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, <alan@lxorguk.ukuu.org.uk>
   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         rtnl_lock();
 265         ipmr_free_table(mrt);
 266         rtnl_unlock();
 267 err1:
 268         fib_rules_unregister(ops);
 269         return err;
 270 }
 271
 272 static void __net_exit ipmr_rules_exit(struct net *net)
 273 {
 274         struct mr_table *mrt, *next;
 275
 276         rtnl_lock();
 277         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 278                 list_del(&mrt->list);
 279                 ipmr_free_table(mrt);
 280         }
 281         fib_rules_unregister(net->ipv4.mr_rules_ops);
 282         rtnl_unlock();
 283 }
 284
 285 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 286                            struct netlink_ext_ack *extack)
 287 {
 288         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR, extack);
 289 }
 290
 291 static unsigned int ipmr_rules_seq_read(struct net *net)
 292 {
 293         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 294 }
 295
 296 bool ipmr_rule_default(const struct fib_rule *rule)
 297 {
 298         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 299 }
 300 EXPORT_SYMBOL(ipmr_rule_default);
 301 #else
 302 #define ipmr_for_each_table(mrt, net) \
 303         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 304
 305 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 306                                            struct mr_table *mrt)
 307 {
 308         if (!mrt)
 309                 return net->ipv4.mrt;
 310         return NULL;
 311 }
 312
 313 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 314 {
 315         return net->ipv4.mrt;
 316 }
 317
 318 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 319                            struct mr_table **mrt)
 320 {
 321         *mrt = net->ipv4.mrt;
 322         return 0;
 323 }
 324
 325 static int __net_init ipmr_rules_init(struct net *net)
 326 {
 327         struct mr_table *mrt;
 328
 329         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 330         if (IS_ERR(mrt))
 331                 return PTR_ERR(mrt);
 332         net->ipv4.mrt = mrt;
 333         return 0;
 334 }
 335
 336 static void __net_exit ipmr_rules_exit(struct net *net)
 337 {
 338         rtnl_lock();
 339         ipmr_free_table(net->ipv4.mrt);
 340         net->ipv4.mrt = NULL;
 341         rtnl_unlock();
 342 }
 343
 344 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 345                            struct netlink_ext_ack *extack)
 346 {
 347         return 0;
 348 }
 349
 350 static unsigned int ipmr_rules_seq_read(struct net *net)
 351 {
 352         return 0;
 353 }
 354
 355 bool ipmr_rule_default(const struct fib_rule *rule)
 356 {
 357         return true;
 358 }
 359 EXPORT_SYMBOL(ipmr_rule_default);
 360 #endif
 361
 362 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 363                                 const void *ptr)
 364 {
 365         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 366         struct mfc_cache *c = (struct mfc_cache *)ptr;
 367
 368         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 369                cmparg->mfc_origin != c->mfc_origin;
 370 }
 371
 372 static const struct rhashtable_params ipmr_rht_params = {
 373         .head_offset = offsetof(struct mr_mfc, mnode),
 374         .key_offset = offsetof(struct mfc_cache, cmparg),
 375         .key_len = sizeof(struct mfc_cache_cmp_arg),
 376         .nelem_hint = 3,
 377         .obj_cmpfn = ipmr_hash_cmp,
 378         .automatic_shrinking = true,
 379 };
 380
 381 static void ipmr_new_table_set(struct mr_table *mrt,
 382                                struct net *net)
 383 {
 384 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 385         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 386 #endif
 387 }
 388
 389 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 390         .mfc_mcastgrp = htonl(INADDR_ANY),
 391         .mfc_origin = htonl(INADDR_ANY),
 392 };
 393
 394 static struct mr_table_ops ipmr_mr_table_ops = {
 395         .rht_params = &ipmr_rht_params,
 396         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 397 };
 398
 399 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 400 {
 401         struct mr_table *mrt;
 402
 403         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 404         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 405                 return ERR_PTR(-EINVAL);
 406
 407         mrt = ipmr_get_table(net, id);
 408         if (mrt)
 409                 return mrt;
 410
 411         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 412                               ipmr_expire_process, ipmr_new_table_set);
 413 }
 414
 415 static void ipmr_free_table(struct mr_table *mrt)
 416 {
 417         del_timer_sync(&mrt->ipmr_expire_timer);
 418         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC |
 419                                  MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC);
 420         rhltable_destroy(&mrt->mfc_hash);
 421         kfree(mrt);
 422 }
 423
 424 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 425
 426 /* Initialize ipmr pimreg/tunnel in_device */
 427 static bool ipmr_init_vif_indev(const struct net_device *dev)
 428 {
 429         struct in_device *in_dev;
 430
 431         ASSERT_RTNL();
 432
 433         in_dev = __in_dev_get_rtnl(dev);
 434         if (!in_dev)
 435                 return false;
 436         ipv4_devconf_setall(in_dev);
 437         neigh_parms_data_state_setall(in_dev->arp_parms);
 438         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 439
 440         return true;
 441 }
 442
 443 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 444 {
 445         struct net_device *tunnel_dev, *new_dev;
 446         struct ip_tunnel_parm p = { };
 447         int err;
 448
 449         tunnel_dev = __dev_get_by_name(net, "tunl0");
 450         if (!tunnel_dev)
 451                 goto out;
 452
 453         p.iph.daddr = v->vifc_rmt_addr.s_addr;
 454         p.iph.saddr = v->vifc_lcl_addr.s_addr;
 455         p.iph.version = 4;
 456         p.iph.ihl = 5;
 457         p.iph.protocol = IPPROTO_IPIP;
 458         sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 459
 460         if (!tunnel_dev->netdev_ops->ndo_tunnel_ctl)
 461                 goto out;
 462         err = tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 463                         SIOCADDTUNNEL);
 464         if (err)
 465                 goto out;
 466
 467         new_dev = __dev_get_by_name(net, p.name);
 468         if (!new_dev)
 469                 goto out;
 470
 471         new_dev->flags |= IFF_MULTICAST;
 472         if (!ipmr_init_vif_indev(new_dev))
 473                 goto out_unregister;
 474         if (dev_open(new_dev, NULL))
 475                 goto out_unregister;
 476         dev_hold(new_dev);
 477         err = dev_set_allmulti(new_dev, 1);
 478         if (err) {
 479                 dev_close(new_dev);
 480                 tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 481                                 SIOCDELTUNNEL);
 482                 dev_put(new_dev);
 483                 new_dev = ERR_PTR(err);
 484         }
 485         return new_dev;
 486
 487 out_unregister:
 488         unregister_netdevice(new_dev);
 489 out:
 490         return ERR_PTR(-ENOBUFS);
 491 }
 492
 493 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 494 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 495 {
 496         struct net *net = dev_net(dev);
 497         struct mr_table *mrt;
 498         struct flowi4 fl4 = {
 499                 .flowi4_oif     = dev->ifindex,
 500                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 501                 .flowi4_mark    = skb->mark,
 502         };
 503         int err;
 504
 505         err = ipmr_fib_lookup(net, &fl4, &mrt);
 506         if (err < 0) {
 507                 kfree_skb(skb);
 508                 return err;
 509         }
 510
 511         read_lock(&mrt_lock);
 512         dev->stats.tx_bytes += skb->len;
 513         dev->stats.tx_packets++;
 514         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 515         read_unlock(&mrt_lock);
 516         kfree_skb(skb);
 517         return NETDEV_TX_OK;
 518 }
 519
 520 static int reg_vif_get_iflink(const struct net_device *dev)
 521 {
 522         return 0;
 523 }
 524
 525 static const struct net_device_ops reg_vif_netdev_ops = {
 526         .ndo_start_xmit = reg_vif_xmit,
 527         .ndo_get_iflink = reg_vif_get_iflink,
 528 };
 529
 530 static void reg_vif_setup(struct net_device *dev)
 531 {
 532         dev->type               = ARPHRD_PIMREG;
 533         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 534         dev->flags              = IFF_NOARP;
 535         dev->netdev_ops         = &reg_vif_netdev_ops;
 536         dev->needs_free_netdev  = true;
 537         dev->features           |= NETIF_F_NETNS_LOCAL;
 538 }
 539
 540 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 541 {
 542         struct net_device *dev;
 543         char name[IFNAMSIZ];
 544
 545         if (mrt->id == RT_TABLE_DEFAULT)
 546                 sprintf(name, "pimreg");
 547         else
 548                 sprintf(name, "pimreg%u", mrt->id);
 549
 550         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 551
 552         if (!dev)
 553                 return NULL;
 554
 555         dev_net_set(dev, net);
 556
 557         if (register_netdevice(dev)) {
 558                 free_netdev(dev);
 559                 return NULL;
 560         }
 561
 562         if (!ipmr_init_vif_indev(dev))
 563                 goto failure;
 564         if (dev_open(dev, NULL))
 565                 goto failure;
 566
 567         dev_hold(dev);
 568
 569         return dev;
 570
 571 failure:
 572         unregister_netdevice(dev);
 573         return NULL;
 574 }
 575
 576 /* called with rcu_read_lock() */
 577 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 578                      unsigned int pimlen)
 579 {
 580         struct net_device *reg_dev = NULL;
 581         struct iphdr *encap;
 582
 583         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 584         /* Check that:
 585          * a. packet is really sent to a multicast group
 586          * b. packet is not a NULL-REGISTER
 587          * c. packet is not truncated
 588          */
 589         if (!ipv4_is_multicast(encap->daddr) ||
 590             encap->tot_len == 0 ||
 591             ntohs(encap->tot_len) + pimlen > skb->len)
 592                 return 1;
 593
 594         read_lock(&mrt_lock);
 595         if (mrt->mroute_reg_vif_num >= 0)
 596                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 597         read_unlock(&mrt_lock);
 598
 599         if (!reg_dev)
 600                 return 1;
 601
 602         skb->mac_header = skb->network_header;
 603         skb_pull(skb, (u8 *)encap - skb->data);
 604         skb_reset_network_header(skb);
 605         skb->protocol = htons(ETH_P_IP);
 606         skb->ip_summed = CHECKSUM_NONE;
 607
 608         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 609
 610         netif_rx(skb);
 611
 612         return NET_RX_SUCCESS;
 613 }
 614 #else
 615 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 616 {
 617         return NULL;
 618 }
 619 #endif
 620
 621 static int call_ipmr_vif_entry_notifiers(struct net *net,
 622                                          enum fib_event_type event_type,
 623                                          struct vif_device *vif,
 624                                          vifi_t vif_index, u32 tb_id)
 625 {
 626         return mr_call_vif_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 627                                      vif, vif_index, tb_id,
 628                                      &net->ipv4.ipmr_seq);
 629 }
 630
 631 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 632                                          enum fib_event_type event_type,
 633                                          struct mfc_cache *mfc, u32 tb_id)
 634 {
 635         return mr_call_mfc_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 636                                      &mfc->_c, tb_id, &net->ipv4.ipmr_seq);
 637 }
 638
 639 /**
 640  *      vif_delete - Delete a VIF entry
 641  *      @mrt: Table to delete from
 642  *      @vifi: VIF identifier to delete
 643  *      @notify: Set to 1, if the caller is a notifier_call
 644  *      @head: if unregistering the VIF, place it on this queue
 645  */
 646 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 647                       struct list_head *head)
 648 {
 649         struct net *net = read_pnet(&mrt->net);
 650         struct vif_device *v;
 651         struct net_device *dev;
 652         struct in_device *in_dev;
 653
 654         if (vifi < 0 || vifi >= mrt->maxvif)
 655                 return -EADDRNOTAVAIL;
 656
 657         v = &mrt->vif_table[vifi];
 658
 659         if (VIF_EXISTS(mrt, vifi))
 660                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 661                                               mrt->id);
 662
 663         write_lock_bh(&mrt_lock);
 664         dev = v->dev;
 665         v->dev = NULL;
 666
 667         if (!dev) {
 668                 write_unlock_bh(&mrt_lock);
 669                 return -EADDRNOTAVAIL;
 670         }
 671
 672         if (vifi == mrt->mroute_reg_vif_num)
 673                 mrt->mroute_reg_vif_num = -1;
 674
 675         if (vifi + 1 == mrt->maxvif) {
 676                 int tmp;
 677
 678                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 679                         if (VIF_EXISTS(mrt, tmp))
 680                                 break;
 681                 }
 682                 mrt->maxvif = tmp+1;
 683         }
 684
 685         write_unlock_bh(&mrt_lock);
 686
 687         dev_set_allmulti(dev, -1);
 688
 689         in_dev = __in_dev_get_rtnl(dev);
 690         if (in_dev) {
 691                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 692                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 693                                             NETCONFA_MC_FORWARDING,
 694                                             dev->ifindex, &in_dev->cnf);
 695                 ip_rt_multicast_event(in_dev);
 696         }
 697
 698         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 699                 unregister_netdevice_queue(dev, head);
 700
 701         dev_put(dev);
 702         return 0;
 703 }
 704
 705 static void ipmr_cache_free_rcu(struct rcu_head *head)
 706 {
 707         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 708
 709         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 710 }
 711
 712 static void ipmr_cache_free(struct mfc_cache *c)
 713 {
 714         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 715 }
 716
 717 /* Destroy an unresolved cache entry, killing queued skbs
 718  * and reporting error to netlink readers.
