net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <[email protected]>
   5  *        Linux Consultancy and Custom Driver Development
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *      Fixes:
  13  *      Michael Chastain        :       Incorrect size of copying.
  14  *      Alan Cox                :       Added the cache manager code
  15  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  16  *      Mike McLagan            :       Routing by source
  17  *      Malcolm Beattie         :       Buffer handling fixes.
  18  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  19  *      SVR Anand               :       Fixed several multicast bugs and problems.
  20  *      Alexey Kuznetsov        :       Status, optimisations and more.
  21  *      Brad Parker             :       Better behaviour on mrouted upcall
  22  *                                      overflow.
  23  *      Carlos Picoto           :       PIMv1 Support
  24  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  25  *                                      Relax this requirement to work with older peers.
  26  *
  27  */
  28
  29 #include <linux/uaccess.h>
  30 #include <linux/types.h>
  31 #include <linux/cache.h>
  32 #include <linux/capability.h>
  33 #include <linux/errno.h>
  34 #include <linux/mm.h>
  35 #include <linux/kernel.h>
  36 #include <linux/fcntl.h>
  37 #include <linux/stat.h>
  38 #include <linux/socket.h>
  39 #include <linux/in.h>
  40 #include <linux/inet.h>
  41 #include <linux/netdevice.h>
  42 #include <linux/inetdevice.h>
  43 #include <linux/igmp.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/mroute.h>
  47 #include <linux/init.h>
  48 #include <linux/if_ether.h>
  49 #include <linux/slab.h>
  50 #include <net/net_namespace.h>
  51 #include <net/ip.h>
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54 #include <net/route.h>
  55 #include <net/icmp.h>
  56 #include <net/udp.h>
  57 #include <net/raw.h>
  58 #include <linux/notifier.h>
  59 #include <linux/if_arp.h>
  60 #include <linux/netfilter_ipv4.h>
  61 #include <linux/compat.h>
  62 #include <linux/export.h>
  63 #include <net/ip_tunnels.h>
  64 #include <net/checksum.h>
  65 #include <net/netlink.h>
  66 #include <net/fib_rules.h>
  67 #include <linux/netconf.h>
  68 #include <net/nexthop.h>
  69 #include <net/switchdev.h>
  70
  71 struct ipmr_rule {
  72         struct fib_rule         common;
  73 };
  74
  75 struct ipmr_result {
  76         struct mr_table         *mrt;
  77 };
  78
  79 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  80  * Note that the changes are semaphored via rtnl_lock.
  81  */
  82
  83 static DEFINE_RWLOCK(mrt_lock);
  84
  85 /* Multicast router control variables */
  86
  87 /* Special spinlock for queue of unresolved entries */
  88 static DEFINE_SPINLOCK(mfc_unres_lock);
  89
  90 /* We return to original Alan's scheme. Hash table of resolved
  91  * entries is changed only in process context and protected
  92  * with weak lock mrt_lock. Queue of unresolved entries is protected
  93  * with strong spinlock mfc_unres_lock.
  94  *
  95  * In this case data path is free of exclusive locks at all.
  96  */
  97
  98 static struct kmem_cache *mrt_cachep __ro_after_init;
  99
 100 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 101 static void ipmr_free_table(struct mr_table *mrt);
 102
 103 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 104                           struct net_device *dev, struct sk_buff *skb,
 105                           struct mfc_cache *cache, int local);
 106 static int ipmr_cache_report(struct mr_table *mrt,
 107                              struct sk_buff *pkt, vifi_t vifi, int assert);
 108 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 109                               struct mr_mfc *c, struct rtmsg *rtm);
 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_get_table(struct net *net, u32 id)
 121 {
 122         struct mr_table *mrt;
 123
 124         ipmr_for_each_table(mrt, net) {
 125                 if (mrt->id == id)
 126                         return mrt;
 127         }
 128         return NULL;
 129 }
 130
 131 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 132                            struct mr_table **mrt)
 133 {
 134         int err;
 135         struct ipmr_result res;
 136         struct fib_lookup_arg arg = {
 137                 .result = &res,
 138                 .flags = FIB_LOOKUP_NOREF,
 139         };
 140
 141         /* update flow if oif or iif point to device enslaved to l3mdev */
 142         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 143
 144         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 145                                flowi4_to_flowi(flp4), 0, &arg);
 146         if (err < 0)
 147                 return err;
 148         *mrt = res.mrt;
 149         return 0;
 150 }
 151
 152 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 153                             int flags, struct fib_lookup_arg *arg)
 154 {
 155         struct ipmr_result *res = arg->result;
 156         struct mr_table *mrt;
 157
 158         switch (rule->action) {
 159         case FR_ACT_TO_TBL:
 160                 break;
 161         case FR_ACT_UNREACHABLE:
 162                 return -ENETUNREACH;
 163         case FR_ACT_PROHIBIT:
 164                 return -EACCES;
 165         case FR_ACT_BLACKHOLE:
 166         default:
 167                 return -EINVAL;
 168         }
 169
 170         arg->table = fib_rule_get_table(rule, arg);
 171
 172         mrt = ipmr_get_table(rule->fr_net, arg->table);
 173         if (!mrt)
 174                 return -EAGAIN;
 175         res->mrt = mrt;
 176         return 0;
 177 }
 178
 179 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 180 {
 181         return 1;
 182 }
 183
 184 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 185         FRA_GENERIC_POLICY,
 186 };
 187
 188 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 189                                struct fib_rule_hdr *frh, struct nlattr **tb)
 190 {
 191         return 0;
 192 }
 193
 194 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 195                              struct nlattr **tb)
 196 {
 197         return 1;
 198 }
 199
 200 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 201                           struct fib_rule_hdr *frh)
 202 {
 203         frh->dst_len = 0;
 204         frh->src_len = 0;
 205         frh->tos     = 0;
 206         return 0;
 207 }
 208
 209 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 210         .family         = RTNL_FAMILY_IPMR,
 211         .rule_size      = sizeof(struct ipmr_rule),
 212         .addr_size      = sizeof(u32),
 213         .action         = ipmr_rule_action,
 214         .match          = ipmr_rule_match,
 215         .configure      = ipmr_rule_configure,
 216         .compare        = ipmr_rule_compare,
 217         .fill           = ipmr_rule_fill,
 218         .nlgroup        = RTNLGRP_IPV4_RULE,
 219         .policy         = ipmr_rule_policy,
 220         .owner          = THIS_MODULE,
 221 };
 222
 223 static int __net_init ipmr_rules_init(struct net *net)
 224 {
 225         struct fib_rules_ops *ops;
 226         struct mr_table *mrt;
 227         int err;
 228
 229         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 230         if (IS_ERR(ops))
 231                 return PTR_ERR(ops);
 232
 233         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 234
 235         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 236         if (IS_ERR(mrt)) {
 237                 err = PTR_ERR(mrt);
 238                 goto err1;
 239         }
 240
 241         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 242         if (err < 0)
 243                 goto err2;
 244
 245         net->ipv4.mr_rules_ops = ops;
 246         return 0;
 247
 248 err2:
 249         ipmr_free_table(mrt);
 250 err1:
 251         fib_rules_unregister(ops);
 252         return err;
 253 }
 254
 255 static void __net_exit ipmr_rules_exit(struct net *net)
 256 {
 257         struct mr_table *mrt, *next;
 258
 259         rtnl_lock();
 260         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 261                 list_del(&mrt->list);
 262                 ipmr_free_table(mrt);
 263         }
 264         fib_rules_unregister(net->ipv4.mr_rules_ops);
 265         rtnl_unlock();
 266 }
 267
 268 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 269 {
 270         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR);
 271 }
 272
 273 static unsigned int ipmr_rules_seq_read(struct net *net)
 274 {
 275         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 276 }
 277
 278 bool ipmr_rule_default(const struct fib_rule *rule)
 279 {
 280         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 281 }
 282 EXPORT_SYMBOL(ipmr_rule_default);
 283 #else
 284 #define ipmr_for_each_table(mrt, net) \
 285         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 286
 287 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 288 {
 289         return net->ipv4.mrt;
 290 }
 291
 292 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 293                            struct mr_table **mrt)
 294 {
 295         *mrt = net->ipv4.mrt;
 296         return 0;
 297 }
 298
 299 static int __net_init ipmr_rules_init(struct net *net)
 300 {
 301         struct mr_table *mrt;
 302
 303         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 304         if (IS_ERR(mrt))
 305                 return PTR_ERR(mrt);
 306         net->ipv4.mrt = mrt;
 307         return 0;
 308 }
 309
 310 static void __net_exit ipmr_rules_exit(struct net *net)
 311 {
 312         rtnl_lock();
 313         ipmr_free_table(net->ipv4.mrt);
 314         net->ipv4.mrt = NULL;
 315         rtnl_unlock();
 316 }
 317
 318 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb)
 319 {
 320         return 0;
 321 }
 322
 323 static unsigned int ipmr_rules_seq_read(struct net *net)
 324 {
 325         return 0;
 326 }
 327
 328 bool ipmr_rule_default(const struct fib_rule *rule)
 329 {
 330         return true;
 331 }
 332 EXPORT_SYMBOL(ipmr_rule_default);
 333 #endif
 334
 335 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 336                                 const void *ptr)
 337 {
 338         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 339         struct mfc_cache *c = (struct mfc_cache *)ptr;
 340
 341         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 342                cmparg->mfc_origin != c->mfc_origin;
 343 }
 344
 345 static const struct rhashtable_params ipmr_rht_params = {
 346         .head_offset = offsetof(struct mr_mfc, mnode),
 347         .key_offset = offsetof(struct mfc_cache, cmparg),
 348         .key_len = sizeof(struct mfc_cache_cmp_arg),
 349         .nelem_hint = 3,
 350         .locks_mul = 1,
 351         .obj_cmpfn = ipmr_hash_cmp,
 352         .automatic_shrinking = true,
 353 };
 354
 355 static void ipmr_new_table_set(struct mr_table *mrt,
 356                                struct net *net)
 357 {
 358 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 359         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 360 #endif
 361 }
 362
 363 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 364         .mfc_mcastgrp = htonl(INADDR_ANY),
 365         .mfc_origin = htonl(INADDR_ANY),
 366 };
 367
 368 static struct mr_table_ops ipmr_mr_table_ops = {
 369         .rht_params = &ipmr_rht_params,
 370         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 371 };
 372
 373 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 374 {
 375         struct mr_table *mrt;
 376
 377         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 378         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 379                 return ERR_PTR(-EINVAL);
 380
 381         mrt = ipmr_get_table(net, id);
 382         if (mrt)
 383                 return mrt;
 384
 385         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 386                               ipmr_expire_process, ipmr_new_table_set);
 387 }
 388
 389 static void ipmr_free_table(struct mr_table *mrt)
 390 {
 391         del_timer_sync(&mrt->ipmr_expire_timer);
 392         mroute_clean_tables(mrt, true);
 393         rhltable_destroy(&mrt->mfc_hash);
 394         kfree(mrt);
 395 }
 396
 397 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 398
 399 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 400 {
 401         struct net *net = dev_net(dev);
 402
 403         dev_close(dev);
 404
 405         dev = __dev_get_by_name(net, "tunl0");
 406         if (dev) {
 407                 const struct net_device_ops *ops = dev->netdev_ops;
 408                 struct ifreq ifr;
 409                 struct ip_tunnel_parm p;
