net/ipv4/ipmr.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      IP multicast routing support for mrouted 3.6/3.8
   4  *
   5  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   6  *        Linux Consultancy and Custom Driver Development
   7  *
   8  *      Fixes:
   9  *      Michael Chastain        :       Incorrect size of copying.
  10  *      Alan Cox                :       Added the cache manager code
  11  *      Alan Cox                :       Fixed the clone/copy bug and device race.
  12  *      Mike McLagan            :       Routing by source
  13  *      Malcolm Beattie         :       Buffer handling fixes.
  14  *      Alexey Kuznetsov        :       Double buffer free and other fixes.
  15  *      SVR Anand               :       Fixed several multicast bugs and problems.
  16  *      Alexey Kuznetsov        :       Status, optimisations and more.
  17  *      Brad Parker             :       Better behaviour on mrouted upcall
  18  *                                      overflow.
  19  *      Carlos Picoto           :       PIMv1 Support
  20  *      Pavlin Ivanov Radoslavov:       PIMv2 Registers must checksum only PIM header
  21  *                                      Relax this requirement to work with older peers.
  22  */
  23
  24 #include <linux/uaccess.h>
  25 #include <linux/types.h>
  26 #include <linux/cache.h>
  27 #include <linux/capability.h>
  28 #include <linux/errno.h>
  29 #include <linux/mm.h>
  30 #include <linux/kernel.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/socket.h>
  34 #include <linux/in.h>
  35 #include <linux/inet.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/inetdevice.h>
  38 #include <linux/igmp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/mroute.h>
  42 #include <linux/init.h>
  43 #include <linux/if_ether.h>
  44 #include <linux/slab.h>
  45 #include <net/net_namespace.h>
  46 #include <net/ip.h>
  47 #include <net/protocol.h>
  48 #include <linux/skbuff.h>
  49 #include <net/route.h>
  50 #include <net/icmp.h>
  51 #include <net/udp.h>
  52 #include <net/raw.h>
  53 #include <linux/notifier.h>
  54 #include <linux/if_arp.h>
  55 #include <linux/netfilter_ipv4.h>
  56 #include <linux/compat.h>
  57 #include <linux/export.h>
  58 #include <linux/rhashtable.h>
  59 #include <net/ip_tunnels.h>
  60 #include <net/checksum.h>
  61 #include <net/netlink.h>
  62 #include <net/fib_rules.h>
  63 #include <linux/netconf.h>
  64 #include <net/rtnh.h>
  65
  66 #include <linux/nospec.h>
  67
  68 struct ipmr_rule {
  69         struct fib_rule         common;
  70 };
  71
  72 struct ipmr_result {
  73         struct mr_table         *mrt;
  74 };
  75
  76 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  77  * Note that the changes are semaphored via rtnl_lock.
  78  */
  79
  80 static DEFINE_RWLOCK(mrt_lock);
  81
  82 /* Multicast router control variables */
  83
  84 /* Special spinlock for queue of unresolved entries */
  85 static DEFINE_SPINLOCK(mfc_unres_lock);
  86
  87 /* We return to original Alan's scheme. Hash table of resolved
  88  * entries is changed only in process context and protected
  89  * with weak lock mrt_lock. Queue of unresolved entries is protected
  90  * with strong spinlock mfc_unres_lock.
  91  *
  92  * In this case data path is free of exclusive locks at all.
  93  */
  94
  95 static struct kmem_cache *mrt_cachep __ro_after_init;
  96
  97 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
  98 static void ipmr_free_table(struct mr_table *mrt);
  99
 100 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 101                           struct net_device *dev, struct sk_buff *skb,
 102                           struct mfc_cache *cache, int local);
 103 static int ipmr_cache_report(struct mr_table *mrt,
 104                              struct sk_buff *pkt, vifi_t vifi, int assert);
 105 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 106                                  int cmd);
 107 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt);
 108 static void mroute_clean_tables(struct mr_table *mrt, int flags);
 109 static void ipmr_expire_process(struct timer_list *t);
 110
 111 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 112 #define ipmr_for_each_table(mrt, net)                                   \
 113         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list,        \
 114                                 lockdep_rtnl_is_held() ||               \
 115                                 list_empty(&net->ipv4.mr_tables))
 116
 117 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 118                                            struct mr_table *mrt)
 119 {
 120         struct mr_table *ret;
 121
 122         if (!mrt)
 123                 ret = list_entry_rcu(net->ipv4.mr_tables.next,
 124                                      struct mr_table, list);
 125         else
 126                 ret = list_entry_rcu(mrt->list.next,
 127                                      struct mr_table, list);
 128
 129         if (&ret->list == &net->ipv4.mr_tables)
 130                 return NULL;
 131         return ret;
 132 }
 133
 134 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 135 {
 136         struct mr_table *mrt;
 137
 138         ipmr_for_each_table(mrt, net) {
 139                 if (mrt->id == id)
 140                         return mrt;
 141         }
 142         return NULL;
 143 }
 144
 145 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 146                            struct mr_table **mrt)
 147 {
 148         int err;
 149         struct ipmr_result res;
 150         struct fib_lookup_arg arg = {
 151                 .result = &res,
 152                 .flags = FIB_LOOKUP_NOREF,
 153         };
 154
 155         /* update flow if oif or iif point to device enslaved to l3mdev */
 156         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 157
 158         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 159                                flowi4_to_flowi(flp4), 0, &arg);
 160         if (err < 0)
 161                 return err;
 162         *mrt = res.mrt;
 163         return 0;
 164 }
 165
 166 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 167                             int flags, struct fib_lookup_arg *arg)
 168 {
 169         struct ipmr_result *res = arg->result;
 170         struct mr_table *mrt;
 171
 172         switch (rule->action) {
 173         case FR_ACT_TO_TBL:
 174                 break;
 175         case FR_ACT_UNREACHABLE:
 176                 return -ENETUNREACH;
 177         case FR_ACT_PROHIBIT:
 178                 return -EACCES;
 179         case FR_ACT_BLACKHOLE:
 180         default:
 181                 return -EINVAL;
 182         }
 183
 184         arg->table = fib_rule_get_table(rule, arg);
 185
 186         mrt = ipmr_get_table(rule->fr_net, arg->table);
 187         if (!mrt)
 188                 return -EAGAIN;
 189         res->mrt = mrt;
 190         return 0;
 191 }
 192
 193 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 194 {
 195         return 1;
 196 }
 197
 198 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 199                                struct fib_rule_hdr *frh, struct nlattr **tb,
 200                                struct netlink_ext_ack *extack)
 201 {
 202         return 0;
 203 }
 204
 205 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 206                              struct nlattr **tb)
 207 {
 208         return 1;
 209 }
 210
 211 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 212                           struct fib_rule_hdr *frh)
 213 {
 214         frh->dst_len = 0;
 215         frh->src_len = 0;
 216         frh->tos     = 0;
 217         return 0;
 218 }
 219
 220 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 221         .family         = RTNL_FAMILY_IPMR,
 222         .rule_size      = sizeof(struct ipmr_rule),
 223         .addr_size      = sizeof(u32),
 224         .action         = ipmr_rule_action,
 225         .match          = ipmr_rule_match,
 226         .configure      = ipmr_rule_configure,
 227         .compare        = ipmr_rule_compare,
 228         .fill           = ipmr_rule_fill,
 229         .nlgroup        = RTNLGRP_IPV4_RULE,
 230         .owner          = THIS_MODULE,
 231 };
 232
 233 static int __net_init ipmr_rules_init(struct net *net)
 234 {
 235         struct fib_rules_ops *ops;
 236         struct mr_table *mrt;
 237         int err;
 238
 239         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 240         if (IS_ERR(ops))
 241                 return PTR_ERR(ops);
 242
 243         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 244
 245         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 246         if (IS_ERR(mrt)) {
 247                 err = PTR_ERR(mrt);
 248                 goto err1;
 249         }
 250
 251         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 252         if (err < 0)
 253                 goto err2;
 254
 255         net->ipv4.mr_rules_ops = ops;
 256         return 0;
 257
 258 err2:
 259         rtnl_lock();
 260         ipmr_free_table(mrt);
 261         rtnl_unlock();
 262 err1:
 263         fib_rules_unregister(ops);
 264         return err;
 265 }
 266
 267 static void __net_exit ipmr_rules_exit(struct net *net)
 268 {
 269         struct mr_table *mrt, *next;
 270
 271         rtnl_lock();
 272         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 273                 list_del(&mrt->list);
 274                 ipmr_free_table(mrt);
 275         }
 276         fib_rules_unregister(net->ipv4.mr_rules_ops);
 277         rtnl_unlock();
 278 }
 279
 280 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 281                            struct netlink_ext_ack *extack)
 282 {
 283         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR, extack);
 284 }
 285
 286 static unsigned int ipmr_rules_seq_read(struct net *net)
 287 {
 288         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 289 }
 290
 291 bool ipmr_rule_default(const struct fib_rule *rule)
 292 {
 293         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 294 }
 295 EXPORT_SYMBOL(ipmr_rule_default);
 296 #else
 297 #define ipmr_for_each_table(mrt, net) \
 298         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 299
 300 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 301                                            struct mr_table *mrt)
 302 {
 303         if (!mrt)
 304                 return net->ipv4.mrt;
 305         return NULL;
 306 }
 307
 308 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 309 {
 310         return net->ipv4.mrt;
 311 }
 312
 313 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 314                            struct mr_table **mrt)
 315 {
 316         *mrt = net->ipv4.mrt;
 317         return 0;
 318 }
 319
 320 static int __net_init ipmr_rules_init(struct net *net)
 321 {
 322         struct mr_table *mrt;
 323
 324         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 325         if (IS_ERR(mrt))
 326                 return PTR_ERR(mrt);
 327         net->ipv4.mrt = mrt;
 328         return 0;
 329 }
 330
 331 static void __net_exit ipmr_rules_exit(struct net *net)
 332 {
 333         rtnl_lock();
 334         ipmr_free_table(net->ipv4.mrt);
 335         net->ipv4.mrt = NULL;
 336         rtnl_unlock();
 337 }
 338
 339 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 340                            struct netlink_ext_ack *extack)
 341 {
 342         return 0;
 343 }
 344
 345 static unsigned int ipmr_rules_seq_read(struct net *net)
 346 {
 347         return 0;
 348 }
 349
 350 bool ipmr_rule_default(const struct fib_rule *rule)
 351 {
 352         return true;
 353 }
 354 EXPORT_SYMBOL(ipmr_rule_default);
 355 #endif
 356
 357 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 358                                 const void *ptr)
 359 {
 360         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 361         struct mfc_cache *c = (struct mfc_cache *)ptr;