 719  */
 720 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 721 {
 722         struct net *net = read_pnet(&mrt->net);
 723         struct sk_buff *skb;
 724         struct nlmsgerr *e;
 725
 726         atomic_dec(&mrt->cache_resolve_queue_len);
 727
 728         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 729                 if (ip_hdr(skb)->version == 0) {
 730                         struct nlmsghdr *nlh = skb_pull(skb,
 731                                                         sizeof(struct iphdr));
 732                         nlh->nlmsg_type = NLMSG_ERROR;
 733                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 734                         skb_trim(skb, nlh->nlmsg_len);
 735                         e = nlmsg_data(nlh);
 736                         e->error = -ETIMEDOUT;
 737                         memset(&e->msg, 0, sizeof(e->msg));
 738
 739                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 740                 } else {
 741                         kfree_skb(skb);
 742                 }
 743         }
 744
 745         ipmr_cache_free(c);
 746 }
 747
 748 /* Timer process for the unresolved queue. */
 749 static void ipmr_expire_process(struct timer_list *t)
 750 {
 751         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 752         struct mr_mfc *c, *next;
 753         unsigned long expires;
 754         unsigned long now;
 755
 756         if (!spin_trylock(&mfc_unres_lock)) {
 757                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 758                 return;
 759         }
 760
 761         if (list_empty(&mrt->mfc_unres_queue))
 762                 goto out;
 763
 764         now = jiffies;
 765         expires = 10*HZ;
 766
 767         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 768                 if (time_after(c->mfc_un.unres.expires, now)) {
 769                         unsigned long interval = c->mfc_un.unres.expires - now;
 770                         if (interval < expires)
 771                                 expires = interval;
 772                         continue;
 773                 }
 774
 775                 list_del(&c->list);
 776                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 777                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 778         }
 779
 780         if (!list_empty(&mrt->mfc_unres_queue))
 781                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 782
 783 out:
 784         spin_unlock(&mfc_unres_lock);
 785 }
 786
 787 /* Fill oifs list. It is called under write locked mrt_lock. */
 788 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 789                                    unsigned char *ttls)
 790 {
 791         int vifi;
 792
 793         cache->mfc_un.res.minvif = MAXVIFS;
 794         cache->mfc_un.res.maxvif = 0;
 795         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 796
 797         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 798                 if (VIF_EXISTS(mrt, vifi) &&
 799                     ttls[vifi] && ttls[vifi] < 255) {
 800                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 801                         if (cache->mfc_un.res.minvif > vifi)
 802                                 cache->mfc_un.res.minvif = vifi;
 803                         if (cache->mfc_un.res.maxvif <= vifi)
 804                                 cache->mfc_un.res.maxvif = vifi + 1;
 805                 }
 806         }
 807         cache->mfc_un.res.lastuse = jiffies;
 808 }
 809
 810 static int vif_add(struct net *net, struct mr_table *mrt,
 811                    struct vifctl *vifc, int mrtsock)
 812 {
 813         struct netdev_phys_item_id ppid = { };
 814         int vifi = vifc->vifc_vifi;
 815         struct vif_device *v = &mrt->vif_table[vifi];
 816         struct net_device *dev;
 817         struct in_device *in_dev;
 818         int err;
 819
 820         /* Is vif busy ? */
 821         if (VIF_EXISTS(mrt, vifi))
 822                 return -EADDRINUSE;
 823
 824         switch (vifc->vifc_flags) {
 825         case VIFF_REGISTER:
 826                 if (!ipmr_pimsm_enabled())
 827                         return -EINVAL;
 828                 /* Special Purpose VIF in PIM
 829                  * All the packets will be sent to the daemon
 830                  */
 831                 if (mrt->mroute_reg_vif_num >= 0)
 832                         return -EADDRINUSE;
 833                 dev = ipmr_reg_vif(net, mrt);
 834                 if (!dev)
 835                         return -ENOBUFS;
 836                 err = dev_set_allmulti(dev, 1);
 837                 if (err) {
 838                         unregister_netdevice(dev);
 839                         dev_put(dev);
 840                         return err;
 841                 }
 842                 break;
 843         case VIFF_TUNNEL:
 844                 dev = ipmr_new_tunnel(net, vifc);
 845                 if (IS_ERR(dev))
 846                         return PTR_ERR(dev);
 847                 break;
 848         case VIFF_USE_IFINDEX:
 849         case 0:
 850                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 851                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 852                         if (dev && !__in_dev_get_rtnl(dev)) {
 853                                 dev_put(dev);
 854                                 return -EADDRNOTAVAIL;
 855                         }
 856                 } else {
 857                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 858                 }
 859                 if (!dev)
 860                         return -EADDRNOTAVAIL;
 861                 err = dev_set_allmulti(dev, 1);
 862                 if (err) {
 863                         dev_put(dev);
 864                         return err;
 865                 }
 866                 break;
 867         default:
 868                 return -EINVAL;
 869         }
 870
 871         in_dev = __in_dev_get_rtnl(dev);
 872         if (!in_dev) {
 873                 dev_put(dev);
 874                 return -EADDRNOTAVAIL;
 875         }
 876         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 877         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 878                                     dev->ifindex, &in_dev->cnf);
 879         ip_rt_multicast_event(in_dev);
 880
 881         /* Fill in the VIF structures */
 882         vif_device_init(v, dev, vifc->vifc_rate_limit,
 883                         vifc->vifc_threshold,
 884                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 885                         (VIFF_TUNNEL | VIFF_REGISTER));
 886
 887         err = dev_get_port_parent_id(dev, &ppid, true);
 888         if (err == 0) {
 889                 memcpy(v->dev_parent_id.id, ppid.id, ppid.id_len);
 890                 v->dev_parent_id.id_len = ppid.id_len;
 891         } else {
 892                 v->dev_parent_id.id_len = 0;
 893         }
 894
 895         v->local = vifc->vifc_lcl_addr.s_addr;
 896         v->remote = vifc->vifc_rmt_addr.s_addr;
 897
 898         /* And finish update writing critical data */
 899         write_lock_bh(&mrt_lock);
 900         v->dev = dev;
 901         if (v->flags & VIFF_REGISTER)
 902                 mrt->mroute_reg_vif_num = vifi;
 903         if (vifi+1 > mrt->maxvif)
 904                 mrt->maxvif = vifi+1;
 905         write_unlock_bh(&mrt_lock);
 906         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 907         return 0;
 908 }
 909
 910 /* called with rcu_read_lock() */
 911 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 912                                          __be32 origin,
 913                                          __be32 mcastgrp)
 914 {
 915         struct mfc_cache_cmp_arg arg = {
 916                         .mfc_mcastgrp = mcastgrp,
 917                         .mfc_origin = origin
 918         };
 919
 920         return mr_mfc_find(mrt, &arg);
 921 }
 922
 923 /* Look for a (*,G) entry */
 924 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 925                                              __be32 mcastgrp, int vifi)
 926 {
 927         struct mfc_cache_cmp_arg arg = {
 928                         .mfc_mcastgrp = mcastgrp,
 929                         .mfc_origin = htonl(INADDR_ANY)
 930         };
 931
 932         if (mcastgrp == htonl(INADDR_ANY))
 933                 return mr_mfc_find_any_parent(mrt, vifi);
 934         return mr_mfc_find_any(mrt, vifi, &arg);
 935 }
 936
 937 /* Look for a (S,G,iif) entry if parent != -1 */
 938 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 939                                                 __be32 origin, __be32 mcastgrp,
 940                                                 int parent)
 941 {
 942         struct mfc_cache_cmp_arg arg = {
 943                         .mfc_mcastgrp = mcastgrp,
 944                         .mfc_origin = origin,
 945         };
 946
 947         return mr_mfc_find_parent(mrt, &arg, parent);
 948 }
 949
 950 /* Allocate a multicast cache entry */
 951 static struct mfc_cache *ipmr_cache_alloc(void)
 952 {
 953         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 954
 955         if (c) {
 956                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 957                 c->_c.mfc_un.res.minvif = MAXVIFS;
 958                 c->_c.free = ipmr_cache_free_rcu;
 959                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 960         }
 961         return c;
 962 }
 963
 964 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 965 {
 966         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 967
 968         if (c) {
 969                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
 970                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
 971         }
 972         return c;
 973 }
 974
 975 /* A cache entry has gone into a resolved state from queued */
 976 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 977                                struct mfc_cache *uc, struct mfc_cache *c)
 978 {
 979         struct sk_buff *skb;
 980         struct nlmsgerr *e;
 981
 982         /* Play the pending entries through our router */
 983         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
 984                 if (ip_hdr(skb)->version == 0) {
 985                         struct nlmsghdr *nlh = skb_pull(skb,
 986                                                         sizeof(struct iphdr));
 987
 988                         if (mr_fill_mroute(mrt, skb, &c->_c,
 989                                            nlmsg_data(nlh)) > 0) {
 990                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 991                                                  (u8 *)nlh;
 992                         } else {
 993                                 nlh->nlmsg_type = NLMSG_ERROR;
 994                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 995                                 skb_trim(skb, nlh->nlmsg_len);
 996                                 e = nlmsg_data(nlh);
 997                                 e->error = -EMSGSIZE;
 998                                 memset(&e->msg, 0, sizeof(e->msg));
 999                         }
1000
1001                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1002                 } else {
1003                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1004                 }
1005         }
1006 }
1007
1008 /* Bounce a cache query up to mrouted and netlink.
