git.rot13.org Git - linux/blob - net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <linux/uaccess.h>
  30 #include <linux/types.h>
  31 #include <linux/cache.h>
  32 #include <linux/capability.h>
  33 #include <linux/errno.h>
  34 #include <linux/mm.h>
  35 #include <linux/kernel.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/stat.h>
  38 #include <linux/socket.h>
  39 #include <linux/in.h>
  40 #include <linux/inet.h>
  41 #include <linux/netdevice.h>
  42 #include <linux/inetdevice.h>
  43 #include <linux/igmp.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/mroute.h>
  47 #include <linux/init.h>
  48 #include <linux/if_ether.h>
  49 #include <linux/slab.h>
  50 #include <net/net_namespace.h>
  51 #include <net/ip.h>
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54 #include <net/route.h>
  55 #include <net/icmp.h>
  56 #include <net/udp.h>
  57 #include <net/raw.h>
  58 #include <linux/notifier.h>
  59 #include <linux/if_arp.h>
  60 #include <linux/netfilter_ipv4.h>
  61 #include <linux/compat.h>
  62 #include <linux/export.h>
  63 #include <linux/rhashtable.h>
  64 #include <net/ip_tunnels.h>
  65 #include <net/checksum.h>
  66 #include <net/netlink.h>
  67 #include <net/fib_rules.h>
  68 #include <linux/netconf.h>
  69 #include <net/nexthop.h>
  70
  71 #include <linux/nospec.h>
  72
  73 struct ipmr_rule {
  74         struct fib_rule         common;
  75 };
  76
  77 struct ipmr_result {
  78         struct mr_table         *mrt;
  79 };
  80
  81 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  82  * Note that the changes are semaphored via rtnl_lock.
  83  */
  84
  85 static DEFINE_RWLOCK(mrt_lock);
  86
  87 /* Multicast router control variables */
  88
  89 /* Special spinlock for queue of unresolved entries */
  90 static DEFINE_SPINLOCK(mfc_unres_lock);
  91
  92 /* We return to original Alan's scheme. Hash table of resolved
  93  * entries is changed only in process context and protected
  94  * with weak lock mrt_lock. Queue of unresolved entries is protected
  95  * with strong spinlock mfc_unres_lock.
  96  *
  97  * In this case data path is free of exclusive locks at all.
  98  */
  99
 100 static struct kmem_cache *mrt_cachep __ro_after_init;
 101
 102 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 103 static void ipmr_free_table(struct mr_table *mrt);
 104
 105 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 106                           struct net_device *dev, struct sk_buff *skb,
 107                           struct mfc_cache *cache, int local);
 108 static int ipmr_cache_report(struct mr_table *mrt,
 109                              struct sk_buff *pkt, vifi_t vifi, int assert);
 110 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 111                                  int cmd);
 112 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt);
 113 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 114 static void ipmr_expire_process(struct timer_list *t);
 115
 116 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 117 #define ipmr_for_each_table(mrt, net) \
 118         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 119
 120 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 121                                            struct mr_table *mrt)
 122 {
 123         struct mr_table *ret;
 124
 125         if (!mrt)
 126                 ret = list_entry_rcu(net->ipv4.mr_tables.next,
 127                                      struct mr_table, list);
 128         else
 129                 ret = list_entry_rcu(mrt->list.next,
 130                                      struct mr_table, list);
 131
 132         if (&ret->list == &net->ipv4.mr_tables)
 133                 return NULL;
 134         return ret;
 135 }
 136
 137 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 138 {
 139         struct mr_table *mrt;
 140
 141         ipmr_for_each_table(mrt, net) {
 142                 if (mrt->id == id)
 143                         return mrt;
 144         }
 145         return NULL;
 146 }
 147
 148 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 149                            struct mr_table **mrt)
 150 {
 151         int err;
 152         struct ipmr_result res;
 153         struct fib_lookup_arg arg = {
 154                 .result = &res,
 155                 .flags = FIB_LOOKUP_NOREF,
 156         };
 157
 158         /* update flow if oif or iif point to device enslaved to l3mdev */
 159         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 160
 161         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 162                                flowi4_to_flowi(flp4), 0, &arg);
 163         if (err < 0)
 164                 return err;
 165         *mrt = res.mrt;
 166         return 0;
 167 }
 168
 169 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 170                             int flags, struct fib_lookup_arg *arg)
 171 {
 172         struct ipmr_result *res = arg->result;
 173         struct mr_table *mrt;
 174
 175         switch (rule->action) {
 176         case FR_ACT_TO_TBL:
 177                 break;
 178         case FR_ACT_UNREACHABLE:
 179                 return -ENETUNREACH;
 180         case FR_ACT_PROHIBIT:
 181                 return -EACCES;
 182         case FR_ACT_BLACKHOLE:
 183         default:
 184                 return -EINVAL;
 185         }
 186
 187         arg->table = fib_rule_get_table(rule, arg);
 188
 189         mrt = ipmr_get_table(rule->fr_net, arg->table);
 190         if (!mrt)
 191                 return -EAGAIN;
 192         res->mrt = mrt;
 193         return 0;
 194 }
 195
 196 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 197 {
 198         return 1;
 199 }
 200
 201 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 202         FRA_GENERIC_POLICY,
 203 };
 204
 205 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 206                                struct fib_rule_hdr *frh, struct nlattr **tb,
 207                                struct netlink_ext_ack *extack)
 208 {
 209         return 0;
 210 }
 211
 212 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 213                              struct nlattr **tb)
 214 {
 215         return 1;
 216 }
 217
 218 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 219                           struct fib_rule_hdr *frh)
 220 {
 221         frh->dst_len = 0;
 222         frh->src_len = 0;
 223         frh->tos     = 0;
 224         return 0;
 225 }
 226
 227 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 228         .family         = RTNL_FAMILY_IPMR,
 229         .rule_size      = sizeof(struct ipmr_rule),
 230         .addr_size      = sizeof(u32),
 231         .action         = ipmr_rule_action,
 232         .match          = ipmr_rule_match,
 233         .configure      = ipmr_rule_configure,
 234         .compare        = ipmr_rule_compare,
 235         .fill           = ipmr_rule_fill,
 236         .nlgroup        = RTNLGRP_IPV4_RULE,
 237         .policy         = ipmr_rule_policy,
 238         .owner          = THIS_MODULE,
 239 };
 240
 241 static int __net_init ipmr_rules_init(struct net *net)
 242 {
 243         struct fib_rules_ops *ops;
 244         struct mr_table *mrt;
 245         int err;
 246
 247         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 248         if (IS_ERR(ops))
 249                 return PTR_ERR(ops);
 250
 251         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 252
 253         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 254         if (IS_ERR(mrt)) {
 255                 err = PTR_ERR(mrt);
 256                 goto err1;
 257         }
 258
 259         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 260         if (err < 0)
 261                 goto err2;
 262
 263         net->ipv4.mr_rules_ops = ops;
 264         return 0;
 265
 266 err2:
 267         ipmr_free_table(mrt);
 268 err1:
 269         fib_rules_unregister(ops);
 270         return err;
 271 }
 272
 273 static void __net_exit ipmr_rules_exit(struct net *net)
 274 {
 275         struct mr_table *mrt, *next;
 276
 277         rtnl_lock();
 278         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 279                 list_del(&mrt->list);
 280                 ipmr_free_table(mrt);
 281         }
 282         fib_rules_unregister(net->ipv4.mr_rules_ops);
 283         rtnl_unlock();
 284 }
 285
 286 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 287 {
 288         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR);
 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 {
 346         return 0;
 347 }
 348
 349 static unsigned int ipmr_rules_seq_read(struct net *net)
 350 {
 351         return 0;
 352 }
 353
 354 bool ipmr_rule_default(const struct fib_rule *rule)
 355 {
 356         return true;
 357 }
 358 EXPORT_SYMBOL(ipmr_rule_default);
 359 #endif
 360
 361 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 362                                 const void *ptr)
 363 {
 364         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 365         struct mfc_cache *c = (struct mfc_cache *)ptr;
 366
 367         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 368                cmparg->mfc_origin != c->mfc_origin;
 369 }
 370
 371 static const struct rhashtable_params ipmr_rht_params = {
 372         .head_offset = offsetof(struct mr_mfc, mnode),
 373         .key_offset = offsetof(struct mfc_cache, cmparg),
 374         .key_len = sizeof(struct mfc_cache_cmp_arg),
 375         .nelem_hint = 3,
 376         .locks_mul = 1,
 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, true);
 419         rhltable_destroy(&mrt->mfc_hash);
 420         kfree(mrt);
 421 }
 422
 423 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 424
 425 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 426 {
 427         struct net *net = dev_net(dev);
 428
 429         dev_close(dev);
 430
 431         dev = __dev_get_by_name(net, "tunl0");
 432         if (dev) {
 433                 const struct net_device_ops *ops = dev->netdev_ops;
 434                 struct ifreq ifr;
 435                 struct ip_tunnel_parm p;
 436
 437                 memset(&p, 0, sizeof(p));
 438                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 439                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 440                 p.iph.version = 4;
 441                 p.iph.ihl = 5;
 442                 p.iph.protocol = IPPROTO_IPIP;
 443                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 444                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 445
 446                 if (ops->ndo_do_ioctl) {
 447                         mm_segment_t oldfs = get_fs();
 448
 449                         set_fs(KERNEL_DS);
 450                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 451                         set_fs(oldfs);
 452                 }
 453         }
 454 }
 455
 456 /* Initialize ipmr pimreg/tunnel in_device */
 457 static bool ipmr_init_vif_indev(const struct net_device *dev)
 458 {
 459         struct in_device *in_dev;
 460
 461         ASSERT_RTNL();
 462
 463         in_dev = __in_dev_get_rtnl(dev);
 464         if (!in_dev)
 465                 return false;
 466         ipv4_devconf_setall(in_dev);
 467         neigh_parms_data_state_setall(in_dev->arp_parms);
 468         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 469
 470         return true;
 471 }
 472
 473 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 474 {
 475         struct net_device  *dev;
 476
 477         dev = __dev_get_by_name(net, "tunl0");
 478
 479         if (dev) {
 480                 const struct net_device_ops *ops = dev->netdev_ops;
 481                 int err;
 482                 struct ifreq ifr;
 483                 struct ip_tunnel_parm p;
 484
 485                 memset(&p, 0, sizeof(p));
 486                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 487                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 488                 p.iph.version = 4;
 489                 p.iph.ihl = 5;
 490                 p.iph.protocol = IPPROTO_IPIP;
 491                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 492                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 493
 494                 if (ops->ndo_do_ioctl) {
 495                         mm_segment_t oldfs = get_fs();
 496
 497                         set_fs(KERNEL_DS);
 498                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 499                         set_fs(oldfs);
 500                 } else {
 501                         err = -EOPNOTSUPP;
 502                 }
 503                 dev = NULL;
 504
 505                 if (err == 0 &&
 506                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 507                         dev->flags |= IFF_MULTICAST;
 508                         if (!ipmr_init_vif_indev(dev))
 509                                 goto failure;
 510                         if (dev_open(dev, NULL))
 511                                 goto failure;
 512                         dev_hold(dev);
 513                 }
 514         }
 515         return dev;
 516
 517 failure:
 518         unregister_netdevice(dev);
 519         return NULL;
 520 }
 521
 522 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 523 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 524 {
 525         struct net *net = dev_net(dev);
 526         struct mr_table *mrt;
 527         struct flowi4 fl4 = {
 528                 .flowi4_oif     = dev->ifindex,
 529                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 530                 .flowi4_mark    = skb->mark,
 531         };
 532         int err;
 533
 534         err = ipmr_fib_lookup(net, &fl4, &mrt);
 535         if (err < 0) {
 536                 kfree_skb(skb);
 537                 return err;
 538         }
 539
 540         read_lock(&mrt_lock);
 541         dev->stats.tx_bytes += skb->len;