 362
 363         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 364                cmparg->mfc_origin != c->mfc_origin;
 365 }
 366
 367 static const struct rhashtable_params ipmr_rht_params = {
 368         .head_offset = offsetof(struct mr_mfc, mnode),
 369         .key_offset = offsetof(struct mfc_cache, cmparg),
 370         .key_len = sizeof(struct mfc_cache_cmp_arg),
 371         .nelem_hint = 3,
 372         .obj_cmpfn = ipmr_hash_cmp,
 373         .automatic_shrinking = true,
 374 };
 375
 376 static void ipmr_new_table_set(struct mr_table *mrt,
 377                                struct net *net)
 378 {
 379 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 380         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 381 #endif
 382 }
 383
 384 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 385         .mfc_mcastgrp = htonl(INADDR_ANY),
 386         .mfc_origin = htonl(INADDR_ANY),
 387 };
 388
 389 static struct mr_table_ops ipmr_mr_table_ops = {
 390         .rht_params = &ipmr_rht_params,
 391         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 392 };
 393
 394 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 395 {
 396         struct mr_table *mrt;
 397
 398         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 399         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 400                 return ERR_PTR(-EINVAL);
 401
 402         mrt = ipmr_get_table(net, id);
 403         if (mrt)
 404                 return mrt;
 405
 406         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 407                               ipmr_expire_process, ipmr_new_table_set);
 408 }
 409
 410 static void ipmr_free_table(struct mr_table *mrt)
 411 {
 412         del_timer_sync(&mrt->ipmr_expire_timer);
 413         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC |
 414                                  MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC);
 415         rhltable_destroy(&mrt->mfc_hash);
 416         kfree(mrt);
 417 }
 418
 419 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 420
 421 /* Initialize ipmr pimreg/tunnel in_device */
 422 static bool ipmr_init_vif_indev(const struct net_device *dev)
 423 {
 424         struct in_device *in_dev;
 425
 426         ASSERT_RTNL();
 427
 428         in_dev = __in_dev_get_rtnl(dev);
 429         if (!in_dev)
 430                 return false;
 431         ipv4_devconf_setall(in_dev);
 432         neigh_parms_data_state_setall(in_dev->arp_parms);
 433         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 434
 435         return true;
 436 }
 437
 438 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 439 {
 440         struct net_device *tunnel_dev, *new_dev;
 441         struct ip_tunnel_parm p = { };
 442         int err;
 443
 444         tunnel_dev = __dev_get_by_name(net, "tunl0");
 445         if (!tunnel_dev)
 446                 goto out;
 447
 448         p.iph.daddr = v->vifc_rmt_addr.s_addr;
 449         p.iph.saddr = v->vifc_lcl_addr.s_addr;
 450         p.iph.version = 4;
 451         p.iph.ihl = 5;
 452         p.iph.protocol = IPPROTO_IPIP;
 453         sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 454
 455         if (!tunnel_dev->netdev_ops->ndo_tunnel_ctl)
 456                 goto out;
 457         err = tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 458                         SIOCADDTUNNEL);
 459         if (err)
 460                 goto out;
 461
 462         new_dev = __dev_get_by_name(net, p.name);
 463         if (!new_dev)
 464                 goto out;
 465
 466         new_dev->flags |= IFF_MULTICAST;
 467         if (!ipmr_init_vif_indev(new_dev))
 468                 goto out_unregister;
 469         if (dev_open(new_dev, NULL))
 470                 goto out_unregister;
 471         dev_hold(new_dev);
 472         err = dev_set_allmulti(new_dev, 1);
 473         if (err) {
 474                 dev_close(new_dev);
 475                 tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 476                                 SIOCDELTUNNEL);
 477                 dev_put(new_dev);
 478                 new_dev = ERR_PTR(err);
 479         }
 480         return new_dev;
 481
 482 out_unregister:
 483         unregister_netdevice(new_dev);
 484 out:
 485         return ERR_PTR(-ENOBUFS);
 486 }
 487
 488 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 489 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 490 {
 491         struct net *net = dev_net(dev);
 492         struct mr_table *mrt;
 493         struct flowi4 fl4 = {
 494                 .flowi4_oif     = dev->ifindex,
 495                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 496                 .flowi4_mark    = skb->mark,
 497         };
 498         int err;
 499
 500         err = ipmr_fib_lookup(net, &fl4, &mrt);
 501         if (err < 0) {
 502                 kfree_skb(skb);
 503                 return err;
 504         }
 505
 506         read_lock(&mrt_lock);
 507         dev->stats.tx_bytes += skb->len;
 508         dev->stats.tx_packets++;
 509         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 510         read_unlock(&mrt_lock);
 511         kfree_skb(skb);
 512         return NETDEV_TX_OK;
 513 }
 514
 515 static int reg_vif_get_iflink(const struct net_device *dev)
 516 {
 517         return 0;
 518 }
 519
 520 static const struct net_device_ops reg_vif_netdev_ops = {
 521         .ndo_start_xmit = reg_vif_xmit,
 522         .ndo_get_iflink = reg_vif_get_iflink,
 523 };
 524
 525 static void reg_vif_setup(struct net_device *dev)
 526 {
 527         dev->type               = ARPHRD_PIMREG;
 528         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 529         dev->flags              = IFF_NOARP;
 530         dev->netdev_ops         = &reg_vif_netdev_ops;
 531         dev->needs_free_netdev  = true;
 532         dev->features           |= NETIF_F_NETNS_LOCAL;
 533 }
 534
 535 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 536 {
 537         struct net_device *dev;
 538         char name[IFNAMSIZ];
 539
 540         if (mrt->id == RT_TABLE_DEFAULT)
 541                 sprintf(name, "pimreg");
 542         else
 543                 sprintf(name, "pimreg%u", mrt->id);
 544
 545         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 546
 547         if (!dev)
 548                 return NULL;
 549
 550         dev_net_set(dev, net);
 551
 552         if (register_netdevice(dev)) {
 553                 free_netdev(dev);
 554                 return NULL;
 555         }
 556
 557         if (!ipmr_init_vif_indev(dev))
 558                 goto failure;
 559         if (dev_open(dev, NULL))
 560                 goto failure;
 561
 562         dev_hold(dev);
 563
 564         return dev;
 565
 566 failure:
 567         unregister_netdevice(dev);
 568         return NULL;
 569 }
 570
 571 /* called with rcu_read_lock() */
 572 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 573                      unsigned int pimlen)
 574 {
 575         struct net_device *reg_dev = NULL;
 576         struct iphdr *encap;
 577
 578         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 579         /* Check that:
 580          * a. packet is really sent to a multicast group
 581          * b. packet is not a NULL-REGISTER
 582          * c. packet is not truncated
 583          */
 584         if (!ipv4_is_multicast(encap->daddr) ||
 585             encap->tot_len == 0 ||
 586             ntohs(encap->tot_len) + pimlen > skb->len)
 587                 return 1;
 588
 589         read_lock(&mrt_lock);
 590         if (mrt->mroute_reg_vif_num >= 0)
 591                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 592         read_unlock(&mrt_lock);
 593
 594         if (!reg_dev)
 595                 return 1;
 596
 597         skb->mac_header = skb->network_header;
 598         skb_pull(skb, (u8 *)encap - skb->data);
 599         skb_reset_network_header(skb);
 600         skb->protocol = htons(ETH_P_IP);
 601         skb->ip_summed = CHECKSUM_NONE;
 602
 603         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 604
 605         netif_rx(skb);
 606
 607         return NET_RX_SUCCESS;
 608 }
 609 #else
 610 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 611 {
 612         return NULL;
 613 }
 614 #endif
 615
 616 static int call_ipmr_vif_entry_notifiers(struct net *net,
 617                                          enum fib_event_type event_type,
 618                                          struct vif_device *vif,
 619                                          vifi_t vif_index, u32 tb_id)
 620 {
 621         return mr_call_vif_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 622                                      vif, vif_index, tb_id,
 623                                      &net->ipv4.ipmr_seq);
 624 }
 625
 626 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 627                                          enum fib_event_type event_type,
 628                                          struct mfc_cache *mfc, u32 tb_id)
 629 {
 630         return mr_call_mfc_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 631                                      &mfc->_c, tb_id, &net->ipv4.ipmr_seq);
 632 }
 633
 634 /**
 635  *      vif_delete - Delete a VIF entry
 636  *      @mrt: Table to delete from
 637  *      @vifi: VIF identifier to delete
 638  *      @notify: Set to 1, if the caller is a notifier_call
 639  *      @head: if unregistering the VIF, place it on this queue
 640  */
 641 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 642                       struct list_head *head)
 643 {
 644         struct net *net = read_pnet(&mrt->net);
 645         struct vif_device *v;
 646         struct net_device *dev;
 647         struct in_device *in_dev;
 648
 649         if (vifi < 0 || vifi >= mrt->maxvif)
 650                 return -EADDRNOTAVAIL;
 651
 652         v = &mrt->vif_table[vifi];
 653
 654         if (VIF_EXISTS(mrt, vifi))
 655                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 656                                               mrt->id);
 657
 658         write_lock_bh(&mrt_lock);
 659         dev = v->dev;
 660         v->dev = NULL;
 661
 662         if (!dev) {
 663                 write_unlock_bh(&mrt_lock);
 664                 return -EADDRNOTAVAIL;
 665         }
 666
 667         if (vifi == mrt->mroute_reg_vif_num)
 668                 mrt->mroute_reg_vif_num = -1;
 669
 670         if (vifi + 1 == mrt->maxvif) {
 671                 int tmp;
 672
 673                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 674                         if (VIF_EXISTS(mrt, tmp))
 675                                 break;
 676                 }
 677                 mrt->maxvif = tmp+1;
 678         }
 679
 680         write_unlock_bh(&mrt_lock);
 681
 682         dev_set_allmulti(dev, -1);
 683
 684         in_dev = __in_dev_get_rtnl(dev);
 685         if (in_dev) {
 686                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 687                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 688                                             NETCONFA_MC_FORWARDING,
 689                                             dev->ifindex, &in_dev->cnf);
 690                 ip_rt_multicast_event(in_dev);
 691         }
 692
 693         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 694                 unregister_netdevice_queue(dev, head);
 695
 696         dev_put_track(dev, &v->dev_tracker);
 697         return 0;
 698 }
 699
 700 static void ipmr_cache_free_rcu(struct rcu_head *head)
 701 {
 702         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 703
 704         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 705 }
 706
 707 static void ipmr_cache_free(struct mfc_cache *c)
 708 {
 709         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 710 }
 711
 712 /* Destroy an unresolved cache entry, killing queued skbs
 713  * and reporting error to netlink readers.
