver(skb);
2166
2167         return 0;
2168
2169 dont_forward:
2170         if (local)
2171                 return ip_local_deliver(skb);
2172         kfree_skb(skb);
2173         return 0;
2174 }
2175
2176 #ifdef CONFIG_IP_PIMSM_V1
2177 /* Handle IGMP messages of PIMv1 */
2178 int pim_rcv_v1(struct sk_buff *skb)
2179 {
2180         struct igmphdr *pim;
2181         struct net *net = dev_net(skb->dev);
2182         struct mr_table *mrt;
2183
2184         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2185                 goto drop;
2186
2187         pim = igmp_hdr(skb);
2188
2189         mrt = ipmr_rt_fib_lookup(net, skb);
2190         if (IS_ERR(mrt))
2191                 goto drop;
2192         if (!mrt->mroute_do_pim ||
2193             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2194                 goto drop;
2195
2196         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2197 drop:
2198                 kfree_skb(skb);
2199         }
2200         return 0;
2201 }
2202 #endif
2203
2204 #ifdef CONFIG_IP_PIMSM_V2
2205 static int pim_rcv(struct sk_buff *skb)
2206 {
2207         struct pimreghdr *pim;
2208         struct net *net = dev_net(skb->dev);
2209         struct mr_table *mrt;
2210
2211         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2212                 goto drop;
2213
2214         pim = (struct pimreghdr *)skb_transport_header(skb);
2215         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2216             (pim->flags & PIM_NULL_REGISTER) ||
2217             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2218              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2219                 goto drop;
2220
2221         mrt = ipmr_rt_fib_lookup(net, skb);
2222         if (IS_ERR(mrt))
2223                 goto drop;
2224         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2225 drop:
2226                 kfree_skb(skb);
2227         }
2228         return 0;
2229 }
2230 #endif
2231
2232 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2233                    __be32 saddr, __be32 daddr,
2234                    struct rtmsg *rtm, u32 portid)
2235 {
2236         struct mfc_cache *cache;
2237         struct mr_table *mrt;
2238         int err;
2239
2240         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2241         if (!mrt)
2242                 return -ENOENT;
2243
2244         rcu_read_lock();
2245         cache = ipmr_cache_find(mrt, saddr, daddr);
2246         if (!cache && skb->dev) {
2247                 int vif = ipmr_find_vif(mrt, skb->dev);
2248
2249                 if (vif >= 0)
2250                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2251         }
2252         if (!cache) {
2253                 struct sk_buff *skb2;
2254                 struct iphdr *iph;
2255                 struct net_device *dev;
2256                 int vif = -1;
2257
2258                 dev = skb->dev;
2259                 read_lock(&mrt_lock);
2260                 if (dev)
2261                         vif = ipmr_find_vif(mrt, dev);
2262                 if (vif < 0) {
2263                         read_unlock(&mrt_lock);
2264                         rcu_read_unlock();
2265                         return -ENODEV;
2266                 }
2267
2268                 skb2 = skb_realloc_headroom(skb, sizeof(struct iphdr));
2269                 if (!skb2) {
2270                         read_unlock(&mrt_lock);
2271                         rcu_read_unlock();
2272                         return -ENOMEM;
2273                 }
2274
2275                 NETLINK_CB(skb2).portid = portid;
2276                 skb_push(skb2, sizeof(struct iphdr));
2277                 skb_reset_network_header(skb2);
2278                 iph = ip_hdr(skb2);
2279                 iph->ihl = sizeof(struct iphdr) >> 2;
2280                 iph->saddr = saddr;
2281                 iph->daddr = daddr;
2282                 iph->version = 0;
2283                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2284                 read_unlock(&mrt_lock);
2285                 rcu_read_unlock();
2286                 return err;
2287         }
2288
2289         read_lock(&mrt_lock);
2290         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2291         read_unlock(&mrt_lock);
2292         rcu_read_unlock();
2293         return err;
2294 }
2295
2296 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2297                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2298                             int flags)
2299 {
2300         struct nlmsghdr *nlh;
2301         struct rtmsg *rtm;
2302         int err;
2303
2304         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2305         if (!nlh)
2306                 return -EMSGSIZE;
2307
2308         rtm = nlmsg_data(nlh);
2309         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2310         rtm->rtm_dst_len  = 32;
2311         rtm->rtm_src_len  = 32;
2312         rtm->rtm_tos      = 0;
2313         rtm->rtm_table    = mrt->id;
2314         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2315                 goto nla_put_failure;
2316         rtm->rtm_type     = RTN_MULTICAST;
2317         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2318         if (c->_c.mfc_flags & MFC_STATIC)
2319                 rtm->rtm_protocol = RTPROT_STATIC;
2320         else
2321                 rtm->rtm_protocol = RTPROT_MROUTED;
2322         rtm->rtm_flags    = 0;
2323
2324         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2325             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2326                 goto nla_put_failure;