 542         dev->stats.tx_packets++;
 543         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 544         read_unlock(&mrt_lock);
 545         kfree_skb(skb);
 546         return NETDEV_TX_OK;
 547 }
 548
 549 static int reg_vif_get_iflink(const struct net_device *dev)
 550 {
 551         return 0;
 552 }
 553
 554 static const struct net_device_ops reg_vif_netdev_ops = {
 555         .ndo_start_xmit = reg_vif_xmit,
 556         .ndo_get_iflink = reg_vif_get_iflink,
 557 };
 558
 559 static void reg_vif_setup(struct net_device *dev)
 560 {
 561         dev->type               = ARPHRD_PIMREG;
 562         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 563         dev->flags              = IFF_NOARP;
 564         dev->netdev_ops         = &reg_vif_netdev_ops;
 565         dev->needs_free_netdev  = true;
 566         dev->features           |= NETIF_F_NETNS_LOCAL;
 567 }
 568
 569 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 570 {
 571         struct net_device *dev;
 572         char name[IFNAMSIZ];
 573
 574         if (mrt->id == RT_TABLE_DEFAULT)
 575                 sprintf(name, "pimreg");
 576         else
 577                 sprintf(name, "pimreg%u", mrt->id);
 578
 579         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 580
 581         if (!dev)
 582                 return NULL;
 583
 584         dev_net_set(dev, net);
 585
 586         if (register_netdevice(dev)) {
 587                 free_netdev(dev);
 588                 return NULL;
 589         }
 590
 591         if (!ipmr_init_vif_indev(dev))
 592                 goto failure;
 593         if (dev_open(dev, NULL))
 594                 goto failure;
 595
 596         dev_hold(dev);
 597
 598         return dev;
 599
 600 failure:
 601         unregister_netdevice(dev);
 602         return NULL;
 603 }
 604
 605 /* called with rcu_read_lock() */
 606 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 607                      unsigned int pimlen)
 608 {
 609         struct net_device *reg_dev = NULL;
 610         struct iphdr *encap;
 611
 612         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 613         /* Check that:
 614          * a. packet is really sent to a multicast group
 615          * b. packet is not a NULL-REGISTER
 616          * c. packet is not truncated
 617          */
 618         if (!ipv4_is_multicast(encap->daddr) ||
 619             encap->tot_len == 0 ||
 620             ntohs(encap->tot_len) + pimlen > skb->len)
 621                 return 1;
 622
 623         read_lock(&mrt_lock);
 624         if (mrt->mroute_reg_vif_num >= 0)
 625                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 626         read_unlock(&mrt_lock);
 627
 628         if (!reg_dev)
 629                 return 1;
 630
 631         skb->mac_header = skb->network_header;
 632         skb_pull(skb, (u8 *)encap - skb->data);
 633         skb_reset_network_header(skb);
 634         skb->protocol = htons(ETH_P_IP);
 635         skb->ip_summed = CHECKSUM_NONE;
 636
 637         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 638
 639         netif_rx(skb);
 640
 641         return NET_RX_SUCCESS;
 642 }
 643 #else
 644 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 645 {
 646         return NULL;
 647 }
 648 #endif
 649
 650 static int call_ipmr_vif_entry_notifiers(struct net *net,
 651                                          enum fib_event_type event_type,
 652                                          struct vif_device *vif,
 653                                          vifi_t vif_index, u32 tb_id)
 654 {
 655         return mr_call_vif_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 656                                      vif, vif_index, tb_id,
 657                                      &net->ipv4.ipmr_seq);
 658 }
 659
 660 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 661                                          enum fib_event_type event_type,
 662                                          struct mfc_cache *mfc, u32 tb_id)
 663 {
 664         return mr_call_mfc_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 665                                      &mfc->_c, tb_id, &net->ipv4.ipmr_seq);
 666 }
 667
 668 /**
 669  *      vif_delete - Delete a VIF entry
 670  *      @notify: Set to 1, if the caller is a notifier_call
 671  */
 672 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 673                       struct list_head *head)
 674 {
 675         struct net *net = read_pnet(&mrt->net);
 676         struct vif_device *v;
 677         struct net_device *dev;
 678         struct in_device *in_dev;
 679
 680         if (vifi < 0 || vifi >= mrt->maxvif)
 681                 return -EADDRNOTAVAIL;
 682
 683         v = &mrt->vif_table[vifi];
 684
 685         if (VIF_EXISTS(mrt, vifi))
 686                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 687                                               mrt->id);
 688
 689         write_lock_bh(&mrt_lock);
 690         dev = v->dev;
 691         v->dev = NULL;
 692
 693         if (!dev) {
 694                 write_unlock_bh(&mrt_lock);
 695                 return -EADDRNOTAVAIL;
 696         }
 697
 698         if (vifi == mrt->mroute_reg_vif_num)
 699                 mrt->mroute_reg_vif_num = -1;
 700
 701         if (vifi + 1 == mrt->maxvif) {
 702                 int tmp;
 703
 704                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 705                         if (VIF_EXISTS(mrt, tmp))
 706                                 break;
 707                 }
 708                 mrt->maxvif = tmp+1;
 709         }
 710
 711         write_unlock_bh(&mrt_lock);
 712
 713         dev_set_allmulti(dev, -1);
 714
 715         in_dev = __in_dev_get_rtnl(dev);
 716         if (in_dev) {
 717                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 718                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 719                                             NETCONFA_MC_FORWARDING,
 720                                             dev->ifindex, &in_dev->cnf);
 721                 ip_rt_multicast_event(in_dev);
 722         }
 723
 724         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 725                 unregister_netdevice_queue(dev, head);
 726
 727         dev_put(dev);
 728         return 0;
 729 }
 730
 731 static void ipmr_cache_free_rcu(struct rcu_head *head)
 732 {
 733         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 734
 735         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 736 }
 737
 738 static void ipmr_cache_free(struct mfc_cache *c)
 739 {
 740         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 741 }
 742
 743 /* Destroy an unresolved cache entry, killing queued skbs
 744  * and reporting error to netlink readers.
 745  */
 746 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 747 {
 748         struct net *net = read_pnet(&mrt->net);
 749         struct sk_buff *skb;
 750         struct nlmsgerr *e;
 751
 752         atomic_dec(&mrt->cache_resolve_queue_len);
 753
 754         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 755                 if (ip_hdr(skb)->version == 0) {
 756                         struct nlmsghdr *nlh = skb_pull(skb,
 757                                                         sizeof(struct iphdr));
 758                         nlh->nlmsg_type = NLMSG_ERROR;
 759                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 760                         skb_trim(skb, nlh->nlmsg_len);
 761                         e = nlmsg_data(nlh);
 762                         e->error = -ETIMEDOUT;
 763                         memset(&e->msg, 0, sizeof(e->msg));
 764
 765                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 766                 } else {
 767                         kfree_skb(skb);
 768                 }
 769         }
 770
 771         ipmr_cache_free(c);
 772 }
 773
 774 /* Timer process for the unresolved queue. */
 775 static void ipmr_expire_process(struct timer_list *t)
 776 {
 777         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 778         struct mr_mfc *c, *next;
 779         unsigned long expires;
 780         unsigned long now;
 781
 782         if (!spin_trylock(&mfc_unres_lock)) {
 783                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 784                 return;
 785         }
 786
 787         if (list_empty(&mrt->mfc_unres_queue))
 788                 goto out;
 789
 790         now = jiffies;
 791         expires = 10*HZ;
 792
 793         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 794                 if (time_after(c->mfc_un.unres.expires, now)) {
 795                         unsigned long interval = c->mfc_un.unres.expires - now;
 796                         if (interval < expires)
 797                                 expires = interval;
 798                         continue;
 799                 }
 800
 801                 list_del(&c->list);
 802                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 803                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 804         }
 805
 806         if (!list_empty(&mrt->mfc_unres_queue))
 807                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 808
 809 out:
 810         spin_unlock(&mfc_unres_lock);
 811 }
 812
 813 /* Fill oifs list. It is called under write locked mrt_lock. */
 814 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 815                                    unsigned char *ttls)
 816 {
 817         int vifi;
 818
 819         cache->mfc_un.res.minvif = MAXVIFS;
 820         cache->mfc_un.res.maxvif = 0;
 821         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 822
 823         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 824                 if (VIF_EXISTS(mrt, vifi) &&
 825                     ttls[vifi] && ttls[vifi] < 255) {
 826                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 827                         if (cache->mfc_un.res.minvif > vifi)
 828                                 cache->mfc_un.res.minvif = vifi;
 829                         if (cache->mfc_un.res.maxvif <= vifi)
 830                                 cache->mfc_un.res.maxvif = vifi + 1;
 831                 }
 832         }
 833         cache->mfc_un.res.lastuse = jiffies;
 834 }
 835
 836 static int vif_add(struct net *net, struct mr_table *mrt,
 837                    struct vifctl *vifc, int mrtsock)
 838 {
 839         struct netdev_phys_item_id ppid = { };
 840         int vifi = vifc->vifc_vifi;
 841         struct vif_device *v = &mrt->vif_table[vifi];
 842         struct net_device *dev;
 843         struct in_device *in_dev;
 844         int err;
 845
 846         /* Is vif busy ? */
 847         if (VIF_EXISTS(mrt, vifi))
 848                 return -EADDRINUSE;
 849
 850         switch (vifc->vifc_flags) {
 851         case VIFF_REGISTER:
 852                 if (!ipmr_pimsm_enabled())
 853                         return -EINVAL;
 854                 /* Special Purpose VIF in PIM
 855                  * All the packets will be sent to the daemon
 856                  */
 857                 if (mrt->mroute_reg_vif_num >= 0)
 858                         return -EADDRINUSE;
 859                 dev = ipmr_reg_vif(net, mrt);
 860                 if (!dev)
 861                         return -ENOBUFS;
 862                 err = dev_set_allmulti(dev, 1);
 863                 if (err) {
 864                         unregister_netdevice(dev);
 865                         dev_put(dev);
 866                         return err;
 867                 }
 868                 break;
 869         case VIFF_TUNNEL:
 870                 dev = ipmr_new_tunnel(net, vifc);
 871                 if (!dev)
 872                         return -ENOBUFS;
 873                 err = dev_set_allmulti(dev, 1);
 874                 if (err) {
 875                         ipmr_del_tunnel(dev, vifc);
 876                         dev_put(dev);
 877                         return err;
 878                 }
 879                 break;
 880         case VIFF_USE_IFINDEX:
 881         case 0:
 882                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 883                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 884                         if (dev && !__in_dev_get_rtnl(dev)) {
 885                                 dev_put(dev);
 886                                 return -EADDRNOTAVAIL;
 887                         }
 888                 } else {
 889                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 890                 }
 891                 if (!dev)
 892                         return -EADDRNOTAVAIL;
 893                 err = dev_set_allmulti(dev, 1);
 894                 if (err) {
 895                         dev_put(dev);
 896                         return err;
 897                 }
 898                 break;
 899         default:
 900                 return -EINVAL;
 901         }
 902
 903         in_dev = __in_dev_get_rtnl(dev);
 904         if (!in_dev) {
 905                 dev_put(dev);
 906                 return -EADDRNOTAVAIL;
 907         }
 908         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 909         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 910                                     dev->ifindex, &in_dev->cnf);
 911         ip_rt_multicast_event(in_dev);
 912
 913         /* Fill in the VIF structures */
 914         vif_device_init(v, dev, vifc->vifc_rate_limit,
 915                         vifc->vifc_threshold,
 916                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 917                         (VIFF_TUNNEL | VIFF_REGISTER));
 918
 919         err = dev_get_port_parent_id(dev, &ppid, true);
 920         if (err == 0) {
 921                 memcpy(v->dev_parent_id.id, ppid.id, ppid.id_len);
 922                 v->dev_parent_id.id_len = ppid.id_len;
 923         } else {
 924                 v->dev_parent_id.id_len = 0;
 925         }