 714  */
 715 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 716 {
 717         struct net *net = read_pnet(&mrt->net);
 718         struct sk_buff *skb;
 719         struct nlmsgerr *e;
 720
 721         atomic_dec(&mrt->cache_resolve_queue_len);
 722
 723         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 724                 if (ip_hdr(skb)->version == 0) {
 725                         struct nlmsghdr *nlh = skb_pull(skb,
 726                                                         sizeof(struct iphdr));
 727                         nlh->nlmsg_type = NLMSG_ERROR;
 728                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 729                         skb_trim(skb, nlh->nlmsg_len);
 730                         e = nlmsg_data(nlh);
 731                         e->error = -ETIMEDOUT;
 732                         memset(&e->msg, 0, sizeof(e->msg));
 733
 734                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 735                 } else {
 736                         kfree_skb(skb);
 737                 }
 738         }
 739
 740         ipmr_cache_free(c);
 741 }
 742
 743 /* Timer process for the unresolved queue. */
 744 static void ipmr_expire_process(struct timer_list *t)
 745 {
 746         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 747         struct mr_mfc *c, *next;
 748         unsigned long expires;
 749         unsigned long now;
 750
 751         if (!spin_trylock(&mfc_unres_lock)) {
 752                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 753                 return;
 754         }
 755
 756         if (list_empty(&mrt->mfc_unres_queue))
 757                 goto out;
 758
 759         now = jiffies;
 760         expires = 10*HZ;
 761
 762         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 763                 if (time_after(c->mfc_un.unres.expires, now)) {
 764                         unsigned long interval = c->mfc_un.unres.expires - now;
 765                         if (interval < expires)
 766                                 expires = interval;
 767                         continue;
 768                 }
 769
 770                 list_del(&c->list);
 771                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 772                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 773         }
 774
 775         if (!list_empty(&mrt->mfc_unres_queue))
 776                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 777
 778 out:
 779         spin_unlock(&mfc_unres_lock);
 780 }
 781
 782 /* Fill oifs list. It is called under write locked mrt_lock. */
 783 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 784                                    unsigned char *ttls)
 785 {
 786         int vifi;
 787
 788         cache->mfc_un.res.minvif = MAXVIFS;
 789         cache->mfc_un.res.maxvif = 0;
 790         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 791
 792         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 793                 if (VIF_EXISTS(mrt, vifi) &&
 794                     ttls[vifi] && ttls[vifi] < 255) {
 795                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 796                         if (cache->mfc_un.res.minvif > vifi)
 797                                 cache->mfc_un.res.minvif = vifi;
 798                         if (cache->mfc_un.res.maxvif <= vifi)
 799                                 cache->mfc_un.res.maxvif = vifi + 1;
 800                 }
 801         }
 802         cache->mfc_un.res.lastuse = jiffies;
 803 }
 804
 805 static int vif_add(struct net *net, struct mr_table *mrt,
 806                    struct vifctl *vifc, int mrtsock)
 807 {
 808         struct netdev_phys_item_id ppid = { };
 809         int vifi = vifc->vifc_vifi;
 810         struct vif_device *v = &mrt->vif_table[vifi];
 811         struct net_device *dev;
 812         struct in_device *in_dev;
 813         int err;
 814
 815         /* Is vif busy ? */
 816         if (VIF_EXISTS(mrt, vifi))
 817                 return -EADDRINUSE;
 818
 819         switch (vifc->vifc_flags) {
 820         case VIFF_REGISTER:
 821                 if (!ipmr_pimsm_enabled())
 822                         return -EINVAL;
 823                 /* Special Purpose VIF in PIM
 824                  * All the packets will be sent to the daemon
 825                  */
 826                 if (mrt->mroute_reg_vif_num >= 0)
 827                         return -EADDRINUSE;
 828                 dev = ipmr_reg_vif(net, mrt);
 829                 if (!dev)
 830                         return -ENOBUFS;
 831                 err = dev_set_allmulti(dev, 1);
 832                 if (err) {
 833                         unregister_netdevice(dev);
 834                         dev_put(dev);
 835                         return err;
 836                 }
 837                 break;
 838         case VIFF_TUNNEL:
 839                 dev = ipmr_new_tunnel(net, vifc);
 840                 if (IS_ERR(dev))
 841                         return PTR_ERR(dev);
 842                 break;
 843         case VIFF_USE_IFINDEX:
 844         case 0:
 845                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 846                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 847                         if (dev && !__in_dev_get_rtnl(dev)) {
 848                                 dev_put(dev);
 849                                 return -EADDRNOTAVAIL;
 850                         }
 851                 } else {
 852                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 853                 }
 854                 if (!dev)
 855                         return -EADDRNOTAVAIL;
 856                 err = dev_set_allmulti(dev, 1);
 857                 if (err) {
 858                         dev_put(dev);
 859                         return err;
 860                 }
 861                 break;
 862         default:
 863                 return -EINVAL;
 864         }
 865
 866         in_dev = __in_dev_get_rtnl(dev);
 867         if (!in_dev) {
 868                 dev_put(dev);
 869                 return -EADDRNOTAVAIL;
 870         }
 871         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 872         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 873                                     dev->ifindex, &in_dev->cnf);
 874         ip_rt_multicast_event(in_dev);
 875
 876         /* Fill in the VIF structures */
 877         vif_device_init(v, dev, vifc->vifc_rate_limit,
 878                         vifc->vifc_threshold,
 879                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 880                         (VIFF_TUNNEL | VIFF_REGISTER));
 881
 882         err = dev_get_port_parent_id(dev, &ppid, true);
 883         if (err == 0) {
 884                 memcpy(v->dev_parent_id.id, ppid.id, ppid.id_len);
 885                 v->dev_parent_id.id_len = ppid.id_len;
 886         } else {
 887                 v->dev_parent_id.id_len = 0;
 888         }
 889
 890         v->local = vifc->vifc_lcl_addr.s_addr;
 891         v->remote = vifc->vifc_rmt_addr.s_addr;
 892
 893         /* And finish update writing critical data */
 894         write_lock_bh(&mrt_lock);
 895         v->dev = dev;
 896         netdev_tracker_alloc(dev, &v->dev_tracker, GFP_ATOMIC);
 897         if (v->flags & VIFF_REGISTER)
 898                 mrt->mroute_reg_vif_num = vifi;
 899         if (vifi+1 > mrt->maxvif)
 900                 mrt->maxvif = vifi+1;
 901         write_unlock_bh(&mrt_lock);
 902         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 903         return 0;
 904 }
 905
 906 /* called with rcu_read_lock() */
 907 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 908                                          __be32 origin,
 909                                          __be32 mcastgrp)
 910 {
 911         struct mfc_cache_cmp_arg arg = {
 912                         .mfc_mcastgrp = mcastgrp,
 913                         .mfc_origin = origin
 914         };
 915
 916         return mr_mfc_find(mrt, &arg);
 917 }
 918
 919 /* Look for a (*,G) entry */
 920 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 921                                              __be32 mcastgrp, int vifi)
 922 {
 923         struct mfc_cache_cmp_arg arg = {
 924                         .mfc_mcastgrp = mcastgrp,
 925                         .mfc_origin = htonl(INADDR_ANY)
 926         };
 927
 928         if (mcastgrp == htonl(INADDR_ANY))
 929                 return mr_mfc_find_any_parent(mrt, vifi);
 930         return mr_mfc_find_any(mrt, vifi, &arg);
 931 }
 932
 933 /* Look for a (S,G,iif) entry if parent != -1 */
 934 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 935                                                 __be32 origin, __be32 mcastgrp,
 936                                                 int parent)
 937 {
 938         struct mfc_cache_cmp_arg arg = {
 939                         .mfc_mcastgrp = mcastgrp,
 940                         .mfc_origin = origin,
 941         };
 942
 943         return mr_mfc_find_parent(mrt, &arg, parent);
 944 }
 945
 946 /* Allocate a multicast cache entry */
 947 static struct mfc_cache *ipmr_cache_alloc(void)
 948 {
 949         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 950
 951         if (c) {
 952                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 953                 c->_c.mfc_un.res.minvif = MAXVIFS;
 954                 c->_c.free = ipmr_cache_free_rcu;
 955                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 956         }
 957         return c;
 958 }
 959
 960 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 961 {
 962         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 963
 964         if (c) {
 965                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
 966                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
 967         }
 968         return c;
 969 }
 970
 971 /* A cache entry has gone into a resolved state from queued */
 972 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 973                                struct mfc_cache *uc, struct mfc_cache *c)
 974 {
 975         struct sk_buff *skb;
 976         struct nlmsgerr *e;
 977
 978         /* Play the pending entries through our router */
 979         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
 980                 if (ip_hdr(skb)->version == 0) {
 981                         struct nlmsghdr *nlh = skb_pull(skb,
 982                                                         sizeof(struct iphdr));
 983
 984                         if (mr_fill_mroute(mrt, skb, &c->_c,
 985                                            nlmsg_data(nlh)) > 0) {
 986                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 987                                                  (u8 *)nlh;
 988                         } else {
 989                                 nlh->nlmsg_type = NLMSG_ERROR;
 990                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 991                                 skb_trim(skb, nlh->nlmsg_len);
 992                                 e = nlmsg_data(nlh);
 993                                 e->error = -EMSGSIZE;
 994                                 memset(&e->msg, 0, sizeof(e->msg));
 995                         }
 996
 997                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 998                 } else {
 999                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1000                 }
1001         }
1002 }
1003
1004 /* Bounce a cache query up to mrouted and netlink.
1005  *
1006  * Called under mrt_lock.
1007  */
1008 static int ipmr_cache_report(struct mr_table *mrt,
1009                              struct sk_buff *pkt, vifi_t vifi, int assert)
1010 {
1011         const int ihl = ip_hdrlen(pkt);
1012         struct sock *mroute_sk;
1013         struct igmphdr *igmp;
1014         struct igmpmsg *msg;
1015         struct sk_buff *skb;
1016         int ret;
1017
1018         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1019                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1020         else
1021                 skb = alloc_skb(128, GFP_ATOMIC);
1022
1023         if (!skb)
1024                 return -ENOBUFS;
1025
1026         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1027                 /* Ugly, but we have no choice with this interface.
1028                  * Duplicate old header, fix ihl, length etc.
1029                  * And all this only to mangle msg->im_msgtype and
1030                  * to set msg->im_mbz to "mbz" :-)
1031                  */
1032                 skb_push(skb, sizeof(struct iphdr));
1033                 skb_reset_network_header(skb);
1034                 skb_reset_transport_header(skb);
1035                 msg = (struct igmpmsg *)skb_network_header(skb);
1036                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1037                 msg->im_msgtype = assert;
1038                 msg->im_mbz = 0;
1039                 if (assert == IGMPMSG_WRVIFWHOLE) {
1040                         msg->im_vif = vifi;
1041                         msg->im_vif_hi = vifi >> 8;
1042                 } else {
1043                         msg->im_vif = mrt->mroute_reg_vif_num;
1044                         msg->im_vif_hi = mrt->mroute_reg_vif_num >> 8;
1045                 }
1046                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1047                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1048                                              sizeof(struct iphdr));
1049         } else {
1050                 /* Copy the IP header */
1051                 skb_set_network_header(skb, skb->len);
1052                 skb_put(skb, ihl);
1053                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1054                 /* Flag to the kernel this is a route add */
1055                 ip_hdr(skb)->protocol = 0;
1056                 msg = (struct igmpmsg *)skb_network_header(skb);
1057                 msg->im_vif = vifi;
1058                 msg->im_vif_hi = vifi >> 8;
1059                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1060                 /* Add our header */
1061                 igmp = skb_put(skb, sizeof(struct igmphdr));
1062                 igmp->type = assert;
1063                 msg->im_msgtype = assert;
1064                 igmp->code = 0;
1065                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1066                 skb->transport_header = skb->network_header;
1067         }
1068
1069         rcu_read_lock();
1070         mroute_sk = rcu_dereference(mrt->mroute_sk);
1071         if (!mroute_sk) {
1072                 rcu_read_unlock();
1073                 kfree_skb(skb);
1074                 return -EINVAL;
1075         }
1076
1077         igmpmsg_netlink_event(mrt, skb);
1078
1079         /* Deliver to mrouted */
1080         ret = sock_queue_rcv_skb(mroute_sk, skb);
1081         rcu_read_unlock();
1082         if (ret < 0) {
1083                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1084                 kfree_skb(skb);
1085         }
1086
1087         return ret;
1088 }
1089
1090 /* Queue a packet for resolution. It gets locked cache entry! */
1091 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1092                                  struct sk_buff *skb, struct net_device *dev)
1093 {
1094         const struct iphdr *iph = ip_hdr(skb);
1095         struct mfc_cache *c;
1096         bool found = false;
1097         int err;
1098
1099         spin_lock_bh(&mfc_unres_lock);
1100         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1101                 if (c->mfc_mcastgrp == iph->daddr &&
1102                     c->mfc_origin == iph->saddr) {
1103                         found = true;
1104                         break;
1105                 }
1106         }
1107
1108         if (!found) {
1109                 /* Create a new entry if allowable */
1110                 c = ipmr_cache_alloc_unres();
1111                 if (!c) {
1112                         spin_unlock_bh(&mfc_unres_lock);
1113
1114                         kfree_skb(skb);
1115                         return -ENOBUFS;
1116                 }
1117
1118                 /* Fill in the new cache entry */
1119                 c->_c.mfc_parent = -1;
1120                 c->mfc_origin   = iph->saddr;
1121                 c->mfc_mcastgrp = iph->daddr;
1122
1123                 /* Reflect first query at mrouted. */
1124                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1125
1126                 if (err < 0) {
1127                         /* If the report failed throw the cache entry
1128                            out - Brad Parker
1129                          */
1130                         spin_unlock_bh(&mfc_unres_lock);
1131
1132                         ipmr_cache_free(c);
1133                         kfree_skb(skb);
1134                         return err;
1135                 }
1136
1137                 atomic_inc(&mrt->cache_resolve_queue_len);
1138                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1139                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1140
1141                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1142                         mod_timer(&mrt->ipmr_expire_timer,
1143                                   c->_c.mfc_un.unres.expires);
1144         }
1145
1146         /* See if we can append the packet */
1147         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1148                 kfree_skb(skb);
1149                 err = -ENOBUFS;
1150         } else {
1151                 if (dev) {
1152                         skb->dev = dev;
1153                         skb->skb_iif = dev->ifindex;
1154                 }
1155                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1156                 err = 0;
1157         }
1158
1159         spin_unlock_bh(&mfc_unres_lock);
1160         return err;
1161 }
1162
1163 /* MFC cache manipulation by user space mroute daemon */
1164
1165 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1166 {
1167         struct net *net = read_pnet(&mrt->net);
1168         struct mfc_cache *c;
1169
1170         /* The entries are added/deleted only under RTNL */
1171         rcu_read_lock();
1172         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1173                                    mfc->mfcc_mcastgrp.s_addr, parent);
1174         rcu_read_unlock();
1175         if (!c)
1176                 return -ENOENT;
1177         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1178         list_del_rcu(&c->_c.list);
1179         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1180         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1181         mr_cache_put(&c->_c);
1182
1183         return 0;
1184 }
1185
1186 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1187                         struct mfcctl *mfc, int mrtsock, int parent)
1188 {
1189         struct mfc_cache *uc, *c;
1190         struct mr_mfc *_uc;
1191         bool found;
1192         int ret;
1193
1194         if (mfc->mfcc_parent >= MAXVIFS)
1195                 return -ENFILE;
1196
1197         /* The entries are added/deleted only under RTNL */
1198         rcu_read_lock();
1199         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1200                                    mfc->mfcc_mcastgrp.s_addr, parent);
1201         rcu_read_unlock();
1202         if (c) {
1203                 write_lock_bh(&mrt_lock);
1204                 c->_c.mfc_parent = mfc->mfcc_parent;
1205                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1206                 if (!mrtsock)
1207                         c->_c.mfc_flags |= MFC_STATIC;
1208                 write_unlock_bh(&mrt_lock);
1209                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1210                                               mrt->id);
1211                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1212                 return 0;
1213         }
1214
1215         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1216             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1217                 return -EINVAL;
1218
1219         c = ipmr_cache_alloc();
1220         if (!c)
1221                 return -ENOMEM;
1222
1223         c->mfc_origin = mfc->mfcc_origin.s_addr;
1224         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1225         c->_c.mfc_parent = mfc->mfcc_parent;
1226         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1227         if (!mrtsock)
1228                 c->_c.mfc_flags |= MFC_STATIC;
1229
1230         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1231                                   ipmr_rht_params);
1232         if (ret) {
1233                 pr_err("ipmr: rhtable insert error %d\n", ret);
1234                 ipmr_cache_free(c);
1235                 return ret;
1236         }
1237         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1238         /* Check to see if we resolved a queued list. If so we
1239          * need to send on the frames and tidy up.