 410
 411                 memset(&p, 0, sizeof(p));
 412                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 413                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 414                 p.iph.version = 4;
 415                 p.iph.ihl = 5;
 416                 p.iph.protocol = IPPROTO_IPIP;
 417                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 418                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 419
 420                 if (ops->ndo_do_ioctl) {
 421                         mm_segment_t oldfs = get_fs();
 422
 423                         set_fs(KERNEL_DS);
 424                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 425                         set_fs(oldfs);
 426                 }
 427         }
 428 }
 429
 430 /* Initialize ipmr pimreg/tunnel in_device */
 431 static bool ipmr_init_vif_indev(const struct net_device *dev)
 432 {
 433         struct in_device *in_dev;
 434
 435         ASSERT_RTNL();
 436
 437         in_dev = __in_dev_get_rtnl(dev);
 438         if (!in_dev)
 439                 return false;
 440         ipv4_devconf_setall(in_dev);
 441         neigh_parms_data_state_setall(in_dev->arp_parms);
 442         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 443
 444         return true;
 445 }
 446
 447 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 448 {
 449         struct net_device  *dev;
 450
 451         dev = __dev_get_by_name(net, "tunl0");
 452
 453         if (dev) {
 454                 const struct net_device_ops *ops = dev->netdev_ops;
 455                 int err;
 456                 struct ifreq ifr;
 457                 struct ip_tunnel_parm p;
 458
 459                 memset(&p, 0, sizeof(p));
 460                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 461                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 462                 p.iph.version = 4;
 463                 p.iph.ihl = 5;
 464                 p.iph.protocol = IPPROTO_IPIP;
 465                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 466                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 467
 468                 if (ops->ndo_do_ioctl) {
 469                         mm_segment_t oldfs = get_fs();
 470
 471                         set_fs(KERNEL_DS);
 472                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 473                         set_fs(oldfs);
 474                 } else {
 475                         err = -EOPNOTSUPP;
 476                 }
 477                 dev = NULL;
 478
 479                 if (err == 0 &&
 480                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 481                         dev->flags |= IFF_MULTICAST;
 482                         if (!ipmr_init_vif_indev(dev))
 483                                 goto failure;
 484                         if (dev_open(dev))
 485                                 goto failure;
 486                         dev_hold(dev);
 487                 }
 488         }
 489         return dev;
 490
 491 failure:
 492         unregister_netdevice(dev);
 493         return NULL;
 494 }
 495
 496 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 497 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 498 {
 499         struct net *net = dev_net(dev);
 500         struct mr_table *mrt;
 501         struct flowi4 fl4 = {
 502                 .flowi4_oif     = dev->ifindex,
 503                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 504                 .flowi4_mark    = skb->mark,
 505         };
 506         int err;
 507
 508         err = ipmr_fib_lookup(net, &fl4, &mrt);
 509         if (err < 0) {
 510                 kfree_skb(skb);
 511                 return err;
 512         }
 513
 514         read_lock(&mrt_lock);
 515         dev->stats.tx_bytes += skb->len;
 516         dev->stats.tx_packets++;
 517         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 518         read_unlock(&mrt_lock);
 519         kfree_skb(skb);
 520         return NETDEV_TX_OK;
 521 }
 522
 523 static int reg_vif_get_iflink(const struct net_device *dev)
 524 {
 525         return 0;
 526 }
 527
 528 static const struct net_device_ops reg_vif_netdev_ops = {
 529         .ndo_start_xmit = reg_vif_xmit,
 530         .ndo_get_iflink = reg_vif_get_iflink,
 531 };
 532
 533 static void reg_vif_setup(struct net_device *dev)
 534 {
 535         dev->type               = ARPHRD_PIMREG;
 536         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 537         dev->flags              = IFF_NOARP;
 538         dev->netdev_ops         = &reg_vif_netdev_ops;
 539         dev->needs_free_netdev  = true;
 540         dev->features           |= NETIF_F_NETNS_LOCAL;
 541 }
 542
 543 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 544 {
 545         struct net_device *dev;
 546         char name[IFNAMSIZ];
 547
 548         if (mrt->id == RT_TABLE_DEFAULT)
 549                 sprintf(name, "pimreg");
 550         else
 551                 sprintf(name, "pimreg%u", mrt->id);
 552
 553         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 554
 555         if (!dev)
 556                 return NULL;
 557
 558         dev_net_set(dev, net);
 559
 560         if (register_netdevice(dev)) {
 561                 free_netdev(dev);
 562                 return NULL;
 563         }
 564
 565         if (!ipmr_init_vif_indev(dev))
 566                 goto failure;
 567         if (dev_open(dev))
 568                 goto failure;
 569
 570         dev_hold(dev);
 571
 572         return dev;
 573
 574 failure:
 575         unregister_netdevice(dev);
 576         return NULL;
 577 }
 578
 579 /* called with rcu_read_lock() */
 580 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 581                      unsigned int pimlen)
 582 {
 583         struct net_device *reg_dev = NULL;
 584         struct iphdr *encap;
 585
 586         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 587         /* Check that:
 588          * a. packet is really sent to a multicast group
 589          * b. packet is not a NULL-REGISTER
 590          * c. packet is not truncated
 591          */
 592         if (!ipv4_is_multicast(encap->daddr) ||
 593             encap->tot_len == 0 ||
 594             ntohs(encap->tot_len) + pimlen > skb->len)
 595                 return 1;
 596
 597         read_lock(&mrt_lock);
 598         if (mrt->mroute_reg_vif_num >= 0)
 599                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 600         read_unlock(&mrt_lock);
 601
 602         if (!reg_dev)
 603                 return 1;
 604
 605         skb->mac_header = skb->network_header;
 606         skb_pull(skb, (u8 *)encap - skb->data);
 607         skb_reset_network_header(skb);
 608         skb->protocol = htons(ETH_P_IP);
 609         skb->ip_summed = CHECKSUM_NONE;
 610
 611         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 612
 613         netif_rx(skb);
 614
 615         return NET_RX_SUCCESS;
 616 }
 617 #else
 618 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 619 {
 620         return NULL;
 621 }
 622 #endif
 623
 624 static int call_ipmr_vif_entry_notifier(struct notifier_block *nb,
 625                                         struct net *net,
 626                                         enum fib_event_type event_type,
 627                                         struct vif_device *vif,
 628                                         vifi_t vif_index, u32 tb_id)
 629 {
 630         struct vif_entry_notifier_info info = {
 631                 .info = {
 632                         .family = RTNL_FAMILY_IPMR,
 633                         .net = net,
 634                 },
 635                 .dev = vif->dev,
 636                 .vif_index = vif_index,
 637                 .vif_flags = vif->flags,
 638                 .tb_id = tb_id,
 639         };
 640
 641         return call_fib_notifier(nb, net, event_type, &info.info);
 642 }
 643
 644 static int call_ipmr_vif_entry_notifiers(struct net *net,
 645                                          enum fib_event_type event_type,
 646                                          struct vif_device *vif,
 647                                          vifi_t vif_index, u32 tb_id)
 648 {
 649         struct vif_entry_notifier_info info = {
 650                 .info = {
 651                         .family = RTNL_FAMILY_IPMR,
 652                         .net = net,
 653                 },
 654                 .dev = vif->dev,
 655                 .vif_index = vif_index,
 656                 .vif_flags = vif->flags,
 657                 .tb_id = tb_id,
 658         };
 659
 660         ASSERT_RTNL();
 661         net->ipv4.ipmr_seq++;
 662         return call_fib_notifiers(net, event_type, &info.info);
 663 }
 664
 665 static int call_ipmr_mfc_entry_notifier(struct notifier_block *nb,
 666                                         struct net *net,
 667                                         enum fib_event_type event_type,
 668                                         struct mfc_cache *mfc, u32 tb_id)
 669 {
 670         struct mfc_entry_notifier_info info = {
 671                 .info = {
 672                         .family = RTNL_FAMILY_IPMR,
 673                         .net = net,
 674                 },
 675                 .mfc = mfc,
 676                 .tb_id = tb_id
 677         };
 678
 679         return call_fib_notifier(nb, net, event_type, &info.info);
 680 }
 681
 682 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 683                                          enum fib_event_type event_type,
 684                                          struct mfc_cache *mfc, u32 tb_id)
 685 {
 686         struct mfc_entry_notifier_info info = {
 687                 .info = {
 688                         .family = RTNL_FAMILY_IPMR,
 689                         .net = net,
 690                 },
 691                 .mfc = mfc,
 692                 .tb_id = tb_id
 693         };
 694
 695         ASSERT_RTNL();
 696         net->ipv4.ipmr_seq++;
 697         return call_fib_notifiers(net, event_type, &info.info);
 698 }
 699
 700 /**
 701  *      vif_delete - Delete a VIF entry
 702  *      @notify: Set to 1, if the caller is a notifier_call
 703  */
 704 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 705                       struct list_head *head)
 706 {
 707         struct net *net = read_pnet(&mrt->net);
 708         struct vif_device *v;
 709         struct net_device *dev;
 710         struct in_device *in_dev;
 711
 712         if (vifi < 0 || vifi >= mrt->maxvif)
 713                 return -EADDRNOTAVAIL;
 714
 715         v = &mrt->vif_table[vifi];
 716
 717         if (VIF_EXISTS(mrt, vifi))
 718                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 719                                               mrt->id);
 720
 721         write_lock_bh(&mrt_lock);
 722         dev = v->dev;
 723         v->dev = NULL;
 724
 725         if (!dev) {
 726                 write_unlock_bh(&mrt_lock);
 727                 return -EADDRNOTAVAIL;
 728         }
 729
 730         if (vifi == mrt->mroute_reg_vif_num)
 731                 mrt->mroute_reg_vif_num = -1;
 732
 733         if (vifi + 1 == mrt->maxvif) {
 734                 int tmp;
 735
 736                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 737                         if (VIF_EXISTS(mrt, tmp))
 738                                 break;
 739                 }
 740                 mrt->maxvif = tmp+1;
 741         }
 742
 743         write_unlock_bh(&mrt_lock);
 744
 745         dev_set_allmulti(dev, -1);
 746
 747         in_dev = __in_dev_get_rtnl(dev);
 748         if (in_dev) {
 749                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 750                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 751                                             NETCONFA_MC_FORWARDING,
 752                                             dev->ifindex, &in_dev->cnf);
 753                 ip_rt_multicast_event(in_dev);
 754         }
 755
 756         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 757                 unregister_netdevice_queue(dev, head);
 758
 759         dev_put(dev);
 760         return 0;
 761 }
 762
 763 static void ipmr_cache_free_rcu(struct rcu_head *head)
 764 {
 765         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 766
 767         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 768 }
 769
 770 void ipmr_cache_free(struct mfc_cache *c)
 771 {
 772         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 773 }
 774 EXPORT_SYMBOL(ipmr_cache_free);
 775
 776 /* Destroy an unresolved cache entry, killing queued skbs
 777  * and reporting error to netlink readers.