 926
 927         v->local = vifc->vifc_lcl_addr.s_addr;
 928         v->remote = vifc->vifc_rmt_addr.s_addr;
 929
 930         /* And finish update writing critical data */
 931         write_lock_bh(&mrt_lock);
 932         v->dev = dev;
 933         if (v->flags & VIFF_REGISTER)
 934                 mrt->mroute_reg_vif_num = vifi;
 935         if (vifi+1 > mrt->maxvif)
 936                 mrt->maxvif = vifi+1;
 937         write_unlock_bh(&mrt_lock);
 938         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 939         return 0;
 940 }
 941
 942 /* called with rcu_read_lock() */
 943 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 944                                          __be32 origin,
 945                                          __be32 mcastgrp)
 946 {
 947         struct mfc_cache_cmp_arg arg = {
 948                         .mfc_mcastgrp = mcastgrp,
 949                         .mfc_origin = origin
 950         };
 951
 952         return mr_mfc_find(mrt, &arg);
 953 }
 954
 955 /* Look for a (*,G) entry */
 956 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 957                                              __be32 mcastgrp, int vifi)
 958 {
 959         struct mfc_cache_cmp_arg arg = {
 960                         .mfc_mcastgrp = mcastgrp,
 961                         .mfc_origin = htonl(INADDR_ANY)
 962         };
 963
 964         if (mcastgrp == htonl(INADDR_ANY))
 965                 return mr_mfc_find_any_parent(mrt, vifi);
 966         return mr_mfc_find_any(mrt, vifi, &arg);
 967 }
 968
 969 /* Look for a (S,G,iif) entry if parent != -1 */
 970 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 971                                                 __be32 origin, __be32 mcastgrp,
 972                                                 int parent)
 973 {
 974         struct mfc_cache_cmp_arg arg = {
 975                         .mfc_mcastgrp = mcastgrp,
 976                         .mfc_origin = origin,
 977         };
 978
 979         return mr_mfc_find_parent(mrt, &arg, parent);
 980 }
 981
 982 /* Allocate a multicast cache entry */
 983 static struct mfc_cache *ipmr_cache_alloc(void)
 984 {
 985         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 986
 987         if (c) {
 988                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 989                 c->_c.mfc_un.res.minvif = MAXVIFS;
 990                 c->_c.free = ipmr_cache_free_rcu;
 991                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 992         }
 993         return c;
 994 }
 995
 996 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 997 {
 998         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 999
1000         if (c) {
1001                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
1002                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
1003         }
1004         return c;
1005 }
1006
1007 /* A cache entry has gone into a resolved state from queued */
1008 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
1009                                struct mfc_cache *uc, struct mfc_cache *c)
1010 {
1011         struct sk_buff *skb;
1012         struct nlmsgerr *e;
1013
1014         /* Play the pending entries through our router */
1015         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
1016                 if (ip_hdr(skb)->version == 0) {
1017                         struct nlmsghdr *nlh = skb_pull(skb,
1018                                                         sizeof(struct iphdr));
1019
1020                         if (mr_fill_mroute(mrt, skb, &c->_c,
1021                                            nlmsg_data(nlh)) > 0) {
1022                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
1023                                                  (u8 *)nlh;
1024                         } else {
1025                                 nlh->nlmsg_type = NLMSG_ERROR;
1026                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
1027                                 skb_trim(skb, nlh->nlmsg_len);
1028                                 e = nlmsg_data(nlh);
1029                                 e->error = -EMSGSIZE;
1030                                 memset(&e->msg, 0, sizeof(e->msg));
1031                         }
1032
1033                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1034                 } else {
1035                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1036                 }
1037         }
1038 }
1039
1040 /* Bounce a cache query up to mrouted and netlink.
1041  *
1042  * Called under mrt_lock.
1043  */
1044 static int ipmr_cache_report(struct mr_table *mrt,
1045                              struct sk_buff *pkt, vifi_t vifi, int assert)
1046 {
1047         const int ihl = ip_hdrlen(pkt);
1048         struct sock *mroute_sk;
1049         struct igmphdr *igmp;
1050         struct igmpmsg *msg;
1051         struct sk_buff *skb;
1052         int ret;
1053
1054         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1055                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1056         else
1057                 skb = alloc_skb(128, GFP_ATOMIC);
1058
1059         if (!skb)
1060                 return -ENOBUFS;
1061
1062         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1063                 /* Ugly, but we have no choice with this interface.
1064                  * Duplicate old header, fix ihl, length etc.
1065                  * And all this only to mangle msg->im_msgtype and
1066                  * to set msg->im_mbz to "mbz" :-)
1067                  */
1068                 skb_push(skb, sizeof(struct iphdr));
1069                 skb_reset_network_header(skb);
1070                 skb_reset_transport_header(skb);
1071                 msg = (struct igmpmsg *)skb_network_header(skb);
1072                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1073                 msg->im_msgtype = assert;
1074                 msg->im_mbz = 0;
1075                 if (assert == IGMPMSG_WRVIFWHOLE)
1076                         msg->im_vif = vifi;
1077                 else
1078                         msg->im_vif = mrt->mroute_reg_vif_num;
1079                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1080                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1081                                              sizeof(struct iphdr));
1082         } else {
1083                 /* Copy the IP header */
1084                 skb_set_network_header(skb, skb->len);
1085                 skb_put(skb, ihl);
1086                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1087                 /* Flag to the kernel this is a route add */
1088                 ip_hdr(skb)->protocol = 0;
1089                 msg = (struct igmpmsg *)skb_network_header(skb);
1090                 msg->im_vif = vifi;
1091                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1092                 /* Add our header */
1093                 igmp = skb_put(skb, sizeof(struct igmphdr));
1094                 igmp->type = assert;
1095                 msg->im_msgtype = assert;
1096                 igmp->code = 0;
1097                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1098                 skb->transport_header = skb->network_header;
1099         }
1100
1101         rcu_read_lock();
1102         mroute_sk = rcu_dereference(mrt->mroute_sk);
1103         if (!mroute_sk) {
1104                 rcu_read_unlock();
1105                 kfree_skb(skb);
1106                 return -EINVAL;
1107         }
1108
1109         igmpmsg_netlink_event(mrt, skb);
1110
1111         /* Deliver to mrouted */
1112         ret = sock_queue_rcv_skb(mroute_sk, skb);
1113         rcu_read_unlock();
1114         if (ret < 0) {
1115                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1116                 kfree_skb(skb);
1117         }
1118
1119         return ret;
1120 }
1121
1122 /* Queue a packet for resolution. It gets locked cache entry! */
1123 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1124                                  struct sk_buff *skb, struct net_device *dev)
1125 {
1126         const struct iphdr *iph = ip_hdr(skb);
1127         struct mfc_cache *c;
1128         bool found = false;
1129         int err;
1130
1131         spin_lock_bh(&mfc_unres_lock);
1132         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1133                 if (c->mfc_mcastgrp == iph->daddr &&
1134                     c->mfc_origin == iph->saddr) {
1135                         found = true;
1136                         break;
1137                 }
1138         }
1139
1140         if (!found) {
1141                 /* Create a new entry if allowable */
1142                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1143                     (c = ipmr_cache_alloc_unres()) == NULL) {
1144                         spin_unlock_bh(&mfc_unres_lock);
1145
1146                         kfree_skb(skb);
1147                         return -ENOBUFS;
1148                 }
1149
1150                 /* Fill in the new cache entry */
1151                 c->_c.mfc_parent = -1;
1152                 c->mfc_origin   = iph->saddr;
1153                 c->mfc_mcastgrp = iph->daddr;
1154
1155                 /* Reflect first query at mrouted. */
1156                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1157
1158                 if (err < 0) {
1159                         /* If the report failed throw the cache entry
1160                            out - Brad Parker
1161                          */
1162                         spin_unlock_bh(&mfc_unres_lock);
1163
1164                         ipmr_cache_free(c);
1165                         kfree_skb(skb);
1166                         return err;
1167                 }
1168
1169                 atomic_inc(&mrt->cache_resolve_queue_len);
1170                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1171                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1172
1173                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1174                         mod_timer(&mrt->ipmr_expire_timer,
1175                                   c->_c.mfc_un.unres.expires);
1176         }
1177
1178         /* See if we can append the packet */
1179         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1180                 kfree_skb(skb);
1181                 err = -ENOBUFS;
1182         } else {
1183                 if (dev) {
1184                         skb->dev = dev;
1185                         skb->skb_iif = dev->ifindex;
1186                 }
1187                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1188                 err = 0;
1189         }
1190
1191         spin_unlock_bh(&mfc_unres_lock);
1192         return err;
1193 }
1194
1195 /* MFC cache manipulation by user space mroute daemon */
1196
1197 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1198 {
1199         struct net *net = read_pnet(&mrt->net);
1200         struct mfc_cache *c;
1201
1202         /* The entries are added/deleted only under RTNL */
1203         rcu_read_lock();
1204         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1205                                    mfc->mfcc_mcastgrp.s_addr, parent);
1206         rcu_read_unlock();
1207         if (!c)
1208                 return -ENOENT;
1209         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1210         list_del_rcu(&c->_c.list);
1211         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1212         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1213         mr_cache_put(&c->_c);
1214
1215         return 0;
1216 }
1217
1218 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1219                         struct mfcctl *mfc, int mrtsock, int parent)
1220 {
1221         struct mfc_cache *uc, *c;
1222         struct mr_mfc *_uc;
1223         bool found;
1224         int ret;
1225
1226         if (mfc->mfcc_parent >= MAXVIFS)
1227                 return -ENFILE;
1228
1229         /* The entries are added/deleted only under RTNL */
1230         rcu_read_lock();
1231         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1232                                    mfc->mfcc_mcastgrp.s_addr, parent);
1233         rcu_read_unlock();
1234         if (c) {
1235                 write_lock_bh(&mrt_lock);
1236                 c->_c.mfc_parent = mfc->mfcc_parent;
1237                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1238                 if (!mrtsock)
1239                         c->_c.mfc_flags |= MFC_STATIC;
1240                 write_unlock_bh(&mrt_lock);
1241                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1242                                               mrt->id);
1243                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1244                 return 0;
1245         }
1246
1247         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1248             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1249                 return -EINVAL;
1250
1251         c = ipmr_cache_alloc();
1252         if (!c)
1253                 return -ENOMEM;
1254
1255         c->mfc_origin = mfc->mfcc_origin.s_addr;
1256         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1257         c->_c.mfc_parent = mfc->mfcc_parent;
1258         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1259         if (!mrtsock)
1260                 c->_c.mfc_flags |= MFC_STATIC;
1261
1262         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1263                                   ipmr_rht_params);
1264         if (ret) {
1265                 pr_err("ipmr: rhtable insert error %d\n", ret);
1266                 ipmr_cache_free(c);
1267                 return ret;
1268         }
1269         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1270         /* Check to see if we resolved a queued list. If so we
1271          * need to send on the frames and tidy up.