2327         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2328         /* do not break the dump if cache is unresolved */
2329         if (err < 0 && err != -ENOENT)
2330                 goto nla_put_failure;
2331
2332         nlmsg_end(skb, nlh);
2333         return 0;
2334
2335 nla_put_failure:
2336         nlmsg_cancel(skb, nlh);
2337         return -EMSGSIZE;
2338 }
2339
2340 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2341                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2342                              int flags)
2343 {
2344         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2345                                 cmd, flags);
2346 }
2347
2348 static size_t mroute_msgsize(bool unresolved, int maxvif)
2349 {
2350         size_t len =
2351                 NLMSG_ALIGN(sizeof(struct rtmsg))
2352                 + nla_total_size(4)     /* RTA_TABLE */
2353                 + nla_total_size(4)     /* RTA_SRC */
2354                 + nla_total_size(4)     /* RTA_DST */
2355                 ;
2356
2357         if (!unresolved)
2358                 len = len
2359                       + nla_total_size(4)       /* RTA_IIF */
2360                       + nla_total_size(0)       /* RTA_MULTIPATH */
2361                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2362                                                 /* RTA_MFC_STATS */
2363                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2364                 ;
2365
2366         return len;
2367 }
2368
2369 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2370                                  int cmd)
2371 {
2372         struct net *net = read_pnet(&mrt->net);
2373         struct sk_buff *skb;
2374         int err = -ENOBUFS;
2375
2376         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2377                                        mrt->maxvif),
2378                         GFP_ATOMIC);
2379         if (!skb)
2380                 goto errout;
2381
2382         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2383         if (err < 0)
2384                 goto errout;
2385
2386         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2387         return;
2388
2389 errout:
2390         kfree_skb(skb);
2391         if (err < 0)
2392                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2393 }
2394
2395 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2396 {
2397         size_t len =
2398                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2399                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2400                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2401                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2402                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2403                 + nla_total_size(4)     /* IPMRA_CREPORT_TABLE */
2404                                         /* IPMRA_CREPORT_PKT */
2405                 + nla_total_size(payloadlen)
2406                 ;
2407
2408         return len;
2409 }
2410
2411 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2412 {
2413         struct net *net = read_pnet(&mrt->net);
2414         struct nlmsghdr *nlh;
2415         struct rtgenmsg *rtgenm;
2416         struct igmpmsg *msg;
2417         struct sk_buff *skb;
2418         struct nlattr *nla;
2419         int payloadlen;
2420
2421         payloadlen = pkt->len - sizeof(struct igmpmsg);
2422         msg = (struct igmpmsg *)skb_network_header(pkt);
2423
2424         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2425         if (!skb)
2426                 goto errout;
2427
2428         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2429                         sizeof(struct rtgenmsg), 0);
2430         if (!nlh)
2431                 goto errout;
2432         rtgenm = nlmsg_data(nlh);
2433         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2434         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2435             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif | (msg->im_vif_hi << 8)) ||
2436             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2437                             msg->im_src.s_addr) ||
2438             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2439                             msg->im_dst.s_addr) ||
2440             nla_put_u32(skb, IPMRA_CREPORT_TABLE, mrt->id))
2441                 goto nla_put_failure;
2442
2443         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2444         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2445                                   nla_data(nla), payloadlen))
2446                 goto nla_put_failure;
2447
2448         nlmsg_end(skb, nlh);
2449
2450         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2451         return;
2452
2453 nla_put_failure:
2454         nlmsg_cancel(skb, nlh);
2455 errout:
2456         kfree_skb(skb);
2457         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2458 }
2459
2460 static int ipmr_rtm_valid_getroute_req(struct sk_buff *skb,
2461                                        const struct nlmsghdr *nlh,
2462                                        struct nlattr **tb,
2463                                        struct netlink_ext_ack *extack)
2464 {
2465         struct rtmsg *rtm;
2466         int i, err;
2467
2468         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2469                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for multicast route get request");
2470                 return -EINVAL;
2471         }
2472
2473         if (!netlink_strict_get_check(skb))
