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         ipmr_free_table(mrt);
 260 err1:
 261         fib_rules_unregister(ops);
 262         return err;
 263 }
 264
 265 static void __net_exit ipmr_rules_exit(struct net *net)
 266 {
 267         struct mr_table *mrt, *next;
 268
 269         ASSERT_RTNL();
 270         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 271                 list_del(&mrt->list);
 272                 ipmr_free_table(mrt);
 273         }
 274         fib_rules_unregister(net->ipv4.mr_rules_ops);
 275 }
 276
 277 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 278                            struct netlink_ext_ack *extack)
 279 {
 280         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR, extack);
 281 }
 282
 283 static unsigned int ipmr_rules_seq_read(struct net *net)
 284 {
 285         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 286 }
 287
 288 bool ipmr_rule_default(const struct fib_rule *rule)
 289 {
 290         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 291 }
 292 EXPORT_SYMBOL(ipmr_rule_default);
 293 #else
 294 #define ipmr_for_each_table(mrt, net) \
 295         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 296
 297 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 298                                            struct mr_table *mrt)
 299 {
 300         if (!mrt)
 301                 return net->ipv4.mrt;
 302         return NULL;
 303 }
 304
 305 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 306 {
 307         return net->ipv4.mrt;
 308 }
 309
 310 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 311                            struct mr_table **mrt)
 312 {
 313         *mrt = net->ipv4.mrt;
 314         return 0;
 315 }
 316
 317 static int __net_init ipmr_rules_init(struct net *net)
 318 {
 319         struct mr_table *mrt;
 320
 321         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 322         if (IS_ERR(mrt))
 323                 return PTR_ERR(mrt);
 324         net->ipv4.mrt = mrt;
 325         return 0;
 326 }
 327
 328 static void __net_exit ipmr_rules_exit(struct net *net)
 329 {
 330         ASSERT_RTNL();
 331         ipmr_free_table(net->ipv4.mrt);
 332         net->ipv4.mrt = NULL;
 333 }
 334
 335 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 336                            struct netlink_ext_ack *extack)
 337 {
 338         return 0;
 339 }
 340
 341 static unsigned int ipmr_rules_seq_read(struct net *net)
 342 {
 343         return 0;
 344 }
 345
 346 bool ipmr_rule_default(const struct fib_rule *rule)
 347 {
 348         return true;
 349 }
 350 EXPORT_SYMBOL(ipmr_rule_default);
 351 #endif
 352
 353 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 354                                 const void *ptr)
 355 {
 356         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 357         struct mfc_cache *c = (struct mfc_cache *)ptr;
 358
 359         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 360                cmparg->mfc_origin != c->mfc_origin;
 361 }
 362
 363 static const struct rhashtable_params ipmr_rht_params = {
 364         .head_offset = offsetof(struct mr_mfc, mnode),
 365         .key_offset = offsetof(struct mfc_cache, cmparg),
 366         .key_len = sizeof(struct mfc_cache_cmp_arg),
 367         .nelem_hint = 3,
 368         .obj_cmpfn = ipmr_hash_cmp,
 369         .automatic_shrinking = true,
 370 };
 371
 372 static void ipmr_new_table_set(struct mr_table *mrt,
 373                                struct net *net)
 374 {
 375 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 376         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 377 #endif
 378 }
 379
 380 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 381         .mfc_mcastgrp = htonl(INADDR_ANY),
 382         .mfc_origin = htonl(INADDR_ANY),
 383 };
 384
 385 static struct mr_table_ops ipmr_mr_table_ops = {
 386         .rht_params = &ipmr_rht_params,
 387         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 388 };
 389
 390 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 391 {
 392         struct mr_table *mrt;
 393
 394         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 395         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 396                 return ERR_PTR(-EINVAL);
 397
 398         mrt = ipmr_get_table(net, id);
 399         if (mrt)
 400                 return mrt;
 401
 402         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 403                               ipmr_expire_process, ipmr_new_table_set);
 404 }
 405
 406 static void ipmr_free_table(struct mr_table *mrt)
 407 {
 408         del_timer_sync(&mrt->ipmr_expire_timer);
 409         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC |
 410                                  MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC);
 411         rhltable_destroy(&mrt->mfc_hash);
 412         kfree(mrt);
 413 }
 414
 415 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 416
 417 /* Initialize ipmr pimreg/tunnel in_device */
 418 static bool ipmr_init_vif_indev(const struct net_device *dev)
 419 {
 420         struct in_device *in_dev;
 421
 422         ASSERT_RTNL();
 423
 424         in_dev = __in_dev_get_rtnl(dev);
 425         if (!in_dev)
 426                 return false;
 427         ipv4_devconf_setall(in_dev);
 428         neigh_parms_data_state_setall(in_dev->arp_parms);
 429         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 430
 431         return true;
 432 }
 433
 434 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 435 {
 436         struct net_device *tunnel_dev, *new_dev;
 437         struct ip_tunnel_parm p = { };
 438         int err;
 439
 440         tunnel_dev = __dev_get_by_name(net, "tunl0");
 441         if (!tunnel_dev)
 442                 goto out;
 443
 444         p.iph.daddr = v->vifc_rmt_addr.s_addr;
 445         p.iph.saddr = v->vifc_lcl_addr.s_addr;
 446         p.iph.version = 4;
 447         p.iph.ihl = 5;
 448         p.iph.protocol = IPPROTO_IPIP;
 449         sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 450
 451         if (!tunnel_dev->netdev_ops->ndo_tunnel_ctl)
 452                 goto out;
 453         err = tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 454                         SIOCADDTUNNEL);
 455         if (err)
 456                 goto out;
 457
 458         new_dev = __dev_get_by_name(net, p.name);
 459         if (!new_dev)
 460                 goto out;
 461
 462         new_dev->flags |= IFF_MULTICAST;
 463         if (!ipmr_init_vif_indev(new_dev))
 464                 goto out_unregister;
 465         if (dev_open(new_dev, NULL))
 466                 goto out_unregister;
 467         dev_hold(new_dev);
 468         err = dev_set_allmulti(new_dev, 1);
 469         if (err) {
 470                 dev_close(new_dev);
 471                 tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 472                                 SIOCDELTUNNEL);
 473                 dev_put(new_dev);
 474                 new_dev = ERR_PTR(err);
 475         }
 476         return new_dev;
 477
 478 out_unregister:
 479         unregister_netdevice(new_dev);
 480 out:
 481         return ERR_PTR(-ENOBUFS);
 482 }
 483
 484 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 485 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 486 {
 487         struct net *net = dev_net(dev);
 488         struct mr_table *mrt;
 489         struct flowi4 fl4 = {
 490                 .flowi4_oif     = dev->ifindex,
 491                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 492                 .flowi4_mark    = skb->mark,
 493         };
 494         int err;
 495
 496         err = ipmr_fib_lookup(net, &fl4, &mrt);
 497         if (err < 0) {
 498                 kfree_skb(skb);
 499                 return err;
 500         }
 501
 502         read_lock(&mrt_lock);
 503         dev->stats.tx_bytes += skb->len;
 504         dev->stats.tx_packets++;
 505         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 506         read_unlock(&mrt_lock);
 507         kfree_skb(skb);
 508         return NETDEV_TX_OK;
 509 }
 510
 511 static int reg_vif_get_iflink(const struct net_device *dev)
 512 {
 513         return 0;
 514 }
 515
 516 static const struct net_device_ops reg_vif_netdev_ops = {
 517         .ndo_start_xmit = reg_vif_xmit,
 518         .ndo_get_iflink = reg_vif_get_iflink,
 519 };
 520
 521 static void reg_vif_setup(struct net_device *dev)
 522 {
 523         dev->type               = ARPHRD_PIMREG;
 524         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 525         dev->flags              = IFF_NOARP;
 526         dev->netdev_ops         = &reg_vif_netdev_ops;
 527         dev->needs_free_netdev  = true;
 528         dev->features           |= NETIF_F_NETNS_LOCAL;
 529 }
 530
 531 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 532 {
 533         struct net_device *dev;
 534         char name[IFNAMSIZ];
 535
 536         if (mrt->id == RT_TABLE_DEFAULT)
 537                 sprintf(name, "pimreg");
 538         else
 539                 sprintf(name, "pimreg%u", mrt->id);
 540
 541         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 542
 543         if (!dev)
 544                 return NULL;
 545
 546         dev_net_set(dev, net);
 547
 548         if (register_netdevice(dev)) {
 549                 free_netdev(dev);
 550                 return NULL;
 551         }
 552
 553         if (!ipmr_init_vif_indev(dev))
 554                 goto failure;
 555         if (dev_open(dev, NULL))
 556                 goto failure;
 557
 558         dev_hold(dev);
 559
 560         return dev;
 561
 562 failure:
 563         unregister_netdevice(dev);
 564         return NULL;
 565 }
 566
 567 /* called with rcu_read_lock() */
 568 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 569                      unsigned int pimlen)
 570 {
 571         struct net_device *reg_dev = NULL;
 572         struct iphdr *encap;
 573
 574         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 575         /* Check that:
 576          * a. packet is really sent to a multicast group
 577          * b. packet is not a NULL-REGISTER
 578          * c. packet is not truncated
 579          */
 580         if (!ipv4_is_multicast(encap->daddr) ||
 581             encap->tot_len == 0 ||
 582             ntohs(encap->tot_len) + pimlen > skb->len)
 583                 return 1;
 584
 585         read_lock(&mrt_lock);
 586         if (mrt->mroute_reg_vif_num >= 0)
 587                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 588         read_unlock(&mrt_lock);
 589
 590         if (!reg_dev)
 591                 return 1;
 592
 593         skb->mac_header = skb->network_header;
 594         skb_pull(skb, (u8 *)encap - skb->data);
 595         skb_reset_network_header(skb);
 596         skb->protocol = htons(ETH_P_IP);
 597         skb->ip_summed = CHECKSUM_NONE;
 598
 599         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 600
 601         netif_rx(skb);
 602
 603         return NET_RX_SUCCESS;
 604 }
 605 #else
 606 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 607 {
 608         return NULL;
 609 }
 610 #endif
 611
 612 static int call_ipmr_vif_entry_notifiers(struct net *net,
 613                                          enum fib_event_type event_type,
 614                                          struct vif_device *vif,
 615                                          vifi_t vif_index, u32 tb_id)
 616 {
 617         return mr_call_vif_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 618                                      vif, vif_index, tb_id,
 619                                      &net->ipv4.ipmr_seq);
 620 }
 621
 622 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 623                                          enum fib_event_type event_type,
 624                                          struct mfc_cache *mfc, u32 tb_id)
 625 {
 626         return mr_call_mfc_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 627                                      &mfc->_c, tb_id, &net->ipv4.ipmr_seq);
 628 }
 629
 630 /**
 631  *      vif_delete - Delete a VIF entry
 632  *      @mrt: Table to delete from
 633  *      @vifi: VIF identifier to delete
 634  *      @notify: Set to 1, if the caller is a notifier_call
 635  *      @head: if unregistering the VIF, place it on this queue
 636  */
 637 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 638                       struct list_head *head)
 639 {
 640         struct net *net = read_pnet(&mrt->net);
 641         struct vif_device *v;
 642         struct net_device *dev;
 643         struct in_device *in_dev;
 644
 645         if (vifi < 0 || vifi >= mrt->maxvif)
 646                 return -EADDRNOTAVAIL;
 647
 648         v = &mrt->vif_table[vifi];
 649
 650         if (VIF_EXISTS(mrt, vifi))
 651                 call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, vifi,
 652                                               mrt->id);
 653
 654         write_lock_bh(&mrt_lock);
 655         dev = v->dev;
 656         v->dev = NULL;
 657
 658         if (!dev) {
 659                 write_unlock_bh(&mrt_lock);
 660                 return -EADDRNOTAVAIL;
 661         }
 662
 663         if (vifi == mrt->mroute_reg_vif_num)
 664                 mrt->mroute_reg_vif_num = -1;
 665
 666         if (vifi + 1 == mrt->maxvif) {
 667                 int tmp;
 668
 669                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 670                         if (VIF_EXISTS(mrt, tmp))
 671                                 break;
 672                 }
 673                 mrt->maxvif = tmp+1;
 674         }
 675
 676         write_unlock_bh(&mrt_lock);
 677
 678         dev_set_allmulti(dev, -1);
 679
 680         in_dev = __in_dev_get_rtnl(dev);
 681         if (in_dev) {
 682                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 683                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 684                                             NETCONFA_MC_FORWARDING,
 685                                             dev->ifindex, &in_dev->cnf);
 686                 ip_rt_multicast_event(in_dev);
 687         }
 688
 689         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 690                 unregister_netdevice_queue(dev, head);
 691
 692         dev_put_track(dev, &v->dev_tracker);
 693         return 0;
 694 }
 695
 696 static void ipmr_cache_free_rcu(struct rcu_head *head)
 697 {
 698         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 699
 700         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 701 }
 702
 703 static void ipmr_cache_free(struct mfc_cache *c)
 704 {
 705         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 706 }
 707
 708 /* Destroy an unresolved cache entry, killing queued skbs
 709  * and reporting error to netlink readers.