1272          */
1273         found = false;
1274         spin_lock_bh(&mfc_unres_lock);
1275         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1276                 uc = (struct mfc_cache *)_uc;
1277                 if (uc->mfc_origin == c->mfc_origin &&
1278                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1279                         list_del(&_uc->list);
1280                         atomic_dec(&mrt->cache_resolve_queue_len);
1281                         found = true;
1282                         break;
1283                 }
1284         }
1285         if (list_empty(&mrt->mfc_unres_queue))
1286                 del_timer(&mrt->ipmr_expire_timer);
1287         spin_unlock_bh(&mfc_unres_lock);
1288
1289         if (found) {
1290                 ipmr_cache_resolve(net, mrt, uc, c);
1291                 ipmr_cache_free(uc);
1292         }
1293         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1294         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1295         return 0;
1296 }
1297
1298 /* Close the multicast socket, and clear the vif tables etc */
1299 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1300 {
1301         struct net *net = read_pnet(&mrt->net);
1302         struct mr_mfc *c, *tmp;
1303         struct mfc_cache *cache;
1304         LIST_HEAD(list);
1305         int i;
1306
1307         /* Shut down all active vif entries */
1308         for (i = 0; i < mrt->maxvif; i++) {
1309                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1310                         continue;
1311                 vif_delete(mrt, i, 0, &list);
1312         }
1313         unregister_netdevice_many(&list);
1314
1315         /* Wipe the cache */
1316         list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1317                 if (!all && (c->mfc_flags & MFC_STATIC))
1318                         continue;
1319                 rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1320                 list_del_rcu(&c->list);
1321                 cache = (struct mfc_cache *)c;
1322                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1323                                               mrt->id);
1324                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1325                 mr_cache_put(c);
1326         }
1327
1328         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1329                 spin_lock_bh(&mfc_unres_lock);
1330                 list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1331                         list_del(&c->list);
1332                         cache = (struct mfc_cache *)c;
1333                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1334                         ipmr_destroy_unres(mrt, cache);
1335                 }
1336                 spin_unlock_bh(&mfc_unres_lock);
1337         }
1338 }
1339
1340 /* called from ip_ra_control(), before an RCU grace period,
1341  * we dont need to call synchronize_rcu() here
1342  */
1343 static void mrtsock_destruct(struct sock *sk)
1344 {
1345         struct net *net = sock_net(sk);
1346         struct mr_table *mrt;
1347
1348         rtnl_lock();
1349         ipmr_for_each_table(mrt, net) {
1350                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1351                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1352                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1353                                                     NETCONFA_MC_FORWARDING,
1354                                                     NETCONFA_IFINDEX_ALL,
1355                                                     net->ipv4.devconf_all);
1356                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1357                         mroute_clean_tables(mrt, false);
1358                 }
1359         }
1360         rtnl_unlock();
1361 }
1362
1363 /* Socket options and virtual interface manipulation. The whole
1364  * virtual interface system is a complete heap, but unfortunately
1365  * that's how BSD mrouted happens to think. Maybe one day with a proper
1366  * MOSPF/PIM router set up we can clean this up.
1367  */
1368
1369 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1370                          unsigned int optlen)
1371 {
1372         struct net *net = sock_net(sk);
1373         int val, ret = 0, parent = 0;
1374         struct mr_table *mrt;
1375         struct vifctl vif;
1376         struct mfcctl mfc;
1377         bool do_wrvifwhole;
1378         u32 uval;
1379
1380         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1381         rtnl_lock();
1382         if (sk->sk_type != SOCK_RAW ||
1383             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1384                 ret = -EOPNOTSUPP;
1385                 goto out_unlock;
1386         }
1387
1388         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1389         if (!mrt) {
1390                 ret = -ENOENT;
1391                 goto out_unlock;
1392         }
1393         if (optname != MRT_INIT) {
1394                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1395                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1396                         ret = -EACCES;
1397                         goto out_unlock;
1398                 }
1399         }
1400
1401         switch (optname) {
1402         case MRT_INIT:
1403                 if (optlen != sizeof(int)) {
1404                         ret = -EINVAL;
1405                         break;
1406                 }
1407                 if (rtnl_dereference(mrt->mroute_sk)) {
1408                         ret = -EADDRINUSE;
1409                         break;
1410                 }
1411
1412                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1413                 if (ret == 0) {
1414                         rcu_assign_pointer(mrt->mroute_sk, sk);
1415                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1416                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1417                                                     NETCONFA_MC_FORWARDING,
1418                                                     NETCONFA_IFINDEX_ALL,
1419                                                     net->ipv4.devconf_all);
1420                 }
1421                 break;
1422         case MRT_DONE:
1423                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1424                         ret = -EACCES;
1425                 } else {
1426                         /* We need to unlock here because mrtsock_destruct takes
1427                          * care of rtnl itself and we can't change that due to
1428                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1429                          */
1430                         rtnl_unlock();
1431                         ret = ip_ra_control(sk, 0, NULL);
1432                         goto out;
1433                 }
1434                 break;
1435         case MRT_ADD_VIF:
1436         case MRT_DEL_VIF:
1437                 if (optlen != sizeof(vif)) {
1438                         ret = -EINVAL;
1439                         break;
1440                 }
1441                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1442                         ret = -EFAULT;
1443                         break;
1444                 }
1445                 if (vif.vifc_vifi >= MAXVIFS) {
1446                         ret = -ENFILE;
1447                         break;
1448                 }
1449                 if (optname == MRT_ADD_VIF) {
1450                         ret = vif_add(net, mrt, &vif,
1451                                       sk == rtnl_dereference(mrt->mroute_sk));
1452                 } else {
1453                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1454                 }
1455                 break;
1456         /* Manipulate the forwarding caches. These live
1457          * in a sort of kernel/user symbiosis.
1458          */
1459         case MRT_ADD_MFC:
1460         case MRT_DEL_MFC:
1461                 parent = -1;
1462                 /* fall through */
1463         case MRT_ADD_MFC_PROXY:
1464         case MRT_DEL_MFC_PROXY:
1465                 if (optlen != sizeof(mfc)) {
1466                         ret = -EINVAL;
1467                         break;
1468                 }
1469                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1470                         ret = -EFAULT;
1471                         break;
1472                 }
1473                 if (parent == 0)
1474                         parent = mfc.mfcc_parent;
1475                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1476                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1477                 else
1478                         ret = ipmr_mfc_add(net, mrt, &mfc,
1479                                            sk == rtnl_dereference(mrt->mroute_sk),
1480                                            parent);
1481                 break;
1482         /* Control PIM assert. */
1483         case MRT_ASSERT:
1484                 if (optlen != sizeof(val)) {
1485                         ret = -EINVAL;
1486                         break;
1487                 }
1488                 if (get_user(val, (int __user *)optval)) {
1489                         ret = -EFAULT;
1490                         break;
1491                 }
1492                 mrt->mroute_do_assert = val;
1493                 break;
1494         case MRT_PIM:
1495                 if (!ipmr_pimsm_enabled()) {
1496                         ret = -ENOPROTOOPT;
1497                         break;
1498                 }
1499                 if (optlen != sizeof(val)) {
1500                         ret = -EINVAL;
1501                         break;
1502                 }
1503                 if (get_user(val, (int __user *)optval)) {
1504                         ret = -EFAULT;
1505                         break;
1506                 }
1507
1508                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1509                 val = !!val;
1510                 if (val != mrt->mroute_do_pim) {
1511                         mrt->mroute_do_pim = val;
1512                         mrt->mroute_do_assert = val;
1513                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1514                 }
1515                 break;
1516         case MRT_TABLE:
1517                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1518                         ret = -ENOPROTOOPT;
1519                         break;
1520                 }
1521                 if (optlen != sizeof(uval)) {
1522                         ret = -EINVAL;
1523                         break;
1524                 }
1525                 if (get_user(uval, (u32 __user *)optval)) {
1526                         ret = -EFAULT;
1527                         break;
1528                 }
1529
1530                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1531                         ret = -EBUSY;
1532                 } else {
1533                         mrt = ipmr_new_table(net, uval);
1534                         if (IS_ERR(mrt))
1535                                 ret = PTR_ERR(mrt);
1536                         else
1537                                 raw_sk(sk)->ipmr_table = uval;
1538                 }
1539                 break;
1540         /* Spurious command, or MRT_VERSION which you cannot set. */
1541         default:
1542                 ret = -ENOPROTOOPT;
1543         }
1544 out_unlock:
1545         rtnl_unlock();
1546 out:
1547         return ret;
1548 }
1549
1550 /* Getsock opt support for the multicast routing system. */
1551 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1552 {
1553         int olr;
1554         int val;
1555         struct net *net = sock_net(sk);
1556         struct mr_table *mrt;
1557
1558         if (sk->sk_type != SOCK_RAW ||
1559             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1560                 return -EOPNOTSUPP;
1561
1562         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1563         if (!mrt)
1564                 return -ENOENT;
1565
1566         switch (optname) {
1567         case MRT_VERSION:
1568                 val = 0x0305;
1569                 break;
1570         case MRT_PIM:
1571                 if (!ipmr_pimsm_enabled())
1572                         return -ENOPROTOOPT;
1573                 val = mrt->mroute_do_pim;
1574                 break;
1575         case MRT_ASSERT:
1576                 val = mrt->mroute_do_assert;
1577                 break;
1578         default:
1579                 return -ENOPROTOOPT;
1580         }
1581
1582         if (get_user(olr, optlen))
1583                 return -EFAULT;
1584         olr = min_t(unsigned int, olr, sizeof(int));
1585         if (olr < 0)
1586                 return -EINVAL;
1587         if (put_user(olr, optlen))
1588                 return -EFAULT;
1589         if (copy_to_user(optval, &val, olr))
1590                 return -EFAULT;
1591         return 0;
1592 }
1593
1594 /* The IP multicast ioctl support routines. */
1595 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1596 {
1597         struct sioc_sg_req sr;
1598         struct sioc_vif_req vr;
1599         struct vif_device *vif;
1600         struct mfc_cache *c;
1601         struct net *net = sock_net(sk);
1602         struct mr_table *mrt;
1603
1604         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1605         if (!mrt)
1606                 return -ENOENT;
1607
1608         switch (cmd) {
1609         case SIOCGETVIFCNT:
1610                 if (copy_from_user(&vr, arg, sizeof(vr)))
1611                         return -EFAULT;
1612                 if (vr.vifi >= mrt->maxvif)
1613                         return -EINVAL;
1614                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1615                 read_lock(&mrt_lock);
1616                 vif = &mrt->vif_table[vr.vifi];
1617                 if (VIF_EXISTS(mrt, vr.vifi)) {
1618                         vr.icount = vif->pkt_in;
1619                         vr.ocount = vif->pkt_out;
1620                         vr.ibytes = vif->bytes_in;
1621                         vr.obytes = vif->bytes_out;