2474                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2475                                               rtm_ipv4_policy, extack);
2476
2477         rtm = nlmsg_data(nlh);
2478         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2479             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2480             rtm->rtm_tos || rtm->rtm_table || rtm->rtm_protocol ||
2481             rtm->rtm_scope || rtm->rtm_type || rtm->rtm_flags) {
2482                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for multicast route get request");
2483                 return -EINVAL;
2484         }
2485
2486         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2487                                             rtm_ipv4_policy, extack);
2488         if (err)
2489                 return err;
2490
2491         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2492             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2493                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2494                 return -EINVAL;
2495         }
2496
2497         for (i = 0; i <= RTA_MAX; i++) {
2498                 if (!tb[i])
2499                         continue;
2500
2501                 switch (i) {
2502                 case RTA_SRC:
2503                 case RTA_DST:
2504                 case RTA_TABLE:
2505                         break;
2506                 default:
2507                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in multicast route get request");
2508                         return -EINVAL;
2509                 }
2510         }
2511
2512         return 0;
2513 }
2514
2515 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2516                              struct netlink_ext_ack *extack)
2517 {
2518         struct net *net = sock_net(in_skb->sk);
2519         struct nlattr *tb[RTA_MAX + 1];
2520         struct sk_buff *skb = NULL;
2521         struct mfc_cache *cache;
2522         struct mr_table *mrt;
2523         __be32 src, grp;
2524         u32 tableid;
2525         int err;
2526
2527         err = ipmr_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2528         if (err < 0)
2529                 goto errout;
2530
2531         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2532         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2533         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2534
2535         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2536         if (!mrt) {
2537                 err = -ENOENT;
2538                 goto errout_free;
2539         }
2540
2541         /* entries are added/deleted only under RTNL */
2542         rcu_read_lock();
2543         cache = ipmr_cache_find(mrt, src, grp);
2544         rcu_read_unlock();
2545         if (!cache) {
2546                 err = -ENOENT;
2547                 goto errout_free;
2548         }
2549
2550         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2551         if (!skb) {
2552                 err = -ENOBUFS;
2553                 goto errout_free;
2554         }
2555
2556         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2557                                nlh->nlmsg_seq, cache,
2558                                RTM_NEWROUTE, 0);
2559         if (err < 0)
2560                 goto errout_free;
2561
2562         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2563
2564 errout:
2565         return err;
2566
2567 errout_free:
2568         kfree_skb(skb);
2569         goto errout;
2570 }
2571
2572 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2573 {
2574         struct fib_dump_filter filter = {};
2575         int err;
2576
2577         if (cb->strict_check) {
2578                 err = ip_valid_fib_dump_req(sock_net(skb->sk), cb->nlh,
2579                                             &filter, cb);
2580                 if (err < 0)
2581                         return err;
2582         }
2583
2584         if (filter.table_id) {
2585                 struct mr_table *mrt;
2586
2587                 mrt = ipmr_get_table(sock_net(skb->sk), filter.table_id);
2588                 if (!mrt) {
2589                         if (rtnl_msg_family(cb->nlh) != RTNL_FAMILY_IPMR)
2590                                 return skb->len;
2591
2592                         NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
2593                         return -ENOENT;
2594                 }
2595                 err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
2596                                     &mfc_unres_lock, &filter);
2597                 return skb->len ? : err;
2598         }
2599
2600         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2601                                 _ipmr_fill_mroute, &mfc_unres_lock, &filter);
2602 }
2603
2604 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2605         [RTA_SRC]       = { .type = NLA_U32 },
2606         [RTA_DST]       = { .type = NLA_U32 },
2607         [RTA_IIF]       = { .type = NLA_U32 },
2608         [RTA_TABLE]     = { .type = NLA_U32 },
2609         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2610 };
2611
2612 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2613 {
2614         switch (rtm_protocol) {
2615         case RTPROT_STATIC:
2616         case RTPROT_MROUTED:
2617                 return true;
2618         }
2619         return false;
2620 }
2621
2622 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2623 {
2624         struct rtnexthop *rtnh = nla_data(nla);
2625         int remaining = nla_len(nla), vifi = 0;
2626
2627         while (rtnh_ok(rtnh, remaining)) {