 778  */
 779 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 780 {
 781         struct net *net = read_pnet(&mrt->net);
 782         struct sk_buff *skb;
 783         struct nlmsgerr *e;
 784
 785         atomic_dec(&mrt->cache_resolve_queue_len);
 786
 787         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 788                 if (ip_hdr(skb)->version == 0) {
 789                         struct nlmsghdr *nlh = skb_pull(skb,
 790                                                         sizeof(struct iphdr));
 791                         nlh->nlmsg_type = NLMSG_ERROR;
 792                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 793                         skb_trim(skb, nlh->nlmsg_len);
 794                         e = nlmsg_data(nlh);
 795                         e->error = -ETIMEDOUT;
 796                         memset(&e->msg, 0, sizeof(e->msg));
 797
 798                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 799                 } else {
 800                         kfree_skb(skb);
 801                 }
 802         }
 803
 804         ipmr_cache_free(c);
 805 }
 806
 807 /* Timer process for the unresolved queue. */
 808 static void ipmr_expire_process(struct timer_list *t)
 809 {
 810         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 811         struct mr_mfc *c, *next;
 812         unsigned long expires;
 813         unsigned long now;
 814
 815         if (!spin_trylock(&mfc_unres_lock)) {
 816                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 817                 return;
 818         }
 819
 820         if (list_empty(&mrt->mfc_unres_queue))
 821                 goto out;
 822
 823         now = jiffies;
 824         expires = 10*HZ;
 825
 826         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 827                 if (time_after(c->mfc_un.unres.expires, now)) {
 828                         unsigned long interval = c->mfc_un.unres.expires - now;
 829                         if (interval < expires)
 830                                 expires = interval;
 831                         continue;
 832                 }
 833
 834                 list_del(&c->list);
 835                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 836                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 837         }
 838
 839         if (!list_empty(&mrt->mfc_unres_queue))
 840                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 841
 842 out:
 843         spin_unlock(&mfc_unres_lock);
 844 }
 845
 846 /* Fill oifs list. It is called under write locked mrt_lock. */
 847 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 848                                    unsigned char *ttls)
 849 {
 850         int vifi;
 851
 852         cache->mfc_un.res.minvif = MAXVIFS;
 853         cache->mfc_un.res.maxvif = 0;
 854         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 855
 856         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 857                 if (VIF_EXISTS(mrt, vifi) &&
 858                     ttls[vifi] && ttls[vifi] < 255) {
 859                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 860                         if (cache->mfc_un.res.minvif > vifi)
 861                                 cache->mfc_un.res.minvif = vifi;
 862                         if (cache->mfc_un.res.maxvif <= vifi)
 863                                 cache->mfc_un.res.maxvif = vifi + 1;
 864                 }
 865         }
 866         cache->mfc_un.res.lastuse = jiffies;
 867 }
 868
 869 static int vif_add(struct net *net, struct mr_table *mrt,
 870                    struct vifctl *vifc, int mrtsock)
 871 {
 872         int vifi = vifc->vifc_vifi;
 873         struct switchdev_attr attr = {
 874                 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
 875         };
 876         struct vif_device *v = &mrt->vif_table[vifi];
 877         struct net_device *dev;
 878         struct in_device *in_dev;
 879         int err;
 880
 881         /* Is vif busy ? */
 882         if (VIF_EXISTS(mrt, vifi))
 883                 return -EADDRINUSE;
 884
 885         switch (vifc->vifc_flags) {
 886         case VIFF_REGISTER:
 887                 if (!ipmr_pimsm_enabled())
 888                         return -EINVAL;
 889                 /* Special Purpose VIF in PIM
 890                  * All the packets will be sent to the daemon
 891                  */
 892                 if (mrt->mroute_reg_vif_num >= 0)
 893                         return -EADDRINUSE;
 894                 dev = ipmr_reg_vif(net, mrt);
 895                 if (!dev)
 896                         return -ENOBUFS;
 897                 err = dev_set_allmulti(dev, 1);
 898                 if (err) {
 899                         unregister_netdevice(dev);
 900                         dev_put(dev);
 901                         return err;
 902                 }
 903                 break;
 904         case VIFF_TUNNEL:
 905                 dev = ipmr_new_tunnel(net, vifc);
 906                 if (!dev)
 907                         return -ENOBUFS;
 908                 err = dev_set_allmulti(dev, 1);
 909                 if (err) {
 910                         ipmr_del_tunnel(dev, vifc);
 911                         dev_put(dev);
 912                         return err;
 913                 }
 914                 break;
 915         case VIFF_USE_IFINDEX:
 916         case 0:
 917                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 918                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 919                         if (dev && !__in_dev_get_rtnl(dev)) {
 920                                 dev_put(dev);
 921                                 return -EADDRNOTAVAIL;
 922                         }
 923                 } else {
 924                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 925                 }
 926                 if (!dev)
 927                         return -EADDRNOTAVAIL;
 928                 err = dev_set_allmulti(dev, 1);
 929                 if (err) {
 930                         dev_put(dev);
 931                         return err;
 932                 }
 933                 break;
 934         default:
 935                 return -EINVAL;
 936         }
 937
 938         in_dev = __in_dev_get_rtnl(dev);
 939         if (!in_dev) {
 940                 dev_put(dev);
 941                 return -EADDRNOTAVAIL;
 942         }
 943         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 944         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 945                                     dev->ifindex, &in_dev->cnf);
 946         ip_rt_multicast_event(in_dev);
 947
 948         /* Fill in the VIF structures */
 949         vif_device_init(v, dev, vifc->vifc_rate_limit,
 950                         vifc->vifc_threshold,
 951                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 952                         (VIFF_TUNNEL | VIFF_REGISTER));
 953
 954         attr.orig_dev = dev;
 955         if (!switchdev_port_attr_get(dev, &attr)) {
 956                 memcpy(v->dev_parent_id.id, attr.u.ppid.id, attr.u.ppid.id_len);
 957                 v->dev_parent_id.id_len = attr.u.ppid.id_len;
 958         } else {
 959                 v->dev_parent_id.id_len = 0;
 960         }
 961
 962         v->local = vifc->vifc_lcl_addr.s_addr;
 963         v->remote = vifc->vifc_rmt_addr.s_addr;
 964
 965         /* And finish update writing critical data */
 966         write_lock_bh(&mrt_lock);
 967         v->dev = dev;
 968         if (v->flags & VIFF_REGISTER)
 969                 mrt->mroute_reg_vif_num = vifi;
 970         if (vifi+1 > mrt->maxvif)
 971                 mrt->maxvif = vifi+1;
 972         write_unlock_bh(&mrt_lock);
 973         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 974         return 0;
 975 }
 976
 977 /* called with rcu_read_lock() */
 978 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 979                                          __be32 origin,
 980                                          __be32 mcastgrp)
 981 {
 982         struct mfc_cache_cmp_arg arg = {
 983                         .mfc_mcastgrp = mcastgrp,
 984                         .mfc_origin = origin
 985         };
 986
 987         return mr_mfc_find(mrt, &arg);
 988 }
 989
 990 /* Look for a (*,G) entry */
 991 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 992                                              __be32 mcastgrp, int vifi)
 993 {
 994         struct mfc_cache_cmp_arg arg = {
 995                         .mfc_mcastgrp = mcastgrp,
 996                         .mfc_origin = htonl(INADDR_ANY)
 997         };
 998
 999         if (mcastgrp == htonl(INADDR_ANY))
1000                 return mr_mfc_find_any_parent(mrt, vifi);
1001         return mr_mfc_find_any(mrt, vifi, &arg);
1002 }
1003
1004 /* Look for a (S,G,iif) entry if parent != -1 */
1005 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
1006                                                 __be32 origin, __be32 mcastgrp,
1007                                                 int parent)
1008 {
1009         struct mfc_cache_cmp_arg arg = {
1010                         .mfc_mcastgrp = mcastgrp,
1011                         .mfc_origin = origin,
1012         };
1013
1014         return mr_mfc_find_parent(mrt, &arg, parent);
1015 }
1016
1017 /* Allocate a multicast cache entry */
1018 static struct mfc_cache *ipmr_cache_alloc(void)
1019 {
1020         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
1021
1022         if (c) {
1023                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
1024                 c->_c.mfc_un.res.minvif = MAXVIFS;
1025                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
1026         }
1027         return c;
1028 }
1029
1030 static struct mfc_cache *ipmr_cache_alloc_unres(void)
1031 {
1032         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
1033
1034         if (c) {
1035                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
1036                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
1037         }
1038         return c;
1039 }
1040
1041 /* A cache entry has gone into a resolved state from queued */
1042 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
1043                                struct mfc_cache *uc, struct mfc_cache *c)
1044 {
1045         struct sk_buff *skb;
1046         struct nlmsgerr *e;
1047
1048         /* Play the pending entries through our router */
1049         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
1050                 if (ip_hdr(skb)->version == 0) {
1051                         struct nlmsghdr *nlh = skb_pull(skb,
1052                                                         sizeof(struct iphdr));
1053
1054                         if (__ipmr_fill_mroute(mrt, skb, &c->_c,
1055                                                nlmsg_data(nlh)) > 0) {
1056                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
1057                                                  (u8 *)nlh;
1058                         } else {
1059                                 nlh->nlmsg_type = NLMSG_ERROR;
1060                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
1061                                 skb_trim(skb, nlh->nlmsg_len);
1062                                 e = nlmsg_data(nlh);
1063                                 e->error = -EMSGSIZE;
1064                                 memset(&e->msg, 0, sizeof(e->msg));
1065                         }
1066
1067                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1068                 } else {
1069                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1070                 }
1071         }
1072 }
1073
1074 /* Bounce a cache query up to mrouted and netlink.
1075  *
1076  * Called under mrt_lock.
1077  */
1078 static int ipmr_cache_report(struct mr_table *mrt,
1079                              struct sk_buff *pkt, vifi_t vifi, int assert)
1080 {
1081         const int ihl = ip_hdrlen(pkt);
1082         struct sock *mroute_sk;
1083         struct igmphdr *igmp;
1084         struct igmpmsg *msg;
1085         struct sk_buff *skb;
1086         int ret;
1087
1088         if (assert == IGMPMSG_WHOLEPKT)
1089                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1090         else
1091                 skb = alloc_skb(128, GFP_ATOMIC);
1092
1093         if (!skb)
1094                 return -ENOBUFS;
1095
1096         if (assert == IGMPMSG_WHOLEPKT) {
1097                 /* Ugly, but we have no choice with this interface.
1098                  * Duplicate old header, fix ihl, length etc.
1099                  * And all this only to mangle msg->im_msgtype and
1100                  * to set msg->im_mbz to "mbz" :-)
1101                  */
1102                 skb_push(skb, sizeof(struct iphdr));
1103                 skb_reset_network_header(skb);
1104                 skb_reset_transport_header(skb);
1105                 msg = (struct igmpmsg *)skb_network_header(skb);
1106                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1107                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
1108                 msg->im_mbz = 0;
1109                 msg->im_vif = mrt->mroute_reg_vif_num;
1110                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1111                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1112                                              sizeof(struct iphdr));
1113         } else {
1114                 /* Copy the IP header */
1115                 skb_set_network_header(skb, skb->len);
1116                 skb_put(skb, ihl);
1117                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1118                 /* Flag to the kernel this is a route add */
1119                 ip_hdr(skb)->protocol = 0;
1120                 msg = (struct igmpmsg *)skb_network_header(skb);
1121                 msg->im_vif = vifi;
1122                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1123                 /* Add our header */
1124                 igmp = skb_put(skb, sizeof(struct igmphdr));
1125                 igmp->type = assert;
1126                 msg->im_msgtype = assert;
1127                 igmp->code = 0;
1128                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1129                 skb->transport_header = skb->network_header;
1130         }
1131
1132         rcu_read_lock();
1133         mroute_sk = rcu_dereference(mrt->mroute_sk);
1134         if (!mroute_sk) {
1135                 rcu_read_unlock();
1136                 kfree_skb(skb);
1137                 return -EINVAL;
1138         }
1139
1140         igmpmsg_netlink_event(mrt, skb);
1141
1142         /* Deliver to mrouted */
1143         ret = sock_queue_rcv_skb(mroute_sk, skb);
1144         rcu_read_unlock();
1145         if (ret < 0) {
1146                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1147                 kfree_skb(skb);
1148         }
1149
1150         return ret;
1151 }
1152
1153 /* Queue a packet for resolution. It gets locked cache entry! */
1154 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1155                                  struct sk_buff *skb, struct net_device *dev)
1156 {
1157         const struct iphdr *iph = ip_hdr(skb);
1158         struct mfc_cache *c;
1159         bool found = false;
1160         int err;
1161
1162         spin_lock_bh(&mfc_unres_lock);
1163         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1164                 if (c->mfc_mcastgrp == iph->daddr &&
1165                     c->mfc_origin == iph->saddr) {
1166                         found = true;
1167                         break;
1168                 }
1169         }
1170
1171         if (!found) {
1172                 /* Create a new entry if allowable */
1173                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1174                     (c = ipmr_cache_alloc_unres()) == NULL) {
1175                         spin_unlock_bh(&mfc_unres_lock);
1176
1177                         kfree_skb(skb);
1178                         return -ENOBUFS;
1179                 }
1180
1181                 /* Fill in the new cache entry */
1182                 c->_c.mfc_parent = -1;
1183                 c->mfc_origin   = iph->saddr;
1184                 c->mfc_mcastgrp = iph->daddr;
1185
1186                 /* Reflect first query at mrouted. */
1187                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1188
1189                 if (err < 0) {
1190                         /* If the report failed throw the cache entry
1191                            out - Brad Parker
1192                          */
1193                         spin_unlock_bh(&mfc_unres_lock);
1194
1195                         ipmr_cache_free(c);
1196                         kfree_skb(skb);
1197                         return err;
1198                 }
1199
1200                 atomic_inc(&mrt->cache_resolve_queue_len);
1201                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1202                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1203
1204                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1205                         mod_timer(&mrt->ipmr_expire_timer,
1206                                   c->_c.mfc_un.unres.expires);
1207         }
1208
1209         /* See if we can append the packet */
1210         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1211                 kfree_skb(skb);
1212                 err = -ENOBUFS;
1213         } else {
1214                 if (dev) {
1215                         skb->dev = dev;
1216                         skb->skb_iif = dev->ifindex;
1217                 }
1218                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1219                 err = 0;
1220         }
1221
1222         spin_unlock_bh(&mfc_unres_lock);
1223         return err;
1224 }
1225
1226 /* MFC cache manipulation by user space mroute daemon */
1227
1228 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1229 {
1230         struct net *net = read_pnet(&mrt->net);
1231         struct mfc_cache *c;
1232
1233         /* The entries are added/deleted only under RTNL */
1234         rcu_read_lock();
1235         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1236                                    mfc->mfcc_mcastgrp.s_addr, parent);
1237         rcu_read_unlock();
1238         if (!c)
1239                 return -ENOENT;
1240         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1241         list_del_rcu(&c->_c.list);
1242         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1243         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1244         ipmr_cache_put(c);
1245
1246         return 0;
1247 }
1248
1249 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1250                         struct mfcctl *mfc, int mrtsock, int parent)
1251 {
1252         struct mfc_cache *uc, *c;
1253         struct mr_mfc *_uc;
1254         bool found;
1255         int ret;
1256
1257         if (mfc->mfcc_parent >= MAXVIFS)
1258                 return -ENFILE;
1259
1260         /* The entries are added/deleted only under RTNL */
1261         rcu_read_lock();
1262         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1263                                    mfc->mfcc_mcastgrp.s_addr, parent);
1264         rcu_read_unlock();
1265         if (c) {
1266                 write_lock_bh(&mrt_lock);
1267                 c->_c.mfc_parent = mfc->mfcc_parent;
1268                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1269                 if (!mrtsock)
1270                         c->_c.mfc_flags |= MFC_STATIC;
1271                 write_unlock_bh(&mrt_lock);
1272                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1273                                               mrt->id);
1274                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1275                 return 0;
1276         }
1277
1278         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1279             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1280                 return -EINVAL;
1281
1282         c = ipmr_cache_alloc();
1283         if (!c)
1284                 return -ENOMEM;
1285
1286         c->mfc_origin = mfc->mfcc_origin.s_addr;
1287         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1288         c->_c.mfc_parent = mfc->mfcc_parent;
1289         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1290         if (!mrtsock)
1291                 c->_c.mfc_flags |= MFC_STATIC;
1292
1293         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1294                                   ipmr_rht_params);
1295         if (ret) {
1296                 pr_err("ipmr: rhtable insert error %d\n", ret);
1297                 ipmr_cache_free(c);
1298                 return ret;
1299         }
1300         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1301         /* Check to see if we resolved a queued list. If so we
1302          * need to send on the frames and tidy up.