1009  *
1010  * Called under mrt_lock.
1011  */
1012 static int ipmr_cache_report(struct mr_table *mrt,
1013                              struct sk_buff *pkt, vifi_t vifi, int assert)
1014 {
1015         const int ihl = ip_hdrlen(pkt);
1016         struct sock *mroute_sk;
1017         struct igmphdr *igmp;
1018         struct igmpmsg *msg;
1019         struct sk_buff *skb;
1020         int ret;
1021
1022         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1023                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1024         else
1025                 skb = alloc_skb(128, GFP_ATOMIC);
1026
1027         if (!skb)
1028                 return -ENOBUFS;
1029
1030         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1031                 /* Ugly, but we have no choice with this interface.
1032                  * Duplicate old header, fix ihl, length etc.
1033                  * And all this only to mangle msg->im_msgtype and
1034                  * to set msg->im_mbz to "mbz" :-)
1035                  */
1036                 skb_push(skb, sizeof(struct iphdr));
1037                 skb_reset_network_header(skb);
1038                 skb_reset_transport_header(skb);
1039                 msg = (struct igmpmsg *)skb_network_header(skb);
1040                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1041                 msg->im_msgtype = assert;
1042                 msg->im_mbz = 0;
1043                 if (assert == IGMPMSG_WRVIFWHOLE) {
1044                         msg->im_vif = vifi;
1045                         msg->im_vif_hi = vifi >> 8;
1046                 } else {
1047                         msg->im_vif = mrt->mroute_reg_vif_num;
1048                         msg->im_vif_hi = mrt->mroute_reg_vif_num >> 8;
1049                 }
1050                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1051                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1052                                              sizeof(struct iphdr));
1053         } else {
1054                 /* Copy the IP header */
1055                 skb_set_network_header(skb, skb->len);
1056                 skb_put(skb, ihl);
1057                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1058                 /* Flag to the kernel this is a route add */
1059                 ip_hdr(skb)->protocol = 0;
1060                 msg = (struct igmpmsg *)skb_network_header(skb);
1061                 msg->im_vif = vifi;
1062                 msg->im_vif_hi = vifi >> 8;
1063                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1064                 /* Add our header */
1065                 igmp = skb_put(skb, sizeof(struct igmphdr));
1066                 igmp->type = assert;
1067                 msg->im_msgtype = assert;
1068                 igmp->code = 0;
1069                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1070                 skb->transport_header = skb->network_header;
1071         }
1072
1073         rcu_read_lock();
1074         mroute_sk = rcu_dereference(mrt->mroute_sk);
1075         if (!mroute_sk) {
1076                 rcu_read_unlock();
1077                 kfree_skb(skb);
1078                 return -EINVAL;
1079         }
1080
1081         igmpmsg_netlink_event(mrt, skb);
1082
1083         /* Deliver to mrouted */
1084         ret = sock_queue_rcv_skb(mroute_sk, skb);
1085         rcu_read_unlock();
1086         if (ret < 0) {
1087                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1088                 kfree_skb(skb);
1089         }
1090
1091         return ret;
1092 }
1093
1094 /* Queue a packet for resolution. It gets locked cache entry! */
1095 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1096                                  struct sk_buff *skb, struct net_device *dev)
1097 {
1098         const struct iphdr *iph = ip_hdr(skb);
1099         struct mfc_cache *c;
1100         bool found = false;
1101         int err;
1102
1103         spin_lock_bh(&mfc_unres_lock);
1104         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1105                 if (c->mfc_mcastgrp == iph->daddr &&
1106                     c->mfc_origin == iph->saddr) {
1107                         found = true;
1108                         break;
1109                 }
1110         }
1111
1112         if (!found) {
1113                 /* Create a new entry if allowable */
1114                 c = ipmr_cache_alloc_unres();
1115                 if (!c) {
1116                         spin_unlock_bh(&mfc_unres_lock);
1117
1118                         kfree_skb(skb);
1119                         return -ENOBUFS;
1120                 }
1121
1122                 /* Fill in the new cache entry */
1123                 c->_c.mfc_parent = -1;
1124                 c->mfc_origin   = iph->saddr;
1125                 c->mfc_mcastgrp = iph->daddr;
1126
1127                 /* Reflect first query at mrouted. */
1128                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1129
1130                 if (err < 0) {
1131                         /* If the report failed throw the cache entry
1132                            out - Brad Parker
1133                          */
1134                         spin_unlock_bh(&mfc_unres_lock);
1135
1136                         ipmr_cache_free(c);
1137                         kfree_skb(skb);
1138                         return err;
1139                 }
1140
1141                 atomic_inc(&mrt->cache_resolve_queue_len);
1142                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1143                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1144
1145                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1146                         mod_timer(&mrt->ipmr_expire_timer,
1147                                   c->_c.mfc_un.unres.expires);
1148         }
1149
1150         /* See if we can append the packet */
1151         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1152                 kfree_skb(skb);
1153                 err = -ENOBUFS;
1154         } else {
1155                 if (dev) {
1156                         skb->dev = dev;
1157                         skb->skb_iif = dev->ifindex;
1158                 }
1159                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1160                 err = 0;
1161         }
1162
1163         spin_unlock_bh(&mfc_unres_lock);
1164         return err;
1165 }
1166
1167 /* MFC cache manipulation by user space mroute daemon */
1168
1169 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1170 {
1171         struct net *net = read_pnet(&mrt->net);
1172         struct mfc_cache *c;
1173
1174         /* The entries are added/deleted only under RTNL */
1175         rcu_read_lock();
1176         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1177                                    mfc->mfcc_mcastgrp.s_addr, parent);
1178         rcu_read_unlock();
1179         if (!c)
1180                 return -ENOENT;
1181         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1182         list_del_rcu(&c->_c.list);
1183         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1184         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1185         mr_cache_put(&c->_c);
1186
1187         return 0;
1188 }
1189
1190 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1191                         struct mfcctl *mfc, int mrtsock, int parent)
1192 {
1193         struct mfc_cache *uc, *c;
1194         struct mr_mfc *_uc;
1195         bool found;
1196         int ret;
1197
1198         if (mfc->mfcc_parent >= MAXVIFS)
1199                 return -ENFILE;
1200
1201         /* The entries are added/deleted only under RTNL */
1202         rcu_read_lock();
1203         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1204                                    mfc->mfcc_mcastgrp.s_addr, parent);
1205         rcu_read_unlock();
1206         if (c) {
1207                 write_lock_bh(&mrt_lock);
1208                 c->_c.mfc_parent = mfc->mfcc_parent;
1209                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1210                 if (!mrtsock)
1211                         c->_c.mfc_flags |= MFC_STATIC;
1212                 write_unlock_bh(&mrt_lock);
1213                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1214                                               mrt->id);
1215                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1216                 return 0;
1217         }
1218
1219         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1220             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1221                 return -EINVAL;
1222
1223         c = ipmr_cache_alloc();
1224         if (!c)
1225                 return -ENOMEM;
1226
1227         c->mfc_origin = mfc->mfcc_origin.s_addr;
1228         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1229         c->_c.mfc_parent = mfc->mfcc_parent;
1230         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1231         if (!mrtsock)
1232                 c->_c.mfc_flags |= MFC_STATIC;
1233
1234         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1235                                   ipmr_rht_params);
1236         if (ret) {
1237                 pr_err("ipmr: rhtable insert error %d\n", ret);
1238                 ipmr_cache_free(c);
1239                 return ret;
1240         }
1241         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1242         /* Check to see if we resolved a queued list. If so we
1243          * need to send on the frames and tidy up.
1244          */
1245         found = false;
1246         spin_lock_bh(&mfc_unres_lock);
1247         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1248                 uc = (struct mfc_cache *)_uc;
1249                 if (uc->mfc_origin == c->mfc_origin &&
1250                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1251                         list_del(&_uc->list);
1252                         atomic_dec(&mrt->cache_resolve_queue_len);
1253                         found = true;
1254                         break;
1255                 }
1256         }
1257         if (list_empty(&mrt->mfc_unres_queue))
1258                 del_timer(&mrt->ipmr_expire_timer);
1259         spin_unlock_bh(&mfc_unres_lock);
1260
1261         if (found) {
1262                 ipmr_cache_resolve(net, mrt, uc, c);
1263                 ipmr_cache_free(uc);
1264         }
1265         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1266         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1267         return 0;
1268 }
1269
1270 /* Close the multicast socket, and clear the vif tables etc */
1271 static void mroute_clean_tables(struct mr_table *mrt, int flags)
1272 {
1273         struct net *net = read_pnet(&mrt->net);
1274         struct mr_mfc *c, *tmp;
1275         struct mfc_cache *cache;
1276         LIST_HEAD(list);
1277         int i;
1278
1279         /* Shut down all active vif entries */
1280         if (flags & (MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC)) {
1281                 for (i = 0; i < mrt->maxvif; i++) {
1282                         if (((mrt->vif_table[i].flags & VIFF_STATIC) &&
1283                              !(flags & MRT_FLUSH_VIFS_STATIC)) ||
1284                             (!(mrt->vif_table[i].flags & VIFF_STATIC) && !(flags & MRT_FLUSH_VIFS)))
1285                                 continue;
1286                         vif_delete(mrt, i, 0, &list);
1287                 }
1288                 unregister_netdevice_many(&list);
1289         }
1290
1291         /* Wipe the cache */
1292         if (flags & (MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC)) {
1293                 list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1294                         if (((c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC_STATIC)) ||
1295                             (!(c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC)))
1296                                 continue;
1297                         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1298                         list_del_rcu(&c->list);
1299                         cache = (struct mfc_cache *)c;
1300                         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1301                                                       mrt->id);
1302                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1303                         mr_cache_put(c);
1304                 }
1305         }
1306
1307         if (flags & MRT_FLUSH_MFC) {
1308                 if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1309                         spin_lock_bh(&mfc_unres_lock);
1310                         list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1311                                 list_del(&c->list);
1312                                 cache = (struct mfc_cache *)c;
1313                                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1314                                 ipmr_destroy_unres(mrt, cache);
1315                         }
1316                         spin_unlock_bh(&mfc_unres_lock);
1317                 }
1318         }
1319 }
1320
1321 /* called from ip_ra_control(), before an RCU grace period,
1322  * we dont need to call synchronize_rcu() here
1323  */
1324 static void mrtsock_destruct(struct sock *sk)
1325 {
1326         struct net *net = sock_net(sk);
1327         struct mr_table *mrt;
1328
1329         rtnl_lock();
1330         ipmr_for_each_table(mrt, net) {
1331                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1332                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1333                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1334                                                     NETCONFA_MC_FORWARDING,
1335                                                     NETCONFA_IFINDEX_ALL,
1336                                                     net->ipv4.devconf_all);
1337                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1338                         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_MFC);
1339                 }
1340         }
1341         rtnl_unlock();
1342 }
1343
1344 /* Socket options and virtual interface manipulation. The whole
1345  * virtual interface system is a complete heap, but unfortunately
1346  * that's how BSD mrouted happens to think. Maybe one day with a proper
1347  * MOSPF/PIM router set up we can clean this up.