2628                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2629                 if (++vifi == MAXVIFS)
2630                         break;
2631                 rtnh = rtnh_next(rtnh, &remaining);
2632         }
2633
2634         return remaining > 0 ? -EINVAL : vifi;
2635 }
2636
2637 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2638 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2639                             struct mfcctl *mfcc, int *mrtsock,
2640                             struct mr_table **mrtret,
2641                             struct netlink_ext_ack *extack)
2642 {
2643         struct net_device *dev = NULL;
2644         u32 tblid = RT_TABLE_DEFAULT;
2645         struct mr_table *mrt;
2646         struct nlattr *attr;
2647         struct rtmsg *rtm;
2648         int ret, rem;
2649
2650         ret = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
2651                                         rtm_ipmr_policy, extack);
2652         if (ret < 0)
2653                 goto out;
2654         rtm = nlmsg_data(nlh);
2655
2656         ret = -EINVAL;
2657         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2658             rtm->rtm_type != RTN_MULTICAST ||
2659             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2660             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2661                 goto out;
2662
2663         memset(mfcc, 0, sizeof(*mfcc));
2664         mfcc->mfcc_parent = -1;
2665         ret = 0;
2666         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2667                 switch (nla_type(attr)) {
2668                 case RTA_SRC:
2669                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2670                         break;
2671                 case RTA_DST:
2672                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2673                         break;
2674                 case RTA_IIF:
2675                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2676                         if (!dev) {
2677                                 ret = -ENODEV;
2678                                 goto out;
2679                         }
2680                         break;
2681                 case RTA_MULTIPATH:
2682                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2683                                 ret = -EINVAL;
2684                                 goto out;
2685                         }
2686                         break;
2687                 case RTA_PREFSRC:
2688                         ret = 1;
2689                         break;
2690                 case RTA_TABLE:
2691                         tblid = nla_get_u32(attr);
2692                         break;
2693                 }
2694         }
2695         mrt = ipmr_get_table(net, tblid);
2696         if (!mrt) {
2697                 ret = -ENOENT;
2698                 goto out;
2699         }
2700         *mrtret = mrt;
2701         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2702         if (dev)
2703                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2704
2705 out:
2706         return ret;
2707 }
2708
2709 /* takes care of both newroute and delroute */
2710 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2711                           struct netlink_ext_ack *extack)
2712 {
2713         struct net *net = sock_net(skb->sk);
2714         int ret, mrtsock, parent;
2715         struct mr_table *tbl;
2716         struct mfcctl mfcc;
2717
2718         mrtsock = 0;
2719         tbl = NULL;
2720         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2721         if (ret < 0)
2722                 return ret;
2723
2724         parent = ret ? mfcc.mfcc_parent : -1;
2725         if (nlh->nlmsg_type == RTM_NEWROUTE)
2726                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2727         else
2728                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2729 }
2730
2731 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2732 {
2733         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2734
2735         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2736             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2737             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2738                         mrt->mroute_reg_vif_num) ||
2739             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2740                        mrt->mroute_do_assert) ||
2741             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2742             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2743                        mrt->mroute_do_wrvifwhole))
2744                 return false;
2745
2746         return true;
2747 }
2748
2749 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2750 {
2751         struct nlattr *vif_nest;
2752         struct vif_device *vif;
2753
2754         /* if the VIF doesn't exist just continue */
2755         if (!VIF_EXISTS(mrt, vifid))
2756                 return true;
2757
2758         vif = &mrt->vif_table[vifid];
2759         vif_nest = nla_nest_start_noflag(skb, IPMRA_VIF);
2760         if (!vif_nest)
2761                 return false;
2762         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2763             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2764             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2765             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2766                               IPMRA_VIFA_PAD) ||
2767             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2768                               IPMRA_VIFA_PAD) ||