1303          */
1304         found = false;
1305         spin_lock_bh(&mfc_unres_lock);
1306         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1307                 uc = (struct mfc_cache *)_uc;
1308                 if (uc->mfc_origin == c->mfc_origin &&
1309                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1310                         list_del(&_uc->list);
1311                         atomic_dec(&mrt->cache_resolve_queue_len);
1312                         found = true;
1313                         break;
1314                 }
1315         }
1316         if (list_empty(&mrt->mfc_unres_queue))
1317                 del_timer(&mrt->ipmr_expire_timer);
1318         spin_unlock_bh(&mfc_unres_lock);
1319
1320         if (found) {
1321                 ipmr_cache_resolve(net, mrt, uc, c);
1322                 ipmr_cache_free(uc);
1323         }
1324         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1325         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1326         return 0;
1327 }
1328
1329 /* Close the multicast socket, and clear the vif tables etc */
1330 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1331 {
1332         struct net *net = read_pnet(&mrt->net);
1333         struct mr_mfc *c, *tmp;
1334         struct mfc_cache *cache;
1335         LIST_HEAD(list);
1336         int i;
1337
1338         /* Shut down all active vif entries */
1339         for (i = 0; i < mrt->maxvif; i++) {
1340                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1341                         continue;
1342                 vif_delete(mrt, i, 0, &list);
1343         }
1344         unregister_netdevice_many(&list);
1345
1346         /* Wipe the cache */
1347         list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1348                 if (!all && (c->mfc_flags & MFC_STATIC))
1349                         continue;
1350                 rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1351                 list_del_rcu(&c->list);
1352                 cache = (struct mfc_cache *)c;
1353                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1354                                               mrt->id);
1355                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1356                 ipmr_cache_put(cache);
1357         }
1358
1359         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1360                 spin_lock_bh(&mfc_unres_lock);
1361                 list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1362                         list_del(&c->list);
1363                         cache = (struct mfc_cache *)c;
1364                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1365                         ipmr_destroy_unres(mrt, cache);
1366                 }
1367                 spin_unlock_bh(&mfc_unres_lock);
1368         }
1369 }
1370
1371 /* called from ip_ra_control(), before an RCU grace period,
1372  * we dont need to call synchronize_rcu() here
1373  */
1374 static void mrtsock_destruct(struct sock *sk)
1375 {
1376         struct net *net = sock_net(sk);
1377         struct mr_table *mrt;
1378
1379         ASSERT_RTNL();
1380         ipmr_for_each_table(mrt, net) {
1381                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1382                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1383                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1384                                                     NETCONFA_MC_FORWARDING,
1385                                                     NETCONFA_IFINDEX_ALL,
1386                                                     net->ipv4.devconf_all);
1387                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1388                         mroute_clean_tables(mrt, false);
1389                 }
1390         }
1391 }
1392
1393 /* Socket options and virtual interface manipulation. The whole
1394  * virtual interface system is a complete heap, but unfortunately
1395  * that's how BSD mrouted happens to think. Maybe one day with a proper
1396  * MOSPF/PIM router set up we can clean this up.
1397  */
1398
1399 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1400                          unsigned int optlen)
1401 {
1402         struct net *net = sock_net(sk);
1403         int val, ret = 0, parent = 0;
1404         struct mr_table *mrt;
1405         struct vifctl vif;
1406         struct mfcctl mfc;
1407         u32 uval;
1408
1409         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1410         rtnl_lock();
1411         if (sk->sk_type != SOCK_RAW ||
1412             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1413                 ret = -EOPNOTSUPP;
1414                 goto out_unlock;
1415         }
1416
1417         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1418         if (!mrt) {
1419                 ret = -ENOENT;
1420                 goto out_unlock;
1421         }
1422         if (optname != MRT_INIT) {
1423                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1424                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1425                         ret = -EACCES;
1426                         goto out_unlock;
1427                 }
1428         }
1429
1430         switch (optname) {
1431         case MRT_INIT:
1432                 if (optlen != sizeof(int)) {
1433                         ret = -EINVAL;
1434                         break;
1435                 }
1436                 if (rtnl_dereference(mrt->mroute_sk)) {
1437                         ret = -EADDRINUSE;
1438                         break;
1439                 }
1440
1441                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1442                 if (ret == 0) {
1443                         rcu_assign_pointer(mrt->mroute_sk, sk);
1444                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1445                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1446                                                     NETCONFA_MC_FORWARDING,
1447                                                     NETCONFA_IFINDEX_ALL,
1448                                                     net->ipv4.devconf_all);
1449                 }
1450                 break;
1451         case MRT_DONE:
1452                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1453                         ret = -EACCES;
1454                 } else {
1455                         ret = ip_ra_control(sk, 0, NULL);
1456                         goto out_unlock;
1457                 }
1458                 break;
1459         case MRT_ADD_VIF:
1460         case MRT_DEL_VIF:
1461                 if (optlen != sizeof(vif)) {
1462                         ret = -EINVAL;
1463                         break;
1464                 }
1465                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1466                         ret = -EFAULT;
1467                         break;
1468                 }
1469                 if (vif.vifc_vifi >= MAXVIFS) {
1470                         ret = -ENFILE;
1471                         break;
1472                 }
1473                 if (optname == MRT_ADD_VIF) {
1474                         ret = vif_add(net, mrt, &vif,
1475                                       sk == rtnl_dereference(mrt->mroute_sk));
1476                 } else {
1477                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1478                 }
1479                 break;
1480         /* Manipulate the forwarding caches. These live
1481          * in a sort of kernel/user symbiosis.
1482          */
1483         case MRT_ADD_MFC:
1484         case MRT_DEL_MFC:
1485                 parent = -1;
1486                 /* fall through */
1487         case MRT_ADD_MFC_PROXY:
1488         case MRT_DEL_MFC_PROXY:
1489                 if (optlen != sizeof(mfc)) {
1490                         ret = -EINVAL;
1491                         break;
1492                 }
1493                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1494                         ret = -EFAULT;
1495                         break;
1496                 }
1497                 if (parent == 0)
1498                         parent = mfc.mfcc_parent;
1499                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1500                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1501                 else
1502                         ret = ipmr_mfc_add(net, mrt, &mfc,
1503                                            sk == rtnl_dereference(mrt->mroute_sk),
1504                                            parent);
1505                 break;
1506         /* Control PIM assert. */
1507         case MRT_ASSERT:
1508                 if (optlen != sizeof(val)) {
1509                         ret = -EINVAL;
1510                         break;
1511                 }
1512                 if (get_user(val, (int __user *)optval)) {
1513                         ret = -EFAULT;
1514                         break;
1515                 }
1516                 mrt->mroute_do_assert = val;
1517                 break;
1518         case MRT_PIM:
1519                 if (!ipmr_pimsm_enabled()) {
1520                         ret = -ENOPROTOOPT;
1521                         break;
1522                 }
1523                 if (optlen != sizeof(val)) {
1524                         ret = -EINVAL;
1525                         break;
1526                 }
1527                 if (get_user(val, (int __user *)optval)) {
1528                         ret = -EFAULT;
1529                         break;
1530                 }
1531
1532                 val = !!val;
1533                 if (val != mrt->mroute_do_pim) {
1534                         mrt->mroute_do_pim = val;
1535                         mrt->mroute_do_assert = val;
1536                 }
1537                 break;
1538         case MRT_TABLE:
1539                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1540                         ret = -ENOPROTOOPT;
1541                         break;
1542                 }
1543                 if (optlen != sizeof(uval)) {
1544                         ret = -EINVAL;
1545                         break;
1546                 }
1547                 if (get_user(uval, (u32 __user *)optval)) {
1548                         ret = -EFAULT;
1549                         break;
1550                 }
1551
1552                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1553                         ret = -EBUSY;
1554                 } else {
1555                         mrt = ipmr_new_table(net, uval);
1556                         if (IS_ERR(mrt))
1557                                 ret = PTR_ERR(mrt);
1558                         else
1559                                 raw_sk(sk)->ipmr_table = uval;
1560                 }
1561                 break;
1562         /* Spurious command, or MRT_VERSION which you cannot set. */
1563         default:
1564                 ret = -ENOPROTOOPT;
1565         }
1566 out_unlock:
1567         rtnl_unlock();
1568         return ret;
1569 }
1570
1571 /* Getsock opt support for the multicast routing system. */
1572 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1573 {
1574         int olr;
1575         int val;
1576         struct net *net = sock_net(sk);
1577         struct mr_table *mrt;
1578
1579         if (sk->sk_type != SOCK_RAW ||
1580             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1581                 return -EOPNOTSUPP;
1582
1583         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1584         if (!mrt)
1585                 return -ENOENT;
1586
1587         switch (optname) {
1588         case MRT_VERSION:
1589                 val = 0x0305;
1590                 break;
1591         case MRT_PIM:
1592                 if (!ipmr_pimsm_enabled())
1593                         return -ENOPROTOOPT;
1594                 val = mrt->mroute_do_pim;
1595                 break;
1596         case MRT_ASSERT:
1597                 val = mrt->mroute_do_assert;
1598                 break;
1599         default:
1600                 return -ENOPROTOOPT;
1601         }
1602
1603         if (get_user(olr, optlen))
1604                 return -EFAULT;
1605         olr = min_t(unsigned int, olr, sizeof(int));
1606         if (olr < 0)
1607                 return -EINVAL;
1608         if (put_user(olr, optlen))
1609                 return -EFAULT;
1610         if (copy_to_user(optval, &val, olr))
1611                 return -EFAULT;
1612         return 0;
1613 }
1614
1615 /* The IP multicast ioctl support routines. */
1616 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1617 {
1618         struct sioc_sg_req sr;
1619         struct sioc_vif_req vr;
1620         struct vif_device *vif;
1621         struct mfc_cache *c;
1622         struct net *net = sock_net(sk);
1623         struct mr_table *mrt;
1624
1625         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1626         if (!mrt)
1627                 return -ENOENT;
1628
1629         switch (cmd) {
1630         case SIOCGETVIFCNT:
1631                 if (copy_from_user(&vr, arg, sizeof(vr)))
1632                         return -EFAULT;
1633                 if (vr.vifi >= mrt->maxvif)
1634                         return -EINVAL;
1635                 read_lock(&mrt_lock);
1636                 vif = &mrt->vif_table[vr.vifi];
1637                 if (VIF_EXISTS(mrt, vr.vifi)) {
1638                         vr.icount = vif->pkt_in;
1639                         vr.ocount = vif->pkt_out;
1640                         vr.ibytes = vif->bytes_in;
1641                         vr.obytes = vif->bytes_out;
1642                         read_unlock(&mrt_lock);
1643
1644                         if (copy_to_user(arg, &vr, sizeof(vr)))
1645                                 return -EFAULT;
1646                         return 0;
1647                 }
1648                 read_unlock(&mrt_lock);