1348  */
1349
1350 int ip_mroute_setsockopt(struct sock *sk, int optname, sockptr_t optval,
1351                          unsigned int optlen)
1352 {
1353         struct net *net = sock_net(sk);
1354         int val, ret = 0, parent = 0;
1355         struct mr_table *mrt;
1356         struct vifctl vif;
1357         struct mfcctl mfc;
1358         bool do_wrvifwhole;
1359         u32 uval;
1360
1361         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1362         rtnl_lock();
1363         if (sk->sk_type != SOCK_RAW ||
1364             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1365                 ret = -EOPNOTSUPP;
1366                 goto out_unlock;
1367         }
1368
1369         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1370         if (!mrt) {
1371                 ret = -ENOENT;
1372                 goto out_unlock;
1373         }
1374         if (optname != MRT_INIT) {
1375                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1376                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1377                         ret = -EACCES;
1378                         goto out_unlock;
1379                 }
1380         }
1381
1382         switch (optname) {
1383         case MRT_INIT:
1384                 if (optlen != sizeof(int)) {
1385                         ret = -EINVAL;
1386                         break;
1387                 }
1388                 if (rtnl_dereference(mrt->mroute_sk)) {
1389                         ret = -EADDRINUSE;
1390                         break;
1391                 }
1392
1393                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1394                 if (ret == 0) {
1395                         rcu_assign_pointer(mrt->mroute_sk, sk);
1396                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1397                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1398                                                     NETCONFA_MC_FORWARDING,
1399                                                     NETCONFA_IFINDEX_ALL,
1400                                                     net->ipv4.devconf_all);
1401                 }
1402                 break;
1403         case MRT_DONE:
1404                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1405                         ret = -EACCES;
1406                 } else {
1407                         /* We need to unlock here because mrtsock_destruct takes
1408                          * care of rtnl itself and we can't change that due to
1409                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1410                          */
1411                         rtnl_unlock();
1412                         ret = ip_ra_control(sk, 0, NULL);
1413                         goto out;
1414                 }
1415                 break;
1416         case MRT_ADD_VIF:
1417         case MRT_DEL_VIF:
1418                 if (optlen != sizeof(vif)) {
1419                         ret = -EINVAL;
1420                         break;
1421                 }
1422                 if (copy_from_sockptr(&vif, optval, sizeof(vif))) {
1423                         ret = -EFAULT;
1424                         break;
1425                 }
1426                 if (vif.vifc_vifi >= MAXVIFS) {
1427                         ret = -ENFILE;
1428                         break;
1429                 }
1430                 if (optname == MRT_ADD_VIF) {
1431                         ret = vif_add(net, mrt, &vif,
1432                                       sk == rtnl_dereference(mrt->mroute_sk));
1433                 } else {
1434                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1435                 }
1436                 break;
1437         /* Manipulate the forwarding caches. These live
1438          * in a sort of kernel/user symbiosis.
1439          */
1440         case MRT_ADD_MFC:
1441         case MRT_DEL_MFC:
1442                 parent = -1;
1443                 fallthrough;
1444         case MRT_ADD_MFC_PROXY:
1445         case MRT_DEL_MFC_PROXY:
1446                 if (optlen != sizeof(mfc)) {
1447                         ret = -EINVAL;
1448                         break;
1449                 }
1450                 if (copy_from_sockptr(&mfc, optval, sizeof(mfc))) {
1451                         ret = -EFAULT;
1452                         break;
1453                 }
1454                 if (parent == 0)
1455                         parent = mfc.mfcc_parent;
1456                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1457                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1458                 else
1459                         ret = ipmr_mfc_add(net, mrt, &mfc,
1460                                            sk == rtnl_dereference(mrt->mroute_sk),
1461                                            parent);
1462                 break;
1463         case MRT_FLUSH:
1464                 if (optlen != sizeof(val)) {
1465                         ret = -EINVAL;
1466                         break;
1467                 }
1468                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1469                         ret = -EFAULT;
1470                         break;
1471                 }
1472                 mroute_clean_tables(mrt, val);
1473                 break;
1474         /* Control PIM assert. */
1475         case MRT_ASSERT:
1476                 if (optlen != sizeof(val)) {
1477                         ret = -EINVAL;
1478                         break;
1479                 }
1480                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1481                         ret = -EFAULT;
1482                         break;
1483                 }
1484                 mrt->mroute_do_assert = val;
1485                 break;
1486         case MRT_PIM:
1487                 if (!ipmr_pimsm_enabled()) {
1488                         ret = -ENOPROTOOPT;
1489                         break;
1490                 }
1491                 if (optlen != sizeof(val)) {
1492                         ret = -EINVAL;
1493                         break;
1494                 }
1495                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1496                         ret = -EFAULT;
1497                         break;
1498                 }
1499
1500                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1501                 val = !!val;
1502                 if (val != mrt->mroute_do_pim) {
1503                         mrt->mroute_do_pim = val;
1504                         mrt->mroute_do_assert = val;
1505                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1506                 }
1507                 break;
1508         case MRT_TABLE:
1509                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1510                         ret = -ENOPROTOOPT;
1511                         break;
1512                 }
1513                 if (optlen != sizeof(uval)) {
1514                         ret = -EINVAL;
1515                         break;
1516                 }
1517                 if (copy_from_sockptr(&uval, optval, sizeof(uval))) {
1518                         ret = -EFAULT;
1519                         break;
1520                 }
1521
1522                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1523                         ret = -EBUSY;
1524                 } else {
1525                         mrt = ipmr_new_table(net, uval);
1526                         if (IS_ERR(mrt))
1527                                 ret = PTR_ERR(mrt);
1528                         else
1529                                 raw_sk(sk)->ipmr_table = uval;
1530                 }
1531                 break;
1532         /* Spurious command, or MRT_VERSION which you cannot set. */
1533         default:
1534                 ret = -ENOPROTOOPT;
1535         }
1536 out_unlock:
1537         rtnl_unlock();
1538 out:
1539         return ret;
1540 }
1541
1542 /* Getsock opt support for the multicast routing system. */
1543 int ip_mroute_getsockopt(struct sock *sk, int optname, sockptr_t optval,
1544                          sockptr_t optlen)
1545 {
1546         int olr;
1547         int val;
1548         struct net *net = sock_net(sk);
1549         struct mr_table *mrt;
1550
1551         if (sk->sk_type != SOCK_RAW ||
1552             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1553                 return -EOPNOTSUPP;
1554
1555         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1556         if (!mrt)
1557                 return -ENOENT;
1558
1559         switch (optname) {
1560         case MRT_VERSION:
1561                 val = 0x0305;
1562                 break;
1563         case MRT_PIM:
1564                 if (!ipmr_pimsm_enabled())
1565                         return -ENOPROTOOPT;
1566                 val = mrt->mroute_do_pim;
1567                 break;
1568         case MRT_ASSERT:
1569                 val = mrt->mroute_do_assert;
1570                 break;
1571         default:
1572                 return -ENOPROTOOPT;
1573         }
1574
1575         if (copy_from_sockptr(&olr, optlen, sizeof(int)))
1576                 return -EFAULT;
1577         if (olr < 0)
1578                 return -EINVAL;
1579
1580         olr = min_t(unsigned int, olr, sizeof(int));
1581
1582         if (copy_to_sockptr(optlen, &olr, sizeof(int)))
1583                 return -EFAULT;
1584         if (copy_to_sockptr(optval, &val, olr))
1585                 return -EFAULT;
1586         return 0;
1587 }
1588
1589 /* The IP multicast ioctl support routines. */
1590 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1591 {
1592         struct sioc_sg_req sr;
1593         struct sioc_vif_req vr;
1594         struct vif_device *vif;
1595         struct mfc_cache *c;
1596         struct net *net = sock_net(sk);
1597         struct mr_table *mrt;
1598
1599         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1600         if (!mrt)
1601                 return -ENOENT;
1602
1603         switch (cmd) {
1604         case SIOCGETVIFCNT:
1605                 if (copy_from_user(&vr, arg, sizeof(vr)))
1606                         return -EFAULT;
1607                 if (vr.vifi >= mrt->maxvif)
1608                         return -EINVAL;
1609                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1610                 read_lock(&mrt_lock);
1611                 vif = &mrt->vif_table[vr.vifi];
1612                 if (VIF_EXISTS(mrt, vr.vifi)) {
1613                         vr.icount = vif->pkt_in;
1614                         vr.ocount = vif->pkt_out;
1615                         vr.ibytes = vif->bytes_in;
1616                         vr.obytes = vif->bytes_out;
1617                         read_unlock(&mrt_lock);
1618
1619                         if (copy_to_user(arg, &vr, sizeof(vr)))
1620                                 return -EFAULT;
1621                         return 0;
1622                 }
1623                 read_unlock(&mrt_lock);
1624                 return -EADDRNOTAVAIL;
1625         case SIOCGETSGCNT:
1626                 if (copy_from_user(&sr, arg, sizeof(sr)))
1627                         return -EFAULT;
1628
1629                 rcu_read_lock();
1630                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1631                 if (c) {
1632                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1633                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1634                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1635                         rcu_read_unlock();
1636
1637                         if (copy_to_user(arg, &sr, sizeof(sr)))
1638                                 return -EFAULT;
1639                         return 0;
1640                 }
1641                 rcu_read_unlock();
1642                 return -EADDRNOTAVAIL;
1643         default:
1644                 return -ENOIOCTLCMD;
1645         }
1646 }
1647
1648 #ifdef CONFIG_COMPAT
1649 struct compat_sioc_sg_req {
1650         struct in_addr src;
1651         struct in_addr grp;
1652         compat_ulong_t pktcnt;
1653         compat_ulong_t bytecnt;
1654         compat_ulong_t wrong_if;
1655 };
1656
1657 struct compat_sioc_vif_req {
1658         vifi_t  vifi;           /* Which iface */
1659         compat_ulong_t icount;
1660         compat_ulong_t ocount;
1661         compat_ulong_t ibytes;
1662         compat_ulong_t obytes;
1663 };
1664
1665 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1666 {
1667         struct compat_sioc_sg_req sr;
1668         struct compat_sioc_vif_req vr;
1669         struct vif_device *vif;
1670         struct mfc_cache *c;
1671         struct net *net = sock_net(sk);
1672         struct mr_table *mrt;
1673
1674         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1675         if (!mrt)
1676                 return -ENOENT;
1677
1678         switch (cmd) {
1679         case SIOCGETVIFCNT:
1680                 if (copy_from_user(&vr, arg, sizeof(vr)))
1681                         return -EFAULT;
1682                 if (vr.vifi >= mrt->maxvif)
1683                         return -EINVAL;
1684                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1685                 read_lock(&mrt_lock);
1686                 vif = &mrt->vif_table[vr.vifi];
1687                 if (VIF_EXISTS(mrt, vr.vifi)) {
1688                         vr.icount = vif->pkt_in;
1689                         vr.ocount = vif->pkt_out;
1690                         vr.ibytes = vif->bytes_in;
1691                         vr.obytes = vif->bytes_out;
1692                         read_unlock(&mrt_lock);
1693
1694                         if (copy_to_user(arg, &vr, sizeof(vr)))
1695                                 return -EFAULT;
1696                         return 0;
1697                 }
1698                 read_unlock(&mrt_lock);
1699                 return -EADDRNOTAVAIL;
1700         case SIOCGETSGCNT:
1701                 if (copy_from_user(&sr, arg, sizeof(sr)))
1702                         return -EFAULT;
1703
1704                 rcu_read_lock();
1705                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1706                 if (c) {
1707                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1708                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1709                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1710                         rcu_read_unlock();
1711
1712                         if (copy_to_user(arg, &sr, sizeof(sr)))
1713                                 return -EFAULT;
1714                         return 0;
1715                 }
1716                 rcu_read_unlock();
1717                 return -EADDRNOTAVAIL;
1718         default:
1719                 return -ENOIOCTLCMD;
1720         }
1721 }
1722 #endif
1723
1724 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1725 {
1726         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1727         struct net *net = dev_net(dev);
1728         struct mr_table *mrt;
1729         struct vif_device *v;
1730         int ct;
1731
1732         if (event != NETDEV_UNREGISTER)
1733                 return NOTIFY_DONE;
1734
1735         ipmr_for_each_table(mrt, net) {
1736                 v = &mrt->vif_table[0];
1737                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1738                         if (v->dev == dev)
1739                                 vif_delete(mrt, ct, 1, NULL);
1740                 }
1741         }
1742         return NOTIFY_DONE;
1743 }
1744
1745 static struct notifier_block ip_mr_notifier = {
1746         .notifier_call = ipmr_device_event,
1747 };
1748
1749 /* Encapsulate a packet by attaching a valid IPIP header to it.