2769             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2770                               IPMRA_VIFA_PAD) ||
2771             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2772                               IPMRA_VIFA_PAD) ||
2773             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2774             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2775                 nla_nest_cancel(skb, vif_nest);
2776                 return false;
2777         }
2778         nla_nest_end(skb, vif_nest);
2779
2780         return true;
2781 }
2782
2783 static int ipmr_valid_dumplink(const struct nlmsghdr *nlh,
2784                                struct netlink_ext_ack *extack)
2785 {
2786         struct ifinfomsg *ifm;
2787
2788         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ifm))) {
2789                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for ipmr link dump");
2790                 return -EINVAL;
2791         }
2792
2793         if (nlmsg_attrlen(nlh, sizeof(*ifm))) {
2794                 NL_SET_ERR_MSG(extack, "Invalid data after header in ipmr link dump");
2795                 return -EINVAL;
2796         }
2797
2798         ifm = nlmsg_data(nlh);
2799         if (ifm->__ifi_pad || ifm->ifi_type || ifm->ifi_flags ||
2800             ifm->ifi_change || ifm->ifi_index) {
2801                 NL_SET_ERR_MSG(extack, "Invalid values in header for ipmr link dump request");
2802                 return -EINVAL;
2803         }
2804
2805         return 0;
2806 }
2807
2808 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2809 {
2810         struct net *net = sock_net(skb->sk);
2811         struct nlmsghdr *nlh = NULL;
2812         unsigned int t = 0, s_t;
2813         unsigned int e = 0, s_e;
2814         struct mr_table *mrt;
2815
2816         if (cb->strict_check) {
2817                 int err = ipmr_valid_dumplink(cb->nlh, cb->extack);
2818
2819                 if (err < 0)
2820                         return err;
2821         }
2822
2823         s_t = cb->args[0];
2824         s_e = cb->args[1];
2825
2826         ipmr_for_each_table(mrt, net) {
2827                 struct nlattr *vifs, *af;
2828                 struct ifinfomsg *hdr;
2829                 u32 i;
2830
2831                 if (t < s_t)
2832                         goto skip_table;
2833                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2834                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2835                                 sizeof(*hdr), NLM_F_MULTI);
2836                 if (!nlh)
2837                         break;
2838
2839                 hdr = nlmsg_data(nlh);
2840                 memset(hdr, 0, sizeof(*hdr));
2841                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2842
2843                 af = nla_nest_start_noflag(skb, IFLA_AF_SPEC);
2844                 if (!af) {
2845                         nlmsg_cancel(skb, nlh);
2846                         goto out;
2847                 }
2848
2849                 if (!ipmr_fill_table(mrt, skb)) {
2850                         nlmsg_cancel(skb, nlh);
2851                         goto out;
2852                 }
2853
2854                 vifs = nla_nest_start_noflag(skb, IPMRA_TABLE_VIFS);
2855                 if (!vifs) {
2856                         nla_nest_end(skb, af);
2857                         nlmsg_end(skb, nlh);
2858                         goto out;
2859                 }
2860                 for (i = 0; i < mrt->maxvif; i++) {
2861                         if (e < s_e)
2862                                 goto skip_entry;
2863                         if (!ipmr_fill_vif(mrt, i, skb)) {
2864                                 nla_nest_end(skb, vifs);
2865                                 nla_nest_end(skb, af);
2866                                 nlmsg_end(skb, nlh);
2867                                 goto out;
2868                         }
2869 skip_entry:
2870                         e++;
2871                 }
2872                 s_e = 0;
2873                 e = 0;
2874                 nla_nest_end(skb, vifs);
2875                 nla_nest_end(skb, af);
2876                 nlmsg_end(skb, nlh);
2877 skip_table:
2878                 t++;
2879         }
2880
2881 out:
2882         cb->args[1] = e;
2883         cb->args[0] = t;
2884
2885         return skb->len;
2886 }
2887
2888 #ifdef CONFIG_PROC_FS
2889 /* The /proc interfaces to multicast routing :
2890  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2891  */
2892
2893 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2894         __acquires(mrt_lock)
2895 {
2896         struct mr_vif_iter *iter = seq->private;
2897         struct net *net = seq_file_net(seq);
2898         struct mr_table *mrt;
2899
2900         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2901         if (!mrt)
2902                 return ERR_PTR(-ENOENT);
2903
2904         iter->mrt = mrt;
2905
2906         read_lock(&mrt_lock);
2907         return mr_vif_seq_start(seq, pos);
2908 }
2909
2910 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2911         __releases(mrt_lock)
2912 {
2913         read_unlock(&mrt_lock);
2914 }
2915
2916 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2917 {
2918         struct mr_vif_iter *iter = seq->private;
2919         struct mr_table *mrt = iter->mrt;
2920
2921         if (v == SEQ_START_TOKEN) {
2922                 seq_puts(seq,
2923                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2924         } else {
2925                 const struct vif_device *vif = v;
2926                 const char *name =  vif->dev ?