1649                 return -EADDRNOTAVAIL;
1650         case SIOCGETSGCNT:
1651                 if (copy_from_user(&sr, arg, sizeof(sr)))
1652                         return -EFAULT;
1653
1654                 rcu_read_lock();
1655                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1656                 if (c) {
1657                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1658                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1659                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1660                         rcu_read_unlock();
1661
1662                         if (copy_to_user(arg, &sr, sizeof(sr)))
1663                                 return -EFAULT;
1664                         return 0;
1665                 }
1666                 rcu_read_unlock();
1667                 return -EADDRNOTAVAIL;
1668         default:
1669                 return -ENOIOCTLCMD;
1670         }
1671 }
1672
1673 #ifdef CONFIG_COMPAT
1674 struct compat_sioc_sg_req {
1675         struct in_addr src;
1676         struct in_addr grp;
1677         compat_ulong_t pktcnt;
1678         compat_ulong_t bytecnt;
1679         compat_ulong_t wrong_if;
1680 };
1681
1682 struct compat_sioc_vif_req {
1683         vifi_t  vifi;           /* Which iface */
1684         compat_ulong_t icount;
1685         compat_ulong_t ocount;
1686         compat_ulong_t ibytes;
1687         compat_ulong_t obytes;
1688 };
1689
1690 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1691 {
1692         struct compat_sioc_sg_req sr;
1693         struct compat_sioc_vif_req vr;
1694         struct vif_device *vif;
1695         struct mfc_cache *c;
1696         struct net *net = sock_net(sk);
1697         struct mr_table *mrt;
1698
1699         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1700         if (!mrt)
1701                 return -ENOENT;
1702
1703         switch (cmd) {
1704         case SIOCGETVIFCNT:
1705                 if (copy_from_user(&vr, arg, sizeof(vr)))
1706                         return -EFAULT;
1707                 if (vr.vifi >= mrt->maxvif)
1708                         return -EINVAL;
1709                 read_lock(&mrt_lock);
1710                 vif = &mrt->vif_table[vr.vifi];
1711                 if (VIF_EXISTS(mrt, vr.vifi)) {
1712                         vr.icount = vif->pkt_in;
1713                         vr.ocount = vif->pkt_out;
1714                         vr.ibytes = vif->bytes_in;
1715                         vr.obytes = vif->bytes_out;
1716                         read_unlock(&mrt_lock);
1717
1718                         if (copy_to_user(arg, &vr, sizeof(vr)))
1719                                 return -EFAULT;
1720                         return 0;
1721                 }
1722                 read_unlock(&mrt_lock);
1723                 return -EADDRNOTAVAIL;
1724         case SIOCGETSGCNT:
1725                 if (copy_from_user(&sr, arg, sizeof(sr)))
1726                         return -EFAULT;
1727
1728                 rcu_read_lock();
1729                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1730                 if (c) {
1731                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1732                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1733                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1734                         rcu_read_unlock();
1735
1736                         if (copy_to_user(arg, &sr, sizeof(sr)))
1737                                 return -EFAULT;
1738                         return 0;
1739                 }
1740                 rcu_read_unlock();
1741                 return -EADDRNOTAVAIL;
1742         default:
1743                 return -ENOIOCTLCMD;
1744         }
1745 }
1746 #endif
1747
1748 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1749 {
1750         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1751         struct net *net = dev_net(dev);
1752         struct mr_table *mrt;
1753         struct vif_device *v;
1754         int ct;
1755
1756         if (event != NETDEV_UNREGISTER)
1757                 return NOTIFY_DONE;
1758
1759         ipmr_for_each_table(mrt, net) {
1760                 v = &mrt->vif_table[0];
1761                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1762                         if (v->dev == dev)
1763                                 vif_delete(mrt, ct, 1, NULL);
1764                 }
1765         }
1766         return NOTIFY_DONE;
1767 }
1768
1769 static struct notifier_block ip_mr_notifier = {
1770         .notifier_call = ipmr_device_event,
1771 };
1772
1773 /* Encapsulate a packet by attaching a valid IPIP header to it.
1774  * This avoids tunnel drivers and other mess and gives us the speed so
1775  * important for multicast video.
1776  */
1777 static void ip_encap(struct net *net, struct sk_buff *skb,
1778                      __be32 saddr, __be32 daddr)
1779 {
1780         struct iphdr *iph;
1781         const struct iphdr *old_iph = ip_hdr(skb);
1782
1783         skb_push(skb, sizeof(struct iphdr));
1784         skb->transport_header = skb->network_header;
1785         skb_reset_network_header(skb);
1786         iph = ip_hdr(skb);
1787
1788         iph->version    =       4;
1789         iph->tos        =       old_iph->tos;
1790         iph->ttl        =       old_iph->ttl;
1791         iph->frag_off   =       0;
1792         iph->daddr      =       daddr;
1793         iph->saddr      =       saddr;
1794         iph->protocol   =       IPPROTO_IPIP;
1795         iph->ihl        =       5;
1796         iph->tot_len    =       htons(skb->len);
1797         ip_select_ident(net, skb, NULL);
1798         ip_send_check(iph);
1799
1800         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1801         nf_reset(skb);
1802 }
1803
1804 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1805                                       struct sk_buff *skb)
1806 {
1807         struct ip_options *opt = &(IPCB(skb)->opt);
1808
1809         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1810         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1811
1812         if (unlikely(opt->optlen))
1813                 ip_forward_options(skb);
1814
1815         return dst_output(net, sk, skb);
1816 }
1817
1818 #ifdef CONFIG_NET_SWITCHDEV
1819 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1820                                    int in_vifi, int out_vifi)
1821 {
1822         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1823         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1824
1825         if (!skb->offload_mr_fwd_mark)
1826                 return false;
1827         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1828                 return false;
1829         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1830                                         &in_vif->dev_parent_id);
1831 }
1832 #else
1833 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1834                                    int in_vifi, int out_vifi)
1835 {
1836         return false;
1837 }
1838 #endif
1839
1840 /* Processing handlers for ipmr_forward */
1841
1842 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1843                             int in_vifi, struct sk_buff *skb,
1844                             struct mfc_cache *c, int vifi)
1845 {
1846         const struct iphdr *iph = ip_hdr(skb);
1847         struct vif_device *vif = &mrt->vif_table[vifi];
1848         struct net_device *dev;
1849         struct rtable *rt;
1850         struct flowi4 fl4;
1851         int    encap = 0;
1852
1853         if (!vif->dev)
1854                 goto out_free;
1855
1856         if (vif->flags & VIFF_REGISTER) {
1857                 vif->pkt_out++;
1858                 vif->bytes_out += skb->len;
1859                 vif->dev->stats.tx_bytes += skb->len;
1860                 vif->dev->stats.tx_packets++;
1861                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1862                 goto out_free;
1863         }
1864
1865         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1866                 goto out_free;
1867
1868         if (vif->flags & VIFF_TUNNEL) {
1869                 rt = ip_route_output_ports(net, &fl4, NULL,
1870                                            vif->remote, vif->local,
1871                                            0, 0,
1872                                            IPPROTO_IPIP,
1873                                            RT_TOS(iph->tos), vif->link);
1874                 if (IS_ERR(rt))
1875                         goto out_free;
1876                 encap = sizeof(struct iphdr);
1877         } else {
1878                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1879                                            0, 0,
1880                                            IPPROTO_IPIP,
1881                                            RT_TOS(iph->tos), vif->link);
1882                 if (IS_ERR(rt))
1883                         goto out_free;
1884         }
1885
1886         dev = rt->dst.dev;
1887
1888         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1889                 /* Do not fragment multicasts. Alas, IPv4 does not
1890                  * allow to send ICMP, so that packets will disappear
1891                  * to blackhole.
1892                  */
1893                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1894                 ip_rt_put(rt);
1895                 goto out_free;
1896         }
1897
1898         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1899
1900         if (skb_cow(skb, encap)) {
1901                 ip_rt_put(rt);
1902                 goto out_free;
1903         }
1904
1905         vif->pkt_out++;
1906         vif->bytes_out += skb->len;
1907
1908         skb_dst_drop(skb);
1909         skb_dst_set(skb, &rt->dst);
1910         ip_decrease_ttl(ip_hdr(skb));
1911
1912         /* FIXME: forward and output firewalls used to be called here.
1913          * What do we do with netfilter? -- RR
1914          */
1915         if (vif->flags & VIFF_TUNNEL) {
1916                 ip_encap(net, skb, vif->local, vif->remote);
1917                 /* FIXME: extra output firewall step used to be here. --RR */
1918                 vif->dev->stats.tx_packets++;
1919                 vif->dev->stats.tx_bytes += skb->len;
1920         }
1921
1922         IPCB(skb)->flags |= IPSKB_FORWARDED;
1923
1924         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1925          * not only before forwarding, but after forwarding on all output
1926          * interfaces. It is clear, if mrouter runs a multicasting
1927          * program, it should receive packets not depending to what interface
1928          * program is joined.
1929          * If we will not make it, the program will have to join on all
1930          * interfaces. On the other hand, multihoming host (or router, but
1931          * not mrouter) cannot join to more than one interface - it will
1932          * result in receiving multiple packets.
1933          */
1934         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1935                 net, NULL, skb, skb->dev, dev,
1936                 ipmr_forward_finish);
1937         return;
1938
1939 out_free:
1940         kfree_skb(skb);
1941 }
1942
1943 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1944 {
1945         int ct;
1946
1947         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1948                 if (mrt->vif_table[ct].dev == dev)
1949                         break;
1950         }
1951         return ct;
1952 }
1953
1954 /* "local" means that we should preserve one skb (for local delivery) */
1955 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1956                           struct net_device *dev, struct sk_buff *skb,
1957                           struct mfc_cache *c, int local)
1958 {
1959         int true_vifi = ipmr_find_vif(mrt, dev);
1960         int psend = -1;
1961         int vif, ct;
1962
1963         vif = c->_c.mfc_parent;
1964         c->_c.mfc_un.res.pkt++;
1965         c->_c.mfc_un.res.bytes += skb->len;
1966         c->_c.mfc_un.res.lastuse = jiffies;
1967
1968         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1969                 struct mfc_cache *cache_proxy;
1970
1971                 /* For an (*,G) entry, we only check that the incomming
1972                  * interface is part of the static tree.
1973                  */
1974                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1975                 if (cache_proxy &&
1976                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1977                         goto forward;
1978         }
1979
1980         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1981         if (mrt->vif_table[vif].dev != dev) {
1982                 if (rt_is_output_route(skb_rtable(skb))) {
1983                         /* It is our own packet, looped back.
1984                          * Very complicated situation...
1985                          *
1986                          * The best workaround until routing daemons will be
1987                          * fixed is not to redistribute packet, if it was
1988                          * send through wrong interface. It means, that
1989                          * multicast applications WILL NOT work for
1990                          * (S,G), which have default multicast route pointing
1991                          * to wrong oif. In any case, it is not a good
1992                          * idea to use multicasting applications on router.
1993                          */
1994                         goto dont_forward;
1995                 }
1996
1997                 c->_c.mfc_un.res.wrong_if++;
1998
1999                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
2000                     /* pimsm uses asserts, when switching from RPT to SPT,
2001                      * so that we cannot check that packet arrived on an oif.