1750  * This avoids tunnel drivers and other mess and gives us the speed so
1751  * important for multicast video.
1752  */
1753 static void ip_encap(struct net *net, struct sk_buff *skb,
1754                      __be32 saddr, __be32 daddr)
1755 {
1756         struct iphdr *iph;
1757         const struct iphdr *old_iph = ip_hdr(skb);
1758
1759         skb_push(skb, sizeof(struct iphdr));
1760         skb->transport_header = skb->network_header;
1761         skb_reset_network_header(skb);
1762         iph = ip_hdr(skb);
1763
1764         iph->version    =       4;
1765         iph->tos        =       old_iph->tos;
1766         iph->ttl        =       old_iph->ttl;
1767         iph->frag_off   =       0;
1768         iph->daddr      =       daddr;
1769         iph->saddr      =       saddr;
1770         iph->protocol   =       IPPROTO_IPIP;
1771         iph->ihl        =       5;
1772         iph->tot_len    =       htons(skb->len);
1773         ip_select_ident(net, skb, NULL);
1774         ip_send_check(iph);
1775
1776         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1777         nf_reset_ct(skb);
1778 }
1779
1780 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1781                                       struct sk_buff *skb)
1782 {
1783         struct ip_options *opt = &(IPCB(skb)->opt);
1784
1785         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1786         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1787
1788         if (unlikely(opt->optlen))
1789                 ip_forward_options(skb);
1790
1791         return dst_output(net, sk, skb);
1792 }
1793
1794 #ifdef CONFIG_NET_SWITCHDEV
1795 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1796                                    int in_vifi, int out_vifi)
1797 {
1798         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1799         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1800
1801         if (!skb->offload_l3_fwd_mark)
1802                 return false;
1803         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1804                 return false;
1805         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1806                                         &in_vif->dev_parent_id);
1807 }
1808 #else
1809 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1810                                    int in_vifi, int out_vifi)
1811 {
1812         return false;
1813 }
1814 #endif
1815
1816 /* Processing handlers for ipmr_forward */
1817
1818 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1819                             int in_vifi, struct sk_buff *skb, int vifi)
1820 {
1821         const struct iphdr *iph = ip_hdr(skb);
1822         struct vif_device *vif = &mrt->vif_table[vifi];
1823         struct net_device *dev;
1824         struct rtable *rt;
1825         struct flowi4 fl4;
1826         int    encap = 0;
1827
1828         if (!vif->dev)
1829                 goto out_free;
1830
1831         if (vif->flags & VIFF_REGISTER) {
1832                 vif->pkt_out++;
1833                 vif->bytes_out += skb->len;
1834                 vif->dev->stats.tx_bytes += skb->len;
1835                 vif->dev->stats.tx_packets++;
1836                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1837                 goto out_free;
1838         }
1839
1840         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1841                 goto out_free;
1842
1843         if (vif->flags & VIFF_TUNNEL) {
1844                 rt = ip_route_output_ports(net, &fl4, NULL,
1845                                            vif->remote, vif->local,
1846                                            0, 0,
1847                                            IPPROTO_IPIP,
1848                                            RT_TOS(iph->tos), vif->link);
1849                 if (IS_ERR(rt))
1850                         goto out_free;
1851                 encap = sizeof(struct iphdr);
1852         } else {
1853                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1854                                            0, 0,
1855                                            IPPROTO_IPIP,
1856                                            RT_TOS(iph->tos), vif->link);
1857                 if (IS_ERR(rt))
1858                         goto out_free;
1859         }
1860
1861         dev = rt->dst.dev;
1862
1863         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1864                 /* Do not fragment multicasts. Alas, IPv4 does not
1865                  * allow to send ICMP, so that packets will disappear
1866                  * to blackhole.
1867                  */
1868                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1869                 ip_rt_put(rt);
1870                 goto out_free;
1871         }
1872
1873         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1874
1875         if (skb_cow(skb, encap)) {
1876                 ip_rt_put(rt);
1877                 goto out_free;
1878         }
1879
1880         vif->pkt_out++;
1881         vif->bytes_out += skb->len;
1882
1883         skb_dst_drop(skb);
1884         skb_dst_set(skb, &rt->dst);
1885         ip_decrease_ttl(ip_hdr(skb));
1886
1887         /* FIXME: forward and output firewalls used to be called here.
1888          * What do we do with netfilter? -- RR
1889          */
1890         if (vif->flags & VIFF_TUNNEL) {
1891                 ip_encap(net, skb, vif->local, vif->remote);
1892                 /* FIXME: extra output firewall step used to be here. --RR */
1893                 vif->dev->stats.tx_packets++;
1894                 vif->dev->stats.tx_bytes += skb->len;
1895         }
1896
1897         IPCB(skb)->flags |= IPSKB_FORWARDED;
1898
1899         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1900          * not only before forwarding, but after forwarding on all output
1901          * interfaces. It is clear, if mrouter runs a multicasting
1902          * program, it should receive packets not depending to what interface
1903          * program is joined.
1904          * If we will not make it, the program will have to join on all
1905          * interfaces. On the other hand, multihoming host (or router, but
1906          * not mrouter) cannot join to more than one interface - it will
1907          * result in receiving multiple packets.
1908          */
1909         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1910                 net, NULL, skb, skb->dev, dev,
1911                 ipmr_forward_finish);
1912         return;
1913
1914 out_free:
1915         kfree_skb(skb);
1916 }
1917
1918 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1919 {
1920         int ct;
1921
1922         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1923                 if (mrt->vif_table[ct].dev == dev)
1924                         break;
1925         }
1926         return ct;
1927 }
1928
1929 /* "local" means that we should preserve one skb (for local delivery) */
1930 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1931                           struct net_device *dev, struct sk_buff *skb,
1932                           struct mfc_cache *c, int local)
1933 {
1934         int true_vifi = ipmr_find_vif(mrt, dev);
1935         int psend = -1;
1936         int vif, ct;
1937
1938         vif = c->_c.mfc_parent;
1939         c->_c.mfc_un.res.pkt++;
1940         c->_c.mfc_un.res.bytes += skb->len;
1941         c->_c.mfc_un.res.lastuse = jiffies;
1942
1943         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1944                 struct mfc_cache *cache_proxy;
1945
1946                 /* For an (*,G) entry, we only check that the incomming
1947                  * interface is part of the static tree.
1948                  */
1949                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1950                 if (cache_proxy &&
1951                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1952                         goto forward;
1953         }
1954
1955         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1956         if (mrt->vif_table[vif].dev != dev) {
1957                 if (rt_is_output_route(skb_rtable(skb))) {
1958                         /* It is our own packet, looped back.
1959                          * Very complicated situation...
1960                          *
1961                          * The best workaround until routing daemons will be
1962                          * fixed is not to redistribute packet, if it was
1963                          * send through wrong interface. It means, that
1964                          * multicast applications WILL NOT work for
1965                          * (S,G), which have default multicast route pointing
1966                          * to wrong oif. In any case, it is not a good
1967                          * idea to use multicasting applications on router.
1968                          */
1969                         goto dont_forward;
1970                 }
1971
1972                 c->_c.mfc_un.res.wrong_if++;
1973
1974                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1975                     /* pimsm uses asserts, when switching from RPT to SPT,
1976                      * so that we cannot check that packet arrived on an oif.
1977                      * It is bad, but otherwise we would need to move pretty
1978                      * large chunk of pimd to kernel. Ough... --ANK
1979                      */
1980                     (mrt->mroute_do_pim ||
1981                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1982                     time_after(jiffies,
1983                                c->_c.mfc_un.res.last_assert +
1984                                MFC_ASSERT_THRESH)) {
1985                         c->_c.mfc_un.res.last_assert = jiffies;
1986                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1987                         if (mrt->mroute_do_wrvifwhole)
1988                                 ipmr_cache_report(mrt, skb, true_vifi,
1989                                                   IGMPMSG_WRVIFWHOLE);
1990                 }
1991                 goto dont_forward;
1992         }
1993
1994 forward:
1995         mrt->vif_table[vif].pkt_in++;
1996         mrt->vif_table[vif].bytes_in += skb->len;
1997
1998         /* Forward the frame */
1999         if (c->mfc_origin == htonl(INADDR_ANY) &&
2000             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
2001                 if (true_vifi >= 0 &&
2002                     true_vifi != c->_c.mfc_parent &&
2003                     ip_hdr(skb)->ttl >
2004                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2005                         /* It's an (*,*) entry and the packet is not coming from
2006                          * the upstream: forward the packet to the upstream
2007                          * only.