 710  */
 711 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 712 {
 713         struct net *net = read_pnet(&mrt->net);
 714         struct sk_buff *skb;
 715         struct nlmsgerr *e;
 716
 717         atomic_dec(&mrt->cache_resolve_queue_len);
 718
 719         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 720                 if (ip_hdr(skb)->version == 0) {
 721                         struct nlmsghdr *nlh = skb_pull(skb,
 722                                                         sizeof(struct iphdr));
 723                         nlh->nlmsg_type = NLMSG_ERROR;
 724                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 725                         skb_trim(skb, nlh->nlmsg_len);
 726                         e = nlmsg_data(nlh);
 727                         e->error = -ETIMEDOUT;
 728                         memset(&e->msg, 0, sizeof(e->msg));
 729
 730                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 731                 } else {
 732                         kfree_skb(skb);
 733                 }
 734         }
 735
 736         ipmr_cache_free(c);
 737 }
 738
 739 /* Timer process for the unresolved queue. */
 740 static void ipmr_expire_process(struct timer_list *t)
 741 {
 742         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 743         struct mr_mfc *c, *next;
 744         unsigned long expires;
 745         unsigned long now;
 746
 747         if (!spin_trylock(&mfc_unres_lock)) {
 748                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 749                 return;
 750         }
 751
 752         if (list_empty(&mrt->mfc_unres_queue))
 753                 goto out;
 754
 755         now = jiffies;
 756         expires = 10*HZ;
 757
 758         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 759                 if (time_after(c->mfc_un.unres.expires, now)) {
 760                         unsigned long interval = c->mfc_un.unres.expires - now;
 761                         if (interval < expires)
 762                                 expires = interval;
 763                         continue;
 764                 }
 765
 766                 list_del(&c->list);
 767                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 768                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 769         }
 770
 771         if (!list_empty(&mrt->mfc_unres_queue))
 772                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 773
 774 out:
 775         spin_unlock(&mfc_unres_lock);
 776 }
 777
 778 /* Fill oifs list. It is called under write locked mrt_lock. */
 779 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 780                                    unsigned char *ttls)
 781 {
 782         int vifi;
 783
 784         cache->mfc_un.res.minvif = MAXVIFS;
 785         cache->mfc_un.res.maxvif = 0;
 786         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 787
 788         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 789                 if (VIF_EXISTS(mrt, vifi) &&
 790                     ttls[vifi] && ttls[vifi] < 255) {
 791                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 792                         if (cache->mfc_un.res.minvif > vifi)
 793                                 cache->mfc_un.res.minvif = vifi;
 794                         if (cache->mfc_un.res.maxvif <= vifi)
 795                                 cache->mfc_un.res.maxvif = vifi + 1;
 796                 }
 797         }
 798         cache->mfc_un.res.lastuse = jiffies;
 799 }
 800
 801 static int vif_add(struct net *net, struct mr_table *mrt,
 802                    struct vifctl *vifc, int mrtsock)
 803 {
 804         struct netdev_phys_item_id ppid = { };
 805         int vifi = vifc->vifc_vifi;
 806         struct vif_device *v = &mrt->vif_table[vifi];
 807         struct net_device *dev;
 808         struct in_device *in_dev;
 809         int err;
 810
 811         /* Is vif busy ? */
 812         if (VIF_EXISTS(mrt, vifi))
 813                 return -EADDRINUSE;
 814
 815         switch (vifc->vifc_flags) {
 816         case VIFF_REGISTER:
 817                 if (!ipmr_pimsm_enabled())
 818                         return -EINVAL;
 819                 /* Special Purpose VIF in PIM
 820                  * All the packets will be sent to the daemon
 821                  */
 822                 if (mrt->mroute_reg_vif_num >= 0)
 823                         return -EADDRINUSE;
 824                 dev = ipmr_reg_vif(net, mrt);
 825                 if (!dev)
 826                         return -ENOBUFS;
 827                 err = dev_set_allmulti(dev, 1);
 828                 if (err) {
 829                         unregister_netdevice(dev);
 830                         dev_put(dev);
 831                         return err;
 832                 }
 833                 break;
 834         case VIFF_TUNNEL:
 835                 dev = ipmr_new_tunnel(net, vifc);
 836                 if (IS_ERR(dev))
 837                         return PTR_ERR(dev);
 838                 break;
 839         case VIFF_USE_IFINDEX:
 840         case 0:
 841                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 842                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 843                         if (dev && !__in_dev_get_rtnl(dev)) {
 844                                 dev_put(dev);
 845                                 return -EADDRNOTAVAIL;
 846                         }
 847                 } else {
 848                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 849                 }
 850                 if (!dev)
 851                         return -EADDRNOTAVAIL;
 852                 err = dev_set_allmulti(dev, 1);
 853                 if (err) {
 854                         dev_put(dev);
 855                         return err;
 856                 }
 857                 break;
 858         default:
 859                 return -EINVAL;
 860         }
 861
 862         in_dev = __in_dev_get_rtnl(dev);
 863         if (!in_dev) {
 864                 dev_put(dev);
 865                 return -EADDRNOTAVAIL;
 866         }
 867         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 868         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 869                                     dev->ifindex, &in_dev->cnf);
 870         ip_rt_multicast_event(in_dev);
 871
 872         /* Fill in the VIF structures */
 873         vif_device_init(v, dev, vifc->vifc_rate_limit,
 874                         vifc->vifc_threshold,
 875                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 876                         (VIFF_TUNNEL | VIFF_REGISTER));
 877
 878         err = dev_get_port_parent_id(dev, &ppid, true);
 879         if (err == 0) {
 880                 memcpy(v->dev_parent_id.id, ppid.id, ppid.id_len);
 881                 v->dev_parent_id.id_len = ppid.id_len;
 882         } else {
 883                 v->dev_parent_id.id_len = 0;
 884         }
 885
 886         v->local = vifc->vifc_lcl_addr.s_addr;
 887         v->remote = vifc->vifc_rmt_addr.s_addr;
 888
 889         /* And finish update writing critical data */
 890         write_lock_bh(&mrt_lock);
 891         v->dev = dev;
 892         netdev_tracker_alloc(dev, &v->dev_tracker, GFP_ATOMIC);
 893         if (v->flags & VIFF_REGISTER)
 894                 mrt->mroute_reg_vif_num = vifi;
 895         if (vifi+1 > mrt->maxvif)
 896                 mrt->maxvif = vifi+1;
 897         write_unlock_bh(&mrt_lock);
 898         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, vifi, mrt->id);
 899         return 0;
 900 }
 901
 902 /* called with rcu_read_lock() */
 903 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 904                                          __be32 origin,
 905                                          __be32 mcastgrp)
 906 {
 907         struct mfc_cache_cmp_arg arg = {
 908                         .mfc_mcastgrp = mcastgrp,
 909                         .mfc_origin = origin
 910         };
 911
 912         return mr_mfc_find(mrt, &arg);
 913 }
 914
 915 /* Look for a (*,G) entry */
 916 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 917                                              __be32 mcastgrp, int vifi)
 918 {
 919         struct mfc_cache_cmp_arg arg = {
 920                         .mfc_mcastgrp = mcastgrp,
 921                         .mfc_origin = htonl(INADDR_ANY)
 922         };
 923
 924         if (mcastgrp == htonl(INADDR_ANY))
 925                 return mr_mfc_find_any_parent(mrt, vifi);
 926         return mr_mfc_find_any(mrt, vifi, &arg);
 927 }
 928
 929 /* Look for a (S,G,iif) entry if parent != -1 */
 930 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 931                                                 __be32 origin, __be32 mcastgrp,
 932                                                 int parent)
 933 {
 934         struct mfc_cache_cmp_arg arg = {
 935                         .mfc_mcastgrp = mcastgrp,
 936                         .mfc_origin = origin,
 937         };
 938
 939         return mr_mfc_find_parent(mrt, &arg, parent);
 940 }
 941
 942 /* Allocate a multicast cache entry */
 943 static struct mfc_cache *ipmr_cache_alloc(void)
 944 {
 945         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 946
 947         if (c) {
 948                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 949                 c->_c.mfc_un.res.minvif = MAXVIFS;
 950                 c->_c.free = ipmr_cache_free_rcu;
 951                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 952         }
 953         return c;
 954 }
 955
 956 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 957 {
 958         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 959
 960         if (c) {
 961                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
 962                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
 963         }
 964         return c;
 965 }
 966
 967 /* A cache entry has gone into a resolved state from queued */
 968 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 969                                struct mfc_cache *uc, struct mfc_cache *c)
 970 {
 971         struct sk_buff *skb;
 972         struct nlmsgerr *e;
 973
 974         /* Play the pending entries through our router */
 975         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
 976                 if (ip_hdr(skb)->version == 0) {
 977                         struct nlmsghdr *nlh = skb_pull(skb,
 978                                                         sizeof(struct iphdr));
 979
 980                         if (mr_fill_mroute(mrt, skb, &c->_c,
 981                                            nlmsg_data(nlh)) > 0) {
 982                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 983                                                  (u8 *)nlh;
 984                         } else {
 985                                 nlh->nlmsg_type = NLMSG_ERROR;
 986                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 987                                 skb_trim(skb, nlh->nlmsg_len);
 988                                 e = nlmsg_data(nlh);
 989                                 e->error = -EMSGSIZE;
 990                                 memset(&e->msg, 0, sizeof(e->msg));
 991                         }
 992
 993                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 994                 } else {
 995                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
 996                 }
 997         }
 998 }
 999
1000 /* Bounce a cache query up to mrouted and netlink.