2002                      * It is bad, but otherwise we would need to move pretty
2003                      * large chunk of pimd to kernel. Ough... --ANK
2004                      */
2005                     (mrt->mroute_do_pim ||
2006                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
2007                     time_after(jiffies,
2008                                c->_c.mfc_un.res.last_assert +
2009                                MFC_ASSERT_THRESH)) {
2010                         c->_c.mfc_un.res.last_assert = jiffies;
2011                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
2012                 }
2013                 goto dont_forward;
2014         }
2015
2016 forward:
2017         mrt->vif_table[vif].pkt_in++;
2018         mrt->vif_table[vif].bytes_in += skb->len;
2019
2020         /* Forward the frame */
2021         if (c->mfc_origin == htonl(INADDR_ANY) &&
2022             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
2023                 if (true_vifi >= 0 &&
2024                     true_vifi != c->_c.mfc_parent &&
2025                     ip_hdr(skb)->ttl >
2026                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2027                         /* It's an (*,*) entry and the packet is not coming from
2028                          * the upstream: forward the packet to the upstream
2029                          * only.
2030                          */
2031                         psend = c->_c.mfc_parent;
2032                         goto last_forward;
2033                 }
2034                 goto dont_forward;
2035         }
2036         for (ct = c->_c.mfc_un.res.maxvif - 1;
2037              ct >= c->_c.mfc_un.res.minvif; ct--) {
2038                 /* For (*,G) entry, don't forward to the incoming interface */
2039                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2040                      ct != true_vifi) &&
2041                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2042                         if (psend != -1) {
2043                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2044
2045                                 if (skb2)
2046                                         ipmr_queue_xmit(net, mrt, true_vifi,
2047                                                         skb2, c, psend);
2048                         }
2049                         psend = ct;
2050                 }
2051         }
2052 last_forward:
2053         if (psend != -1) {
2054                 if (local) {
2055                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2056
2057                         if (skb2)
2058                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2059                                                 c, psend);
2060                 } else {
2061                         ipmr_queue_xmit(net, mrt, true_vifi, skb, c, psend);
2062                         return;
2063                 }
2064         }
2065
2066 dont_forward:
2067         if (!local)
2068                 kfree_skb(skb);
2069 }
2070
2071 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2072 {
2073         struct rtable *rt = skb_rtable(skb);
2074         struct iphdr *iph = ip_hdr(skb);
2075         struct flowi4 fl4 = {
2076                 .daddr = iph->daddr,
2077                 .saddr = iph->saddr,
2078                 .flowi4_tos = RT_TOS(iph->tos),
2079                 .flowi4_oif = (rt_is_output_route(rt) ?
2080                                skb->dev->ifindex : 0),
2081                 .flowi4_iif = (rt_is_output_route(rt) ?
2082                                LOOPBACK_IFINDEX :
2083                                skb->dev->ifindex),
2084                 .flowi4_mark = skb->mark,
2085         };
2086         struct mr_table *mrt;
2087         int err;
2088
2089         err = ipmr_fib_lookup(net, &fl4, &mrt);
2090         if (err)
2091                 return ERR_PTR(err);
2092         return mrt;
2093 }
2094
2095 /* Multicast packets for forwarding arrive here
2096  * Called with rcu_read_lock();
2097  */
2098 int ip_mr_input(struct sk_buff *skb)
2099 {
2100         struct mfc_cache *cache;
2101         struct net *net = dev_net(skb->dev);
2102         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2103         struct mr_table *mrt;
2104         struct net_device *dev;
2105
2106         /* skb->dev passed in is the loX master dev for vrfs.
2107          * As there are no vifs associated with loopback devices,
2108          * get the proper interface that does have a vif associated with it.
2109          */
2110         dev = skb->dev;
2111         if (netif_is_l3_master(skb->dev)) {
2112                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2113                 if (!dev) {
2114                         kfree_skb(skb);
2115                         return -ENODEV;
2116                 }
2117         }
2118
2119         /* Packet is looped back after forward, it should not be
2120          * forwarded second time, but still can be delivered locally.
2121          */
2122         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2123                 goto dont_forward;
2124
2125         mrt = ipmr_rt_fib_lookup(net, skb);
2126         if (IS_ERR(mrt)) {
2127                 kfree_skb(skb);
2128                 return PTR_ERR(mrt);
2129         }
2130         if (!local) {
2131                 if (IPCB(skb)->opt.router_alert) {
2132                         if (ip_call_ra_chain(skb))
2133                                 return 0;
2134                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2135                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2136                          * Cisco IOS <= 11.2(8)) do not put router alert
2137                          * option to IGMP packets destined to routable
2138                          * groups. It is very bad, because it means
2139                          * that we can forward NO IGMP messages.
2140                          */
2141                         struct sock *mroute_sk;
2142
2143                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2144                         if (mroute_sk) {
2145                                 nf_reset(skb);
2146                                 raw_rcv(mroute_sk, skb);
2147                                 return 0;
2148                         }
2149                     }
2150         }
2151
2152         /* already under rcu_read_lock() */
2153         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2154         if (!cache) {
2155                 int vif = ipmr_find_vif(mrt, dev);
2156
2157                 if (vif >= 0)
2158                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2159                                                     vif);
2160         }
2161
2162         /* No usable cache entry */
2163         if (!cache) {
2164                 int vif;
2165
2166                 if (local) {
2167                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2168                         ip_local_deliver(skb);
2169                         if (!skb2)
2170                                 return -ENOBUFS;
2171                         skb = skb2;
2172                 }
2173
2174                 read_lock(&mrt_lock);
2175                 vif = ipmr_find_vif(mrt, dev);
2176                 if (vif >= 0) {
2177                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2178                         read_unlock(&mrt_lock);
2179
2180                         return err2;
2181                 }
2182                 read_unlock(&mrt_lock);
2183                 kfree_skb(skb);
2184                 return -ENODEV;
2185         }
2186
2187         read_lock(&mrt_lock);
2188         ip_mr_forward(net, mrt, dev, skb, cache, local);
2189         read_unlock(&mrt_lock);
2190
2191         if (local)
2192                 return ip_local_deliver(skb);
2193
2194         return 0;
2195
2196 dont_forward:
2197         if (local)
2198                 return ip_local_deliver(skb);
2199         kfree_skb(skb);
2200         return 0;
2201 }
2202
2203 #ifdef CONFIG_IP_PIMSM_V1
2204 /* Handle IGMP messages of PIMv1 */
2205 int pim_rcv_v1(struct sk_buff *skb)
2206 {
2207         struct igmphdr *pim;
2208         struct net *net = dev_net(skb->dev);
2209         struct mr_table *mrt;
2210
2211         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2212                 goto drop;
2213
2214         pim = igmp_hdr(skb);
2215
2216         mrt = ipmr_rt_fib_lookup(net, skb);
2217         if (IS_ERR(mrt))
2218                 goto drop;
2219         if (!mrt->mroute_do_pim ||
2220             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2221                 goto drop;
2222
2223         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2224 drop:
2225                 kfree_skb(skb);
2226         }
2227         return 0;
2228 }
2229 #endif
2230
2231 #ifdef CONFIG_IP_PIMSM_V2
2232 static int pim_rcv(struct sk_buff *skb)
2233 {
2234         struct pimreghdr *pim;
2235         struct net *net = dev_net(skb->dev);
2236         struct mr_table *mrt;
2237
2238         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2239                 goto drop;
2240
2241         pim = (struct pimreghdr *)skb_transport_header(skb);
2242         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2243             (pim->flags & PIM_NULL_REGISTER) ||
2244             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2245              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2246                 goto drop;
2247
2248         mrt = ipmr_rt_fib_lookup(net, skb);
2249         if (IS_ERR(mrt))
2250                 goto drop;
2251         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2252 drop:
2253                 kfree_skb(skb);
2254         }
2255         return 0;
2256 }
2257 #endif
2258
2259 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2260                               struct mr_mfc *c, struct rtmsg *rtm)
2261 {
2262         struct rta_mfc_stats mfcs;
2263         struct nlattr *mp_attr;
2264         struct rtnexthop *nhp;
2265         unsigned long lastuse;
2266         int ct;
2267
2268         /* If cache is unresolved, don't try to parse IIF and OIF */
2269         if (c->mfc_parent >= MAXVIFS) {
2270                 rtm->rtm_flags |= RTNH_F_UNRESOLVED;
2271                 return -ENOENT;
2272         }
2273
2274         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2275             nla_put_u32(skb, RTA_IIF,
2276                         mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2277                 return -EMSGSIZE;
2278
2279         if (c->mfc_flags & MFC_OFFLOAD)
2280                 rtm->rtm_flags |= RTNH_F_OFFLOAD;
2281
2282         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2283                 return -EMSGSIZE;
2284
2285         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2286                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2287                         struct vif_device *vif;
2288
2289                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2290                                 nla_nest_cancel(skb, mp_attr);
2291                                 return -EMSGSIZE;
2292                         }
2293
2294                         nhp->rtnh_flags = 0;
2295                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2296                         vif = &mrt->vif_table[ct];
2297                         nhp->rtnh_ifindex = vif->dev->ifindex;
2298                         nhp->rtnh_len = sizeof(*nhp);
2299                 }
2300         }
2301
2302         nla_nest_end(skb, mp_attr);
2303
2304         lastuse = READ_ONCE(c->mfc_un.res.lastuse);
2305         lastuse = time_after_eq(jiffies, lastuse) ? jiffies - lastuse : 0;
2306
2307         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2308         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2309         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2310         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) ||
2311             nla_put_u64_64bit(skb, RTA_EXPIRES, jiffies_to_clock_t(lastuse),
2312                               RTA_PAD))
2313                 return -EMSGSIZE;
2314
2315         rtm->rtm_type = RTN_MULTICAST;
2316         return 1;
2317 }
2318
2319 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2320                    __be32 saddr, __be32 daddr,
2321                    struct rtmsg *rtm, u32 portid)
2322 {
2323         struct mfc_cache *cache;
2324         struct mr_table *mrt;
2325         int err;
2326
2327         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2328         if (!mrt)
2329                 return -ENOENT;
2330
2331         rcu_read_lock();
2332         cache = ipmr_cache_find(mrt, saddr, daddr);
2333         if (!cache && skb->dev) {
2334                 int vif = ipmr_find_vif(mrt, skb->dev);
2335
2336                 if (vif >= 0)
2337                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2338         }
2339         if (!cache) {
2340                 struct sk_buff *skb2;
2341                 struct iphdr *iph;
2342                 struct net_device *dev;
2343                 int vif = -1;
2344
2345                 dev = skb->dev;
2346                 read_lock(&mrt_lock);
2347                 if (dev)
2348                         vif = ipmr_find_vif(mrt, dev);
2349                 if (vif < 0) {
2350                         read_unlock(&mrt_lock);
2351                         rcu_read_unlock();
2352                         return -ENODEV;
2353                 }
2354                 skb2 = skb_clone(skb, GFP_ATOMIC);
2355                 if (!skb2) {
2356                         read_unlock(&mrt_lock);
2357                         rcu_read_unlock();
2358                         return -ENOMEM;
2359                 }
2360
2361                 NETLINK_CB(skb2).portid = portid;
2362                 skb_push(skb2, sizeof(struct iphdr));
2363                 skb_reset_network_header(skb2);
2364                 iph = ip_hdr(skb2);
2365                 iph->ihl = sizeof(struct iphdr) >> 2;
2366                 iph->saddr = saddr;
2367                 iph->daddr = daddr;
2368                 iph->version = 0;
2369                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2370                 read_unlock(&mrt_lock);
2371                 rcu_read_unlock();
2372                 return err;
2373         }
2374
2375         read_lock(&mrt_lock);
2376         err = __ipmr_fill_mroute(mrt, skb, &cache->_c, rtm);
2377         read_unlock(&mrt_lock);
2378         rcu_read_unlock();
2379         return err;
2380 }
2381
2382 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2383                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2384                             int flags)
2385 {
2386         struct nlmsghdr *nlh;
2387         struct rtmsg *rtm;
2388         int err;
2389
2390         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2391         if (!nlh)
2392                 return -EMSGSIZE;
2393
2394         rtm = nlmsg_data(nlh);
2395         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2396         rtm->rtm_dst_len  = 32;
2397         rtm->rtm_src_len  = 32;
2398         rtm->rtm_tos      = 0;
2399         rtm->rtm_table    = mrt->id;
2400         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2401                 goto nla_put_failure;
2402         rtm->rtm_type     = RTN_MULTICAST;
2403         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2404         if (c->_c.mfc_flags & MFC_STATIC)
2405                 rtm->rtm_protocol = RTPROT_STATIC;
2406         else
2407                 rtm->rtm_protocol = RTPROT_MROUTED;
2408         rtm->rtm_flags    = 0;
2409
2410         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2411             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2412                 goto nla_put_failure;
2413         err = __ipmr_fill_mroute(mrt, skb, &c->_c, rtm);
2414         /* do not break the dump if cache is unresolved */
2415         if (err < 0 && err != -ENOENT)
2416                 goto nla_put_failure;
2417
2418         nlmsg_end(skb, nlh);
2419         return 0;
2420
2421 nla_put_failure:
2422         nlmsg_cancel(skb, nlh);
2423         return -EMSGSIZE;
2424 }
2425
2426 static size_t mroute_msgsize(bool unresolved, int maxvif)
2427 {
2428         size_t len =
2429                 NLMSG_ALIGN(sizeof(struct rtmsg))
2430                 + nla_total_size(4)     /* RTA_TABLE */
2431                 + nla_total_size(4)     /* RTA_SRC */
2432                 + nla_total_size(4)     /* RTA_DST */
2433                 ;
2434
2435         if (!unresolved)
2436                 len = len
2437                       + nla_total_size(4)       /* RTA_IIF */
2438                       + nla_total_size(0)       /* RTA_MULTIPATH */
2439                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2440                                                 /* RTA_MFC_STATS */
2441                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2442                 ;
2443
2444         return len;
2445 }
2446
2447 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2448                                  int cmd)
2449 {
2450         struct net *net = read_pnet(&mrt->net);
2451         struct sk_buff *skb;
2452         int err = -ENOBUFS;
2453
2454         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2455                                        mrt->maxvif),
2456                         GFP_ATOMIC);
2457         if (!skb)
2458                 goto errout;
2459
2460         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2461         if (err < 0)
2462                 goto errout;
2463