1240          */
1241         found = false;
1242         spin_lock_bh(&mfc_unres_lock);
1243         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1244                 uc = (struct mfc_cache *)_uc;
1245                 if (uc->mfc_origin == c->mfc_origin &&
1246                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1247                         list_del(&_uc->list);
1248                         atomic_dec(&mrt->cache_resolve_queue_len);
1249                         found = true;
1250                         break;
1251                 }
1252         }
1253         if (list_empty(&mrt->mfc_unres_queue))
1254                 del_timer(&mrt->ipmr_expire_timer);
1255         spin_unlock_bh(&mfc_unres_lock);
1256
1257         if (found) {
1258                 ipmr_cache_resolve(net, mrt, uc, c);
1259                 ipmr_cache_free(uc);
1260         }
1261         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1262         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1263         return 0;
1264 }
1265
1266 /* Close the multicast socket, and clear the vif tables etc */
1267 static void mroute_clean_tables(struct mr_table *mrt, int flags)
1268 {
1269         struct net *net = read_pnet(&mrt->net);
1270         struct mr_mfc *c, *tmp;
1271         struct mfc_cache *cache;
1272         LIST_HEAD(list);
1273         int i;
1274
1275         /* Shut down all active vif entries */
1276         if (flags & (MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC)) {
1277                 for (i = 0; i < mrt->maxvif; i++) {
1278                         if (((mrt->vif_table[i].flags & VIFF_STATIC) &&
1279                              !(flags & MRT_FLUSH_VIFS_STATIC)) ||
1280                             (!(mrt->vif_table[i].flags & VIFF_STATIC) && !(flags & MRT_FLUSH_VIFS)))
1281                                 continue;
1282                         vif_delete(mrt, i, 0, &list);
1283                 }
1284                 unregister_netdevice_many(&list);
1285         }
1286
1287         /* Wipe the cache */
1288         if (flags & (MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC)) {
1289                 list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1290                         if (((c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC_STATIC)) ||
1291                             (!(c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC)))
1292                                 continue;
1293                         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1294                         list_del_rcu(&c->list);
1295                         cache = (struct mfc_cache *)c;
1296                         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1297                                                       mrt->id);
1298                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1299                         mr_cache_put(c);
1300                 }
1301         }
1302
1303         if (flags & MRT_FLUSH_MFC) {
1304                 if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1305                         spin_lock_bh(&mfc_unres_lock);
1306                         list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1307                                 list_del(&c->list);
1308                                 cache = (struct mfc_cache *)c;
1309                                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1310                                 ipmr_destroy_unres(mrt, cache);
1311                         }
1312                         spin_unlock_bh(&mfc_unres_lock);
1313                 }
1314         }
1315 }
1316
1317 /* called from ip_ra_control(), before an RCU grace period,
1318  * we don't need to call synchronize_rcu() here
1319  */
1320 static void mrtsock_destruct(struct sock *sk)
1321 {
1322         struct net *net = sock_net(sk);
1323         struct mr_table *mrt;
1324
1325         rtnl_lock();
1326         ipmr_for_each_table(mrt, net) {
1327                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1328                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1329                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1330                                                     NETCONFA_MC_FORWARDING,
1331                                                     NETCONFA_IFINDEX_ALL,
1332                                                     net->ipv4.devconf_all);
1333                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1334                         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_MFC);
1335                 }
1336         }
1337         rtnl_unlock();
1338 }
1339
1340 /* Socket options and virtual interface manipulation. The whole
1341  * virtual interface system is a complete heap, but unfortunately
1342  * that's how BSD mrouted happens to think. Maybe one day with a proper
1343  * MOSPF/PIM router set up we can clean this up.
1344  */
1345
1346 int ip_mroute_setsockopt(struct sock *sk, int optname, sockptr_t optval,
1347                          unsigned int optlen)
1348 {
1349         struct net *net = sock_net(sk);
1350         int val, ret = 0, parent = 0;
1351         struct mr_table *mrt;
1352         struct vifctl vif;
1353         struct mfcctl mfc;
1354         bool do_wrvifwhole;
1355         u32 uval;
1356
1357         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1358         rtnl_lock();
1359         if (sk->sk_type != SOCK_RAW ||
1360             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1361                 ret = -EOPNOTSUPP;
1362                 goto out_unlock;
1363         }
1364
1365         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1366         if (!mrt) {
1367                 ret = -ENOENT;
1368                 goto out_unlock;
1369         }
1370         if (optname != MRT_INIT) {
1371                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1372                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1373                         ret = -EACCES;
1374                         goto out_unlock;
1375                 }
1376         }
1377
1378         switch (optname) {
1379         case MRT_INIT:
1380                 if (optlen != sizeof(int)) {
1381                         ret = -EINVAL;
1382                         break;
1383                 }
1384                 if (rtnl_dereference(mrt->mroute_sk)) {
1385                         ret = -EADDRINUSE;
1386                         break;
1387                 }
1388
1389                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1390                 if (ret == 0) {
1391                         rcu_assign_pointer(mrt->mroute_sk, sk);
1392                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1393                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1394                                                     NETCONFA_MC_FORWARDING,
1395                                                     NETCONFA_IFINDEX_ALL,
1396                                                     net->ipv4.devconf_all);
1397                 }
1398                 break;
1399         case MRT_DONE:
1400                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1401                         ret = -EACCES;
1402                 } else {
1403                         /* We need to unlock here because mrtsock_destruct takes
1404                          * care of rtnl itself and we can't change that due to
1405                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1406                          */
1407                         rtnl_unlock();
1408                         ret = ip_ra_control(sk, 0, NULL);
1409                         goto out;
1410                 }
1411                 break;
1412         case MRT_ADD_VIF:
1413         case MRT_DEL_VIF:
1414                 if (optlen != sizeof(vif)) {
1415                         ret = -EINVAL;
1416                         break;
1417                 }
1418                 if (copy_from_sockptr(&vif, optval, sizeof(vif))) {
1419                         ret = -EFAULT;
1420                         break;
1421                 }
1422                 if (vif.vifc_vifi >= MAXVIFS) {
1423                         ret = -ENFILE;
1424                         break;
1425                 }
1426                 if (optname == MRT_ADD_VIF) {
1427                         ret = vif_add(net, mrt, &vif,
1428                                       sk == rtnl_dereference(mrt->mroute_sk));
1429                 } else {
1430                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1431                 }
1432                 break;
1433         /* Manipulate the forwarding caches. These live
1434          * in a sort of kernel/user symbiosis.