1001  *
1002  * Called under mrt_lock.
1003  */
1004 static int ipmr_cache_report(struct mr_table *mrt,
1005                              struct sk_buff *pkt, vifi_t vifi, int assert)
1006 {
1007         const int ihl = ip_hdrlen(pkt);
1008         struct sock *mroute_sk;
1009         struct igmphdr *igmp;
1010         struct igmpmsg *msg;
1011         struct sk_buff *skb;
1012         int ret;
1013
1014         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1015                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1016         else
1017                 skb = alloc_skb(128, GFP_ATOMIC);
1018
1019         if (!skb)
1020                 return -ENOBUFS;
1021
1022         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1023                 /* Ugly, but we have no choice with this interface.
1024                  * Duplicate old header, fix ihl, length etc.
1025                  * And all this only to mangle msg->im_msgtype and
1026                  * to set msg->im_mbz to "mbz" :-)
1027                  */
1028                 skb_push(skb, sizeof(struct iphdr));
1029                 skb_reset_network_header(skb);
1030                 skb_reset_transport_header(skb);
1031                 msg = (struct igmpmsg *)skb_network_header(skb);
1032                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1033                 msg->im_msgtype = assert;
1034                 msg->im_mbz = 0;
1035                 if (assert == IGMPMSG_WRVIFWHOLE) {
1036                         msg->im_vif = vifi;
1037                         msg->im_vif_hi = vifi >> 8;
1038                 } else {
1039                         msg->im_vif = mrt->mroute_reg_vif_num;
1040                         msg->im_vif_hi = mrt->mroute_reg_vif_num >> 8;
1041                 }
1042                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1043                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1044                                              sizeof(struct iphdr));
1045         } else {
1046                 /* Copy the IP header */
1047                 skb_set_network_header(skb, skb->len);
1048                 skb_put(skb, ihl);
1049                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1050                 /* Flag to the kernel this is a route add */
1051                 ip_hdr(skb)->protocol = 0;
1052                 msg = (struct igmpmsg *)skb_network_header(skb);
1053                 msg->im_vif = vifi;
1054                 msg->im_vif_hi = vifi >> 8;
1055                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1056                 /* Add our header */
1057                 igmp = skb_put(skb, sizeof(struct igmphdr));
1058                 igmp->type = assert;
1059                 msg->im_msgtype = assert;
1060                 igmp->code = 0;
1061                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1062                 skb->transport_header = skb->network_header;
1063         }
1064
1065         rcu_read_lock();
1066         mroute_sk = rcu_dereference(mrt->mroute_sk);
1067         if (!mroute_sk) {
1068                 rcu_read_unlock();
1069                 kfree_skb(skb);
1070                 return -EINVAL;
1071         }
1072
1073         igmpmsg_netlink_event(mrt, skb);
1074
1075         /* Deliver to mrouted */
1076         ret = sock_queue_rcv_skb(mroute_sk, skb);
1077         rcu_read_unlock();
1078         if (ret < 0) {
1079                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1080                 kfree_skb(skb);
1081         }
1082
1083         return ret;
1084 }
1085
1086 /* Queue a packet for resolution. It gets locked cache entry! */
1087 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1088                                  struct sk_buff *skb, struct net_device *dev)
1089 {
1090         const struct iphdr *iph = ip_hdr(skb);
1091         struct mfc_cache *c;
1092         bool found = false;
1093         int err;
1094
1095         spin_lock_bh(&mfc_unres_lock);
1096         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1097                 if (c->mfc_mcastgrp == iph->daddr &&
1098                     c->mfc_origin == iph->saddr) {
1099                         found = true;
1100                         break;
1101                 }
1102         }
1103
1104         if (!found) {
1105                 /* Create a new entry if allowable */
1106                 c = ipmr_cache_alloc_unres();
1107                 if (!c) {
1108                         spin_unlock_bh(&mfc_unres_lock);
1109
1110                         kfree_skb(skb);
1111                         return -ENOBUFS;
1112                 }
1113
1114                 /* Fill in the new cache entry */
1115                 c->_c.mfc_parent = -1;
1116                 c->mfc_origin   = iph->saddr;
1117                 c->mfc_mcastgrp = iph->daddr;
1118
1119                 /* Reflect first query at mrouted. */
1120                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1121
1122                 if (err < 0) {
1123                         /* If the report failed throw the cache entry
1124                            out - Brad Parker
1125                          */
1126                         spin_unlock_bh(&mfc_unres_lock);
1127
1128                         ipmr_cache_free(c);
1129                         kfree_skb(skb);
1130                         return err;
1131                 }
1132
1133                 atomic_inc(&mrt->cache_resolve_queue_len);
1134                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1135                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1136
1137                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1138                         mod_timer(&mrt->ipmr_expire_timer,
1139                                   c->_c.mfc_un.unres.expires);
1140         }
1141
1142         /* See if we can append the packet */
1143         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1144                 kfree_skb(skb);
1145                 err = -ENOBUFS;
1146         } else {
1147                 if (dev) {
1148                         skb->dev = dev;
1149                         skb->skb_iif = dev->ifindex;
1150                 }
1151                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1152                 err = 0;
1153         }
1154
1155         spin_unlock_bh(&mfc_unres_lock);
1156         return err;
1157 }
1158
1159 /* MFC cache manipulation by user space mroute daemon */
1160
1161 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1162 {
1163         struct net *net = read_pnet(&mrt->net);
1164         struct mfc_cache *c;
1165
1166         /* The entries are added/deleted only under RTNL */
1167         rcu_read_lock();
1168         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1169                                    mfc->mfcc_mcastgrp.s_addr, parent);
1170         rcu_read_unlock();
1171         if (!c)
1172                 return -ENOENT;
1173         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1174         list_del_rcu(&c->_c.list);
1175         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1176         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1177         mr_cache_put(&c->_c);
1178
1179         return 0;
1180 }
1181
1182 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1183                         struct mfcctl *mfc, int mrtsock, int parent)
1184 {
1185         struct mfc_cache *uc, *c;
1186         struct mr_mfc *_uc;
1187         bool found;
1188         int ret;
1189
1190         if (mfc->mfcc_parent >= MAXVIFS)
1191                 return -ENFILE;
1192
1193         /* The entries are added/deleted only under RTNL */
1194         rcu_read_lock();
1195         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1196                                    mfc->mfcc_mcastgrp.s_addr, parent);
1197         rcu_read_unlock();
1198         if (c) {
1199                 write_lock_bh(&mrt_lock);
1200                 c->_c.mfc_parent = mfc->mfcc_parent;
1201                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1202                 if (!mrtsock)
1203                         c->_c.mfc_flags |= MFC_STATIC;
1204                 write_unlock_bh(&mrt_lock);
1205                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1206                                               mrt->id);
1207                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1208                 return 0;
1209         }
1210
1211         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1212             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1213                 return -EINVAL;
1214
1215         c = ipmr_cache_alloc();
1216         if (!c)
1217                 return -ENOMEM;
1218
1219         c->mfc_origin = mfc->mfcc_origin.s_addr;
1220         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1221         c->_c.mfc_parent = mfc->mfcc_parent;
1222         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1223         if (!mrtsock)
1224                 c->_c.mfc_flags |= MFC_STATIC;
1225
1226         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1227                                   ipmr_rht_params);
1228         if (ret) {
1229                 pr_err("ipmr: rhtable insert error %d\n", ret);
1230                 ipmr_cache_free(c);
1231                 return ret;
1232         }
1233         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1234         /* Check to see if we resolved a queued list. If so we
1235          * need to send on the frames and tidy up.