2464         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2465         return;
2466
2467 errout:
2468         kfree_skb(skb);
2469         if (err < 0)
2470                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2471 }
2472
2473 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2474 {
2475         size_t len =
2476                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2477                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2478                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2479                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2480                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2481                                         /* IPMRA_CREPORT_PKT */
2482                 + nla_total_size(payloadlen)
2483                 ;
2484
2485         return len;
2486 }
2487
2488 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2489 {
2490         struct net *net = read_pnet(&mrt->net);
2491         struct nlmsghdr *nlh;
2492         struct rtgenmsg *rtgenm;
2493         struct igmpmsg *msg;
2494         struct sk_buff *skb;
2495         struct nlattr *nla;
2496         int payloadlen;
2497
2498         payloadlen = pkt->len - sizeof(struct igmpmsg);
2499         msg = (struct igmpmsg *)skb_network_header(pkt);
2500
2501         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2502         if (!skb)
2503                 goto errout;
2504
2505         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2506                         sizeof(struct rtgenmsg), 0);
2507         if (!nlh)
2508                 goto errout;
2509         rtgenm = nlmsg_data(nlh);
2510         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2511         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2512             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif) ||
2513             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2514                             msg->im_src.s_addr) ||
2515             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2516                             msg->im_dst.s_addr))
2517                 goto nla_put_failure;
2518
2519         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2520         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2521                                   nla_data(nla), payloadlen))
2522                 goto nla_put_failure;
2523
2524         nlmsg_end(skb, nlh);
2525
2526         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2527         return;
2528
2529 nla_put_failure:
2530         nlmsg_cancel(skb, nlh);
2531 errout:
2532         kfree_skb(skb);
2533         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2534 }
2535
2536 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2537                              struct netlink_ext_ack *extack)
2538 {
2539         struct net *net = sock_net(in_skb->sk);
2540         struct nlattr *tb[RTA_MAX + 1];
2541         struct sk_buff *skb = NULL;
2542         struct mfc_cache *cache;
2543         struct mr_table *mrt;
2544         struct rtmsg *rtm;
2545         __be32 src, grp;
2546         u32 tableid;
2547         int err;
2548
2549         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2550                           rtm_ipv4_policy, extack);
2551         if (err < 0)
2552                 goto errout;
2553
2554         rtm = nlmsg_data(nlh);
2555
2556         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2557         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2558         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2559
2560         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2561         if (!mrt) {
2562                 err = -ENOENT;
2563                 goto errout_free;
2564         }
2565
2566         /* entries are added/deleted only under RTNL */
2567         rcu_read_lock();
2568         cache = ipmr_cache_find(mrt, src, grp);
2569         rcu_read_unlock();
2570         if (!cache) {
2571                 err = -ENOENT;
2572                 goto errout_free;
2573         }
2574
2575         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2576         if (!skb) {
2577                 err = -ENOBUFS;
2578                 goto errout_free;
2579         }
2580
2581         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2582                                nlh->nlmsg_seq, cache,
2583                                RTM_NEWROUTE, 0);
2584         if (err < 0)
2585                 goto errout_free;
2586
2587         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2588
2589 errout:
2590         return err;
2591
2592 errout_free:
2593         kfree_skb(skb);
2594         goto errout;
2595 }
2596
2597 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2598 {
2599         struct net *net = sock_net(skb->sk);
2600         unsigned int t = 0, s_t;
2601         unsigned int e = 0, s_e;
2602         struct mr_table *mrt;
2603         struct mr_mfc *mfc;
2604
2605         s_t = cb->args[0];
2606         s_e = cb->args[1];
2607
2608         rcu_read_lock();
2609         ipmr_for_each_table(mrt, net) {
2610                 if (t < s_t)
2611                         goto next_table;
2612                 list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list) {
2613                         if (e < s_e)
2614                                 goto next_entry;
2615                         if (ipmr_fill_mroute(mrt, skb,
2616                                              NETLINK_CB(cb->skb).portid,
2617                                              cb->nlh->nlmsg_seq,
2618                                              (struct mfc_cache *)mfc,
2619                                              RTM_NEWROUTE, NLM_F_MULTI) < 0)
2620                                 goto done;
2621 next_entry:
2622                         e++;
2623                 }
2624                 e = 0;
2625                 s_e = 0;
2626
2627                 spin_lock_bh(&mfc_unres_lock);
2628                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2629                         if (e < s_e)
2630                                 goto next_entry2;
2631                         if (ipmr_fill_mroute(mrt, skb,
2632                                              NETLINK_CB(cb->skb).portid,
2633                                              cb->nlh->nlmsg_seq,
2634                                              (struct mfc_cache *)mfc,
2635                                              RTM_NEWROUTE, NLM_F_MULTI) < 0) {
2636                                 spin_unlock_bh(&mfc_unres_lock);
2637                                 goto done;
2638                         }
2639 next_entry2:
2640                         e++;
2641                 }
2642                 spin_unlock_bh(&mfc_unres_lock);
2643                 e = 0;
2644                 s_e = 0;
2645 next_table:
2646                 t++;
2647         }
2648 done:
2649         rcu_read_unlock();
2650
2651         cb->args[1] = e;
2652         cb->args[0] = t;
2653
2654         return skb->len;
2655 }
2656
2657 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2658         [RTA_SRC]       = { .type = NLA_U32 },
2659         [RTA_DST]       = { .type = NLA_U32 },
2660         [RTA_IIF]       = { .type = NLA_U32 },
2661         [RTA_TABLE]     = { .type = NLA_U32 },
2662         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2663 };
2664
2665 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2666 {
2667         switch (rtm_protocol) {
2668         case RTPROT_STATIC:
2669         case RTPROT_MROUTED:
2670                 return true;
2671         }
2672         return false;
2673 }
2674
2675 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2676 {
2677         struct rtnexthop *rtnh = nla_data(nla);
2678         int remaining = nla_len(nla), vifi = 0;
2679
2680         while (rtnh_ok(rtnh, remaining)) {
2681                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2682                 if (++vifi == MAXVIFS)
2683                         break;
2684                 rtnh = rtnh_next(rtnh, &remaining);
2685         }
2686
2687         return remaining > 0 ? -EINVAL : vifi;
2688 }
2689
2690 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2691 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2692                             struct mfcctl *mfcc, int *mrtsock,
2693                             struct mr_table **mrtret,
2694                             struct netlink_ext_ack *extack)
2695 {
2696         struct net_device *dev = NULL;
2697         u32 tblid = RT_TABLE_DEFAULT;
2698         struct mr_table *mrt;
2699         struct nlattr *attr;
2700         struct rtmsg *rtm;
2701         int ret, rem;
2702
2703         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy,
2704                              extack);
2705         if (ret < 0)
2706                 goto out;
2707         rtm = nlmsg_data(nlh);
2708
2709         ret = -EINVAL;
2710         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2711             rtm->rtm_type != RTN_MULTICAST ||
2712             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2713             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2714                 goto out;
2715
2716         memset(mfcc, 0, sizeof(*mfcc));
2717         mfcc->mfcc_parent = -1;
2718         ret = 0;
2719         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2720                 switch (nla_type(attr)) {
2721                 case RTA_SRC:
2722                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2723                         break;
2724                 case RTA_DST:
2725                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2726                         break;
2727                 case RTA_IIF:
2728                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2729                         if (!dev) {
2730                                 ret = -ENODEV;
2731                                 goto out;
2732                         }
2733                         break;
2734                 case RTA_MULTIPATH:
2735                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2736                                 ret = -EINVAL;
2737                                 goto out;
2738                         }
2739                         break;
2740                 case RTA_PREFSRC:
2741                         ret = 1;
2742                         break;
2743                 case RTA_TABLE:
2744                         tblid = nla_get_u32(attr);
2745                         break;
2746                 }
2747         }
2748         mrt = ipmr_get_table(net, tblid);
2749         if (!mrt) {
2750                 ret = -ENOENT;
2751                 goto out;
2752         }
2753         *mrtret = mrt;
2754         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2755         if (dev)
2756                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2757
2758 out:
2759         return ret;
2760 }
2761
2762 /* takes care of both newroute and delroute */
2763 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2764                           struct netlink_ext_ack *extack)
2765 {
2766         struct net *net = sock_net(skb->sk);
2767         int ret, mrtsock, parent;
2768         struct mr_table *tbl;
2769         struct mfcctl mfcc;
2770
2771         mrtsock = 0;
2772         tbl = NULL;
2773         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2774         if (ret < 0)
2775                 return ret;
2776
2777         parent = ret ? mfcc.mfcc_parent : -1;
2778         if (nlh->nlmsg_type == RTM_NEWROUTE)
2779                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2780         else
2781                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2782 }
2783
2784 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2785 {
2786         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2787
2788         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2789             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2790             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2791                         mrt->mroute_reg_vif_num) ||
2792             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2793                        mrt->mroute_do_assert) ||
2794             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim))
2795                 return false;
2796
2797         return true;
2798 }
2799
2800 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2801 {
2802         struct nlattr *vif_nest;
2803         struct vif_device *vif;
2804
2805         /* if the VIF doesn't exist just continue */
2806         if (!VIF_EXISTS(mrt, vifid))
2807                 return true;
2808
2809         vif = &mrt->vif_table[vifid];
2810         vif_nest = nla_nest_start(skb, IPMRA_VIF);
2811         if (!vif_nest)
2812                 return false;
2813         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2814             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2815             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2816             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2817                               IPMRA_VIFA_PAD) ||
2818             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2819                               IPMRA_VIFA_PAD) ||
2820             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2821                               IPMRA_VIFA_PAD) ||
2822             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2823                               IPMRA_VIFA_PAD) ||
2824             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2825             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2826                 nla_nest_cancel(skb, vif_nest);
2827                 return false;
2828         }
2829         nla_nest_end(skb, vif_nest);
2830
2831         return true;
2832 }
2833
2834 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2835 {
2836         struct net *net = sock_net(skb->sk);
2837         struct nlmsghdr *nlh = NULL;
2838         unsigned int t = 0, s_t;
2839         unsigned int e = 0, s_e;
2840         struct mr_table *mrt;
2841
2842         s_t = cb->args[0];
2843         s_e = cb->args[1];
2844
2845         ipmr_for_each_table(mrt, net) {
2846                 struct nlattr *vifs, *af;
2847                 struct ifinfomsg *hdr;
2848                 u32 i;
2849
2850                 if (t < s_t)
2851                         goto skip_table;
2852                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2853                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2854                                 sizeof(*hdr), NLM_F_MULTI);
2855                 if (!nlh)
2856                         break;
2857
2858                 hdr = nlmsg_data(nlh);
2859                 memset(hdr, 0, sizeof(*hdr));
2860                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2861
2862                 af = nla_nest_start(skb, IFLA_AF_SPEC);
2863                 if (!af) {
2864                         nlmsg_cancel(skb, nlh);
2865                         goto out;
2866                 }
2867
2868                 if (!ipmr_fill_table(mrt, skb)) {
2869                         nlmsg_cancel(skb, nlh);
2870                         goto out;
2871                 }
2872
2873                 vifs = nla_nest_start(skb, IPMRA_TABLE_VIFS);
2874                 if (!vifs) {
2875                         nla_nest_end(skb, af);
2876                         nlmsg_end(skb, nlh);
2877                         goto out;
2878                 }
2879                 for (i = 0; i < mrt->maxvif; i++) {
2880                         if (e < s_e)
2881                                 goto skip_entry;
2882                         if (!ipmr_fill_vif(mrt, i, skb)) {
2883                                 nla_nest_end(skb, vifs);
2884                                 nla_nest_end(skb, af);
2885                                 nlmsg_end(skb, nlh);
2886                                 goto out;
2887                         }
2888 skip_entry:
2889                         e++;
2890                 }
2891                 s_e = 0;
2892                 e = 0;
2893                 nla_nest_end(skb, vifs);
2894                 nla_nest_end(skb, af);
2895                 nlmsg_end(skb, nlh);
2896 skip_table:
2897                 t++;
2898         }
2899
2900 out:
2901         cb->args[1] = e;
2902         cb->args[0] = t;
2903
2904         return skb->len;
2905 }
2906
2907 #ifdef CONFIG_PROC_FS
2908 /* The /proc interfaces to multicast routing :
2909  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2910  */
2911
2912 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2913         __acquires(mrt_lock)
2914 {
2915         struct mr_vif_iter *iter = seq->private;
2916         struct net *net = seq_file_net(seq);
2917         struct mr_table *mrt;
2918
2919         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2920         if (!mrt)
2921                 return ERR_PTR(-ENOENT);
2922
2923         iter->mrt = mrt;
2924
2925         read_lock(&mrt_lock);
2926         return mr_vif_seq_start(seq, pos);
2927 }
2928
2929 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2930         __releases(mrt_lock)
2931 {
2932         read_unlock(&mrt_lock);
2933 }
2934
2935 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2936 {
2937         struct mr_vif_iter *iter = seq->private;
2938         struct mr_table *mrt = iter->mrt;
2939
2940         if (v == SEQ_START_TOKEN) {
2941                 seq_puts(seq,
2942                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2943         } else {
2944                 const struct vif_device *vif = v;
2945                 const char *name =  vif->dev ?