1435          */
1436         case MRT_ADD_MFC:
1437         case MRT_DEL_MFC:
1438                 parent = -1;
1439                 fallthrough;
1440         case MRT_ADD_MFC_PROXY:
1441         case MRT_DEL_MFC_PROXY:
1442                 if (optlen != sizeof(mfc)) {
1443                         ret = -EINVAL;
1444                         break;
1445                 }
1446                 if (copy_from_sockptr(&mfc, optval, sizeof(mfc))) {
1447                         ret = -EFAULT;
1448                         break;
1449                 }
1450                 if (parent == 0)
1451                         parent = mfc.mfcc_parent;
1452                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1453                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1454                 else
1455                         ret = ipmr_mfc_add(net, mrt, &mfc,
1456                                            sk == rtnl_dereference(mrt->mroute_sk),
1457                                            parent);
1458                 break;
1459         case MRT_FLUSH:
1460                 if (optlen != sizeof(val)) {
1461                         ret = -EINVAL;
1462                         break;
1463                 }
1464                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1465                         ret = -EFAULT;
1466                         break;
1467                 }
1468                 mroute_clean_tables(mrt, val);
1469                 break;
1470         /* Control PIM assert. */
1471         case MRT_ASSERT:
1472                 if (optlen != sizeof(val)) {
1473                         ret = -EINVAL;
1474                         break;
1475                 }
1476                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1477                         ret = -EFAULT;
1478                         break;
1479                 }
1480                 mrt->mroute_do_assert = val;
1481                 break;
1482         case MRT_PIM:
1483                 if (!ipmr_pimsm_enabled()) {
1484                         ret = -ENOPROTOOPT;
1485                         break;
1486                 }
1487                 if (optlen != sizeof(val)) {
1488                         ret = -EINVAL;
1489                         break;
1490                 }
1491                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1492                         ret = -EFAULT;
1493                         break;
1494                 }
1495
1496                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1497                 val = !!val;
1498                 if (val != mrt->mroute_do_pim) {
1499                         mrt->mroute_do_pim = val;
1500                         mrt->mroute_do_assert = val;
1501                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1502                 }
1503                 break;
1504         case MRT_TABLE:
1505                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1506                         ret = -ENOPROTOOPT;
1507                         break;
1508                 }
1509                 if (optlen != sizeof(uval)) {
1510                         ret = -EINVAL;
1511                         break;
1512                 }
1513                 if (copy_from_sockptr(&uval, optval, sizeof(uval))) {
1514                         ret = -EFAULT;
1515                         break;
1516                 }
1517
1518                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1519                         ret = -EBUSY;
1520                 } else {
1521                         mrt = ipmr_new_table(net, uval);
1522                         if (IS_ERR(mrt))
1523                                 ret = PTR_ERR(mrt);
1524                         else
1525                                 raw_sk(sk)->ipmr_table = uval;
1526                 }
1527                 break;
1528         /* Spurious command, or MRT_VERSION which you cannot set. */
1529         default:
1530                 ret = -ENOPROTOOPT;
1531         }
1532 out_unlock:
1533         rtnl_unlock();
1534 out:
1535         return ret;
1536 }
1537
1538 /* Getsock opt support for the multicast routing system. */
1539 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1540 {
1541         int olr;
1542         int val;
1543         struct net *net = sock_net(sk);
1544         struct mr_table *mrt;
1545
1546         if (sk->sk_type != SOCK_RAW ||
1547             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1548                 return -EOPNOTSUPP;
1549
1550         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1551         if (!mrt)
1552                 return -ENOENT;
1553
1554         switch (optname) {
1555         case MRT_VERSION:
1556                 val = 0x0305;
1557                 break;
1558         case MRT_PIM:
1559                 if (!ipmr_pimsm_enabled())
1560                         return -ENOPROTOOPT;
1561                 val = mrt->mroute_do_pim;
1562                 break;
1563         case MRT_ASSERT:
1564                 val = mrt->mroute_do_assert;
1565                 break;
1566         default:
1567                 return -ENOPROTOOPT;
1568         }
1569
1570         if (get_user(olr, optlen))
1571                 return -EFAULT;
1572         olr = min_t(unsigned int, olr, sizeof(int));
1573         if (olr < 0)
1574                 return -EINVAL;
1575         if (put_user(olr, optlen))
1576                 return -EFAULT;
1577         if (copy_to_user(optval, &val, olr))
1578                 return -EFAULT;
1579         return 0;
1580 }
1581
1582 /* The IP multicast ioctl support routines. */
1583 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1584 {
1585         struct sioc_sg_req sr;
1586         struct sioc_vif_req vr;
1587         struct vif_device *vif;
1588         struct mfc_cache *c;
1589         struct net *net = sock_net(sk);
1590         struct mr_table *mrt;
1591
1592         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1593         if (!mrt)
1594                 return -ENOENT;
1595
1596         switch (cmd) {
1597         case SIOCGETVIFCNT:
1598                 if (copy_from_user(&vr, arg, sizeof(vr)))
1599                         return -EFAULT;
1600                 if (vr.vifi >= mrt->maxvif)
1601                         return -EINVAL;
1602                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1603                 read_lock(&mrt_lock);
1604                 vif = &mrt->vif_table[vr.vifi];
1605                 if (VIF_EXISTS(mrt, vr.vifi)) {
1606                         vr.icount = vif->pkt_in;
1607                         vr.ocount = vif->pkt_out;
1608                         vr.ibytes = vif->bytes_in;
1609                         vr.obytes = vif->bytes_out;
1610                         read_unlock(&mrt_lock);
1611
1612                         if (copy_to_user(arg, &vr, sizeof(vr)))
1613                                 return -EFAULT;
1614                         return 0;
1615                 }
1616                 read_unlock(&mrt_lock);
1617                 return -EADDRNOTAVAIL;
1618         case SIOCGETSGCNT:
1619                 if (copy_from_user(&sr, arg, sizeof(sr)))
1620                         return -EFAULT;
1621
1622                 rcu_read_lock();
1623                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1624                 if (c) {
1625                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1626                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1627                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1628                         rcu_read_unlock();
1629
1630                         if (copy_to_user(arg, &sr, sizeof(sr)))
1631                                 return -EFAULT;
1632                         return 0;
1633                 }
1634                 rcu_read_unlock();
1635                 return -EADDRNOTAVAIL;
1636         default:
1637                 return -ENOIOCTLCMD;
1638         }
1639 }
1640
1641 #ifdef CONFIG_COMPAT
1642 struct compat_sioc_sg_req {
1643         struct in_addr src;
1644         struct in_addr grp;
1645         compat_ulong_t pktcnt;
1646         compat_ulong_t bytecnt;
1647         compat_ulong_t wrong_if;
1648 };
1649
1650 struct compat_sioc_vif_req {
1651         vifi_t  vifi;           /* Which iface */
1652         compat_ulong_t icount;
1653         compat_ulong_t ocount;
1654         compat_ulong_t ibytes;
1655         compat_ulong_t obytes;
1656 };
1657
1658 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1659 {
1660         struct compat_sioc_sg_req sr;
1661         struct compat_sioc_vif_req vr;
1662         struct vif_device *vif;
1663         struct mfc_cache *c;
1664         struct net *net = sock_net(sk);
1665         struct mr_table *mrt;
1666
1667         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1668         if (!mrt)
1669                 return -ENOENT;
1670
1671         switch (cmd) {
1672         case SIOCGETVIFCNT:
1673                 if (copy_from_user(&vr, arg, sizeof(vr)))
1674                         return -EFAULT;
1675                 if (vr.vifi >= mrt->maxvif)
1676                         return -EINVAL;
1677                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1678                 read_lock(&mrt_lock);
1679                 vif = &mrt->vif_table[vr.vifi];
1680                 if (VIF_EXISTS(mrt, vr.vifi)) {
1681                         vr.icount = vif->pkt_in;
1682                         vr.ocount = vif->pkt_out;
1683                         vr.ibytes = vif->bytes_in;
1684                         vr.obytes = vif->bytes_out;
1685                         read_unlock(&mrt_lock);
1686
1687                         if (copy_to_user(arg, &vr, sizeof(vr)))
1688                                 return -EFAULT;
1689                         return 0;
1690                 }
1691                 read_unlock(&mrt_lock);
1692                 return -EADDRNOTAVAIL;
1693         case SIOCGETSGCNT:
1694                 if (copy_from_user(&sr, arg, sizeof(sr)))
1695                         return -EFAULT;
1696
1697                 rcu_read_lock();
1698                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1699                 if (c) {
1700                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1701                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1702                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1703                         rcu_read_unlock();
1704
1705                         if (copy_to_user(arg, &sr, sizeof(sr)))
1706                                 return -EFAULT;
1707                         return 0;
1708                 }
1709                 rcu_read_unlock();
1710                 return -EADDRNOTAVAIL;
1711         default:
1712                 return -ENOIOCTLCMD;
1713         }
1714 }
1715 #endif
1716
1717 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1718 {
1719         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1720         struct net *net = dev_net(dev);
1721         struct mr_table *mrt;
1722         struct vif_device *v;
1723         int ct;
1724
1725         if (event != NETDEV_UNREGISTER)
1726                 return NOTIFY_DONE;
1727
1728         ipmr_for_each_table(mrt, net) {
1729                 v = &mrt->vif_table[0];
1730                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1731                         if (v->dev == dev)
1732                                 vif_delete(mrt, ct, 1, NULL);
1733                 }
1734         }
1735         return NOTIFY_DONE;
1736 }
1737
1738 static struct notifier_block ip_mr_notifier = {
1739         .notifier_call = ipmr_device_event,
1740 };
1741
1742 /* Encapsulate a packet by attaching a valid IPIP header to it.
1743  * This avoids tunnel drivers and other mess and gives us the speed so
1744  * important for multicast video.
1745  */
1746 static void ip_encap(struct net *net, struct sk_buff *skb,
1747                      __be32 saddr, __be32 daddr)
1748 {
1749         struct iphdr *iph;
1750         const struct iphdr *old_iph = ip_hdr(skb);
1751
1752         skb_push(skb, sizeof(struct iphdr));
1753         skb->transport_header = skb->network_header;
1754         skb_reset_network_header(skb);
1755         iph = ip_hdr(skb);
1756
1757         iph->version    =       4;
1758         iph->tos        =       old_iph->tos;
1759         iph->ttl        =       old_iph->ttl;
1760         iph->frag_off   =       0;
1761         iph->daddr      =       daddr;
1762         iph->saddr      =       saddr;
1763         iph->protocol   =       IPPROTO_IPIP;
1764         iph->ihl        =       5;
1765         iph->tot_len    =       htons(skb->len);
1766         ip_select_ident(net, skb, NULL);
1767         ip_send_check(iph);
1768
1769         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1770         nf_reset_ct(skb);
1771 }
1772
1773 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1774                                       struct sk_buff *skb)
1775 {
1776         struct ip_options *opt = &(IPCB(skb)->opt);
1777
1778         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1779         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1780
1781         if (unlikely(opt->optlen))
1782                 ip_forward_options(skb);
1783
1784         return dst_output(net, sk, skb);
1785 }
1786
1787 #ifdef CONFIG_NET_SWITCHDEV
1788 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1789                                    int in_vifi, int out_vifi)
1790 {
1791         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1792         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1793
1794         if (!skb->offload_l3_fwd_mark)
1795                 return false;
1796         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1797                 return false;
1798         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1799                                         &in_vif->dev_parent_id);
1800 }
1801 #else
1802 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1803                                    int in_vifi, int out_vifi)
1804 {
1805         return false;
1806 }
1807 #endif
1808
1809 /* Processing handlers for ipmr_forward */
1810
1811 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1812                             int in_vifi, struct sk_buff *skb, int vifi)
1813 {
1814         const struct iphdr *iph = ip_hdr(skb);
1815         struct vif_device *vif = &mrt->vif_table[vifi];
1816         struct net_device *dev;
1817         struct rtable *rt;
1818         struct flowi4 fl4;
1819         int    encap = 0;
1820
1821         if (!vif->dev)
1822                 goto out_free;
1823
1824         if (vif->flags & VIFF_REGISTER) {
1825                 vif->pkt_out++;
1826                 vif->bytes_out += skb->len;
1827                 vif->dev->stats.tx_bytes += skb->len;
1828                 vif->dev->stats.tx_packets++;
1829                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1830                 goto out_free;
1831         }
1832
1833         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1834                 goto out_free;
1835
1836         if (vif->flags & VIFF_TUNNEL) {
1837                 rt = ip_route_output_ports(net, &fl4, NULL,
1838                                            vif->remote, vif->local,
1839                                            0, 0,
1840                                            IPPROTO_IPIP,
1841                                            RT_TOS(iph->tos), vif->link);
1842                 if (IS_ERR(rt))
1843                         goto out_free;
1844                 encap = sizeof(struct iphdr);
1845         } else {
1846                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1847                                            0, 0,
1848                                            IPPROTO_IPIP,
1849                                            RT_TOS(iph->tos), vif->link);
1850                 if (IS_ERR(rt))
1851                         goto out_free;
1852         }
1853
1854         dev = rt->dst.dev;
1855
1856         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1857                 /* Do not fragment multicasts. Alas, IPv4 does not
1858                  * allow to send ICMP, so that packets will disappear
1859                  * to blackhole.
1860                  */
1861                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1862                 ip_rt_put(rt);
1863                 goto out_free;
1864         }
1865
1866         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1867
1868         if (skb_cow(skb, encap)) {
1869                 ip_rt_put(rt);
1870                 goto out_free;
1871         }
1872
1873         vif->pkt_out++;
1874         vif->bytes_out += skb->len;
1875
1876         skb_dst_drop(skb);
1877         skb_dst_set(skb, &rt->dst);
1878         ip_decrease_ttl(ip_hdr(skb));
1879
1880         /* FIXME: forward and output firewalls used to be called here.
1881          * What do we do with netfilter? -- RR
1882          */
1883         if (vif->flags & VIFF_TUNNEL) {
1884                 ip_encap(net, skb, vif->local, vif->remote);
1885                 /* FIXME: extra output firewall step used to be here. --RR */
1886                 vif->dev->stats.tx_packets++;
1887                 vif->dev->stats.tx_bytes += skb->len;
1888         }
1889
1890         IPCB(skb)->flags |= IPSKB_FORWARDED;
1891
1892         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1893          * not only before forwarding, but after forwarding on all output
1894          * interfaces. It is clear, if mrouter runs a multicasting
1895          * program, it should receive packets not depending to what interface
1896          * program is joined.
1897          * If we will not make it, the program will have to join on all
1898          * interfaces. On the other hand, multihoming host (or router, but
1899          * not mrouter) cannot join to more than one interface - it will
1900          * result in receiving multiple packets.
1901          */
1902         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1903                 net, NULL, skb, skb->dev, dev,
1904                 ipmr_forward_finish);
1905         return;
1906
1907 out_free:
1908         kfree_skb(skb);
1909 }
1910
1911 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1912 {
1913         int ct;
1914
1915         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1916                 if (mrt->vif_table[ct].dev == dev)
1917                         break;
1918         }
1919         return ct;
1920 }
1921
1922 /* "local" means that we should preserve one skb (for local delivery) */
1923 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1924                           struct net_device *dev, struct sk_buff *skb,
1925                           struct mfc_cache *c, int local)
1926 {
1927         int true_vifi = ipmr_find_vif(mrt, dev);
1928         int psend = -1;
1929         int vif, ct;
1930
1931         vif = c->_c.mfc_parent;
1932         c->_c.mfc_un.res.pkt++;
1933         c->_c.mfc_un.res.bytes += skb->len;
1934         c->_c.mfc_un.res.lastuse = jiffies;
1935
1936         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1937                 struct mfc_cache *cache_proxy;
1938
1939                 /* For an (*,G) entry, we only check that the incoming
1940                  * interface is part of the static tree.