1236          */
1237         found = false;
1238         spin_lock_bh(&mfc_unres_lock);
1239         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1240                 uc = (struct mfc_cache *)_uc;
1241                 if (uc->mfc_origin == c->mfc_origin &&
1242                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1243                         list_del(&_uc->list);
1244                         atomic_dec(&mrt->cache_resolve_queue_len);
1245                         found = true;
1246                         break;
1247                 }
1248         }
1249         if (list_empty(&mrt->mfc_unres_queue))
1250                 del_timer(&mrt->ipmr_expire_timer);
1251         spin_unlock_bh(&mfc_unres_lock);
1252
1253         if (found) {
1254                 ipmr_cache_resolve(net, mrt, uc, c);
1255                 ipmr_cache_free(uc);
1256         }
1257         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1258         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1259         return 0;
1260 }
1261
1262 /* Close the multicast socket, and clear the vif tables etc */
1263 static void mroute_clean_tables(struct mr_table *mrt, int flags)
1264 {
1265         struct net *net = read_pnet(&mrt->net);
1266         struct mr_mfc *c, *tmp;
1267         struct mfc_cache *cache;
1268         LIST_HEAD(list);
1269         int i;
1270
1271         /* Shut down all active vif entries */
1272         if (flags & (MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC)) {
1273                 for (i = 0; i < mrt->maxvif; i++) {
1274                         if (((mrt->vif_table[i].flags & VIFF_STATIC) &&
1275                              !(flags & MRT_FLUSH_VIFS_STATIC)) ||
1276                             (!(mrt->vif_table[i].flags & VIFF_STATIC) && !(flags & MRT_FLUSH_VIFS)))
1277                                 continue;
1278                         vif_delete(mrt, i, 0, &list);
1279                 }
1280                 unregister_netdevice_many(&list);
1281         }
1282
1283         /* Wipe the cache */
1284         if (flags & (MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC)) {
1285                 list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1286                         if (((c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC_STATIC)) ||
1287                             (!(c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC)))
1288                                 continue;
1289                         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1290                         list_del_rcu(&c->list);
1291                         cache = (struct mfc_cache *)c;
1292                         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1293                                                       mrt->id);
1294                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1295                         mr_cache_put(c);
1296                 }
1297         }
1298
1299         if (flags & MRT_FLUSH_MFC) {
1300                 if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1301                         spin_lock_bh(&mfc_unres_lock);
1302                         list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1303                                 list_del(&c->list);
1304                                 cache = (struct mfc_cache *)c;
1305                                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1306                                 ipmr_destroy_unres(mrt, cache);
1307                         }
1308                         spin_unlock_bh(&mfc_unres_lock);
1309                 }
1310         }
1311 }
1312
1313 /* called from ip_ra_control(), before an RCU grace period,
1314  * we don't need to call synchronize_rcu() here
1315  */
1316 static void mrtsock_destruct(struct sock *sk)
1317 {
1318         struct net *net = sock_net(sk);
1319         struct mr_table *mrt;
1320
1321         rtnl_lock();
1322         ipmr_for_each_table(mrt, net) {
1323                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1324                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1325                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1326                                                     NETCONFA_MC_FORWARDING,
1327                                                     NETCONFA_IFINDEX_ALL,
1328                                                     net->ipv4.devconf_all);
1329                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1330                         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_MFC);
1331                 }
1332         }
1333         rtnl_unlock();
1334 }
1335
1336 /* Socket options and virtual interface manipulation. The whole
1337  * virtual interface system is a complete heap, but unfortunately
1338  * that's how BSD mrouted happens to think. Maybe one day with a proper
1339  * MOSPF/PIM router set up we can clean this up.
1340  */
1341
1342 int ip_mroute_setsockopt(struct sock *sk, int optname, sockptr_t optval,
1343                          unsigned int optlen)
1344 {
1345         struct net *net = sock_net(sk);
1346         int val, ret = 0, parent = 0;
1347         struct mr_table *mrt;
1348         struct vifctl vif;
1349         struct mfcctl mfc;
1350         bool do_wrvifwhole;
1351         u32 uval;
1352
1353         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1354         rtnl_lock();
1355         if (sk->sk_type != SOCK_RAW ||
1356             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1357                 ret = -EOPNOTSUPP;
1358                 goto out_unlock;
1359         }
1360
1361         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1362         if (!mrt) {
1363                 ret = -ENOENT;
1364                 goto out_unlock;
1365         }
1366         if (optname != MRT_INIT) {
1367                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1368                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1369                         ret = -EACCES;
1370                         goto out_unlock;
1371                 }
1372         }
1373
1374         switch (optname) {
1375         case MRT_INIT:
1376                 if (optlen != sizeof(int)) {
1377                         ret = -EINVAL;
1378                         break;
1379                 }
1380                 if (rtnl_dereference(mrt->mroute_sk)) {
1381                         ret = -EADDRINUSE;
1382                         break;
1383                 }
1384
1385                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1386                 if (ret == 0) {
1387                         rcu_assign_pointer(mrt->mroute_sk, sk);
1388                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1389                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1390                                                     NETCONFA_MC_FORWARDING,
1391                                                     NETCONFA_IFINDEX_ALL,
1392                                                     net->ipv4.devconf_all);
1393                 }
1394                 break;
1395         case MRT_DONE:
1396                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1397                         ret = -EACCES;
1398                 } else {
1399                         /* We need to unlock here because mrtsock_destruct takes
1400                          * care of rtnl itself and we can't change that due to
1401                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1402                          */
1403                         rtnl_unlock();
1404                         ret = ip_ra_control(sk, 0, NULL);
1405                         goto out;
1406                 }
1407                 break;
1408         case MRT_ADD_VIF:
1409         case MRT_DEL_VIF:
1410                 if (optlen != sizeof(vif)) {
1411                         ret = -EINVAL;
1412                         break;
1413                 }
1414                 if (copy_from_sockptr(&vif, optval, sizeof(vif))) {
1415                         ret = -EFAULT;
1416                         break;
1417                 }
1418                 if (vif.vifc_vifi >= MAXVIFS) {
1419                         ret = -ENFILE;
1420                         break;
1421                 }
1422                 if (optname == MRT_ADD_VIF) {
1423                         ret = vif_add(net, mrt, &vif,
1424                                       sk == rtnl_dereference(mrt->mroute_sk));
1425                 } else {
1426                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1427                 }
1428                 break;
1429         /* Manipulate the forwarding caches. These live
1430          * in a sort of kernel/user symbiosis.
1431          */
1432         case MRT_ADD_MFC:
1433         case MRT_DEL_MFC:
1434                 parent = -1;
1435                 fallthrough;
1436         case MRT_ADD_MFC_PROXY:
1437         case MRT_DEL_MFC_PROXY:
1438                 if (optlen != sizeof(mfc)) {
1439                         ret = -EINVAL;
1440                         break;
1441                 }
1442                 if (copy_from_sockptr(&mfc, optval, sizeof(mfc))) {
1443                         ret = -EFAULT;
1444                         break;
1445                 }
1446                 if (parent == 0)
1447                         parent = mfc.mfcc_parent;
1448                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1449                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1450                 else
1451                         ret = ipmr_mfc_add(net, mrt, &mfc,
1452                                            sk == rtnl_dereference(mrt->mroute_sk),
1453                                            parent);
1454                 break;
1455         case MRT_FLUSH:
1456                 if (optlen != sizeof(val)) {
1457                         ret = -EINVAL;
1458                         break;
1459                 }
1460                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1461                         ret = -EFAULT;
1462                         break;
1463                 }
1464                 mroute_clean_tables(mrt, val);
1465                 break;
1466         /* Control PIM assert. */
1467         case MRT_ASSERT:
1468                 if (optlen != sizeof(val)) {
1469                         ret = -EINVAL;
1470                         break;
1471                 }
1472                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1473                         ret = -EFAULT;
1474                         break;
1475                 }
1476                 mrt->mroute_do_assert = val;
1477                 break;
1478         case MRT_PIM:
1479                 if (!ipmr_pimsm_enabled()) {
1480                         ret = -ENOPROTOOPT;
1481                         break;
1482                 }
1483                 if (optlen != sizeof(val)) {
1484                         ret = -EINVAL;
1485                         break;
1486                 }
1487                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1488                         ret = -EFAULT;
1489                         break;
1490                 }
1491
1492                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1493                 val = !!val;
1494                 if (val != mrt->mroute_do_pim) {
1495                         mrt->mroute_do_pim = val;
1496                         mrt->mroute_do_assert = val;
1497                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1498                 }
1499                 break;
1500         case MRT_TABLE:
1501                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1502                         ret = -ENOPROTOOPT;
1503                         break;
1504                 }
1505                 if (optlen != sizeof(uval)) {
1506                         ret = -EINVAL;
1507                         break;
1508                 }
1509                 if (copy_from_sockptr(&uval, optval, sizeof(uval))) {
1510                         ret = -EFAULT;
1511                         break;
1512                 }
1513
1514                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1515                         ret = -EBUSY;
1516                 } else {
1517                         mrt = ipmr_new_table(net, uval);
1518                         if (IS_ERR(mrt))
1519                                 ret = PTR_ERR(mrt);
1520                         else
1521                                 raw_sk(sk)->ipmr_table = uval;
1522                 }
1523                 break;
1524         /* Spurious command, or MRT_VERSION which you cannot set. */
1525         default:
1526                 ret = -ENOPROTOOPT;
1527         }
1528 out_unlock:
1529         rtnl_unlock();
1530 out:
1531         return ret;
1532 }
1533
1534 /* Getsock opt support for the multicast routing system. */
1535 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1536 {
1537         int olr;
1538         int val;
1539         struct net *net = sock_net(sk);
1540         struct mr_table *mrt;
1541
1542         if (sk->sk_type != SOCK_RAW ||
1543             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1544                 return -EOPNOTSUPP;
1545
1546         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1547         if (!mrt)
1548                 return -ENOENT;
1549
1550         switch (optname) {
1551         case MRT_VERSION:
1552                 val = 0x0305;
1553                 break;
1554         case MRT_PIM:
1555                 if (!ipmr_pimsm_enabled())
1556                         return -ENOPROTOOPT;
1557                 val = mrt->mroute_do_pim;
1558                 break;
1559         case MRT_ASSERT:
1560                 val = mrt->mroute_do_assert;
1561                 break;
1562         default:
1563                 return -ENOPROTOOPT;
1564         }
1565
1566         if (get_user(olr, optlen))
1567                 return -EFAULT;
1568         olr = min_t(unsigned int, olr, sizeof(int));
1569         if (olr < 0)