2927                                     vif->dev->name : "none";
2928
2929                 seq_printf(seq,
2930                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2931                            vif - mrt->vif_table,
2932                            name, vif->bytes_in, vif->pkt_in,
2933                            vif->bytes_out, vif->pkt_out,
2934                            vif->flags, vif->local, vif->remote);
2935         }
2936         return 0;
2937 }
2938
2939 static const struct seq_operations ipmr_vif_seq_ops = {
2940         .start = ipmr_vif_seq_start,
2941         .next  = mr_vif_seq_next,
2942         .stop  = ipmr_vif_seq_stop,
2943         .show  = ipmr_vif_seq_show,
2944 };
2945
2946 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2947 {
2948         struct net *net = seq_file_net(seq);
2949         struct mr_table *mrt;
2950
2951         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2952         if (!mrt)
2953                 return ERR_PTR(-ENOENT);
2954
2955         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2956 }
2957
2958 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2959 {
2960         int n;
2961
2962         if (v == SEQ_START_TOKEN) {
2963                 seq_puts(seq,
2964                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2965         } else {
2966                 const struct mfc_cache *mfc = v;
2967                 const struct mr_mfc_iter *it = seq->private;
2968                 const struct mr_table *mrt = it->mrt;
2969
2970                 seq_printf(seq, "%08X %08X %-3hd",
2971                            (__force u32) mfc->mfc_mcastgrp,
2972                            (__force u32) mfc->mfc_origin,
2973                            mfc->_c.mfc_parent);
2974
2975                 if (it->cache != &mrt->mfc_unres_queue) {
2976                         seq_printf(seq, " %8lu %8lu %8lu",
2977                                    mfc->_c.mfc_un.res.pkt,
2978                                    mfc->_c.mfc_un.res.bytes,
2979                                    mfc->_c.mfc_un.res.wrong_if);
2980                         for (n = mfc->_c.mfc_un.res.minvif;
2981                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2982                                 if (VIF_EXISTS(mrt, n) &&
2983                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2984                                         seq_printf(seq,
2985                                            " %2d:%-3d",
2986                                            n, mfc->_c.mfc_un.res.ttls[n]);
2987                         }
2988                 } else {
2989                         /* unresolved mfc_caches don't contain
2990                          * pkt, bytes and wrong_if values
2991                          */
2992                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2993                 }
2994                 seq_putc(seq, '\n');
2995         }
2996         return 0;
2997 }
2998
2999 static const struct seq_operations ipmr_mfc_seq_ops = {
3000         .start = ipmr_mfc_seq_start,
3001         .next  = mr_mfc_seq_next,
3002         .stop  = mr_mfc_seq_stop,
3003         .show  = ipmr_mfc_seq_show,
3004 };
3005 #endif
3006
3007 #ifdef CONFIG_IP_PIMSM_V2
3008 static const struct net_protocol pim_protocol = {
3009         .handler        =       pim_rcv,
3010         .netns_ok       =       1,
3011 };
3012 #endif
3013
3014 static unsigned int ipmr_seq_read(struct net *net)
3015 {
3016         ASSERT_RTNL();
3017
3018         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3019 }
3020
3021 static int ipmr_dump(struct net *net, struct notifier_block *nb,
3022                      struct netlink_ext_ack *extack)
3023 {
3024         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
3025                        ipmr_mr_table_iter, &mrt_lock, extack);
3026 }
3027
3028 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3029         .family         = RTNL_FAMILY_IPMR,