1622                         read_unlock(&mrt_lock);
1623
1624                         if (copy_to_user(arg, &vr, sizeof(vr)))
1625                                 return -EFAULT;
1626                         return 0;
1627                 }
1628                 read_unlock(&mrt_lock);
1629                 return -EADDRNOTAVAIL;
1630         case SIOCGETSGCNT:
1631                 if (copy_from_user(&sr, arg, sizeof(sr)))
1632                         return -EFAULT;
1633
1634                 rcu_read_lock();
1635                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1636                 if (c) {
1637                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1638                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1639                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1640                         rcu_read_unlock();
1641
1642                         if (copy_to_user(arg, &sr, sizeof(sr)))
1643                                 return -EFAULT;
1644                         return 0;
1645                 }
1646                 rcu_read_unlock();
1647                 return -EADDRNOTAVAIL;
1648         default:
1649                 return -ENOIOCTLCMD;
1650         }
1651 }
1652
1653 #ifdef CONFIG_COMPAT
1654 struct compat_sioc_sg_req {
1655         struct in_addr src;
1656         struct in_addr grp;
1657         compat_ulong_t pktcnt;
1658         compat_ulong_t bytecnt;
1659         compat_ulong_t wrong_if;
1660 };
1661
1662 struct compat_sioc_vif_req {
1663         vifi_t  vifi;           /* Which iface */
1664         compat_ulong_t icount;
1665         compat_ulong_t ocount;
1666         compat_ulong_t ibytes;
1667         compat_ulong_t obytes;
1668 };
1669
1670 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1671 {
1672         struct compat_sioc_sg_req sr;
1673         struct compat_sioc_vif_req vr;
1674         struct vif_device *vif;
1675         struct mfc_cache *c;
1676         struct net *net = sock_net(sk);
1677         struct mr_table *mrt;
1678
1679         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1680         if (!mrt)
1681                 return -ENOENT;
1682
1683         switch (cmd) {
1684         case SIOCGETVIFCNT:
1685                 if (copy_from_user(&vr, arg, sizeof(vr)))
1686                         return -EFAULT;
1687                 if (vr.vifi >= mrt->maxvif)
1688                         return -EINVAL;
1689                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1690                 read_lock(&mrt_lock);
1691                 vif = &mrt->vif_table[vr.vifi];
1692                 if (VIF_EXISTS(mrt, vr.vifi)) {
1693                         vr.icount = vif->pkt_in;
1694                         vr.ocount = vif->pkt_out;
1695                         vr.ibytes = vif->bytes_in;
1696                         vr.obytes = vif->bytes_out;
1697                         read_unlock(&mrt_lock);
1698
1699                         if (copy_to_user(arg, &vr, sizeof(vr)))
1700                                 return -EFAULT;
1701                         return 0;
1702                 }
1703                 read_unlock(&mrt_lock);
1704                 return -EADDRNOTAVAIL;
1705         case SIOCGETSGCNT:
1706                 if (copy_from_user(&sr, arg, sizeof(sr)))
1707                         return -EFAULT;
1708
1709                 rcu_read_lock();
1710                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1711                 if (c) {
1712                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1713                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1714                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1715                         rcu_read_unlock();
1716
1717                         if (copy_to_user(arg, &sr, sizeof(sr)))
1718                                 return -EFAULT;
1719                         return 0;
1720                 }
1721                 rcu_read_unlock();
1722                 return -EADDRNOTAVAIL;
1723         default:
1724                 return -ENOIOCTLCMD;
1725         }
1726 }
1727 #endif
1728
1729 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1730 {
1731         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1732         struct net *net = dev_net(dev);
1733         struct mr_table *mrt;
1734         struct vif_device *v;
1735         int ct;
1736
1737         if (event != NETDEV_UNREGISTER)
1738                 return NOTIFY_DONE;
1739
1740         ipmr_for_each_table(mrt, net) {
1741                 v = &mrt->vif_table[0];
1742                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1743                         if (v->dev == dev)
1744                                 vif_delete(mrt, ct, 1, NULL);
1745                 }
1746         }
1747         return NOTIFY_DONE;
1748 }
1749
1750 static struct notifier_block ip_mr_notifier = {
1751         .notifier_call = ipmr_device_event,
1752 };
1753
1754 /* Encapsulate a packet by attaching a valid IPIP header to it.
1755  * This avoids tunnel drivers and other mess and gives us the speed so
1756  * important for multicast video.
1757  */
1758 static void ip_encap(struct net *net, struct sk_buff *skb,
1759                      __be32 saddr, __be32 daddr)
1760 {
1761         struct iphdr *iph;
1762         const struct iphdr *old_iph = ip_hdr(skb);
1763
1764         skb_push(skb, sizeof(struct iphdr));
1765         skb->transport_header = skb->network_header;
1766         skb_reset_network_header(skb);
1767         iph = ip_hdr(skb);
1768
1769         iph->version    =       4;
1770         iph->tos        =       old_iph->tos;
1771         iph->ttl        =       old_iph->ttl;
1772         iph->frag_off   =       0;
1773         iph->daddr      =       daddr;
1774         iph->saddr      =       saddr;
1775         iph->protocol   =       IPPROTO_IPIP;
1776         iph->ihl        =       5;
1777         iph->tot_len    =       htons(skb->len);
1778         ip_select_ident(net, skb, NULL);
1779         ip_send_check(iph);
1780
1781         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1782         nf_reset(skb);
1783 }
1784
1785 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1786                                       struct sk_buff *skb)
1787 {
1788         struct ip_options *opt = &(IPCB(skb)->opt);
1789
1790         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1791         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1792
1793         if (unlikely(opt->optlen))
1794                 ip_forward_options(skb);
1795
1796         return dst_output(net, sk, skb);
1797 }
1798
1799 #ifdef CONFIG_NET_SWITCHDEV
1800 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1801                                    int in_vifi, int out_vifi)
1802 {
1803         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1804         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1805
1806         if (!skb->offload_l3_fwd_mark)
1807                 return false;
1808         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1809                 return false;
1810         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1811                                         &in_vif->dev_parent_id);
1812 }
1813 #else
1814 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1815                                    int in_vifi, int out_vifi)
1816 {
1817         return false;
1818 }
1819 #endif
1820
1821 /* Processing handlers for ipmr_forward */
1822
1823 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1824                             int in_vifi, struct sk_buff *skb, int vifi)
1825 {
1826         const struct iphdr *iph = ip_hdr(skb);
1827         struct vif_device *vif = &mrt->vif_table[vifi];
1828         struct net_device *dev;
1829         struct rtable *rt;
1830         struct flowi4 fl4;
1831         int    encap = 0;
1832
1833         if (!vif->dev)
1834                 goto out_free;
1835
1836         if (vif->flags & VIFF_REGISTER) {
1837                 vif->pkt_out++;
1838                 vif->bytes_out += skb->len;
1839                 vif->dev->stats.tx_bytes += skb->len;
1840                 vif->dev->stats.tx_packets++;
1841                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1842                 goto out_free;
1843         }
1844
1845         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1846                 goto out_free;
1847
1848         if (vif->flags & VIFF_TUNNEL) {
1849                 rt = ip_route_output_ports(net, &fl4, NULL,
1850                                            vif->remote, vif->local,
1851                                            0, 0,
1852                                            IPPROTO_IPIP,
1853                                            RT_TOS(iph->tos), vif->link);
1854                 if (IS_ERR(rt))
1855                         goto out_free;
1856                 encap = sizeof(struct iphdr);
1857         } else {
1858                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1859                                            0, 0,
1860                                            IPPROTO_IPIP,
1861                                            RT_TOS(iph->tos), vif->link);
1862                 if (IS_ERR(rt))
1863                         goto out_free;
1864         }
1865
1866         dev = rt->dst.dev;
1867
1868         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1869                 /* Do not fragment multicasts. Alas, IPv4 does not
1870                  * allow to send ICMP, so that packets will disappear
1871                  * to blackhole.
1872                  */
1873                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1874                 ip_rt_put(rt);
1875                 goto out_free;
1876         }
1877
1878         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1879
1880         if (skb_cow(skb, encap)) {
1881                 ip_rt_put(rt);
1882                 goto out_free;
1883         }
1884
1885         vif->pkt_out++;
1886         vif->bytes_out += skb->len;
1887
1888         skb_dst_drop(skb);
1889         skb_dst_set(skb, &rt->dst);
1890         ip_decrease_ttl(ip_hdr(skb));
1891
1892         /* FIXME: forward and output firewalls used to be called here.
1893          * What do we do with netfilter? -- RR
1894          */
1895         if (vif->flags & VIFF_TUNNEL) {
1896                 ip_encap(net, skb, vif->local, vif->remote);
1897                 /* FIXME: extra output firewall step used to be here. --RR */
1898                 vif->dev->stats.tx_packets++;
1899                 vif->dev->stats.tx_bytes += skb->len;
1900         }
1901
1902         IPCB(skb)->flags |= IPSKB_FORWARDED;
1903
1904         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1905          * not only before forwarding, but after forwarding on all output
1906          * interfaces. It is clear, if mrouter runs a multicasting
1907          * program, it should receive packets not depending to what interface
1908          * program is joined.
1909          * If we will not make it, the program will have to join on all
1910          * interfaces. On the other hand, multihoming host (or router, but
1911          * not mrouter) cannot join to more than one interface - it will
1912          * result in receiving multiple packets.
1913          */
1914         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1915                 net, NULL, skb, skb->dev, dev,
1916                 ipmr_forward_finish);
1917         return;
1918
1919 out_free:
1920         kfree_skb(skb);
1921 }
1922
1923 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1924 {
1925         int ct;
1926
1927         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1928                 if (mrt->vif_table[ct].dev == dev)
1929                         break;
1930         }
1931         return ct;
1932 }
1933
1934 /* "local" means that we should preserve one skb (for local delivery) */
1935 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1936                           struct net_device *dev, struct sk_buff *skb,
1937                           struct mfc_cache *c, int local)
1938 {
1939         int true_vifi = ipmr_find_vif(mrt, dev);
1940         int psend = -1;
1941         int vif, ct;
1942
1943         vif = c->_c.mfc_parent;
1944         c->_c.mfc_un.res.pkt++;
1945         c->_c.mfc_un.res.bytes += skb->len;
1946         c->_c.mfc_un.res.lastuse = jiffies;
1947
1948         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1949                 struct mfc_cache *cache_proxy;
1950
1951                 /* For an (*,G) entry, we only check that the incomming
1952                  * interface is part of the static tree.
1953                  */
1954                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1955                 if (cache_proxy &&
1956                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1957                         goto forward;
1958         }
1959
1960         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1961         if (mrt->vif_table[vif].dev != dev) {
1962                 if (rt_is_output_route(skb_rtable(skb))) {
1963                         /* It is our own packet, looped back.
1964                          * Very complicated situation...
1965                          *
1966                          * The best workaround until routing daemons will be
1967                          * fixed is not to redistribute packet, if it was
1968                          * send through wrong interface. It means, that
1969                          * multicast applications WILL NOT work for
1970                          * (S,G), which have default multicast route pointing
1971                          * to wrong oif. In any case, it is not a good
1972                          * idea to use multicasting applications on router.
1973                          */
1974                         goto dont_forward;
1975                 }
1976
1977                 c->_c.mfc_un.res.wrong_if++;
1978
1979                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1980                     /* pimsm uses asserts, when switching from RPT to SPT,
1981                      * so that we cannot check that packet arrived on an oif.