1570                 return -EINVAL;
1571         if (put_user(olr, optlen))
1572                 return -EFAULT;
1573         if (copy_to_user(optval, &val, olr))
1574                 return -EFAULT;
1575         return 0;
1576 }
1577
1578 /* The IP multicast ioctl support routines. */
1579 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1580 {
1581         struct sioc_sg_req sr;
1582         struct sioc_vif_req vr;
1583         struct vif_device *vif;
1584         struct mfc_cache *c;
1585         struct net *net = sock_net(sk);
1586         struct mr_table *mrt;
1587
1588         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1589         if (!mrt)
1590                 return -ENOENT;
1591
1592         switch (cmd) {
1593         case SIOCGETVIFCNT:
1594                 if (copy_from_user(&vr, arg, sizeof(vr)))
1595                         return -EFAULT;
1596                 if (vr.vifi >= mrt->maxvif)
1597                         return -EINVAL;
1598                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1599                 read_lock(&mrt_lock);
1600                 vif = &mrt->vif_table[vr.vifi];
1601                 if (VIF_EXISTS(mrt, vr.vifi)) {
1602                         vr.icount = vif->pkt_in;
1603                         vr.ocount = vif->pkt_out;
1604                         vr.ibytes = vif->bytes_in;
1605                         vr.obytes = vif->bytes_out;
1606                         read_unlock(&mrt_lock);
1607
1608                         if (copy_to_user(arg, &vr, sizeof(vr)))
1609                                 return -EFAULT;
1610                         return 0;
1611                 }
1612                 read_unlock(&mrt_lock);
1613                 return -EADDRNOTAVAIL;
1614         case SIOCGETSGCNT:
1615                 if (copy_from_user(&sr, arg, sizeof(sr)))
1616                         return -EFAULT;
1617
1618                 rcu_read_lock();
1619                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1620                 if (c) {
1621                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1622                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1623                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1624                         rcu_read_unlock();
1625
1626                         if (copy_to_user(arg, &sr, sizeof(sr)))
1627                                 return -EFAULT;
1628                         return 0;
1629                 }
1630                 rcu_read_unlock();
1631                 return -EADDRNOTAVAIL;
1632         default:
1633                 return -ENOIOCTLCMD;
1634         }
1635 }
1636
1637 #ifdef CONFIG_COMPAT
1638 struct compat_sioc_sg_req {
1639         struct in_addr src;
1640         struct in_addr grp;
1641         compat_ulong_t pktcnt;
1642         compat_ulong_t bytecnt;
1643         compat_ulong_t wrong_if;
1644 };
1645
1646 struct compat_sioc_vif_req {
1647         vifi_t  vifi;           /* Which iface */
1648         compat_ulong_t icount;
1649         compat_ulong_t ocount;
1650         compat_ulong_t ibytes;
1651         compat_ulong_t obytes;
1652 };
1653
1654 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1655 {
1656         struct compat_sioc_sg_req sr;
1657         struct compat_sioc_vif_req vr;
1658         struct vif_device *vif;
1659         struct mfc_cache *c;
1660         struct net *net = sock_net(sk);
1661         struct mr_table *mrt;
1662
1663         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1664         if (!mrt)
1665                 return -ENOENT;
1666
1667         switch (cmd) {
1668         case SIOCGETVIFCNT:
1669                 if (copy_from_user(&vr, arg, sizeof(vr)))
1670                         return -EFAULT;
1671                 if (vr.vifi >= mrt->maxvif)
1672                         return -EINVAL;
1673                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1674                 read_lock(&mrt_lock);
1675                 vif = &mrt->vif_table[vr.vifi];
1676                 if (VIF_EXISTS(mrt, vr.vifi)) {
1677                         vr.icount = vif->pkt_in;
1678                         vr.ocount = vif->pkt_out;
1679                         vr.ibytes = vif->bytes_in;
1680                         vr.obytes = vif->bytes_out;
1681                         read_unlock(&mrt_lock);
1682
1683                         if (copy_to_user(arg, &vr, sizeof(vr)))
1684                                 return -EFAULT;
1685                         return 0;
1686                 }
1687                 read_unlock(&mrt_lock);
1688                 return -EADDRNOTAVAIL;
1689         case SIOCGETSGCNT:
1690                 if (copy_from_user(&sr, arg, sizeof(sr)))
1691                         return -EFAULT;
1692
1693                 rcu_read_lock();
1694                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1695                 if (c) {
1696                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1697                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1698                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1699                         rcu_read_unlock();
1700
1701                         if (copy_to_user(arg, &sr, sizeof(sr)))
1702                                 return -EFAULT;
1703                         return 0;
1704                 }
1705                 rcu_read_unlock();
1706                 return -EADDRNOTAVAIL;
1707         default:
1708                 return -ENOIOCTLCMD;
1709         }
1710 }
1711 #endif
1712
1713 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1714 {
1715         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1716         struct net *net = dev_net(dev);
1717         struct mr_table *mrt;
1718         struct vif_device *v;
1719         int ct;
1720
1721         if (event != NETDEV_UNREGISTER)
1722                 return NOTIFY_DONE;
1723
1724         ipmr_for_each_table(mrt, net) {
1725                 v = &mrt->vif_table[0];
1726                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1727                         if (v->dev == dev)
1728                                 vif_delete(mrt, ct, 1, NULL);
1729                 }
1730         }
1731         return NOTIFY_DONE;
1732 }
1733
1734 static struct notifier_block ip_mr_notifier = {
1735         .notifier_call = ipmr_device_event,
1736 };
1737
1738 /* Encapsulate a packet by attaching a valid IPIP header to it.
1739  * This avoids tunnel drivers and other mess and gives us the speed so
1740  * important for multicast video.
1741  */
1742 static void ip_encap(struct net *net, struct sk_buff *skb,
1743                      __be32 saddr, __be32 daddr)
1744 {
1745         struct iphdr *iph;
1746         const struct iphdr *old_iph = ip_hdr(skb);
1747
1748         skb_push(skb, sizeof(struct iphdr));
1749         skb->transport_header = skb->network_header;
1750         skb_reset_network_header(skb);
1751         iph = ip_hdr(skb);
1752
1753         iph->version    =       4;
1754         iph->tos        =       old_iph->tos;
1755         iph->ttl        =       old_iph->ttl;
1756         iph->frag_off   =       0;
1757         iph->daddr      =       daddr;
1758         iph->saddr      =       saddr;
1759         iph->protocol   =       IPPROTO_IPIP;
1760         iph->ihl        =       5;
1761         iph->tot_len    =       htons(skb->len);
1762         ip_select_ident(net, skb, NULL);
1763         ip_send_check(iph);
1764
1765         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1766         nf_reset_ct(skb);
1767 }
1768
1769 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1770                                       struct sk_buff *skb)
1771 {
1772         struct ip_options *opt = &(IPCB(skb)->opt);
1773
1774         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1775         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1776
1777         if (unlikely(opt->optlen))
1778                 ip_forward_options(skb);
1779
1780         return dst_output(net, sk, skb);
1781 }
1782
1783 #ifdef CONFIG_NET_SWITCHDEV
1784 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1785                                    int in_vifi, int out_vifi)
1786 {
1787         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1788         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1789
1790         if (!skb->offload_l3_fwd_mark)
1791                 return false;
1792         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1793                 return false;
1794         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1795                                         &in_vif->dev_parent_id);
1796 }
1797 #else
1798 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1799                                    int in_vifi, int out_vifi)
1800 {
1801         return false;
1802 }
1803 #endif
1804
1805 /* Processing handlers for ipmr_forward */
1806
1807 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1808                             int in_vifi, struct sk_buff *skb, int vifi)
1809 {
1810         const struct iphdr *iph = ip_hdr(skb);
1811         struct vif_device *vif = &mrt->vif_table[vifi];
1812         struct net_device *dev;
1813         struct rtable *rt;
1814         struct flowi4 fl4;
1815         int    encap = 0;
1816
1817         if (!vif->dev)
1818                 goto out_free;
1819
1820         if (vif->flags & VIFF_REGISTER) {
1821                 vif->pkt_out++;
1822                 vif->bytes_out += skb->len;
1823                 vif->dev->stats.tx_bytes += skb->len;
1824                 vif->dev->stats.tx_packets++;
1825                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1826                 goto out_free;
1827         }
1828
1829         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1830                 goto out_free;
1831
1832         if (vif->flags & VIFF_TUNNEL) {
1833                 rt = ip_route_output_ports(net, &fl4, NULL,
1834                                            vif->remote, vif->local,
1835                                            0, 0,
1836                                            IPPROTO_IPIP,
1837                                            RT_TOS(iph->tos), vif->link);
1838                 if (IS_ERR(rt))
1839                         goto out_free;
1840                 encap = sizeof(struct iphdr);
1841         } else {
1842                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1843                                            0, 0,
1844                                            IPPROTO_IPIP,
1845                                            RT_TOS(iph->tos), vif->link);
1846                 if (IS_ERR(rt))
1847                         goto out_free;
1848         }
1849
1850         dev = rt->dst.dev;
1851
1852         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1853                 /* Do not fragment multicasts. Alas, IPv4 does not
1854                  * allow to send ICMP, so that packets will disappear
1855                  * to blackhole.
1856                  */
1857                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1858                 ip_rt_put(rt);
1859                 goto out_free;
1860         }
1861
1862         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1863
1864         if (skb_cow(skb, encap)) {
1865                 ip_rt_put(rt);
1866                 goto out_free;
1867         }
1868
1869         vif->pkt_out++;
1870         vif->bytes_out += skb->len;
1871
1872         skb_dst_drop(skb);
1873         skb_dst_set(skb, &rt->dst);
1874         ip_decrease_ttl(ip_hdr(skb));
1875
1876         /* FIXME: forward and output firewalls used to be called here.
1877          * What do we do with netfilter? -- RR
1878          */
1879         if (vif->flags & VIFF_TUNNEL) {
1880                 ip_encap(net, skb, vif->local, vif->remote);
1881                 /* FIXME: extra output firewall step used to be here. --RR */
1882                 vif->dev->stats.tx_packets++;
1883                 vif->dev->stats.tx_bytes += skb->len;
1884         }
1885
1886         IPCB(skb)->flags |= IPSKB_FORWARDED;
1887
1888         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1889          * not only before forwarding, but after forwarding on all output
1890          * interfaces. It is clear, if mrouter runs a multicasting
1891          * program, it should receive packets not depending to what interface
1892          * program is joined.
1893          * If we will not make it, the program will have to join on all
1894          * interfaces. On the other hand, multihoming host (or router, but
1895          * not mrouter) cannot join to more than one interface - it will
1896          * result in receiving multiple packets.
1897          */
1898         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1899                 net, NULL, skb, skb->dev, dev,
1900                 ipmr_forward_finish);
1901         return;
1902
1903 out_free:
1904         kfree_skb(skb);
1905 }
1906
1907 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1908 {
1909         int ct;
1910
1911         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1912                 if (mrt->vif_table[ct].dev == dev)
1913                         break;
1914         }
1915         return ct;
1916 }
1917
1918 /* "local" means that we should preserve one skb (for local delivery) */
1919 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1920                           struct net_device *dev, struct sk_buff *skb,
1921                           struct mfc_cache *c, int local)
1922 {
1923         int true_vifi = ipmr_find_vif(mrt, dev);
1924         int psend = -1;
1925         int vif, ct;
1926
1927         vif = c->_c.mfc_parent;
1928         c->_c.mfc_un.res.pkt++;
1929         c->_c.mfc_un.res.bytes += skb->len;
1930         c->_c.mfc_un.res.lastuse = jiffies;
1931
1932         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1933                 struct mfc_cache *cache_proxy;
1934
1935                 /* For an (*,G) entry, we only check that the incoming
1936                  * interface is part of the static tree.