3030         .fib_seq_read   = ipmr_seq_read,
3031         .fib_dump       = ipmr_dump,
3032         .owner          = THIS_MODULE,
3033 };
3034
3035 static int __net_init ipmr_notifier_init(struct net *net)
3036 {
3037         struct fib_notifier_ops *ops;
3038
3039         net->ipv4.ipmr_seq = 0;
3040
3041         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3042         if (IS_ERR(ops))
3043                 return PTR_ERR(ops);
3044         net->ipv4.ipmr_notifier_ops = ops;
3045
3046         return 0;
3047 }
3048
3049 static void __net_exit ipmr_notifier_exit(struct net *net)
3050 {
3051         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3052         net->ipv4.ipmr_notifier_ops = NULL;
3053 }
3054
3055 /* Setup for IP multicast routing */
3056 static int __net_init ipmr_net_init(struct net *net)
3057 {
3058         int err;
3059
3060         err = ipmr_notifier_init(net);
3061         if (err)
3062                 goto ipmr_notifier_fail;
3063
3064         err = ipmr_rules_init(net);
3065         if (err < 0)
3066                 goto ipmr_rules_fail;
3067
3068 #ifdef CONFIG_PROC_FS
3069         err = -ENOMEM;
3070         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
3071                         sizeof(struct mr_vif_iter)))
3072                 goto proc_vif_fail;
3073         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
3074                         sizeof(struct mr_mfc_iter)))
3075                 goto proc_cache_fail;
3076 #endif
3077         return 0;
3078
3079 #ifdef CONFIG_PROC_FS
3080 proc_cache_fail:
3081         remove_proc_entry("ip_mr_vif", net->proc_net);
3082 proc_vif_fail:
3083         ipmr_rules_exit(net);
3084 #endif
3085 ipmr_rules_fail:
3086         ipmr_notifier_exit(net);
3087 ipmr_notifier_fail:
3088         return err;
3089 }
3090
3091 static void __net_exit ipmr_net_exit(struct net *net)
3092 {
3093 #ifdef CONFIG_PROC_FS
3094         remove_proc_entry("ip_mr_cache", net->proc_net);
3095         remove_proc_entry("ip_mr_vif", net->proc_net);
3096 #endif
3097         ipmr_notifier_exit(net);
3098         ipmr_rules_exit(net);
3099 }
3100
3101 static struct pernet_operations ipmr_net_ops = {
3102         .init = ipmr_net_init,
3103         .exit = ipmr_net_exit,
3104 };
3105
3106 int __init ip_mr_init(void)
3107 {
3108         int err;
3109
3110         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3111                                        sizeof(struct mfc_cache),
3112                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3113                                        NULL);
3114
3115         err = register_pernet_subsys(&ipmr_net_ops);
3116         if (err)
3117                 goto reg_pernet_fail;
3118
3119         err = register_netdevice_notifier(&ip_mr_notifier);
3120         if (err)
3121                 goto reg_notif_fail;
3122 #ifdef CONFIG_IP_PIMSM_V2
3123         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3124                 pr_err("%s: can't add PIM protocol\n", __func__);
3125                 err = -EAGAIN;
3126                 goto add_proto_fail;
3127         }
3128 #endif
3129         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3130                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3131         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3132                       ipmr_rtm_route, NULL, 0);
3133         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3134                       ipmr_rtm_route, NULL, 0);
3135
3136         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3137                       NULL, ipmr_rtm_dumplink, 0);
3138         return 0;
3139
3140 #ifdef CONFIG_IP_PIMSM_V2
3141 add_proto_fail:
3142         unregister_netdevice_notifier(&ip_mr_notifier);
3143 #endif
3144 reg_notif_fail:
3145         unregister_pernet_subsys(&ipmr_net_ops);
3146 reg_pernet_fail:
3147         kmem_cache_destroy(mrt_cachep);
3148         return err;
3149 }