1941                  */
1942                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1943                 if (cache_proxy &&
1944                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1945                         goto forward;
1946         }
1947
1948         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1949         if (mrt->vif_table[vif].dev != dev) {
1950                 if (rt_is_output_route(skb_rtable(skb))) {
1951                         /* It is our own packet, looped back.
1952                          * Very complicated situation...
1953                          *
1954                          * The best workaround until routing daemons will be
1955                          * fixed is not to redistribute packet, if it was
1956                          * send through wrong interface. It means, that
1957                          * multicast applications WILL NOT work for
1958                          * (S,G), which have default multicast route pointing
1959                          * to wrong oif. In any case, it is not a good
1960                          * idea to use multicasting applications on router.
1961                          */
1962                         goto dont_forward;
1963                 }
1964
1965                 c->_c.mfc_un.res.wrong_if++;
1966
1967                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1968                     /* pimsm uses asserts, when switching from RPT to SPT,
1969                      * so that we cannot check that packet arrived on an oif.
1970                      * It is bad, but otherwise we would need to move pretty
1971                      * large chunk of pimd to kernel. Ough... --ANK
1972                      */
1973                     (mrt->mroute_do_pim ||
1974                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1975                     time_after(jiffies,
1976                                c->_c.mfc_un.res.last_assert +
1977                                MFC_ASSERT_THRESH)) {
1978                         c->_c.mfc_un.res.last_assert = jiffies;
1979                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1980                         if (mrt->mroute_do_wrvifwhole)
1981                                 ipmr_cache_report(mrt, skb, true_vifi,
1982                                                   IGMPMSG_WRVIFWHOLE);
1983                 }
1984                 goto dont_forward;
1985         }
1986
1987 forward:
1988         mrt->vif_table[vif].pkt_in++;
1989         mrt->vif_table[vif].bytes_in += skb->len;
1990
1991         /* Forward the frame */
1992         if (c->mfc_origin == htonl(INADDR_ANY) &&
1993             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
1994                 if (true_vifi >= 0 &&
1995                     true_vifi != c->_c.mfc_parent &&
1996                     ip_hdr(skb)->ttl >
1997                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
1998                         /* It's an (*,*) entry and the packet is not coming from
1999                          * the upstream: forward the packet to the upstream
2000                          * only.
2001                          */
2002                         psend = c->_c.mfc_parent;
2003                         goto last_forward;
2004                 }
2005                 goto dont_forward;
2006         }
2007         for (ct = c->_c.mfc_un.res.maxvif - 1;
2008              ct >= c->_c.mfc_un.res.minvif; ct--) {
2009                 /* For (*,G) entry, don't forward to the incoming interface */
2010                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2011                      ct != true_vifi) &&
2012                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2013                         if (psend != -1) {
2014                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2015
2016                                 if (skb2)
2017                                         ipmr_queue_xmit(net, mrt, true_vifi,
2018                                                         skb2, psend);
2019                         }
2020                         psend = ct;
2021                 }
2022         }
2023 last_forward:
2024         if (psend != -1) {
2025                 if (local) {
2026                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2027
2028                         if (skb2)
2029                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2030                                                 psend);
2031                 } else {
2032                         ipmr_queue_xmit(net, mrt, true_vifi, skb, psend);
2033                         return;
2034                 }
2035         }
2036
2037 dont_forward:
2038         if (!local)
2039                 kfree_skb(skb);
2040 }
2041
2042 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2043 {
2044         struct rtable *rt = skb_rtable(skb);
2045         struct iphdr *iph = ip_hdr(skb);
2046         struct flowi4 fl4 = {
2047                 .daddr = iph->daddr,
2048                 .saddr = iph->saddr,
2049                 .flowi4_tos = RT_TOS(iph->tos),
2050                 .flowi4_oif = (rt_is_output_route(rt) ?
2051                                skb->dev->ifindex : 0),
2052                 .flowi4_iif = (rt_is_output_route(rt) ?
2053                                LOOPBACK_IFINDEX :
2054                                skb->dev->ifindex),
2055                 .flowi4_mark = skb->mark,
2056         };
2057         struct mr_table *mrt;
2058         int err;
2059
2060         err = ipmr_fib_lookup(net, &fl4, &mrt);
2061         if (err)
2062                 return ERR_PTR(err);
2063         return mrt;
2064 }
2065
2066 /* Multicast packets for forwarding arrive here
2067  * Called with rcu_read_lock();
2068  */
2069 int ip_mr_input(struct sk_buff *skb)
2070 {
2071         struct mfc_cache *cache;
2072         struct net *net = dev_net(skb->dev);
2073         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2074         struct mr_table *mrt;
2075         struct net_device *dev;
2076
2077         /* skb->dev passed in is the loX master dev for vrfs.
2078          * As there are no vifs associated with loopback devices,
2079          * get the proper interface that does have a vif associated with it.
2080          */
2081         dev = skb->dev;
2082         if (netif_is_l3_master(skb->dev)) {
2083                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2084                 if (!dev) {
2085                         kfree_skb(skb);
2086                         return -ENODEV;
2087                 }
2088         }
2089
2090         /* Packet is looped back after forward, it should not be
2091          * forwarded second time, but still can be delivered locally.
2092          */
2093         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2094                 goto dont_forward;
2095
2096         mrt = ipmr_rt_fib_lookup(net, skb);
2097         if (IS_ERR(mrt)) {
2098                 kfree_skb(skb);
2099                 return PTR_ERR(mrt);
2100         }
2101         if (!local) {
2102                 if (IPCB(skb)->opt.router_alert) {
2103                         if (ip_call_ra_chain(skb))
2104                                 return 0;
2105                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2106                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2107                          * Cisco IOS <= 11.2(8)) do not put router alert
2108                          * option to IGMP packets destined to routable
2109                          * groups. It is very bad, because it means
2110                          * that we can forward NO IGMP messages.
2111                          */
2112                         struct sock *mroute_sk;
2113
2114                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2115                         if (mroute_sk) {
2116                                 nf_reset_ct(skb);
2117                                 raw_rcv(mroute_sk, skb);
2118                                 return 0;
2119                         }
2120                 }
2121         }
2122
2123         /* already under rcu_read_lock() */
2124         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2125         if (!cache) {
2126                 int vif = ipmr_find_vif(mrt, dev);
2127
2128                 if (vif >= 0)
2129                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2130                                                     vif);
2131         }
2132
2133         /* No usable cache entry */
2134         if (!cache) {
2135                 int vif;
2136
2137                 if (local) {
2138                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2139                         ip_local_deliver(skb);
2140                         if (!skb2)
2141                                 return -ENOBUFS;
2142                         skb = skb2;
2143                 }
2144
2145                 read_lock(&mrt_lock);
2146                 vif = ipmr_find_vif(mrt, dev);
2147                 if (vif >= 0) {
2148                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2149                         read_unlock(&mrt_lock);
2150
2151                         return err2;
2152                 }
2153                 read_unlock(&mrt_lock);
2154                 kfree_skb(skb);
2155                 return -ENODEV;
2156         }
2157
2158         read_lock(&mrt_lock);
2159         ip_mr_forward(net, mrt, dev, skb, cache, local);
2160         read_unlock(&mrt_lock);
2161
2162         if (local)
2163                 return ip_local_deliver(skb);
2164
2165         return 0;
2166
2167 dont_forward:
2168         if (local)
2169                 return ip_local_deliver(skb);
2170         kfree_skb(skb);
2171         return 0;
2172 }
2173
2174 #ifdef CONFIG_IP_PIMSM_V1
2175 /* Handle IGMP messages of PIMv1 */
2176 int pim_rcv_v1(struct sk_buff *skb)
2177 {
2178         struct igmphdr *pim;
2179         struct net *net = dev_net(skb->dev);
2180         struct mr_table *mrt;
2181
2182         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2183                 goto drop;
2184
2185         pim = igmp_hdr(skb);
2186
2187         mrt = ipmr_rt_fib_lookup(net, skb);
2188         if (IS_ERR(mrt))
2189                 goto drop;
2190         if (!mrt->mroute_do_pim ||
2191             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2192                 goto drop;
2193
2194         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2195 drop:
2196                 kfree_skb(skb);
2197         }
2198         return 0;
2199 }
2200 #endif
2201
2202 #ifdef CONFIG_IP_PIMSM_V2
2203 static int pim_rcv(struct sk_buff *skb)
2204 {
2205         struct pimreghdr *pim;
2206         struct net *net = dev_net(skb->dev);
2207         struct mr_table *mrt;
2208
2209         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2210                 goto drop;
2211
2212         pim = (struct pimreghdr *)skb_transport_header(skb);
2213         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2214             (pim->flags & PIM_NULL_REGISTER) ||
2215             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2216              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2217                 goto drop;
2218
2219         mrt = ipmr_rt_fib_lookup(net, skb);
2220         if (IS_ERR(mrt))
2221                 goto drop;
2222         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2223 drop:
2224                 kfree_skb(skb);
2225         }
2226         return 0;
2227 }
2228 #endif
2229
2230 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2231                    __be32 saddr, __be32 daddr,
2232                    struct rtmsg *rtm, u32 portid)
2233 {
2234         struct mfc_cache *cache;
2235         struct mr_table *mrt;
2236         int err;
2237
2238         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2239         if (!mrt)
2240                 return -ENOENT;
2241
2242         rcu_read_lock();
2243         cache = ipmr_cache_find(mrt, saddr, daddr);
2244         if (!cache && skb->dev) {
2245                 int vif = ipmr_find_vif(mrt, skb->dev);
2246
2247                 if (vif >= 0)
2248                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2249         }
2250         if (!cache) {
2251                 struct sk_buff *skb2;
2252                 struct iphdr *iph;
2253                 struct net_device *dev;
2254                 int vif = -1;
2255
2256                 dev = skb->dev;
2257                 read_lock(&mrt_lock);
2258                 if (dev)
2259                         vif = ipmr_find_vif(mrt, dev);
2260                 if (vif < 0) {
2261                         read_unlock(&mrt_lock);
2262                         rcu_read_unlock();
2263                         return -ENODEV;
2264                 }
2265
2266                 skb2 = skb_realloc_headroom(skb, sizeof(struct iphdr));
2267                 if (!skb2) {
2268                         read_unlock(&mrt_lock);
2269                         rcu_read_unlock();
2270                         return -ENOMEM;
2271                 }
2272
2273                 NETLINK_CB(skb2).portid = portid;
2274                 skb_push(skb2, sizeof(struct iphdr));
2275                 skb_reset_network_header(skb2);
2276                 iph = ip_hdr(skb2);
2277                 iph->ihl = sizeof(struct iphdr) >> 2;
2278                 iph->saddr = saddr;
2279                 iph->daddr = daddr;
2280                 iph->version = 0;
2281                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2282                 read_unlock(&mrt_lock);
2283                 rcu_read_unlock();
2284                 return err;
2285         }
2286
2287         read_lock(&mrt_lock);
2288         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2289         read_unlock(&mrt_lock);
2290         rcu_read_unlock();
2291         return err;
2292 }
2293
2294 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2295                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2296                             int flags)
2297 {
2298         struct nlmsghdr *nlh;
2299         struct rtmsg *rtm;
2300         int err;
2301
2302         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2303         if (!nlh)
2304                 return -EMSGSIZE;
2305
2306         rtm = nlmsg_data(nlh);
2307         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2308         rtm->rtm_dst_len  = 32;
2309         rtm->rtm_src_len  = 32;
2310         rtm->rtm_tos      = 0;
2311         rtm->rtm_table    = mrt->id;
2312         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2313                 goto nla_put_failure;
2314         rtm->rtm_type     = RTN_MULTICAST;
2315         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2316         if (c->_c.mfc_flags & MFC_STATIC)
2317                 rtm->rtm_protocol = RTPROT_STATIC;
2318         else
2319                 rtm->rtm_protocol = RTPROT_MROUTED;
2320         rtm->rtm_flags    = 0;
2321
2322         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2323             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2324                 goto nla_put_failure;