1937                  */
1938                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1939                 if (cache_proxy &&
1940                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1941                         goto forward;
1942         }
1943
1944         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1945         if (mrt->vif_table[vif].dev != dev) {
1946                 if (rt_is_output_route(skb_rtable(skb))) {
1947                         /* It is our own packet, looped back.
1948                          * Very complicated situation...
1949                          *
1950                          * The best workaround until routing daemons will be
1951                          * fixed is not to redistribute packet, if it was
1952                          * send through wrong interface. It means, that
1953                          * multicast applications WILL NOT work for
1954                          * (S,G), which have default multicast route pointing
1955                          * to wrong oif. In any case, it is not a good
1956                          * idea to use multicasting applications on router.
1957                          */
1958                         goto dont_forward;
1959                 }
1960
1961                 c->_c.mfc_un.res.wrong_if++;
1962
1963                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1964                     /* pimsm uses asserts, when switching from RPT to SPT,
1965                      * so that we cannot check that packet arrived on an oif.
1966                      * It is bad, but otherwise we would need to move pretty
1967                      * large chunk of pimd to kernel. Ough... --ANK
1968                      */
1969                     (mrt->mroute_do_pim ||
1970                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1971                     time_after(jiffies,
1972                                c->_c.mfc_un.res.last_assert +
1973                                MFC_ASSERT_THRESH)) {
1974                         c->_c.mfc_un.res.last_assert = jiffies;
1975                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1976                         if (mrt->mroute_do_wrvifwhole)
1977                                 ipmr_cache_report(mrt, skb, true_vifi,
1978                                                   IGMPMSG_WRVIFWHOLE);
1979                 }
1980                 goto dont_forward;
1981         }
1982
1983 forward:
1984         mrt->vif_table[vif].pkt_in++;
1985         mrt->vif_table[vif].bytes_in += skb->len;
1986
1987         /* Forward the frame */
1988         if (c->mfc_origin == htonl(INADDR_ANY) &&
1989             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
1990                 if (true_vifi >= 0 &&
1991                     true_vifi != c->_c.mfc_parent &&
1992                     ip_hdr(skb)->ttl >
1993                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
1994                         /* It's an (*,*) entry and the packet is not coming from
1995                          * the upstream: forward the packet to the upstream
1996                          * only.
1997                          */
1998                         psend = c->_c.mfc_parent;
1999                         goto last_forward;
2000                 }
2001                 goto dont_forward;
2002         }
2003         for (ct = c->_c.mfc_un.res.maxvif - 1;
2004              ct >= c->_c.mfc_un.res.minvif; ct--) {
2005                 /* For (*,G) entry, don't forward to the incoming interface */
2006                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2007                      ct != true_vifi) &&
2008                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2009                         if (psend != -1) {
2010                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2011
2012                                 if (skb2)
2013                                         ipmr_queue_xmit(net, mrt, true_vifi,
2014                                                         skb2, psend);
2015                         }
2016                         psend = ct;
2017                 }
2018         }
2019 last_forward:
2020         if (psend != -1) {
2021                 if (local) {
2022                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2023
2024                         if (skb2)
2025                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2026                                                 psend);
2027                 } else {
2028                         ipmr_queue_xmit(net, mrt, true_vifi, skb, psend);
2029                         return;
2030                 }
2031         }
2032
2033 dont_forward:
2034         if (!local)
2035                 kfree_skb(skb);
2036 }
2037
2038 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2039 {
2040         struct rtable *rt = skb_rtable(skb);
2041         struct iphdr *iph = ip_hdr(skb);
2042         struct flowi4 fl4 = {
2043                 .daddr = iph->daddr,
2044                 .saddr = iph->saddr,
2045                 .flowi4_tos = RT_TOS(iph->tos),
2046                 .flowi4_oif = (rt_is_output_route(rt) ?
2047                                skb->dev->ifindex : 0),
2048                 .flowi4_iif = (rt_is_output_route(rt) ?
2049                                LOOPBACK_IFINDEX :
2050                                skb->dev->ifindex),
2051                 .flowi4_mark = skb->mark,
2052         };
2053         struct mr_table *mrt;
2054         int err;
2055
2056         err = ipmr_fib_lookup(net, &fl4, &mrt);
2057         if (err)
2058                 return ERR_PTR(err);
2059         return mrt;
2060 }
2061
2062 /* Multicast packets for forwarding arrive here
2063  * Called with rcu_read_lock();
2064  */
2065 int ip_mr_input(struct sk_buff *skb)
2066 {
2067         struct mfc_cache *cache;
2068         struct net *net = dev_net(skb->dev);
2069         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2070         struct mr_table *mrt;
2071         struct net_device *dev;
2072
2073         /* skb->dev passed in is the loX master dev for vrfs.
2074          * As there are no vifs associated with loopback devices,
2075          * get the proper interface that does have a vif associated with it.
2076          */
2077         dev = skb->dev;
2078         if (netif_is_l3_master(skb->dev)) {
2079                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2080                 if (!dev) {
2081                         kfree_skb(skb);
2082                         return -ENODEV;
2083                 }
2084         }
2085
2086         /* Packet is looped back after forward, it should not be
2087          * forwarded second time, but still can be delivered locally.
2088          */
2089         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2090                 goto dont_forward;
2091
2092         mrt = ipmr_rt_fib_lookup(net, skb);
2093         if (IS_ERR(mrt)) {
2094                 kfree_skb(skb);
2095                 return PTR_ERR(mrt);
2096         }
2097         if (!local) {
2098                 if (IPCB(skb)->opt.router_alert) {
2099                         if (ip_call_ra_chain(skb))
2100                                 return 0;
2101                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2102                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2103                          * Cisco IOS <= 11.2(8)) do not put router alert
2104                          * option to IGMP packets destined to routable
2105                          * groups. It is very bad, because it means
2106                          * that we can forward NO IGMP messages.
2107                          */
2108                         struct sock *mroute_sk;
2109
2110                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2111                         if (mroute_sk) {
2112                                 nf_reset_ct(skb);
2113                                 raw_rcv(mroute_sk, skb);
2114                                 return 0;
2115                         }
2116                 }
2117         }
2118
2119         /* already under rcu_read_lock() */
2120         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2121         if (!cache) {
2122                 int vif = ipmr_find_vif(mrt, dev);
2123
2124                 if (vif >= 0)
2125                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2126                                                     vif);
2127         }
2128
2129         /* No usable cache entry */
2130         if (!cache) {
2131                 int vif;
2132
2133                 if (local) {
2134                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2135                         ip_local_deliver(skb);
2136                         if (!skb2)
2137                                 return -ENOBUFS;
2138                         skb = skb2;
2139                 }
2140
2141                 read_lock(&mrt_lock);
2142                 vif = ipmr_find_vif(mrt, dev);
2143                 if (vif >= 0) {
2144                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2145                         read_unlock(&mrt_lock);
2146
2147                         return err2;
2148                 }
2149                 read_unlock(&mrt_lock);
2150                 kfree_skb(skb);
2151                 return -ENODEV;
2152         }
2153
2154         read_lock(&mrt_lock);
2155         ip_mr_forward(net, mrt, dev, skb, cache, local);
2156         read_unlock(&mrt_lock);
2157
2158         if (local)
2159                 return ip_local_deliver(skb);
2160
2161         return 0;
2162
2163 dont_forward:
2164         if (local)
2165                 return ip_local_deliver(skb);
2166         kfree_skb(skb);
2167         return 0;
2168 }
2169
2170 #ifdef CONFIG_IP_PIMSM_V1
2171 /* Handle IGMP messages of PIMv1 */
2172 int pim_rcv_v1(struct sk_buff *skb)
2173 {
2174         struct igmphdr *pim;
2175         struct net *net = dev_net(skb->dev);
2176         struct mr_table *mrt;
2177
2178         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2179                 goto drop;
2180
2181         pim = igmp_hdr(skb);
2182
2183         mrt = ipmr_rt_fib_lookup(net, skb);
2184         if (IS_ERR(mrt))
2185                 goto drop;
2186         if (!mrt->mroute_do_pim ||
2187             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2188                 goto drop;
2189
2190         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2191 drop:
2192                 kfree_skb(skb);
2193         }
2194         return 0;
2195 }
2196 #endif
2197
2198 #ifdef CONFIG_IP_PIMSM_V2
2199 static int pim_rcv(struct sk_buff *skb)
2200 {
2201         struct pimreghdr *pim;
2202         struct net *net = dev_net(skb->dev);
2203         struct mr_table *mrt;
2204
2205         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2206                 goto drop;
2207
2208         pim = (struct pimreghdr *)skb_transport_header(skb);
2209         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2210             (pim->flags & PIM_NULL_REGISTER) ||
2211             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2212              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2213                 goto drop;
2214
2215         mrt = ipmr_rt_fib_lookup(net, skb);
2216         if (IS_ERR(mrt))
2217                 goto drop;
2218         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2219 drop:
2220                 kfree_skb(skb);
2221         }
2222         return 0;
2223 }
2224 #endif
2225
2226 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2227                    __be32 saddr, __be32 daddr,
2228                    struct rtmsg *rtm, u32 portid)
2229 {
2230         struct mfc_cache *cache;
2231         struct mr_table *mrt;
2232         int err;
2233
2234         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2235         if (!mrt)
2236                 return -ENOENT;
2237
2238         rcu_read_lock();
2239         cache = ipmr_cache_find(mrt, saddr, daddr);
2240         if (!cache && skb->dev) {
2241                 int vif = ipmr_find_vif(mrt, skb->dev);
2242
2243                 if (vif >= 0)
2244                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2245         }
2246         if (!cache) {
2247                 struct sk_buff *skb2;
2248                 struct iphdr *iph;
2249                 struct net_device *dev;
2250                 int vif = -1;
2251
2252                 dev = skb->dev;
2253                 read_lock(&mrt_lock);
2254                 if (dev)
2255                         vif = ipmr_find_vif(mrt, dev);
2256                 if (vif < 0) {
2257                         read_unlock(&mrt_lock);
2258                         rcu_read_unlock();
2259                         return -ENODEV;
2260                 }
2261
2262                 skb2 = skb_realloc_headroom(skb, sizeof(struct iphdr));
2263                 if (!skb2) {
2264                         read_unlock(&mrt_lock);
2265                         rcu_read_unlock();
2266                         return -ENOMEM;
2267                 }
2268
2269                 NETLINK_CB(skb2).portid = portid;
2270                 skb_push(skb2, sizeof(struct iphdr));
2271                 skb_reset_network_header(skb2);
2272                 iph = ip_hdr(skb2);
2273                 iph->ihl = sizeof(struct iphdr) >> 2;
2274                 iph->saddr = saddr;
2275                 iph->daddr = daddr;
2276                 iph->version = 0;