2008                          */
2009                         psend = c->_c.mfc_parent;
2010                         goto last_forward;
2011                 }
2012                 goto dont_forward;
2013         }
2014         for (ct = c->_c.mfc_un.res.maxvif - 1;
2015              ct >= c->_c.mfc_un.res.minvif; ct--) {
2016                 /* For (*,G) entry, don't forward to the incoming interface */
2017                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2018                      ct != true_vifi) &&
2019                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2020                         if (psend != -1) {
2021                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2022
2023                                 if (skb2)
2024                                         ipmr_queue_xmit(net, mrt, true_vifi,
2025                                                         skb2, psend);
2026                         }
2027                         psend = ct;
2028                 }
2029         }
2030 last_forward:
2031         if (psend != -1) {
2032                 if (local) {
2033                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2034
2035                         if (skb2)
2036                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2037                                                 psend);
2038                 } else {
2039                         ipmr_queue_xmit(net, mrt, true_vifi, skb, psend);
2040                         return;
2041                 }
2042         }
2043
2044 dont_forward:
2045         if (!local)
2046                 kfree_skb(skb);
2047 }
2048
2049 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2050 {
2051         struct rtable *rt = skb_rtable(skb);
2052         struct iphdr *iph = ip_hdr(skb);
2053         struct flowi4 fl4 = {
2054                 .daddr = iph->daddr,
2055                 .saddr = iph->saddr,
2056                 .flowi4_tos = RT_TOS(iph->tos),
2057                 .flowi4_oif = (rt_is_output_route(rt) ?
2058                                skb->dev->ifindex : 0),
2059                 .flowi4_iif = (rt_is_output_route(rt) ?
2060                                LOOPBACK_IFINDEX :
2061                                skb->dev->ifindex),
2062                 .flowi4_mark = skb->mark,
2063         };
2064         struct mr_table *mrt;
2065         int err;
2066
2067         err = ipmr_fib_lookup(net, &fl4, &mrt);
2068         if (err)
2069                 return ERR_PTR(err);
2070         return mrt;
2071 }
2072
2073 /* Multicast packets for forwarding arrive here
2074  * Called with rcu_read_lock();
2075  */
2076 int ip_mr_input(struct sk_buff *skb)
2077 {
2078         struct mfc_cache *cache;
2079         struct net *net = dev_net(skb->dev);
2080         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2081         struct mr_table *mrt;
2082         struct net_device *dev;
2083
2084         /* skb->dev passed in is the loX master dev for vrfs.
2085          * As there are no vifs associated with loopback devices,
2086          * get the proper interface that does have a vif associated with it.
2087          */
2088         dev = skb->dev;
2089         if (netif_is_l3_master(skb->dev)) {
2090                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2091                 if (!dev) {
2092                         kfree_skb(skb);
2093                         return -ENODEV;
2094                 }
2095         }
2096
2097         /* Packet is looped back after forward, it should not be
2098          * forwarded second time, but still can be delivered locally.
2099          */
2100         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2101                 goto dont_forward;
2102
2103         mrt = ipmr_rt_fib_lookup(net, skb);
2104         if (IS_ERR(mrt)) {
2105                 kfree_skb(skb);
2106                 return PTR_ERR(mrt);
2107         }
2108         if (!local) {
2109                 if (IPCB(skb)->opt.router_alert) {
2110                         if (ip_call_ra_chain(skb))
2111                                 return 0;
2112                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2113                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2114                          * Cisco IOS <= 11.2(8)) do not put router alert
2115                          * option to IGMP packets destined to routable
2116                          * groups. It is very bad, because it means
2117                          * that we can forward NO IGMP messages.
2118                          */
2119                         struct sock *mroute_sk;
2120
2121                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2122                         if (mroute_sk) {
2123                                 nf_reset_ct(skb);
2124                                 raw_rcv(mroute_sk, skb);
2125                                 return 0;
2126                         }
2127                     }
2128         }
2129
2130         /* already under rcu_read_lock() */
2131         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2132         if (!cache) {
2133                 int vif = ipmr_find_vif(mrt, dev);
2134
2135                 if (vif >= 0)
2136                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2137                                                     vif);
2138         }
2139
2140         /* No usable cache entry */
2141         if (!cache) {
2142                 int vif;
2143
2144                 if (local) {
2145                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2146                         ip_local_deliver(skb);
2147                         if (!skb2)
2148                                 return -ENOBUFS;
2149                         skb = skb2;
2150                 }
2151
2152                 read_lock(&mrt_lock);
2153                 vif = ipmr_find_vif(mrt, dev);
2154                 if (vif >= 0) {
2155                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2156                         read_unlock(&mrt_lock);
2157
2158                         return err2;
2159                 }
2160                 read_unlock(&mrt_lock);
2161                 kfree_skb(skb);
2162                 return -ENODEV;
2163         }
2164
2165         read_lock(&mrt_lock);
2166         ip_mr_forward(net, mrt, dev, skb, cache, local);
2167         read_unlock(&mrt_lock);
2168
2169         if (local)
2170                 return ip_local_deliver(skb);
2171
2172         return 0;
2173
2174 dont_forward:
2175         if (local)
2176                 return ip_local_deliver(skb);
2177         kfree_skb(skb);
2178         return 0;
2179 }
2180
2181 #ifdef CONFIG_IP_PIMSM_V1
2182 /* Handle IGMP messages of PIMv1 */
2183 int pim_rcv_v1(struct sk_buff *skb)
2184 {
2185         struct igmphdr *pim;
2186         struct net *net = dev_net(skb->dev);
2187         struct mr_table *mrt;
2188
2189         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2190                 goto drop;
2191
2192         pim = igmp_hdr(skb);
2193
2194         mrt = ipmr_rt_fib_lookup(net, skb);
2195         if (IS_ERR(mrt))
2196                 goto drop;
2197         if (!mrt->mroute_do_pim ||
2198             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2199                 goto drop;
2200
2201         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2202 drop:
2203                 kfree_skb(skb);
2204         }
2205         return 0;
2206 }
2207 #endif
2208
2209 #ifdef CONFIG_IP_PIMSM_V2
2210 static int pim_rcv(struct sk_buff *skb)
2211 {
2212         struct pimreghdr *pim;
2213         struct net *net = dev_net(skb->dev);
2214         struct mr_table *mrt;
2215
2216         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2217                 goto drop;
2218
2219         pim = (struct pimreghdr *)skb_transport_header(skb);
2220         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2221             (pim->flags & PIM_NULL_REGISTER) ||
2222             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2223              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2224                 goto drop;
2225
2226         mrt = ipmr_rt_fib_lookup(net, skb);
2227         if (IS_ERR(mrt))
2228                 goto drop;
2229         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2230 drop:
2231                 kfree_skb(skb);
2232         }
2233         return 0;
2234 }
2235 #endif
2236
2237 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2238                    __be32 saddr, __be32 daddr,
2239                    struct rtmsg *rtm, u32 portid)
2240 {
2241         struct mfc_cache *cache;
2242         struct mr_table *mrt;
2243         int err;
2244
2245         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2246         if (!mrt)
2247                 return -ENOENT;
2248
2249         rcu_read_lock();
2250         cache = ipmr_cache_find(mrt, saddr, daddr);
2251         if (!cache && skb->dev) {
2252                 int vif = ipmr_find_vif(mrt, skb->dev);
2253
2254                 if (vif >= 0)
2255                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2256         }
2257         if (!cache) {
2258                 struct sk_buff *skb2;
2259                 struct iphdr *iph;
2260                 struct net_device *dev;
2261                 int vif = -1;
2262
2263                 dev = skb->dev;
2264                 read_lock(&mrt_lock);
2265                 if (dev)
2266                         vif = ipmr_find_vif(mrt, dev);
2267                 if (vif < 0) {
2268                         read_unlock(&mrt_lock);
2269                         rcu_read_unlock();
2270                         return -ENODEV;
2271                 }
2272
2273                 skb2 = skb_realloc_headroom(skb, sizeof(struct iphdr));
2274                 if (!skb2) {
2275                         read_unlock(&mrt_lock);
2276                         rcu_read_unlock();
2277                         return -ENOMEM;
2278                 }
2279
2280                 NETLINK_CB(skb2).portid = portid;
2281                 skb_push(skb2, sizeof(struct iphdr));
2282                 skb_reset_network_header(skb2);
2283                 iph = ip_hdr(skb2);
2284                 iph->ihl = sizeof(struct iphdr) >> 2;
2285                 iph->saddr = saddr;
2286                 iph->daddr = daddr;
2287                 iph->version = 0;
2288                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2289                 read_unlock(&mrt_lock);
2290                 rcu_read_unlock();
2291                 return err;
2292         }
2293
2294         read_lock(&mrt_lock);
2295         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2296         read_unlock(&mrt_lock);
2297         rcu_read_unlock();
2298         return err;
2299 }
2300
2301 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2302                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2303                             int flags)
2304 {
2305         struct nlmsghdr *nlh;
2306         struct rtmsg *rtm;
2307         int err;
2308
2309         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2310         if (!nlh)
2311                 return -EMSGSIZE;
2312
2313         rtm = nlmsg_data(nlh);
2314         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2315         rtm->rtm_dst_len  = 32;
2316         rtm->rtm_src_len  = 32;
2317         rtm->rtm_tos      = 0;
2318         rtm->rtm_table    = mrt->id;
2319         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2320                 goto nla_put_failure;
2321         rtm->rtm_type     = RTN_MULTICAST;
2322         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2323         if (c->_c.mfc_flags & MFC_STATIC)
2324                 rtm->rtm_protocol = RTPROT_STATIC;
2325         else
2326                 rtm->rtm_protocol = RTPROT_MROUTED;
2327         rtm->rtm_flags    = 0;
2328
2329         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2330             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2331                 goto nla_put_failure;
2332         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2333         /* do not break the dump if cache is unresolved */
2334         if (err < 0 && err != -ENOENT)
2335                 goto nla_put_failure;
2336
2337         nlmsg_end(skb, nlh);
2338         return 0;
2339
2340 nla_put_failure:
2341         nlmsg_cancel(skb, nlh);
2342         return -EMSGSIZE;
2343 }
2344
2345 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2346                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2347                              int flags)
2348 {
2349         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2350                                 cmd, flags);
2351 }
2352
2353 static size_t mroute_msgsize(bool unresolved, int maxvif)
2354 {
2355         size_t len =
2356                 NLMSG_ALIGN(sizeof(struct rtmsg))
2357                 + nla_total_size(4)     /* RTA_TABLE */
2358                 + nla_total_size(4)     /* RTA_SRC */
2359                 + nla_total_size(4)     /* RTA_DST */
2360                 ;
2361
2362         if (!unresolved)
2363                 len = len
2364                       + nla_total_size(4)       /* RTA_IIF */
2365                       + nla_total_size(0)       /* RTA_MULTIPATH */
2366                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2367                                                 /* RTA_MFC_STATS */
2368                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2369                 ;
2370
2371         return len;
2372 }
2373
2374 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2375                                  int cmd)
2376 {
2377         struct net *net = read_pnet(&mrt->net);
2378         struct sk_buff *skb;
2379         int err = -ENOBUFS;
2380
2381         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2382                                        mrt->maxvif),
2383                         GFP_ATOMIC);
2384         if (!skb)
2385                 goto errout;
2386
2387         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2388         if (err < 0)