2946                                     vif->dev->name : "none";
2947
2948                 seq_printf(seq,
2949                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2950                            vif - mrt->vif_table,
2951                            name, vif->bytes_in, vif->pkt_in,
2952                            vif->bytes_out, vif->pkt_out,
2953                            vif->flags, vif->local, vif->remote);
2954         }
2955         return 0;
2956 }
2957
2958 static const struct seq_operations ipmr_vif_seq_ops = {
2959         .start = ipmr_vif_seq_start,
2960         .next  = mr_vif_seq_next,
2961         .stop  = ipmr_vif_seq_stop,
2962         .show  = ipmr_vif_seq_show,
2963 };
2964
2965 static int ipmr_vif_open(struct inode *inode, struct file *file)
2966 {
2967         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2968                             sizeof(struct mr_vif_iter));
2969 }
2970
2971 static const struct file_operations ipmr_vif_fops = {
2972         .open    = ipmr_vif_open,
2973         .read    = seq_read,
2974         .llseek  = seq_lseek,
2975         .release = seq_release_net,
2976 };
2977
2978 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2979 {
2980         struct net *net = seq_file_net(seq);
2981         struct mr_table *mrt;
2982
2983         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2984         if (!mrt)
2985                 return ERR_PTR(-ENOENT);
2986
2987         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2988 }
2989
2990 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2991 {
2992         int n;
2993
2994         if (v == SEQ_START_TOKEN) {
2995                 seq_puts(seq,
2996                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2997         } else {
2998                 const struct mfc_cache *mfc = v;
2999                 const struct mr_mfc_iter *it = seq->private;
3000                 const struct mr_table *mrt = it->mrt;
3001
3002                 seq_printf(seq, "%08X %08X %-3hd",
3003                            (__force u32) mfc->mfc_mcastgrp,
3004                            (__force u32) mfc->mfc_origin,
3005                            mfc->_c.mfc_parent);
3006
3007                 if (it->cache != &mrt->mfc_unres_queue) {
3008                         seq_printf(seq, " %8lu %8lu %8lu",
3009                                    mfc->_c.mfc_un.res.pkt,
3010                                    mfc->_c.mfc_un.res.bytes,
3011                                    mfc->_c.mfc_un.res.wrong_if);
3012                         for (n = mfc->_c.mfc_un.res.minvif;
3013                              n < mfc->_c.mfc_un.res.maxvif; n++) {
3014                                 if (VIF_EXISTS(mrt, n) &&
3015                                     mfc->_c.mfc_un.res.ttls[n] < 255)
3016                                         seq_printf(seq,
3017                                            " %2d:%-3d",
3018                                            n, mfc->_c.mfc_un.res.ttls[n]);
3019                         }
3020                 } else {
3021                         /* unresolved mfc_caches don't contain
3022                          * pkt, bytes and wrong_if values
3023                          */
3024                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
3025                 }
3026                 seq_putc(seq, '\n');
3027         }
3028         return 0;
3029 }
3030
3031 static const struct seq_operations ipmr_mfc_seq_ops = {
3032         .start = ipmr_mfc_seq_start,
3033         .next  = mr_mfc_seq_next,
3034         .stop  = mr_mfc_seq_stop,
3035         .show  = ipmr_mfc_seq_show,
3036 };
3037
3038 static int ipmr_mfc_open(struct inode *inode, struct file *file)
3039 {
3040         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
3041                             sizeof(struct mr_mfc_iter));
3042 }
3043
3044 static const struct file_operations ipmr_mfc_fops = {
3045         .open    = ipmr_mfc_open,
3046         .read    = seq_read,
3047         .llseek  = seq_lseek,
3048         .release = seq_release_net,
3049 };
3050 #endif
3051
3052 #ifdef CONFIG_IP_PIMSM_V2
3053 static const struct net_protocol pim_protocol = {
3054         .handler        =       pim_rcv,
3055         .netns_ok       =       1,
3056 };
3057 #endif
3058
3059 static unsigned int ipmr_seq_read(struct net *net)
3060 {
3061         ASSERT_RTNL();
3062
3063         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3064 }
3065
3066 static int ipmr_dump(struct net *net, struct notifier_block *nb)
3067 {
3068         struct mr_table *mrt;
3069         int err;
3070
3071         err = ipmr_rules_dump(net, nb);
3072         if (err)
3073                 return err;
3074
3075         ipmr_for_each_table(mrt, net) {
3076                 struct vif_device *v = &mrt->vif_table[0];
3077                 struct mr_mfc *mfc;
3078                 int vifi;
3079
3080                 /* Notifiy on table VIF entries */
3081                 read_lock(&mrt_lock);
3082                 for (vifi = 0; vifi < mrt->maxvif; vifi++, v++) {
3083                         if (!v->dev)
3084                                 continue;
3085
3086                         call_ipmr_vif_entry_notifier(nb, net, FIB_EVENT_VIF_ADD,
3087                                                      v, vifi, mrt->id);
3088                 }
3089                 read_unlock(&mrt_lock);
3090
3091                 /* Notify on table MFC entries */
3092                 list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list)
3093                         call_ipmr_mfc_entry_notifier(nb, net,
3094                                                      FIB_EVENT_ENTRY_ADD,
3095                                                      (struct mfc_cache *)mfc,
3096                                                      mrt->id);
3097         }
3098
3099         return 0;
3100 }
3101
3102 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3103         .family         = RTNL_FAMILY_IPMR,
3104         .fib_seq_read   = ipmr_seq_read,
3105         .fib_dump       = ipmr_dump,
3106         .owner          = THIS_MODULE,
3107 };
3108
3109 static int __net_init ipmr_notifier_init(struct net *net)
3110 {
3111         struct fib_notifier_ops *ops;
3112
3113         net->ipv4.ipmr_seq = 0;
3114
3115         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3116         if (IS_ERR(ops))
3117                 return PTR_ERR(ops);
3118         net->ipv4.ipmr_notifier_ops = ops;
3119
3120         return 0;
3121 }
3122
3123 static void __net_exit ipmr_notifier_exit(struct net *net)
3124 {
3125         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3126         net->ipv4.ipmr_notifier_ops = NULL;
3127 }
3128
3129 /* Setup for IP multicast routing */
3130 static int __net_init ipmr_net_init(struct net *net)
3131 {
3132         int err;
3133
3134         err = ipmr_notifier_init(net);
3135         if (err)
3136                 goto ipmr_notifier_fail;
3137
3138         err = ipmr_rules_init(net);
3139         if (err < 0)
3140                 goto ipmr_rules_fail;
3141
3142 #ifdef CONFIG_PROC_FS
3143         err = -ENOMEM;
3144         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
3145                 goto proc_vif_fail;
3146         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
3147                 goto proc_cache_fail;
3148 #endif
3149         return 0;
3150
3151 #ifdef CONFIG_PROC_FS
3152 proc_cache_fail:
3153         remove_proc_entry("ip_mr_vif", net->proc_net);
3154 proc_vif_fail:
3155         ipmr_rules_exit(net);
3156 #endif
3157 ipmr_rules_fail:
3158         ipmr_notifier_exit(net);
3159 ipmr_notifier_fail:
3160         return err;
3161 }
3162
3163 static void __net_exit ipmr_net_exit(struct net *net)
3164 {
3165 #ifdef CONFIG_PROC_FS
3166         remove_proc_entry("ip_mr_cache", net->proc_net);
3167         remove_proc_entry("ip_mr_vif", net->proc_net);
3168 #endif
3169         ipmr_notifier_exit(net);
3170         ipmr_rules_exit(net);
3171 }
3172
3173 static struct pernet_operations ipmr_net_ops = {
3174         .init = ipmr_net_init,
3175         .exit = ipmr_net_exit,
3176         .async = true,
3177 };
3178
3179 int __init ip_mr_init(void)
3180 {
3181         int err;
3182
3183         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3184                                        sizeof(struct mfc_cache),
3185                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3186                                        NULL);
3187
3188         err = register_pernet_subsys(&ipmr_net_ops);
3189         if (err)
3190                 goto reg_pernet_fail;
3191
3192         err = register_netdevice_notifier(&ip_mr_notifier);
3193         if (err)
3194                 goto reg_notif_fail;
3195 #ifdef CONFIG_IP_PIMSM_V2
3196         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3197                 pr_err("%s: can't add PIM protocol\n", __func__);
3198                 err = -EAGAIN;
3199                 goto add_proto_fail;
3200         }
3201 #endif
3202         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3203                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3204         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3205                       ipmr_rtm_route, NULL, 0);
3206         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3207                       ipmr_rtm_route, NULL, 0);
3208
3209         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3210                       NULL, ipmr_rtm_dumplink, 0);
3211         return 0;
3212
3213 #ifdef CONFIG_IP_PIMSM_V2
3214 add_proto_fail:
3215         unregister_netdevice_notifier(&ip_mr_notifier);
3216 #endif
3217 reg_notif_fail:
3218         unregister_pernet_subsys(&ipmr_net_ops);
3219 reg_pernet_fail:
3220         kmem_cache_destroy(mrt_cachep);
3221         return err;
3222 }