2277                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2278                 read_unlock(&mrt_lock);
2279                 rcu_read_unlock();
2280                 return err;
2281         }
2282
2283         read_lock(&mrt_lock);
2284         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2285         read_unlock(&mrt_lock);
2286         rcu_read_unlock();
2287         return err;
2288 }
2289
2290 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2291                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2292                             int flags)
2293 {
2294         struct nlmsghdr *nlh;
2295         struct rtmsg *rtm;
2296         int err;
2297
2298         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2299         if (!nlh)
2300                 return -EMSGSIZE;
2301
2302         rtm = nlmsg_data(nlh);
2303         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2304         rtm->rtm_dst_len  = 32;
2305         rtm->rtm_src_len  = 32;
2306         rtm->rtm_tos      = 0;
2307         rtm->rtm_table    = mrt->id;
2308         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2309                 goto nla_put_failure;
2310         rtm->rtm_type     = RTN_MULTICAST;
2311         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2312         if (c->_c.mfc_flags & MFC_STATIC)
2313                 rtm->rtm_protocol = RTPROT_STATIC;
2314         else
2315                 rtm->rtm_protocol = RTPROT_MROUTED;
2316         rtm->rtm_flags    = 0;
2317
2318         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2319             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2320                 goto nla_put_failure;
2321         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2322         /* do not break the dump if cache is unresolved */
2323         if (err < 0 && err != -ENOENT)
2324                 goto nla_put_failure;
2325
2326         nlmsg_end(skb, nlh);
2327         return 0;
2328
2329 nla_put_failure:
2330         nlmsg_cancel(skb, nlh);
2331         return -EMSGSIZE;
2332 }
2333
2334 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2335                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2336                              int flags)
2337 {
2338         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2339                                 cmd, flags);
2340 }
2341
2342 static size_t mroute_msgsize(bool unresolved, int maxvif)
2343 {
2344         size_t len =
2345                 NLMSG_ALIGN(sizeof(struct rtmsg))
2346                 + nla_total_size(4)     /* RTA_TABLE */
2347                 + nla_total_size(4)     /* RTA_SRC */
2348                 + nla_total_size(4)     /* RTA_DST */
2349                 ;
2350
2351         if (!unresolved)
2352                 len = len
2353                       + nla_total_size(4)       /* RTA_IIF */
2354                       + nla_total_size(0)       /* RTA_MULTIPATH */
2355                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2356                                                 /* RTA_MFC_STATS */
2357                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2358                 ;
2359
2360         return len;
2361 }
2362
2363 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2364                                  int cmd)
2365 {
2366         struct net *net = read_pnet(&mrt->net);
2367         struct sk_buff *skb;
2368         int err = -ENOBUFS;
2369
2370         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2371                                        mrt->maxvif),
2372                         GFP_ATOMIC);
2373         if (!skb)
2374                 goto errout;
2375
2376         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2377         if (err < 0)
2378                 goto errout;
2379
2380         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2381         return;
2382
2383 errout:
2384         kfree_skb(skb);
2385         if (err < 0)
2386                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2387 }
2388
2389 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2390 {
2391         size_t len =
2392                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2393                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2394                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2395                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2396                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2397                 + nla_total_size(4)     /* IPMRA_CREPORT_TABLE */
2398                                         /* IPMRA_CREPORT_PKT */
2399                 + nla_total_size(payloadlen)
2400                 ;
2401
2402         return len;
2403 }
2404
2405 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2406 {
2407         struct net *net = read_pnet(&mrt->net);
2408         struct nlmsghdr *nlh;
2409         struct rtgenmsg *rtgenm;
2410         struct igmpmsg *msg;
2411         struct sk_buff *skb;
2412         struct nlattr *nla;
2413         int payloadlen;
2414
2415         payloadlen = pkt->len - sizeof(struct igmpmsg);
2416         msg = (struct igmpmsg *)skb_network_header(pkt);
2417
2418         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2419         if (!skb)
2420                 goto errout;
2421
2422         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2423                         sizeof(struct rtgenmsg), 0);
2424         if (!nlh)
2425                 goto errout;
2426         rtgenm = nlmsg_data(nlh);
2427         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2428         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2429             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif | (msg->im_vif_hi << 8)) ||
2430             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2431                             msg->im_src.s_addr) ||
2432             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2433                             msg->im_dst.s_addr) ||
2434             nla_put_u32(skb, IPMRA_CREPORT_TABLE, mrt->id))
2435                 goto nla_put_failure;
2436
2437         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2438         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2439                                   nla_data(nla), payloadlen))
2440                 goto nla_put_failure;
2441
2442         nlmsg_end(skb, nlh);
2443
2444         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2445         return;
2446
2447 nla_put_failure:
2448         nlmsg_cancel(skb, nlh);
2449 errout:
2450         kfree_skb(skb);
2451         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2452 }
2453
2454 static int ipmr_rtm_valid_getroute_req(struct sk_buff *skb,
2455                                        const struct nlmsghdr *nlh,
2456                                        struct nlattr **tb,
2457                                        struct netlink_ext_ack *extack)
2458 {
2459         struct rtmsg *rtm;
2460         int i, err;
2461
2462         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2463                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for multicast route get request");
2464                 return -EINVAL;
2465         }
2466
2467         if (!netlink_strict_get_check(skb))
2468                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2469                                               rtm_ipv4_policy, extack);
2470
2471         rtm = nlmsg_data(nlh);
2472         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2473             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2474             rtm->rtm_tos || rtm->rtm_table || rtm->rtm_protocol ||
2475             rtm->rtm_scope || rtm->rtm_type || rtm->rtm_flags) {
2476                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for multicast route get request");
2477                 return -EINVAL;
2478         }
2479
2480         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2481                                             rtm_ipv4_policy, extack);
2482         if (err)
2483                 return err;
2484
2485         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2486             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2487                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2488                 return -EINVAL;
2489         }
2490
2491         for (i = 0; i <= RTA_MAX; i++) {
2492                 if (!tb[i])
2493                         continue;
2494
2495                 switch (i) {
2496                 case RTA_SRC:
2497                 case RTA_DST:
2498                 case RTA_TABLE:
2499                         break;
2500                 default:
2501                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in multicast route get request");
2502                         return -EINVAL;
2503                 }
2504         }
2505
2506         return 0;
2507 }
2508
2509 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2510                              struct netlink_ext_ack *extack)
2511 {
2512         struct net *net = sock_net(in_skb->sk);
2513         struct nlattr *tb[RTA_MAX + 1];
2514         struct sk_buff *skb = NULL;
2515         struct mfc_cache *cache;
2516         struct mr_table *mrt;
2517         __be32 src, grp;
2518         u32 tableid;
2519         int err;
2520
2521         err = ipmr_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2522         if (err < 0)
2523                 goto errout;
2524
2525         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2526         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2527         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2528
2529         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2530         if (!mrt) {
2531                 err = -ENOENT;
2532                 goto errout_free;
2533         }
2534
2535         /* entries are added/deleted only under RTNL */
2536         rcu_read_lock();
2537         cache = ipmr_cache_find(mrt, src, grp);
2538         rcu_read_unlock();
2539         if (!cache) {
2540                 err = -ENOENT;
2541                 goto errout_free;
2542         }
2543
2544         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2545         if (!skb) {
2546                 err = -ENOBUFS;
2547                 goto errout_free;
2548         }
2549
2550         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2551                                nlh->nlmsg_seq, cache,
2552                                RTM_NEWROUTE, 0);
2553         if (err < 0)
2554                 goto errout_free;
2555
2556         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2557
2558 errout:
2559         return err;
2560
2561 errout_free:
2562         kfree_skb(skb);
2563         goto errout;
2564 }
2565
2566 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2567 {
2568         struct fib_dump_filter filter = {};
2569         int err;
2570
2571         if (cb->strict_check) {
2572                 err = ip_valid_fib_dump_req(sock_net(skb->sk), cb->nlh,
2573                                             &filter, cb);
2574                 if (err < 0)
2575                         return err;
2576         }
2577
2578         if (filter.table_id) {
2579                 struct mr_table *mrt;
2580
2581                 mrt = ipmr_get_table(sock_net(skb->sk), filter.table_id);
2582                 if (!mrt) {
2583                         if (rtnl_msg_family(cb->nlh) != RTNL_FAMILY_IPMR)
2584                                 return skb->len;
2585
2586                         NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
2587                         return -ENOENT;
2588                 }
2589                 err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
2590                                     &mfc_unres_lock, &filter);
2591                 return skb->len ? : err;
2592         }
2593
2594         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2595                                 _ipmr_fill_mroute, &mfc_unres_lock, &filter);
2596 }
2597
2598 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2599         [RTA_SRC]       = { .type = NLA_U32 },
2600         [RTA_DST]       = { .type = NLA_U32 },
2601         [RTA_IIF]       = { .type = NLA_U32 },
2602         [RTA_TABLE]     = { .type = NLA_U32 },
2603         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2604 };
2605
2606 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2607 {
2608         switch (rtm_protocol) {
2609         case RTPROT_STATIC:
2610         case RTPROT_MROUTED:
2611                 return true;
2612         }
2613         return false;
2614 }
2615
2616 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2617 {
2618         struct rtnexthop *rtnh = nla_data(nla);
2619         int remaining = nla_len(nla), vifi = 0;
2620
2621         while (rtnh_ok(rtnh, remaining)) {
2622                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2623                 if (++vifi == MAXVIFS)
2624                         break;
2625                 rtnh = rtnh_next(rtnh, &remaining);
2626         }
2627
2628         return remaining > 0 ? -EINVAL : vifi;
2629 }
2630
2631 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2632 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2633                             struct mfcctl *mfcc, int *mrtsock,
2634                             struct mr_table **mrtret,
2635                             struct netlink_ext_ack *extack)
2636 {
2637         struct net_device *dev = NULL;
2638         u32 tblid = RT_TABLE_DEFAULT;
2639         struct mr_table *mrt;