2389                 goto errout;
2390
2391         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2392         return;
2393
2394 errout:
2395         kfree_skb(skb);
2396         if (err < 0)
2397                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2398 }
2399
2400 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2401 {
2402         size_t len =
2403                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2404                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2405                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2406                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2407                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2408                 + nla_total_size(4)     /* IPMRA_CREPORT_TABLE */
2409                                         /* IPMRA_CREPORT_PKT */
2410                 + nla_total_size(payloadlen)
2411                 ;
2412
2413         return len;
2414 }
2415
2416 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2417 {
2418         struct net *net = read_pnet(&mrt->net);
2419         struct nlmsghdr *nlh;
2420         struct rtgenmsg *rtgenm;
2421         struct igmpmsg *msg;
2422         struct sk_buff *skb;
2423         struct nlattr *nla;
2424         int payloadlen;
2425
2426         payloadlen = pkt->len - sizeof(struct igmpmsg);
2427         msg = (struct igmpmsg *)skb_network_header(pkt);
2428
2429         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2430         if (!skb)
2431                 goto errout;
2432
2433         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2434                         sizeof(struct rtgenmsg), 0);
2435         if (!nlh)
2436                 goto errout;
2437         rtgenm = nlmsg_data(nlh);
2438         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2439         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2440             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif | (msg->im_vif_hi << 8)) ||
2441             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2442                             msg->im_src.s_addr) ||
2443             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2444                             msg->im_dst.s_addr) ||
2445             nla_put_u32(skb, IPMRA_CREPORT_TABLE, mrt->id))
2446                 goto nla_put_failure;
2447
2448         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2449         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2450                                   nla_data(nla), payloadlen))
2451                 goto nla_put_failure;
2452
2453         nlmsg_end(skb, nlh);
2454
2455         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2456         return;
2457
2458 nla_put_failure:
2459         nlmsg_cancel(skb, nlh);
2460 errout:
2461         kfree_skb(skb);
2462         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2463 }
2464
2465 static int ipmr_rtm_valid_getroute_req(struct sk_buff *skb,
2466                                        const struct nlmsghdr *nlh,
2467                                        struct nlattr **tb,
2468                                        struct netlink_ext_ack *extack)
2469 {
2470         struct rtmsg *rtm;
2471         int i, err;
2472
2473         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2474                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for multicast route get request");
2475                 return -EINVAL;
2476         }
2477
2478         if (!netlink_strict_get_check(skb))
2479                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2480                                               rtm_ipv4_policy, extack);
2481
2482         rtm = nlmsg_data(nlh);
2483         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2484             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2485             rtm->rtm_tos || rtm->rtm_table || rtm->rtm_protocol ||
2486             rtm->rtm_scope || rtm->rtm_type || rtm->rtm_flags) {
2487                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for multicast route get request");
2488                 return -EINVAL;
2489         }
2490
2491         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2492                                             rtm_ipv4_policy, extack);
2493         if (err)
2494                 return err;
2495
2496         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2497             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2498                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2499                 return -EINVAL;
2500         }
2501
2502         for (i = 0; i <= RTA_MAX; i++) {
2503                 if (!tb[i])
2504                         continue;
2505
2506                 switch (i) {
2507                 case RTA_SRC:
2508                 case RTA_DST:
2509                 case RTA_TABLE:
2510                         break;
2511                 default:
2512                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in multicast route get request");
2513                         return -EINVAL;
2514                 }
2515         }
2516
2517         return 0;
2518 }
2519
2520 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2521                              struct netlink_ext_ack *extack)
2522 {
2523         struct net *net = sock_net(in_skb->sk);
2524         struct nlattr *tb[RTA_MAX + 1];
2525         struct sk_buff *skb = NULL;
2526         struct mfc_cache *cache;
2527         struct mr_table *mrt;
2528         __be32 src, grp;
2529         u32 tableid;
2530         int err;
2531
2532         err = ipmr_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2533         if (err < 0)
2534                 goto errout;
2535
2536         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2537         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2538         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2539
2540         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2541         if (!mrt) {
2542                 err = -ENOENT;
2543                 goto errout_free;
2544         }
2545
2546         /* entries are added/deleted only under RTNL */
2547         rcu_read_lock();
2548         cache = ipmr_cache_find(mrt, src, grp);
2549         rcu_read_unlock();
2550         if (!cache) {
2551                 err = -ENOENT;
2552                 goto errout_free;
2553         }
2554
2555         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2556         if (!skb) {
2557                 err = -ENOBUFS;
2558                 goto errout_free;
2559         }
2560
2561         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2562                                nlh->nlmsg_seq, cache,
2563                                RTM_NEWROUTE, 0);
2564         if (err < 0)
2565                 goto errout_free;
2566
2567         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2568
2569 errout:
2570         return err;
2571
2572 errout_free:
2573         kfree_skb(skb);
2574         goto errout;
2575 }
2576
2577 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2578 {
2579         struct fib_dump_filter filter = {};
2580         int err;
2581
2582         if (cb->strict_check) {
2583                 err = ip_valid_fib_dump_req(sock_net(skb->sk), cb->nlh,
2584                                             &filter, cb);
2585                 if (err < 0)
2586                         return err;
2587         }
2588
2589         if (filter.table_id) {
2590                 struct mr_table *mrt;
2591
2592                 mrt = ipmr_get_table(sock_net(skb->sk), filter.table_id);
2593                 if (!mrt) {
2594                         if (rtnl_msg_family(cb->nlh) != RTNL_FAMILY_IPMR)
2595                                 return skb->len;
2596
2597                         NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
2598                         return -ENOENT;
2599                 }
2600                 err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
2601                                     &mfc_unres_lock, &filter);
2602                 return skb->len ? : err;
2603         }
2604
2605         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2606                                 _ipmr_fill_mroute, &mfc_unres_lock, &filter);
2607 }
2608
2609 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2610         [RTA_SRC]       = { .type = NLA_U32 },
2611         [RTA_DST]       = { .type = NLA_U32 },
2612         [RTA_IIF]       = { .type = NLA_U32 },
2613         [RTA_TABLE]     = { .type = NLA_U32 },
2614         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2615 };
2616
2617 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2618 {
2619         switch (rtm_protocol) {
2620         case RTPROT_STATIC:
2621         case RTPROT_MROUTED:
2622                 return true;
2623         }
2624         return false;
2625 }
2626
2627 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2628 {
2629         struct rtnexthop *rtnh = nla_data(nla);
2630         int remaining = nla_len(nla), vifi = 0;
2631
2632         while (rtnh_ok(rtnh, remaining)) {
2633                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2634                 if (++vifi == MAXVIFS)
2635                         break;
2636                 rtnh = rtnh_next(rtnh, &remaining);
2637         }
2638
2639         return remaining > 0 ? -EINVAL : vifi;
2640 }
2641
2642 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2643 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2644                             struct mfcctl *mfcc, int *mrtsock,
2645                             struct mr_table **mrtret,
2646                             struct netlink_ext_ack *extack)
2647 {
2648         struct net_device *dev = NULL;
2649         u32 tblid = RT_TABLE_DEFAULT;
2650         struct mr_table *mrt;
2651         struct nlattr *attr;
2652         struct rtmsg *rtm;
2653         int ret, rem;
2654
2655         ret = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
2656                                         rtm_ipmr_policy, extack);
2657         if (ret < 0)
2658                 goto out;
2659         rtm = nlmsg_data(nlh);
2660
2661         ret = -EINVAL;
2662         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2663             rtm->rtm_type != RTN_MULTICAST ||
2664             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2665             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2666                 goto out;
2667
2668         memset(mfcc, 0, sizeof(*mfcc));
2669         mfcc->mfcc_parent = -1;
2670         ret = 0;
2671         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2672                 switch (nla_type(attr)) {
2673                 case RTA_SRC:
2674                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2675                         break;
2676                 case RTA_DST:
2677                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2678                         break;
2679                 case RTA_IIF:
2680                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2681                         if (!dev) {
2682                                 ret = -ENODEV;
2683                                 goto out;
2684                         }
2685                         break;
2686                 case RTA_MULTIPATH:
2687                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2688                                 ret = -EINVAL;
2689                                 goto out;
2690                         }
2691                         break;
2692                 case RTA_PREFSRC:
2693                         ret = 1;
2694                         break;
2695                 case RTA_TABLE:
2696                         tblid = nla_get_u32(attr);
2697                         break;
2698                 }
2699         }
2700         mrt = ipmr_get_table(net, tblid);
2701         if (!mrt) {
2702                 ret = -ENOENT;
2703                 goto out;
2704         }
2705         *mrtret = mrt;
2706         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2707         if (dev)
2708                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2709
2710 out:
2711         return ret;
2712 }
2713
2714 /* takes care of both newroute and delroute */
2715 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2716                           struct netlink_ext_ack *extack)
2717 {
2718         struct net *net = sock_net(skb->sk);
2719         int ret, mrtsock, parent;
2720         struct mr_table *tbl;
2721         struct mfcctl mfcc;
2722
2723         mrtsock = 0;
2724         tbl = NULL;
2725         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2726         if (ret < 0)
2727                 return ret;
2728
2729         parent = ret ? mfcc.mfcc_parent : -1;
2730         if (nlh->nlmsg_type == RTM_NEWROUTE)
2731                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2732         else
2733                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2734 }
2735
2736 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2737 {
2738         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2739
2740         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2741             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2742             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2743                         mrt->mroute_reg_vif_num) ||
2744             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2745                        mrt->mroute_do_assert) ||
2746             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2747             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2748                        mrt->mroute_do_wrvifwhole))
2749                 return false;
2750
2751         return true;
2752 }
2753
2754 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2755 {
2756         struct nlattr *vif_nest;
2757         struct vif_device *vif;
2758
2759         /* if the VIF doesn't exist just continue */
2760         if (!VIF_EXISTS(mrt, vifid))