1982                      * It is bad, but otherwise we would need to move pretty
1983                      * large chunk of pimd to kernel. Ough... --ANK
1984                      */
1985                     (mrt->mroute_do_pim ||
1986                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1987                     time_after(jiffies,
1988                                c->_c.mfc_un.res.last_assert +
1989                                MFC_ASSERT_THRESH)) {
1990                         c->_c.mfc_un.res.last_assert = jiffies;
1991                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1992                         if (mrt->mroute_do_wrvifwhole)
1993                                 ipmr_cache_report(mrt, skb, true_vifi,
1994                                                   IGMPMSG_WRVIFWHOLE);
1995                 }
1996                 goto dont_forward;
1997         }
1998
1999 forward:
2000         mrt->vif_table[vif].pkt_in++;
2001         mrt->vif_table[vif].bytes_in += skb->len;
2002
2003         /* Forward the frame */
2004         if (c->mfc_origin == htonl(INADDR_ANY) &&
2005             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
2006                 if (true_vifi >= 0 &&
2007                     true_vifi != c->_c.mfc_parent &&
2008                     ip_hdr(skb)->ttl >
2009                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2010                         /* It's an (*,*) entry and the packet is not coming from
2011                          * the upstream: forward the packet to the upstream
2012                          * only.
2013                          */
2014                         psend = c->_c.mfc_parent;
2015                         goto last_forward;
2016                 }
2017                 goto dont_forward;
2018         }
2019         for (ct = c->_c.mfc_un.res.maxvif - 1;
2020              ct >= c->_c.mfc_un.res.minvif; ct--) {
2021                 /* For (*,G) entry, don't forward to the incoming interface */
2022                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2023                      ct != true_vifi) &&
2024                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2025                         if (psend != -1) {
2026                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2027
2028                                 if (skb2)
2029                                         ipmr_queue_xmit(net, mrt, true_vifi,
2030                                                         skb2, psend);
2031                         }
2032                         psend = ct;
2033                 }
2034         }
2035 last_forward:
2036         if (psend != -1) {
2037                 if (local) {
2038                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2039
2040                         if (skb2)
2041                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2042                                                 psend);
2043                 } else {
2044                         ipmr_queue_xmit(net, mrt, true_vifi, skb, psend);
2045                         return;
2046                 }
2047         }
2048
2049 dont_forward:
2050         if (!local)
2051                 kfree_skb(skb);
2052 }
2053
2054 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2055 {
2056         struct rtable *rt = skb_rtable(skb);
2057         struct iphdr *iph = ip_hdr(skb);
2058         struct flowi4 fl4 = {
2059                 .daddr = iph->daddr,
2060                 .saddr = iph->saddr,
2061                 .flowi4_tos = RT_TOS(iph->tos),
2062                 .flowi4_oif = (rt_is_output_route(rt) ?
2063                                skb->dev->ifindex : 0),
2064                 .flowi4_iif = (rt_is_output_route(rt) ?
2065                                LOOPBACK_IFINDEX :
2066                                skb->dev->ifindex),
2067                 .flowi4_mark = skb->mark,
2068         };
2069         struct mr_table *mrt;
2070         int err;
2071
2072         err = ipmr_fib_lookup(net, &fl4, &mrt);
2073         if (err)
2074                 return ERR_PTR(err);
2075         return mrt;
2076 }
2077
2078 /* Multicast packets for forwarding arrive here
2079  * Called with rcu_read_lock();
2080  */
2081 int ip_mr_input(struct sk_buff *skb)
2082 {
2083         struct mfc_cache *cache;
2084         struct net *net = dev_net(skb->dev);
2085         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2086         struct mr_table *mrt;
2087         struct net_device *dev;
2088
2089         /* skb->dev passed in is the loX master dev for vrfs.
2090          * As there are no vifs associated with loopback devices,
2091          * get the proper interface that does have a vif associated with it.
2092          */
2093         dev = skb->dev;
2094         if (netif_is_l3_master(skb->dev)) {
2095                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2096                 if (!dev) {
2097                         kfree_skb(skb);
2098                         return -ENODEV;
2099                 }
2100         }
2101
2102         /* Packet is looped back after forward, it should not be
2103          * forwarded second time, but still can be delivered locally.
2104          */
2105         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2106                 goto dont_forward;
2107
2108         mrt = ipmr_rt_fib_lookup(net, skb);
2109         if (IS_ERR(mrt)) {
2110                 kfree_skb(skb);
2111                 return PTR_ERR(mrt);
2112         }
2113         if (!local) {
2114                 if (IPCB(skb)->opt.router_alert) {
2115                         if (ip_call_ra_chain(skb))
2116                                 return 0;
2117                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2118                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2119                          * Cisco IOS <= 11.2(8)) do not put router alert
2120                          * option to IGMP packets destined to routable
2121                          * groups. It is very bad, because it means
2122                          * that we can forward NO IGMP messages.
2123                          */
2124                         struct sock *mroute_sk;
2125
2126                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2127                         if (mroute_sk) {
2128                                 nf_reset(skb);
2129                                 raw_rcv(mroute_sk, skb);
2130                                 return 0;
2131                         }
2132                     }
2133         }
2134
2135         /* already under rcu_read_lock() */
2136         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2137         if (!cache) {
2138                 int vif = ipmr_find_vif(mrt, dev);
2139
2140                 if (vif >= 0)
2141                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2142                                                     vif);
2143         }
2144
2145         /* No usable cache entry */
2146         if (!cache) {
2147                 int vif;
2148
2149                 if (local) {
2150                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2151                         ip_local_deliver(skb);
2152                         if (!skb2)
2153                                 return -ENOBUFS;
2154                         skb = skb2;
2155                 }
2156
2157                 read_lock(&mrt_lock);
2158                 vif = ipmr_find_vif(mrt, dev);
2159                 if (vif >= 0) {
2160                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2161                         read_unlock(&mrt_lock);
2162
2163                         return err2;
2164                 }
2165                 read_unlock(&mrt_lock);
2166                 kfree_skb(skb);
2167                 return -ENODEV;
2168         }
2169
2170         read_lock(&mrt_lock);
2171         ip_mr_forward(net, mrt, dev, skb, cache, local);
2172         read_unlock(&mrt_lock);
2173
2174         if (local)
2175                 return ip_local_deliver(skb);
2176
2177         return 0;
2178
2179 dont_forward:
2180         if (local)
2181                 return ip_local_deliver(skb);
2182         kfree_skb(skb);
2183         return 0;
2184 }
2185
2186 #ifdef CONFIG_IP_PIMSM_V1
2187 /* Handle IGMP messages of PIMv1 */
2188 int pim_rcv_v1(struct sk_buff *skb)
2189 {
2190         struct igmphdr *pim;
2191         struct net *net = dev_net(skb->dev);
2192         struct mr_table *mrt;
2193
2194         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2195                 goto drop;
2196
2197         pim = igmp_hdr(skb);
2198
2199         mrt = ipmr_rt_fib_lookup(net, skb);
2200         if (IS_ERR(mrt))
2201                 goto drop;
2202         if (!mrt->mroute_do_pim ||
2203             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2204                 goto drop;
2205
2206         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2207 drop:
2208                 kfree_skb(skb);
2209         }
2210         return 0;
2211 }
2212 #endif
2213
2214 #ifdef CONFIG_IP_PIMSM_V2
2215 static int pim_rcv(struct sk_buff *skb)
2216 {
2217         struct pimreghdr *pim;
2218         struct net *net = dev_net(skb->dev);
2219         struct mr_table *mrt;
2220
2221         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2222                 goto drop;
2223
2224         pim = (struct pimreghdr *)skb_transport_header(skb);
2225         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2226             (pim->flags & PIM_NULL_REGISTER) ||
2227             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2228              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2229                 goto drop;
2230
2231         mrt = ipmr_rt_fib_lookup(net, skb);
2232         if (IS_ERR(mrt))
2233                 goto drop;
2234         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2235 drop:
2236                 kfree_skb(skb);
2237         }
2238         return 0;
2239 }
2240 #endif
2241
2242 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2243                    __be32 saddr, __be32 daddr,
2244                    struct rtmsg *rtm, u32 portid)
2245 {
2246         struct mfc_cache *cache;
2247         struct mr_table *mrt;
2248         int err;
2249
2250         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2251         if (!mrt)
2252                 return -ENOENT;
2253
2254         rcu_read_lock();
2255         cache = ipmr_cache_find(mrt, saddr, daddr);
2256         if (!cache && skb->dev) {
2257                 int vif = ipmr_find_vif(mrt, skb->dev);
2258
2259                 if (vif >= 0)
2260                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2261         }
2262         if (!cache) {
2263                 struct sk_buff *skb2;
2264                 struct iphdr *iph;
2265                 struct net_device *dev;
2266                 int vif = -1;
2267
2268                 dev = skb->dev;
2269                 read_lock(&mrt_lock);
2270                 if (dev)
2271                         vif = ipmr_find_vif(mrt, dev);
2272                 if (vif < 0) {
2273                         read_unlock(&mrt_lock);
2274                         rcu_read_unlock();
2275                         return -ENODEV;
2276                 }
2277                 skb2 = skb_clone(skb, GFP_ATOMIC);
2278                 if (!skb2) {
2279                         read_unlock(&mrt_lock);
2280                         rcu_read_unlock();
2281                         return -ENOMEM;
2282                 }
2283
2284                 NETLINK_CB(skb2).portid = portid;
2285                 skb_push(skb2, sizeof(struct iphdr));
2286                 skb_reset_network_header(skb2);
2287                 iph = ip_hdr(skb2);
2288                 iph->ihl = sizeof(struct iphdr) >> 2;
2289                 iph->saddr = saddr;
2290                 iph->daddr = daddr;
2291                 iph->version = 0;
2292                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2293                 read_unlock(&mrt_lock);
2294                 rcu_read_unlock();
2295                 return err;
2296         }
2297
2298         read_lock(&mrt_lock);
2299         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2300         read_unlock(&mrt_lock);
2301         rcu_read_unlock();
2302         return err;
2303 }
2304
2305 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2306                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2307                             int flags)
2308 {
2309         struct nlmsghdr *nlh;
2310         struct rtmsg *rtm;
2311         int err;
2312
2313         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2314         if (!nlh)
2315                 return -EMSGSIZE;
2316
2317         rtm = nlmsg_data(nlh);
2318         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2319         rtm->rtm_dst_len  = 32;
2320         rtm->rtm_src_len  = 32;
2321         rtm->rtm_tos      = 0;
2322         rtm->rtm_table    = mrt->id;
2323         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2324                 goto nla_put_failure;
2325         rtm->rtm_type     = RTN_MULTICAST;
2326         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2327         if (c->_c.mfc_flags & MFC_STATIC)
2328                 rtm->rtm_protocol = RTPROT_STATIC;
2329         else
2330                 rtm->rtm_protocol = RTPROT_MROUTED;
2331         rtm->rtm_flags    = 0;
2332
2333         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2334             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2335                 goto nla_put_failure;
2336         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2337         /* do not break the dump if cache is unresolved */
2338         if (err < 0 && err != -ENOENT)
2339                 goto nla_put_failure;
2340
2341         nlmsg_end(skb, nlh);
2342         return 0;
2343
2344 nla_put_failure:
2345         nlmsg_cancel(skb, nlh);
2346         return -EMSGSIZE;
2347 }
2348
2349 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2350                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2351                              int flags)
2352 {
2353         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2354                                 cmd, flags);
2355 }
2356
2357 static size_t mroute_msgsize(bool unresolved, int maxvif)
2358 {
2359         size_t len =
2360                 NLMSG_ALIGN(sizeof(struct rtmsg))
2361                 + nla_total_size(4)     /* RTA_TABLE */