2325         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2326         /* do not break the dump if cache is unresolved */
2327         if (err < 0 && err != -ENOENT)
2328                 goto nla_put_failure;
2329
2330         nlmsg_end(skb, nlh);
2331         return 0;
2332
2333 nla_put_failure:
2334         nlmsg_cancel(skb, nlh);
2335         return -EMSGSIZE;
2336 }
2337
2338 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2339                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2340                              int flags)
2341 {
2342         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2343                                 cmd, flags);
2344 }
2345
2346 static size_t mroute_msgsize(bool unresolved, int maxvif)
2347 {
2348         size_t len =
2349                 NLMSG_ALIGN(sizeof(struct rtmsg))
2350                 + nla_total_size(4)     /* RTA_TABLE */
2351                 + nla_total_size(4)     /* RTA_SRC */
2352                 + nla_total_size(4)     /* RTA_DST */
2353                 ;
2354
2355         if (!unresolved)
2356                 len = len
2357                       + nla_total_size(4)       /* RTA_IIF */
2358                       + nla_total_size(0)       /* RTA_MULTIPATH */
2359                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2360                                                 /* RTA_MFC_STATS */
2361                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2362                 ;
2363
2364         return len;
2365 }
2366
2367 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2368                                  int cmd)
2369 {
2370         struct net *net = read_pnet(&mrt->net);
2371         struct sk_buff *skb;
2372         int err = -ENOBUFS;
2373
2374         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2375                                        mrt->maxvif),
2376                         GFP_ATOMIC);
2377         if (!skb)
2378                 goto errout;
2379
2380         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2381         if (err < 0)
2382                 goto errout;
2383
2384         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2385         return;
2386
2387 errout:
2388         kfree_skb(skb);
2389         if (err < 0)
2390                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2391 }
2392
2393 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2394 {
2395         size_t len =
2396                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2397                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2398                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2399                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2400                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2401                 + nla_total_size(4)     /* IPMRA_CREPORT_TABLE */
2402                                         /* IPMRA_CREPORT_PKT */
2403                 + nla_total_size(payloadlen)
2404                 ;
2405
2406         return len;
2407 }
2408
2409 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2410 {
2411         struct net *net = read_pnet(&mrt->net);
2412         struct nlmsghdr *nlh;
2413         struct rtgenmsg *rtgenm;
2414         struct igmpmsg *msg;
2415         struct sk_buff *skb;
2416         struct nlattr *nla;
2417         int payloadlen;
2418
2419         payloadlen = pkt->len - sizeof(struct igmpmsg);
2420         msg = (struct igmpmsg *)skb_network_header(pkt);
2421
2422         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2423         if (!skb)
2424                 goto errout;
2425
2426         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2427                         sizeof(struct rtgenmsg), 0);
2428         if (!nlh)
2429                 goto errout;
2430         rtgenm = nlmsg_data(nlh);
2431         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2432         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2433             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif | (msg->im_vif_hi << 8)) ||
2434             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2435                             msg->im_src.s_addr) ||
2436             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2437                             msg->im_dst.s_addr) ||
2438             nla_put_u32(skb, IPMRA_CREPORT_TABLE, mrt->id))
2439                 goto nla_put_failure;
2440
2441         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2442         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2443                                   nla_data(nla), payloadlen))
2444                 goto nla_put_failure;
2445
2446         nlmsg_end(skb, nlh);
2447
2448         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2449         return;
2450
2451 nla_put_failure:
2452         nlmsg_cancel(skb, nlh);
2453 errout:
2454         kfree_skb(skb);
2455         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2456 }
2457
2458 static int ipmr_rtm_valid_getroute_req(struct sk_buff *skb,
2459                                        const struct nlmsghdr *nlh,
2460                                        struct nlattr **tb,
2461                                        struct netlink_ext_ack *extack)
2462 {
2463         struct rtmsg *rtm;
2464         int i, err;
2465
2466         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2467                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for multicast route get request");
2468                 return -EINVAL;
2469         }
2470
2471         if (!netlink_strict_get_check(skb))
2472                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2473                                               rtm_ipv4_policy, extack);
2474
2475         rtm = nlmsg_data(nlh);
2476         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2477             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2478             rtm->rtm_tos || rtm->rtm_table || rtm->rtm_protocol ||
2479             rtm->rtm_scope || rtm->rtm_type || rtm->rtm_flags) {
2480                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for multicast route get request");
2481                 return -EINVAL;
2482         }
2483
2484         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2485                                             rtm_ipv4_policy, extack);
2486         if (err)
2487                 return err;
2488
2489         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2490             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2491                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2492                 return -EINVAL;
2493         }
2494
2495         for (i = 0; i <= RTA_MAX; i++) {
2496                 if (!tb[i])
2497                         continue;
2498
2499                 switch (i) {
2500                 case RTA_SRC:
2501                 case RTA_DST:
2502                 case RTA_TABLE:
2503                         break;
2504                 default:
2505                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in multicast route get request");
2506                         return -EINVAL;
2507                 }
2508         }
2509
2510         return 0;
2511 }
2512
2513 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2514                              struct netlink_ext_ack *extack)
2515 {
2516         struct net *net = sock_net(in_skb->sk);
2517         struct nlattr *tb[RTA_MAX + 1];
2518         struct sk_buff *skb = NULL;
2519         struct mfc_cache *cache;
2520         struct mr_table *mrt;
2521         __be32 src, grp;
2522         u32 tableid;
2523         int err;
2524
2525         err = ipmr_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2526         if (err < 0)
2527                 goto errout;
2528
2529         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2530         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2531         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2532
2533         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2534         if (!mrt) {
2535                 err = -ENOENT;
2536                 goto errout_free;
2537         }
2538
2539         /* entries are added/deleted only under RTNL */
2540         rcu_read_lock();
2541         cache = ipmr_cache_find(mrt, src, grp);
2542         rcu_read_unlock();
2543         if (!cache) {
2544                 err = -ENOENT;
2545                 goto errout_free;
2546         }
2547
2548         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2549         if (!skb) {
2550                 err = -ENOBUFS;
2551                 goto errout_free;
2552         }
2553
2554         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2555                                nlh->nlmsg_seq, cache,
2556                                RTM_NEWROUTE, 0);
2557         if (err < 0)
2558                 goto errout_free;
2559
2560         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2561
2562 errout:
2563         return err;
2564
2565 errout_free:
2566         kfree_skb(skb);
2567         goto errout;
2568 }
2569
2570 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2571 {
2572         struct fib_dump_filter filter = {};
2573         int err;
2574
2575         if (cb->strict_check) {
2576                 err = ip_valid_fib_dump_req(sock_net(skb->sk), cb->nlh,
2577                                             &filter, cb);
2578                 if (err < 0)
2579                         return err;
2580         }
2581
2582         if (filter.table_id) {
2583                 struct mr_table *mrt;
2584
2585                 mrt = ipmr_get_table(sock_net(skb->sk), filter.table_id);
2586                 if (!mrt) {
2587                         if (rtnl_msg_family(cb->nlh) != RTNL_FAMILY_IPMR)
2588                                 return skb->len;
2589
2590                         NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
2591                         return -ENOENT;
2592                 }
2593                 err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
2594                                     &mfc_unres_lock, &filter);
2595                 return skb->len ? : err;
2596         }
2597
2598         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2599                                 _ipmr_fill_mroute, &mfc_unres_lock, &filter);
2600 }
2601
2602 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2603         [RTA_SRC]       = { .type = NLA_U32 },
2604         [RTA_DST]       = { .type = NLA_U32 },
2605         [RTA_IIF]       = { .type = NLA_U32 },
2606         [RTA_TABLE]     = { .type = NLA_U32 },
2607         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2608 };
2609
2610 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2611 {
2612         switch (rtm_protocol) {
2613         case RTPROT_STATIC:
2614         case RTPROT_MROUTED:
2615                 return true;
2616         }
2617         return false;
2618 }
2619
2620 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2621 {
2622         struct rtnexthop *rtnh = nla_data(nla);
2623         int remaining = nla_len(nla), vifi = 0;
2624
2625         while (rtnh_ok(rtnh, remaining)) {
2626                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2627                 if (++vifi == MAXVIFS)
2628                         break;
2629                 rtnh = rtnh_next(rtnh, &remaining);
2630         }
2631
2632         return remaining > 0 ? -EINVAL : vifi;
2633 }
2634
2635 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2636 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2637                             struct mfcctl *mfcc, int *mrtsock,
2638                             struct mr_table **mrtret,
2639                             struct netlink_ext_ack *extack)
2640 {
2641         struct net_device *dev = NULL;
2642         u32 tblid = RT_TABLE_DEFAULT;
2643         struct mr_table *mrt;
2644         struct nlattr *attr;
2645         struct rtmsg *rtm;
2646         int ret, rem;
2647
2648         ret = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
2649                                         rtm_ipmr_policy, extack);
2650         if (ret < 0)
2651                 goto out;
2652         rtm = nlmsg_data(nlh);
2653
2654         ret = -EINVAL;
2655         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2656             rtm->rtm_type != RTN_MULTICAST ||
2657             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2658             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2659                 goto out;
2660
2661         memset(mfcc, 0, sizeof(*mfcc));
2662         mfcc->mfcc_parent = -1;
2663         ret = 0;
2664         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2665                 switch (nla_type(attr)) {
2666                 case RTA_SRC:
2667                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2668                         break;
2669                 case RTA_DST:
2670                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2671                         break;
2672                 case RTA_IIF:
2673                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2674                         if (!dev) {
2675                                 ret = -ENODEV;
2676                                 goto out;
2677                         }
2678                         break;
2679                 case RTA_MULTIPATH:
2680                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2681                                 ret = -EINVAL;
2682                                 goto out;
2683                         }
2684                         break;
2685                 case RTA_PREFSRC:
2686                         ret = 1;
2687                         break;
2688                 case RTA_TABLE:
2689                         tblid = nla_get_u32(attr);
2690                         break;
2691                 }
2692         }
2693         mrt = ipmr_get_table(net, tblid);
2694         if (!mrt) {
2695                 ret = -ENOENT;
2696                 goto out;
2697         }
2698         *mrtret = mrt;
2699         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2700         if (dev)
2701                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2702
2703 out:
2704         return ret;
2705 }
2706
2707 /* takes care of both newroute and delroute */
2708 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2709                           struct netlink_ext_ack *extack)
2710 {
2711         struct net *net = sock_net(skb->sk);
2712         int ret, mrtsock, parent;
2713         struct mr_table *tbl;
2714         struct mfcctl mfcc;
2715
2716         mrtsock = 0;
2717         tbl = NULL;
2718         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2719         if (ret < 0)
2720                 return ret;
2721
2722         parent = ret ? mfcc.mfcc_parent : -1;
2723         if (nlh->nlmsg_type == RTM_NEWROUTE)
2724                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2725         else
2726                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2727 }