2761                 return true;
2762
2763         vif = &mrt->vif_table[vifid];
2764         vif_nest = nla_nest_start_noflag(skb, IPMRA_VIF);
2765         if (!vif_nest)
2766                 return false;
2767         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2768             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2769             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2770             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2771                               IPMRA_VIFA_PAD) ||
2772             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2773                               IPMRA_VIFA_PAD) ||
2774             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2775                               IPMRA_VIFA_PAD) ||
2776             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2777                               IPMRA_VIFA_PAD) ||
2778             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2779             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2780                 nla_nest_cancel(skb, vif_nest);
2781                 return false;
2782         }
2783         nla_nest_end(skb, vif_nest);
2784
2785         return true;
2786 }
2787
2788 static int ipmr_valid_dumplink(const struct nlmsghdr *nlh,
2789                                struct netlink_ext_ack *extack)
2790 {
2791         struct ifinfomsg *ifm;
2792
2793         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ifm))) {
2794                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for ipmr link dump");
2795                 return -EINVAL;
2796         }
2797
2798         if (nlmsg_attrlen(nlh, sizeof(*ifm))) {
2799                 NL_SET_ERR_MSG(extack, "Invalid data after header in ipmr link dump");
2800                 return -EINVAL;
2801         }
2802
2803         ifm = nlmsg_data(nlh);
2804         if (ifm->__ifi_pad || ifm->ifi_type || ifm->ifi_flags ||
2805             ifm->ifi_change || ifm->ifi_index) {
2806                 NL_SET_ERR_MSG(extack, "Invalid values in header for ipmr link dump request");
2807                 return -EINVAL;
2808         }
2809
2810         return 0;
2811 }
2812
2813 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2814 {
2815         struct net *net = sock_net(skb->sk);
2816         struct nlmsghdr *nlh = NULL;
2817         unsigned int t = 0, s_t;
2818         unsigned int e = 0, s_e;
2819         struct mr_table *mrt;
2820
2821         if (cb->strict_check) {
2822                 int err = ipmr_valid_dumplink(cb->nlh, cb->extack);
2823
2824                 if (err < 0)
2825                         return err;
2826         }
2827
2828         s_t = cb->args[0];
2829         s_e = cb->args[1];
2830
2831         ipmr_for_each_table(mrt, net) {
2832                 struct nlattr *vifs, *af;
2833                 struct ifinfomsg *hdr;
2834                 u32 i;
2835
2836                 if (t < s_t)
2837                         goto skip_table;
2838                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2839                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2840                                 sizeof(*hdr), NLM_F_MULTI);
2841                 if (!nlh)
2842                         break;
2843
2844                 hdr = nlmsg_data(nlh);
2845                 memset(hdr, 0, sizeof(*hdr));
2846                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2847
2848                 af = nla_nest_start_noflag(skb, IFLA_AF_SPEC);
2849                 if (!af) {
2850                         nlmsg_cancel(skb, nlh);
2851                         goto out;
2852                 }
2853
2854                 if (!ipmr_fill_table(mrt, skb)) {
2855                         nlmsg_cancel(skb, nlh);
2856                         goto out;
2857                 }
2858
2859                 vifs = nla_nest_start_noflag(skb, IPMRA_TABLE_VIFS);
2860                 if (!vifs) {
2861                         nla_nest_end(skb, af);
2862                         nlmsg_end(skb, nlh);
2863                         goto out;
2864                 }
2865                 for (i = 0; i < mrt->maxvif; i++) {
2866                         if (e < s_e)
2867                                 goto skip_entry;
2868                         if (!ipmr_fill_vif(mrt, i, skb)) {
2869                                 nla_nest_end(skb, vifs);
2870                                 nla_nest_end(skb, af);
2871                                 nlmsg_end(skb, nlh);
2872                                 goto out;
2873                         }
2874 skip_entry:
2875                         e++;
2876                 }
2877                 s_e = 0;
2878                 e = 0;
2879                 nla_nest_end(skb, vifs);
2880                 nla_nest_end(skb, af);
2881                 nlmsg_end(skb, nlh);
2882 skip_table:
2883                 t++;
2884         }
2885
2886 out:
2887         cb->args[1] = e;
2888         cb->args[0] = t;
2889
2890         return skb->len;
2891 }
2892
2893 #ifdef CONFIG_PROC_FS
2894 /* The /proc interfaces to multicast routing :
2895  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2896  */
2897
2898 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2899         __acquires(mrt_lock)
2900 {
2901         struct mr_vif_iter *iter = seq->private;
2902         struct net *net = seq_file_net(seq);
2903         struct mr_table *mrt;
2904
2905         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2906         if (!mrt)
2907                 return ERR_PTR(-ENOENT);
2908
2909         iter->mrt = mrt;
2910
2911         read_lock(&mrt_lock);
2912         return mr_vif_seq_start(seq, pos);
2913 }
2914
2915 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2916         __releases(mrt_lock)
2917 {
2918         read_unlock(&mrt_lock);
2919 }
2920
2921 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2922 {
2923         struct mr_vif_iter *iter = seq->private;
2924         struct mr_table *mrt = iter->mrt;
2925
2926         if (v == SEQ_START_TOKEN) {
2927                 seq_puts(seq,
2928                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2929         } else {
2930                 const struct vif_device *vif = v;
2931                 const char *name =  vif->dev ?
2932                                     vif->dev->name : "none";
2933
2934                 seq_printf(seq,
2935                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2936                            vif - mrt->vif_table,
2937                            name, vif->bytes_in, vif->pkt_in,
2938                            vif->bytes_out, vif->pkt_out,
2939                            vif->flags, vif->local, vif->remote);
2940         }
2941         return 0;
2942 }
2943
2944 static const struct seq_operations ipmr_vif_seq_ops = {
2945         .start = ipmr_vif_seq_start,
2946         .next  = mr_vif_seq_next,
2947         .stop  = ipmr_vif_seq_stop,
2948         .show  = ipmr_vif_seq_show,
2949 };
2950
2951 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2952 {
2953         struct net *net = seq_file_net(seq);
2954         struct mr_table *mrt;
2955
2956         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2957         if (!mrt)
2958                 return ERR_PTR(-ENOENT);
2959
2960         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2961 }
2962
2963 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2964 {
2965         int n;
2966
2967         if (v == SEQ_START_TOKEN) {
2968                 seq_puts(seq,
2969                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2970         } else {
2971                 const struct mfc_cache *mfc = v;
2972                 const struct mr_mfc_iter *it = seq->private;
2973                 const struct mr_table *mrt = it->mrt;
2974
2975                 seq_printf(seq, "%08X %08X %-3hd",
2976                            (__force u32) mfc->mfc_mcastgrp,
2977                            (__force u32) mfc->mfc_origin,
2978                            mfc->_c.mfc_parent);
2979
2980                 if (it->cache != &mrt->mfc_unres_queue) {
2981                         seq_printf(seq, " %8lu %8lu %8lu",
2982                                    mfc->_c.mfc_un.res.pkt,
2983                                    mfc->_c.mfc_un.res.bytes,
2984                                    mfc->_c.mfc_un.res.wrong_if);
2985                         for (n = mfc->_c.mfc_un.res.minvif;
2986                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2987                                 if (VIF_EXISTS(mrt, n) &&
2988                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2989                                         seq_printf(seq,
2990                                            " %2d:%-3d",
2991                                            n, mfc->_c.mfc_un.res.ttls[n]);
2992                         }
2993                 } else {
2994                         /* unresolved mfc_caches don't contain
2995                          * pkt, bytes and wrong_if values
2996                          */
2997                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2998                 }
2999                 seq_putc(seq, '\n');
3000         }
3001         return 0;
3002 }
3003
3004 static const struct seq_operations ipmr_mfc_seq_ops = {
3005         .start = ipmr_mfc_seq_start,
3006         .next  = mr_mfc_seq_next,
3007         .stop  = mr_mfc_seq_stop,
3008         .show  = ipmr_mfc_seq_show,
3009 };
3010 #endif
3011
3012 #ifdef CONFIG_IP_PIMSM_V2
3013 static const struct net_protocol pim_protocol = {
3014         .handler        =       pim_rcv,
3015         .netns_ok       =       1,
3016 };
3017 #endif
3018
3019 static unsigned int ipmr_seq_read(struct net *net)
3020 {
3021         ASSERT_RTNL();
3022
3023         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3024 }
3025
3026 static int ipmr_dump(struct net *net, struct notifier_block *nb,
3027                      struct netlink_ext_ack *extack)
3028 {
3029         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
3030                        ipmr_mr_table_iter, &mrt_lock, extack);
3031 }
3032
3033 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3034         .family         = RTNL_FAMILY_IPMR,
3035         .fib_seq_read   = ipmr_seq_read,
3036         .fib_dump       = ipmr_dump,
3037         .owner          = THIS_MODULE,
3038 };
3039
3040 static int __net_init ipmr_notifier_init(struct net *net)
3041 {
3042         struct fib_notifier_ops *ops;
3043
3044         net->ipv4.ipmr_seq = 0;
3045
3046         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3047         if (IS_ERR(ops))
3048                 return PTR_ERR(ops);
3049         net->ipv4.ipmr_notifier_ops = ops;
3050
3051         return 0;
3052 }
3053
3054 static void __net_exit ipmr_notifier_exit(struct net *net)
3055 {
3056         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3057         net->ipv4.ipmr_notifier_ops = NULL;
3058 }
3059
3060 /* Setup for IP multicast routing */
3061 static int __net_init ipmr_net_init(struct net *net)
3062 {
3063         int err;
3064
3065         err = ipmr_notifier_init(net);
3066         if (err)
3067                 goto ipmr_notifier_fail;
3068
3069         err = ipmr_rules_init(net);
3070         if (err < 0)
3071                 goto ipmr_rules_fail;
3072
3073 #ifdef CONFIG_PROC_FS
3074         err = -ENOMEM;
3075         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
3076                         sizeof(struct mr_vif_iter)))
3077                 goto proc_vif_fail;
3078         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
3079                         sizeof(struct mr_mfc_iter)))
3080                 goto proc_cache_fail;
3081 #endif
3082         return 0;
3083
3084 #ifdef CONFIG_PROC_FS
3085 proc_cache_fail:
3086         remove_proc_entry("ip_mr_vif", net->proc_net);
3087 proc_vif_fail:
3088         ipmr_rules_exit(net);
3089 #endif
3090 ipmr_rules_fail:
3091         ipmr_notifier_exit(net);
3092 ipmr_notifier_fail:
3093         return err;
3094 }
3095
3096 static void __net_exit ipmr_net_exit(struct net *net)
3097 {
3098 #ifdef CONFIG_PROC_FS
3099         remove_proc_entry("ip_mr_cache", net->proc_net);
3100         remove_proc_entry("ip_mr_vif", net->proc_net);
3101 #endif
3102         ipmr_notifier_exit(net);
3103         ipmr_rules_exit(net);
3104 }
3105
3106 static struct pernet_operations ipmr_net_ops = {
3107         .init = ipmr_net_init,
3108         .exit = ipmr_net_exit,
3109 };
3110
3111 int __init ip_mr_init(void)
3112 {
3113         int err;
3114
3115         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3116                                        sizeof(struct mfc_cache),
3117                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3118                                        NULL);
3119
3120         err = register_pernet_subsys(&ipmr_net_ops);
3121         if (err)
3122                 goto reg_pernet_fail;
3123
3124         err = register_netdevice_notifier(&ip_mr_notifier);
3125         if (err)
3126                 goto reg_notif_fail;
3127 #ifdef CONFIG_IP_PIMSM_V2
3128         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3129                 pr_err("%s: can't add PIM protocol\n", __func__);
3130                 err = -EAGAIN;
3131                 goto add_proto_fail;
3132         }
3133 #endif
3134         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3135                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3136         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3137                       ipmr_rtm_route, NULL, 0);
3138         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3139                       ipmr_rtm_route, NULL, 0);
3140
3141         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3142                       NULL, ipmr_rtm_dumplink, 0);
3143         return 0;
3144
3145 #ifdef CONFIG_IP_PIMSM_V2
3146 add_proto_fail:
3147         unregister_netdevice_notifier(&ip_mr_notifier);
3148 #endif
3149 reg_notif_fail:
3150         unregister_pernet_subsys(&ipmr_net_ops);
3151 reg_pernet_fail:
3152         kmem_cache_destroy(mrt_cachep);
3153         return err;
3154 }