2362                 + nla_total_size(4)     /* RTA_SRC */
2363                 + nla_total_size(4)     /* RTA_DST */
2364                 ;
2365
2366         if (!unresolved)
2367                 len = len
2368                       + nla_total_size(4)       /* RTA_IIF */
2369                       + nla_total_size(0)       /* RTA_MULTIPATH */
2370                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2371                                                 /* RTA_MFC_STATS */
2372                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2373                 ;
2374
2375         return len;
2376 }
2377
2378 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2379                                  int cmd)
2380 {
2381         struct net *net = read_pnet(&mrt->net);
2382         struct sk_buff *skb;
2383         int err = -ENOBUFS;
2384
2385         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2386                                        mrt->maxvif),
2387                         GFP_ATOMIC);
2388         if (!skb)
2389                 goto errout;
2390
2391         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2392         if (err < 0)
2393                 goto errout;
2394
2395         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2396         return;
2397
2398 errout:
2399         kfree_skb(skb);
2400         if (err < 0)
2401                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2402 }
2403
2404 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2405 {
2406         size_t len =
2407                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2408                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2409                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2410                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2411                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2412                                         /* IPMRA_CREPORT_PKT */
2413                 + nla_total_size(payloadlen)
2414                 ;
2415
2416         return len;
2417 }
2418
2419 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2420 {
2421         struct net *net = read_pnet(&mrt->net);
2422         struct nlmsghdr *nlh;
2423         struct rtgenmsg *rtgenm;
2424         struct igmpmsg *msg;
2425         struct sk_buff *skb;
2426         struct nlattr *nla;
2427         int payloadlen;
2428
2429         payloadlen = pkt->len - sizeof(struct igmpmsg);
2430         msg = (struct igmpmsg *)skb_network_header(pkt);
2431
2432         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2433         if (!skb)
2434                 goto errout;
2435
2436         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2437                         sizeof(struct rtgenmsg), 0);
2438         if (!nlh)
2439                 goto errout;
2440         rtgenm = nlmsg_data(nlh);
2441         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2442         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2443             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif) ||
2444             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2445                             msg->im_src.s_addr) ||
2446             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2447                             msg->im_dst.s_addr))
2448                 goto nla_put_failure;
2449
2450         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2451         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2452                                   nla_data(nla), payloadlen))
2453                 goto nla_put_failure;
2454
2455         nlmsg_end(skb, nlh);
2456
2457         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2458         return;
2459
2460 nla_put_failure:
2461         nlmsg_cancel(skb, nlh);
2462 errout:
2463         kfree_skb(skb);
2464         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2465 }
2466
2467 static int ipmr_rtm_valid_getroute_req(struct sk_buff *skb,
2468                                        const struct nlmsghdr *nlh,
2469                                        struct nlattr **tb,
2470                                        struct netlink_ext_ack *extack)
2471 {
2472         struct rtmsg *rtm;
2473         int i, err;
2474
2475         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2476                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for multicast route get request");
2477                 return -EINVAL;
2478         }
2479
2480         if (!netlink_strict_get_check(skb))
2481                 return nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2482                                    rtm_ipv4_policy, extack);
2483
2484         rtm = nlmsg_data(nlh);
2485         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2486             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2487             rtm->rtm_tos || rtm->rtm_table || rtm->rtm_protocol ||
2488             rtm->rtm_scope || rtm->rtm_type || rtm->rtm_flags) {
2489                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for multicast route get request");
2490                 return -EINVAL;
2491         }
2492
2493         err = nlmsg_parse_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2494                                  rtm_ipv4_policy, extack);
2495         if (err)
2496                 return err;
2497
2498         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2499             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2500                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2501                 return -EINVAL;
2502         }
2503
2504         for (i = 0; i <= RTA_MAX; i++) {
2505                 if (!tb[i])
2506                         continue;
2507
2508                 switch (i) {
2509                 case RTA_SRC:
2510                 case RTA_DST:
2511                 case RTA_TABLE:
2512                         break;
2513                 default:
2514                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in multicast route get request");
2515                         return -EINVAL;
2516                 }
2517         }
2518
2519         return 0;
2520 }
2521
2522 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2523                              struct netlink_ext_ack *extack)
2524 {
2525         struct net *net = sock_net(in_skb->sk);
2526         struct nlattr *tb[RTA_MAX + 1];
2527         struct sk_buff *skb = NULL;
2528         struct mfc_cache *cache;
2529         struct mr_table *mrt;
2530         __be32 src, grp;
2531         u32 tableid;
2532         int err;
2533
2534         err = ipmr_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2535         if (err < 0)
2536                 goto errout;
2537
2538         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2539         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2540         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2541
2542         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2543         if (!mrt) {
2544                 err = -ENOENT;
2545                 goto errout_free;
2546         }
2547
2548         /* entries are added/deleted only under RTNL */
2549         rcu_read_lock();
2550         cache = ipmr_cache_find(mrt, src, grp);
2551         rcu_read_unlock();
2552         if (!cache) {
2553                 err = -ENOENT;
2554                 goto errout_free;
2555         }
2556
2557         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2558         if (!skb) {
2559                 err = -ENOBUFS;
2560                 goto errout_free;
2561         }
2562
2563         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2564                                nlh->nlmsg_seq, cache,
2565                                RTM_NEWROUTE, 0);
2566         if (err < 0)
2567                 goto errout_free;
2568
2569         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2570
2571 errout:
2572         return err;
2573
2574 errout_free:
2575         kfree_skb(skb);
2576         goto errout;
2577 }
2578
2579 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2580 {
2581         struct fib_dump_filter filter = {};
2582         int err;
2583
2584         if (cb->strict_check) {
2585                 err = ip_valid_fib_dump_req(sock_net(skb->sk), cb->nlh,
2586                                             &filter, cb);
2587                 if (err < 0)
2588                         return err;
2589         }
2590
2591         if (filter.table_id) {
2592                 struct mr_table *mrt;
2593
2594                 mrt = ipmr_get_table(sock_net(skb->sk), filter.table_id);
2595                 if (!mrt) {
2596                         if (filter.dump_all_families)
2597                                 return skb->len;
2598
2599                         NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
2600                         return -ENOENT;
2601                 }
2602                 err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
2603                                     &mfc_unres_lock, &filter);
2604                 return skb->len ? : err;
2605         }
2606
2607         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2608                                 _ipmr_fill_mroute, &mfc_unres_lock, &filter);
2609 }
2610
2611 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2612         [RTA_SRC]       = { .type = NLA_U32 },
2613         [RTA_DST]       = { .type = NLA_U32 },
2614         [RTA_IIF]       = { .type = NLA_U32 },
2615         [RTA_TABLE]     = { .type = NLA_U32 },
2616         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2617 };
2618
2619 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2620 {
2621         switch (rtm_protocol) {
2622         case RTPROT_STATIC:
2623         case RTPROT_MROUTED:
2624                 return true;
2625         }
2626         return false;
2627 }
2628
2629 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2630 {
2631         struct rtnexthop *rtnh = nla_data(nla);
2632         int remaining = nla_len(nla), vifi = 0;
2633
2634         while (rtnh_ok(rtnh, remaining)) {
2635                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2636                 if (++vifi == MAXVIFS)
2637                         break;
2638                 rtnh = rtnh_next(rtnh, &remaining);
2639         }
2640
2641         return remaining > 0 ? -EINVAL : vifi;
2642 }
2643
2644 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2645 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2646                             struct mfcctl *mfcc, int *mrtsock,
2647                             struct mr_table **mrtret,
2648                             struct netlink_ext_ack *extack)
2649 {
2650         struct net_device *dev = NULL;
2651         u32 tblid = RT_TABLE_DEFAULT;
2652         struct mr_table *mrt;
2653         struct nlattr *attr;
2654         struct rtmsg *rtm;
2655         int ret, rem;
2656
2657         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy,
2658                              extack);
2659         if (ret < 0)
2660                 goto out;
2661         rtm = nlmsg_data(nlh);
2662
2663         ret = -EINVAL;
2664         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2665             rtm->rtm_type != RTN_MULTICAST ||
2666             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2667             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2668                 goto out;
2669
2670         memset(mfcc, 0, sizeof(*mfcc));
2671         mfcc->mfcc_parent = -1;
2672         ret = 0;
2673         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2674                 switch (nla_type(attr)) {
2675                 case RTA_SRC:
2676                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2677                         break;
2678                 case RTA_DST:
2679                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2680                         break;
2681                 case RTA_IIF:
2682                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2683                         if (!dev) {
2684                                 ret = -ENODEV;
2685                                 goto out;
2686                         }
2687                         break;
2688                 case RTA_MULTIPATH:
2689                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2690                                 ret = -EINVAL;
2691                                 goto out;
2692                         }
2693                         break;
2694                 case RTA_PREFSRC:
2695                         ret = 1;
2696                         break;
2697                 case RTA_TABLE:
2698                         tblid = nla_get_u32(attr);
2699                         break;
2700                 }
2701         }
2702         mrt = ipmr_get_table(net, tblid);
2703         if (!mrt) {
2704                 ret = -ENOENT;
2705                 goto out;
2706         }
2707         *mrtret = mrt;
2708         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2709         if (dev)
2710                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2711
2712 out:
2713         return ret;
2714 }
2715
2716 /* takes care of both newroute and delroute */
2717 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2718                           struct netlink_ext_ack *extack)
2719 {
2720         struct net *net = sock_net(skb->sk);
2721         int ret, mrtsock, parent;
2722         struct mr_table *tbl;
2723         struct mfcctl mfcc;
2724
2725         mrtsock = 0;
2726         tbl = NULL;
2727         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2728         if (ret < 0)
2729                 return ret;
2730
2731         parent = ret ? mfcc.mfcc_parent : -1;
2732         if (nlh->nlmsg_type == RTM_NEWROUTE)
2733                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2734         else
2735                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2736 }
2737
2738 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2739 {
2740         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2741
2742         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2743             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2744             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2745                         mrt->mroute_reg_vif_num) ||
2746             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2747                        mrt->mroute_do_assert) ||