2728
2729 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2730 {
2731         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2732
2733         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2734             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2735             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2736                         mrt->mroute_reg_vif_num) ||
2737             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2738                        mrt->mroute_do_assert) ||
2739             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2740             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2741                        mrt->mroute_do_wrvifwhole))
2742                 return false;
2743
2744         return true;
2745 }
2746
2747 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2748 {
2749         struct nlattr *vif_nest;
2750         struct vif_device *vif;
2751
2752         /* if the VIF doesn't exist just continue */
2753         if (!VIF_EXISTS(mrt, vifid))
2754                 return true;
2755
2756         vif = &mrt->vif_table[vifid];
2757         vif_nest = nla_nest_start_noflag(skb, IPMRA_VIF);
2758         if (!vif_nest)
2759                 return false;
2760         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2761             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2762             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2763             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2764                               IPMRA_VIFA_PAD) ||
2765             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2766                               IPMRA_VIFA_PAD) ||
2767             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2768                               IPMRA_VIFA_PAD) ||
2769             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2770                               IPMRA_VIFA_PAD) ||
2771             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2772             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2773                 nla_nest_cancel(skb, vif_nest);
2774                 return false;
2775         }
2776         nla_nest_end(skb, vif_nest);
2777
2778         return true;
2779 }
2780
2781 static int ipmr_valid_dumplink(const struct nlmsghdr *nlh,
2782                                struct netlink_ext_ack *extack)
2783 {
2784         struct ifinfomsg *ifm;
2785
2786         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ifm))) {
2787                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for ipmr link dump");
2788                 return -EINVAL;
2789         }
2790
2791         if (nlmsg_attrlen(nlh, sizeof(*ifm))) {
2792                 NL_SET_ERR_MSG(extack, "Invalid data after header in ipmr link dump");
2793                 return -EINVAL;
2794         }
2795
2796         ifm = nlmsg_data(nlh);
2797         if (ifm->__ifi_pad || ifm->ifi_type || ifm->ifi_flags ||
2798             ifm->ifi_change || ifm->ifi_index) {
2799                 NL_SET_ERR_MSG(extack, "Invalid values in header for ipmr link dump request");
2800                 return -EINVAL;
2801         }
2802
2803         return 0;
2804 }
2805
2806 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2807 {
2808         struct net *net = sock_net(skb->sk);
2809         struct nlmsghdr *nlh = NULL;
2810         unsigned int t = 0, s_t;
2811         unsigned int e = 0, s_e;
2812         struct mr_table *mrt;
2813
2814         if (cb->strict_check) {
2815                 int err = ipmr_valid_dumplink(cb->nlh, cb->extack);
2816
2817                 if (err < 0)
2818                         return err;
2819         }
2820
2821         s_t = cb->args[0];
2822         s_e = cb->args[1];
2823
2824         ipmr_for_each_table(mrt, net) {
2825                 struct nlattr *vifs, *af;
2826                 struct ifinfomsg *hdr;
2827                 u32 i;
2828
2829                 if (t < s_t)
2830                         goto skip_table;
2831                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2832                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2833                                 sizeof(*hdr), NLM_F_MULTI);
2834                 if (!nlh)
2835                         break;
2836
2837                 hdr = nlmsg_data(nlh);
2838                 memset(hdr, 0, sizeof(*hdr));
2839                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2840
2841                 af = nla_nest_start_noflag(skb, IFLA_AF_SPEC);
2842                 if (!af) {
2843                         nlmsg_cancel(skb, nlh);
2844                         goto out;
2845                 }
2846
2847                 if (!ipmr_fill_table(mrt, skb)) {
2848                         nlmsg_cancel(skb, nlh);
2849                         goto out;
2850                 }
2851
2852                 vifs = nla_nest_start_noflag(skb, IPMRA_TABLE_VIFS);
2853                 if (!vifs) {
2854                         nla_nest_end(skb, af);
2855                         nlmsg_end(skb, nlh);
2856                         goto out;
2857                 }
2858                 for (i = 0; i < mrt->maxvif; i++) {
2859                         if (e < s_e)
2860                                 goto skip_entry;
2861                         if (!ipmr_fill_vif(mrt, i, skb)) {
2862                                 nla_nest_end(skb, vifs);
2863                                 nla_nest_end(skb, af);
2864                                 nlmsg_end(skb, nlh);
2865                                 goto out;
2866                         }
2867 skip_entry:
2868                         e++;
2869                 }
2870                 s_e = 0;
2871                 e = 0;
2872                 nla_nest_end(skb, vifs);
2873                 nla_nest_end(skb, af);
2874                 nlmsg_end(skb, nlh);
2875 skip_table:
2876                 t++;
2877         }
2878
2879 out:
2880         cb->args[1] = e;
2881         cb->args[0] = t;
2882
2883         return skb->len;
2884 }
2885
2886 #ifdef CONFIG_PROC_FS
2887 /* The /proc interfaces to multicast routing :
2888  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2889  */
2890
2891 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2892         __acquires(mrt_lock)
2893 {
2894         struct mr_vif_iter *iter = seq->private;
2895         struct net *net = seq_file_net(seq);
2896         struct mr_table *mrt;
2897
2898         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2899         if (!mrt)
2900                 return ERR_PTR(-ENOENT);
2901
2902         iter->mrt = mrt;
2903
2904         read_lock(&mrt_lock);
2905         return mr_vif_seq_start(seq, pos);
2906 }
2907
2908 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2909         __releases(mrt_lock)
2910 {
2911         read_unlock(&mrt_lock);
2912 }
2913
2914 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2915 {
2916         struct mr_vif_iter *iter = seq->private;
2917         struct mr_table *mrt = iter->mrt;
2918
2919         if (v == SEQ_START_TOKEN) {
2920                 seq_puts(seq,
2921                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2922         } else {
2923                 const struct vif_device *vif = v;
2924                 const char *name =  vif->dev ?
2925                                     vif->dev->name : "none";
2926
2927                 seq_printf(seq,
2928                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2929                            vif - mrt->vif_table,
2930                            name, vif->bytes_in, vif->pkt_in,
2931                            vif->bytes_out, vif->pkt_out,
2932                            vif->flags, vif->local, vif->remote);
2933         }
2934         return 0;
2935 }
2936
2937 static const struct seq_operations ipmr_vif_seq_ops = {
2938         .start = ipmr_vif_seq_start,
2939         .next  = mr_vif_seq_next,
2940         .stop  = ipmr_vif_seq_stop,
2941         .show  = ipmr_vif_seq_show,
2942 };
2943
2944 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2945 {
2946         struct net *net = seq_file_net(seq);
2947         struct mr_table *mrt;
2948
2949         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2950         if (!mrt)
2951                 return ERR_PTR(-ENOENT);
2952
2953         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2954 }
2955
2956 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2957 {
2958         int n;
2959
2960         if (v == SEQ_START_TOKEN) {
2961                 seq_puts(seq,
2962                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2963         } else {
2964                 const struct mfc_cache *mfc = v;
2965                 const struct mr_mfc_iter *it = seq->private;
2966                 const struct mr_table *mrt = it->mrt;
2967
2968                 seq_printf(seq, "%08X %08X %-3hd",
2969                            (__force u32) mfc->mfc_mcastgrp,
2970                            (__force u32) mfc->mfc_origin,
2971                            mfc->_c.mfc_parent);
2972
2973                 if (it->cache != &mrt->mfc_unres_queue) {
2974                         seq_printf(seq, " %8lu %8lu %8lu",
2975                                    mfc->_c.mfc_un.res.pkt,
2976                                    mfc->_c.mfc_un.res.bytes,
2977                                    mfc->_c.mfc_un.res.wrong_if);
2978                         for (n = mfc->_c.mfc_un.res.minvif;
2979                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2980                                 if (VIF_EXISTS(mrt, n) &&
2981                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2982                                         seq_printf(seq,
2983                                            " %2d:%-3d",
2984                                            n, mfc->_c.mfc_un.res.ttls[n]);
2985                         }
2986                 } else {
2987                         /* unresolved mfc_caches don't contain
2988                          * pkt, bytes and wrong_if values
2989                          */
2990                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2991                 }
2992                 seq_putc(seq, '\n');
2993         }
2994         return 0;
2995 }
2996
2997 static const struct seq_operations ipmr_mfc_seq_ops = {
2998         .start = ipmr_mfc_seq_start,
2999         .next  = mr_mfc_seq_next,
3000         .stop  = mr_mfc_seq_stop,
3001         .show  = ipmr_mfc_seq_show,
3002 };
3003 #endif
3004
3005 #ifdef CONFIG_IP_PIMSM_V2
3006 static const struct net_protocol pim_protocol = {
3007         .handler        =       pim_rcv,
3008 };
3009 #endif
3010
3011 static unsigned int ipmr_seq_read(struct net *net)
3012 {
3013         ASSERT_RTNL();
3014
3015         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3016 }
3017
3018 static int ipmr_dump(struct net *net, struct notifier_block *nb,
3019                      struct netlink_ext_ack *extack)
3020 {
3021         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
3022                        ipmr_mr_table_iter, &mrt_lock, extack);
3023 }
3024
3025 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3026         .family         = RTNL_FAMILY_IPMR,
3027         .fib_seq_read   = ipmr_seq_read,
3028         .fib_dump       = ipmr_dump,
3029         .owner          = THIS_MODULE,
3030 };
3031
3032 static int __net_init ipmr_notifier_init(struct net *net)
3033 {
3034         struct fib_notifier_ops *ops;
3035
3036         net->ipv4.ipmr_seq = 0;
3037
3038         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3039         if (IS_ERR(ops))
3040                 return PTR_ERR(ops);
3041         net->ipv4.ipmr_notifier_ops = ops;
3042
3043         return 0;
3044 }
3045
3046 static void __net_exit ipmr_notifier_exit(struct net *net)
3047 {
3048         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3049         net->ipv4.ipmr_notifier_ops = NULL;
3050 }
3051
3052 /* Setup for IP multicast routing */
3053 static int __net_init ipmr_net_init(struct net *net)
3054 {
3055         int err;
3056
3057         err = ipmr_notifier_init(net);
3058         if (err)
3059                 goto ipmr_notifier_fail;
3060
3061         err = ipmr_rules_init(net);
3062         if (err < 0)
3063                 goto ipmr_rules_fail;
3064
3065 #ifdef CONFIG_PROC_FS
3066         err = -ENOMEM;
3067         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
3068                         sizeof(struct mr_vif_iter)))
3069                 goto proc_vif_fail;
3070         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
3071                         sizeof(struct mr_mfc_iter)))
3072                 goto proc_cache_fail;
3073 #endif
3074         return 0;
3075
3076 #ifdef CONFIG_PROC_FS
3077 proc_cache_fail:
3078         remove_proc_entry("ip_mr_vif", net->proc_net);
3079 proc_vif_fail:
3080         ipmr_rules_exit(net);
3081 #endif
3082 ipmr_rules_fail:
3083         ipmr_notifier_exit(net);
3084 ipmr_notifier_fail:
3085         return err;
3086 }
3087
3088 static void __net_exit ipmr_net_exit(struct net *net)
3089 {
3090 #ifdef CONFIG_PROC_FS
3091         remove_proc_entry("ip_mr_cache", net->proc_net);
3092         remove_proc_entry("ip_mr_vif", net->proc_net);
3093 #endif
3094         ipmr_notifier_exit(net);
3095         ipmr_rules_exit(net);
3096 }
3097
3098 static struct pernet_operations ipmr_net_ops = {
3099         .init = ipmr_net_init,
3100         .exit = ipmr_net_exit,
3101 };
3102
3103 int __init ip_mr_init(void)
3104 {
3105         int err;
3106
3107         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3108                                        sizeof(struct mfc_cache),
3109                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3110                                        NULL);
3111
3112         err = register_pernet_subsys(&ipmr_net_ops);
3113         if (err)
3114                 goto reg_pernet_fail;
3115
3116         err = register_netdevice_notifier(&ip_mr_notifier);
3117         if (err)
3118                 goto reg_notif_fail;
3119 #ifdef CONFIG_IP_PIMSM_V2
3120         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3121                 pr_err("%s: can't add PIM protocol\n", __func__);
3122                 err = -EAGAIN;
3123                 goto add_proto_fail;
3124         }
3125 #endif
3126         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3127                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3128         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3129                       ipmr_rtm_route, NULL, 0);
3130         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3131                       ipmr_rtm_route, NULL, 0);
3132
3133         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3134                       NULL, ipmr_rtm_dumplink, 0);
3135         return 0;
3136
3137 #ifdef CONFIG_IP_PIMSM_V2
3138 add_proto_fail:
3139         unregister_netdevice_notifier(&ip_mr_notifier);
3140 #endif
3141 reg_notif_fail:
3142         unregister_pernet_subsys(&ipmr_net_ops);
3143 reg_pernet_fail:
3144         kmem_cache_destroy(mrt_cachep);
3145         return err;
3146 }