2748             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2749             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2750                        mrt->mroute_do_wrvifwhole))
2751                 return false;
2752
2753         return true;
2754 }
2755
2756 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2757 {
2758         struct nlattr *vif_nest;
2759         struct vif_device *vif;
2760
2761         /* if the VIF doesn't exist just continue */
2762         if (!VIF_EXISTS(mrt, vifid))
2763                 return true;
2764
2765         vif = &mrt->vif_table[vifid];
2766         vif_nest = nla_nest_start(skb, IPMRA_VIF);
2767         if (!vif_nest)
2768                 return false;
2769         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2770             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2771             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2772             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2773                               IPMRA_VIFA_PAD) ||
2774             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2775                               IPMRA_VIFA_PAD) ||
2776             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2777                               IPMRA_VIFA_PAD) ||
2778             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2779                               IPMRA_VIFA_PAD) ||
2780             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2781             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2782                 nla_nest_cancel(skb, vif_nest);
2783                 return false;
2784         }
2785         nla_nest_end(skb, vif_nest);
2786
2787         return true;
2788 }
2789
2790 static int ipmr_valid_dumplink(const struct nlmsghdr *nlh,
2791                                struct netlink_ext_ack *extack)
2792 {
2793         struct ifinfomsg *ifm;
2794
2795         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ifm))) {
2796                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for ipmr link dump");
2797                 return -EINVAL;
2798         }
2799
2800         if (nlmsg_attrlen(nlh, sizeof(*ifm))) {
2801                 NL_SET_ERR_MSG(extack, "Invalid data after header in ipmr link dump");
2802                 return -EINVAL;
2803         }
2804
2805         ifm = nlmsg_data(nlh);
2806         if (ifm->__ifi_pad || ifm->ifi_type || ifm->ifi_flags ||
2807             ifm->ifi_change || ifm->ifi_index) {
2808                 NL_SET_ERR_MSG(extack, "Invalid values in header for ipmr link dump request");
2809                 return -EINVAL;
2810         }
2811
2812         return 0;
2813 }
2814
2815 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2816 {
2817         struct net *net = sock_net(skb->sk);
2818         struct nlmsghdr *nlh = NULL;
2819         unsigned int t = 0, s_t;
2820         unsigned int e = 0, s_e;
2821         struct mr_table *mrt;
2822
2823         if (cb->strict_check) {
2824                 int err = ipmr_valid_dumplink(cb->nlh, cb->extack);
2825
2826                 if (err < 0)
2827                         return err;
2828         }
2829
2830         s_t = cb->args[0];
2831         s_e = cb->args[1];
2832
2833         ipmr_for_each_table(mrt, net) {
2834                 struct nlattr *vifs, *af;
2835                 struct ifinfomsg *hdr;
2836                 u32 i;
2837
2838                 if (t < s_t)
2839                         goto skip_table;
2840                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2841                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2842                                 sizeof(*hdr), NLM_F_MULTI);
2843                 if (!nlh)
2844                         break;
2845
2846                 hdr = nlmsg_data(nlh);
2847                 memset(hdr, 0, sizeof(*hdr));
2848                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2849
2850                 af = nla_nest_start(skb, IFLA_AF_SPEC);
2851                 if (!af) {
2852                         nlmsg_cancel(skb, nlh);
2853                         goto out;
2854                 }
2855
2856                 if (!ipmr_fill_table(mrt, skb)) {
2857                         nlmsg_cancel(skb, nlh);
2858                         goto out;
2859                 }
2860
2861                 vifs = nla_nest_start(skb, IPMRA_TABLE_VIFS);
2862                 if (!vifs) {
2863                         nla_nest_end(skb, af);
2864                         nlmsg_end(skb, nlh);
2865                         goto out;
2866                 }
2867                 for (i = 0; i < mrt->maxvif; i++) {
2868                         if (e < s_e)
2869                                 goto skip_entry;
2870                         if (!ipmr_fill_vif(mrt, i, skb)) {
2871                                 nla_nest_end(skb, vifs);
2872                                 nla_nest_end(skb, af);
2873                                 nlmsg_end(skb, nlh);
2874                                 goto out;
2875                         }
2876 skip_entry:
2877                         e++;
2878                 }
2879                 s_e = 0;
2880                 e = 0;
2881                 nla_nest_end(skb, vifs);
2882                 nla_nest_end(skb, af);
2883                 nlmsg_end(skb, nlh);
2884 skip_table:
2885                 t++;
2886         }
2887
2888 out:
2889         cb->args[1] = e;
2890         cb->args[0] = t;
2891
2892         return skb->len;
2893 }
2894
2895 #ifdef CONFIG_PROC_FS
2896 /* The /proc interfaces to multicast routing :
2897  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2898  */
2899
2900 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2901         __acquires(mrt_lock)
2902 {
2903         struct mr_vif_iter *iter = seq->private;
2904         struct net *net = seq_file_net(seq);
2905         struct mr_table *mrt;
2906
2907         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2908         if (!mrt)
2909                 return ERR_PTR(-ENOENT);
2910
2911         iter->mrt = mrt;
2912
2913         read_lock(&mrt_lock);
2914         return mr_vif_seq_start(seq, pos);
2915 }
2916
2917 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2918         __releases(mrt_lock)
2919 {
2920         read_unlock(&mrt_lock);
2921 }
2922
2923 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2924 {
2925         struct mr_vif_iter *iter = seq->private;
2926         struct mr_table *mrt = iter->mrt;
2927
2928         if (v == SEQ_START_TOKEN) {
2929                 seq_puts(seq,
2930                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2931         } else {
2932                 const struct vif_device *vif = v;
2933                 const char *name =  vif->dev ?
2934                                     vif->dev->name : "none";
2935
2936                 seq_printf(seq,
2937                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2938                            vif - mrt->vif_table,
2939                            name, vif->bytes_in, vif->pkt_in,
2940                            vif->bytes_out, vif->pkt_out,
2941                            vif->flags, vif->local, vif->remote);
2942         }
2943         return 0;
2944 }
2945
2946 static const struct seq_operations ipmr_vif_seq_ops = {
2947         .start = ipmr_vif_seq_start,
2948         .next  = mr_vif_seq_next,
2949         .stop  = ipmr_vif_seq_stop,
2950         .show  = ipmr_vif_seq_show,
2951 };
2952
2953 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2954 {
2955         struct net *net = seq_file_net(seq);
2956         struct mr_table *mrt;
2957
2958         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2959         if (!mrt)
2960                 return ERR_PTR(-ENOENT);
2961
2962         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2963 }
2964
2965 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2966 {
2967         int n;
2968
2969         if (v == SEQ_START_TOKEN) {
2970                 seq_puts(seq,
2971                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2972         } else {
2973                 const struct mfc_cache *mfc = v;
2974                 const struct mr_mfc_iter *it = seq->private;
2975                 const struct mr_table *mrt = it->mrt;
2976
2977                 seq_printf(seq, "%08X %08X %-3hd",
2978                            (__force u32) mfc->mfc_mcastgrp,
2979                            (__force u32) mfc->mfc_origin,
2980                            mfc->_c.mfc_parent);
2981
2982                 if (it->cache != &mrt->mfc_unres_queue) {
2983                         seq_printf(seq, " %8lu %8lu %8lu",
2984                                    mfc->_c.mfc_un.res.pkt,
2985                                    mfc->_c.mfc_un.res.bytes,
2986                                    mfc->_c.mfc_un.res.wrong_if);
2987                         for (n = mfc->_c.mfc_un.res.minvif;
2988                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2989                                 if (VIF_EXISTS(mrt, n) &&
2990                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2991                                         seq_printf(seq,
2992                                            " %2d:%-3d",
2993                                            n, mfc->_c.mfc_un.res.ttls[n]);
2994                         }
2995                 } else {
2996                         /* unresolved mfc_caches don't contain
2997                          * pkt, bytes and wrong_if values
2998                          */
2999                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
3000                 }
3001                 seq_putc(seq, '\n');
3002         }
3003         return 0;
3004 }
3005
3006 static const struct seq_operations ipmr_mfc_seq_ops = {
3007         .start = ipmr_mfc_seq_start,
3008         .next  = mr_mfc_seq_next,
3009         .stop  = mr_mfc_seq_stop,
3010         .show  = ipmr_mfc_seq_show,
3011 };
3012 #endif
3013
3014 #ifdef CONFIG_IP_PIMSM_V2
3015 static const struct net_protocol pim_protocol = {
3016         .handler        =       pim_rcv,
3017         .netns_ok       =       1,
3018 };
3019 #endif
3020
3021 static unsigned int ipmr_seq_read(struct net *net)
3022 {
3023         ASSERT_RTNL();
3024
3025         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3026 }
3027
3028 static int ipmr_dump(struct net *net, struct notifier_block *nb)
3029 {
3030         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
3031                        ipmr_mr_table_iter, &mrt_lock);
3032 }
3033
3034 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3035         .family         = RTNL_FAMILY_IPMR,
3036         .fib_seq_read   = ipmr_seq_read,
3037         .fib_dump       = ipmr_dump,
3038         .owner          = THIS_MODULE,
3039 };
3040
3041 static int __net_init ipmr_notifier_init(struct net *net)
3042 {
3043         struct fib_notifier_ops *ops;
3044
3045         net->ipv4.ipmr_seq = 0;
3046
3047         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3048         if (IS_ERR(ops))
3049                 return PTR_ERR(ops);
3050         net->ipv4.ipmr_notifier_ops = ops;
3051
3052         return 0;
3053 }
3054
3055 static void __net_exit ipmr_notifier_exit(struct net *net)
3056 {
3057         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3058         net->ipv4.ipmr_notifier_ops = NULL;
3059 }
3060
3061 /* Setup for IP multicast routing */
3062 static int __net_init ipmr_net_init(struct net *net)
3063 {
3064         int err;
3065
3066         err = ipmr_notifier_init(net);
3067         if (err)
3068                 goto ipmr_notifier_fail;
3069
3070         err = ipmr_rules_init(net);
3071         if (err < 0)
3072                 goto ipmr_rules_fail;
3073
3074 #ifdef CONFIG_PROC_FS
3075         err = -ENOMEM;
3076         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
3077                         sizeof(struct mr_vif_iter)))
3078                 goto proc_vif_fail;
3079         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
3080                         sizeof(struct mr_mfc_iter)))
3081                 goto proc_cache_fail;
3082 #endif
3083         return 0;
3084
3085 #ifdef CONFIG_PROC_FS
3086 proc_cache_fail:
3087         remove_proc_entry("ip_mr_vif", net->proc_net);
3088 proc_vif_fail:
3089         ipmr_rules_exit(net);
3090 #endif
3091 ipmr_rules_fail:
3092         ipmr_notifier_exit(net);
3093 ipmr_notifier_fail:
3094         return err;
3095 }
3096
3097 static void __net_exit ipmr_net_exit(struct net *net)
3098 {
3099 #ifdef CONFIG_PROC_FS
3100         remove_proc_entry("ip_mr_cache", net->proc_net);
3101         remove_proc_entry("ip_mr_vif", net->proc_net);
3102 #endif
3103         ipmr_notifier_exit(net);
3104         ipmr_rules_exit(net);
3105 }
3106
3107 static struct pernet_operations ipmr_net_ops = {
3108         .init = ipmr_net_init,
3109         .exit = ipmr_net_exit,
3110 };
3111
3112 int __init ip_mr_init(void)
3113 {
3114         int err;
3115
3116         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3117                                        sizeof(struct mfc_cache),
3118                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3119                                        NULL);
3120
3121         err = register_pernet_subsys(&ipmr_net_ops);
3122         if (err)
3123                 goto reg_pernet_fail;
3124
3125         err = register_netdevice_notifier(&ip_mr_notifier);
3126         if (err)
3127                 goto reg_notif_fail;
3128 #ifdef CONFIG_IP_PIMSM_V2
3129         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3130                 pr_err("%s: can't add PIM protocol\n", __func__);
3131                 err = -EAGAIN;
3132                 goto add_proto_fail;
3133         }
3134 #endif
3135         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3136                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3137         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3138                       ipmr_rtm_route, NULL, 0);
3139         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3140                       ipmr_rtm_route, NULL, 0);
3141
3142         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3143                       NULL, ipmr_rtm_dumplink, 0);
3144         return 0;
3145
3146 #ifdef CONFIG_IP_PIMSM_V2
3147 add_proto_fail:
3148         unregister_netdevice_notifier(&ip_mr_notifier);
3149 #endif
3150 reg_notif_fail:
3151         unregister_pernet_subsys(&ipmr_net_ops);
3152 reg_pernet_fail:
3153         kmem_cache_destroy(mrt_cachep);
3154         return err;
3155 }