2640         struct nlattr *attr;
2641         struct rtmsg *rtm;
2642         int ret, rem;
2643
2644         ret = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
2645                                         rtm_ipmr_policy, extack);
2646         if (ret < 0)
2647                 goto out;
2648         rtm = nlmsg_data(nlh);
2649
2650         ret = -EINVAL;
2651         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2652             rtm->rtm_type != RTN_MULTICAST ||
2653             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2654             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2655                 goto out;
2656
2657         memset(mfcc, 0, sizeof(*mfcc));
2658         mfcc->mfcc_parent = -1;
2659         ret = 0;
2660         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2661                 switch (nla_type(attr)) {
2662                 case RTA_SRC:
2663                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2664                         break;
2665                 case RTA_DST:
2666                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2667                         break;
2668                 case RTA_IIF:
2669                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2670                         if (!dev) {
2671                                 ret = -ENODEV;
2672                                 goto out;
2673                         }
2674                         break;
2675                 case RTA_MULTIPATH:
2676                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2677                                 ret = -EINVAL;
2678                                 goto out;
2679                         }
2680                         break;
2681                 case RTA_PREFSRC:
2682                         ret = 1;
2683                         break;
2684                 case RTA_TABLE:
2685                         tblid = nla_get_u32(attr);
2686                         break;
2687                 }
2688         }
2689         mrt = ipmr_get_table(net, tblid);
2690         if (!mrt) {
2691                 ret = -ENOENT;
2692                 goto out;
2693         }
2694         *mrtret = mrt;
2695         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2696         if (dev)
2697                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2698
2699 out:
2700         return ret;
2701 }
2702
2703 /* takes care of both newroute and delroute */
2704 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2705                           struct netlink_ext_ack *extack)
2706 {
2707         struct net *net = sock_net(skb->sk);
2708         int ret, mrtsock, parent;
2709         struct mr_table *tbl;
2710         struct mfcctl mfcc;
2711
2712         mrtsock = 0;
2713         tbl = NULL;
2714         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2715         if (ret < 0)
2716                 return ret;
2717
2718         parent = ret ? mfcc.mfcc_parent : -1;
2719         if (nlh->nlmsg_type == RTM_NEWROUTE)
2720                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2721         else
2722                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2723 }
2724
2725 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2726 {
2727         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2728
2729         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2730             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2731             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2732                         mrt->mroute_reg_vif_num) ||
2733             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2734                        mrt->mroute_do_assert) ||
2735             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2736             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2737                        mrt->mroute_do_wrvifwhole))
2738                 return false;
2739
2740         return true;
2741 }
2742
2743 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2744 {
2745         struct nlattr *vif_nest;
2746         struct vif_device *vif;
2747
2748         /* if the VIF doesn't exist just continue */
2749         if (!VIF_EXISTS(mrt, vifid))
2750                 return true;
2751
2752         vif = &mrt->vif_table[vifid];
2753         vif_nest = nla_nest_start_noflag(skb, IPMRA_VIF);
2754         if (!vif_nest)
2755                 return false;
2756         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2757             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2758             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2759             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2760                               IPMRA_VIFA_PAD) ||
2761             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2762                               IPMRA_VIFA_PAD) ||
2763             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2764                               IPMRA_VIFA_PAD) ||
2765             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2766                               IPMRA_VIFA_PAD) ||
2767             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2768             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2769                 nla_nest_cancel(skb, vif_nest);
2770                 return false;
2771         }
2772         nla_nest_end(skb, vif_nest);
2773
2774         return true;
2775 }
2776
2777 static int ipmr_valid_dumplink(const struct nlmsghdr *nlh,
2778                                struct netlink_ext_ack *extack)
2779 {
2780         struct ifinfomsg *ifm;
2781
2782         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ifm))) {
2783                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for ipmr link dump");
2784                 return -EINVAL;
2785         }
2786
2787         if (nlmsg_attrlen(nlh, sizeof(*ifm))) {
2788                 NL_SET_ERR_MSG(extack, "Invalid data after header in ipmr link dump");
2789                 return -EINVAL;
2790         }
2791
2792         ifm = nlmsg_data(nlh);
2793         if (ifm->__ifi_pad || ifm->ifi_type || ifm->ifi_flags ||
2794             ifm->ifi_change || ifm->ifi_index) {
2795                 NL_SET_ERR_MSG(extack, "Invalid values in header for ipmr link dump request");
2796                 return -EINVAL;
2797         }
2798
2799         return 0;
2800 }
2801
2802 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2803 {
2804         struct net *net = sock_net(skb->sk);
2805         struct nlmsghdr *nlh = NULL;
2806         unsigned int t = 0, s_t;
2807         unsigned int e = 0, s_e;
2808         struct mr_table *mrt;
2809
2810         if (cb->strict_check) {
2811                 int err = ipmr_valid_dumplink(cb->nlh, cb->extack);
2812
2813                 if (err < 0)
2814                         return err;
2815         }
2816
2817         s_t = cb->args[0];
2818         s_e = cb->args[1];
2819
2820         ipmr_for_each_table(mrt, net) {
2821                 struct nlattr *vifs, *af;
2822                 struct ifinfomsg *hdr;
2823                 u32 i;
2824
2825                 if (t < s_t)
2826                         goto skip_table;
2827                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2828                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2829                                 sizeof(*hdr), NLM_F_MULTI);
2830                 if (!nlh)
2831                         break;
2832
2833                 hdr = nlmsg_data(nlh);
2834                 memset(hdr, 0, sizeof(*hdr));
2835                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2836
2837                 af = nla_nest_start_noflag(skb, IFLA_AF_SPEC);
2838                 if (!af) {
2839                         nlmsg_cancel(skb, nlh);
2840                         goto out;
2841                 }
2842
2843                 if (!ipmr_fill_table(mrt, skb)) {
2844                         nlmsg_cancel(skb, nlh);
2845                         goto out;
2846                 }
2847
2848                 vifs = nla_nest_start_noflag(skb, IPMRA_TABLE_VIFS);
2849                 if (!vifs) {
2850                         nla_nest_end(skb, af);
2851                         nlmsg_end(skb, nlh);
2852                         goto out;
2853                 }
2854                 for (i = 0; i < mrt->maxvif; i++) {
2855                         if (e < s_e)
2856                                 goto skip_entry;
2857                         if (!ipmr_fill_vif(mrt, i, skb)) {
2858                                 nla_nest_end(skb, vifs);
2859                                 nla_nest_end(skb, af);
2860                                 nlmsg_end(skb, nlh);
2861                                 goto out;
2862                         }
2863 skip_entry:
2864                         e++;
2865                 }
2866                 s_e = 0;
2867                 e = 0;
2868                 nla_nest_end(skb, vifs);
2869                 nla_nest_end(skb, af);
2870                 nlmsg_end(skb, nlh);
2871 skip_table:
2872                 t++;
2873         }
2874
2875 out:
2876         cb->args[1] = e;
2877         cb->args[0] = t;
2878
2879         return skb->len;
2880 }
2881
2882 #ifdef CONFIG_PROC_FS
2883 /* The /proc interfaces to multicast routing :
2884  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2885  */
2886
2887 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2888         __acquires(mrt_lock)
2889 {
2890         struct mr_vif_iter *iter = seq->private;
2891         struct net *net = seq_file_net(seq);
2892         struct mr_table *mrt;
2893
2894         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2895         if (!mrt)
2896                 return ERR_PTR(-ENOENT);
2897
2898         iter->mrt = mrt;
2899
2900         read_lock(&mrt_lock);
2901         return mr_vif_seq_start(seq, pos);
2902 }
2903
2904 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2905         __releases(mrt_lock)
2906 {
2907         read_unlock(&mrt_lock);
2908 }
2909
2910 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2911 {
2912         struct mr_vif_iter *iter = seq->private;
2913         struct mr_table *mrt = iter->mrt;
2914
2915         if (v == SEQ_START_TOKEN) {
2916                 seq_puts(seq,
2917                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2918         } else {
2919                 const struct vif_device *vif = v;
2920                 const char *name =  vif->dev ?
2921                                     vif->dev->name : "none";
2922
2923                 seq_printf(seq,
2924                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2925                            vif - mrt->vif_table,
2926                            name, vif->bytes_in, vif->pkt_in,
2927                            vif->bytes_out, vif->pkt_out,
2928                            vif->flags, vif->local, vif->remote);
2929         }
2930         return 0;
2931 }
2932
2933 static const struct seq_operations ipmr_vif_seq_ops = {
2934         .start = ipmr_vif_seq_start,
2935         .next  = mr_vif_seq_next,
2936         .stop  = ipmr_vif_seq_stop,
2937         .show  = ipmr_vif_seq_show,
2938 };
2939
2940 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2941 {
2942         struct net *net = seq_file_net(seq);
2943         struct mr_table *mrt;
2944
2945         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2946         if (!mrt)
2947                 return ERR_PTR(-ENOENT);
2948
2949         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2950 }
2951
2952 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2953 {
2954         int n;
2955
2956         if (v == SEQ_START_TOKEN) {
2957                 seq_puts(seq,
2958                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2959         } else {
2960                 const struct mfc_cache *mfc = v;
2961                 const struct mr_mfc_iter *it = seq->private;
2962                 const struct mr_table *mrt = it->mrt;
2963
2964                 seq_printf(seq, "%08X %08X %-3hd",
2965                            (__force u32) mfc->mfc_mcastgrp,
2966                            (__force u32) mfc->mfc_origin,
2967                            mfc->_c.mfc_parent);
2968
2969                 if (it->cache != &mrt->mfc_unres_queue) {
2970                         seq_printf(seq, " %8lu %8lu %8lu",
2971                                    mfc->_c.mfc_un.res.pkt,
2972                                    mfc->_c.mfc_un.res.bytes,
2973                                    mfc->_c.mfc_un.res.wrong_if);
2974                         for (n = mfc->_c.mfc_un.res.minvif;
2975                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2976                                 if (VIF_EXISTS(mrt, n) &&
2977                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2978                                         seq_printf(seq,
2979                                            " %2d:%-3d",
2980                                            n, mfc->_c.mfc_un.res.ttls[n]);
2981                         }
2982                 } else {
2983                         /* unresolved mfc_caches don't contain
2984                          * pkt, bytes and wrong_if values
2985                          */
2986                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2987                 }
2988                 seq_putc(seq, '\n');
2989         }
2990         return 0;
2991 }
2992
2993 static const struct seq_operations ipmr_mfc_seq_ops = {
2994         .start = ipmr_mfc_seq_start,
2995         .next  = mr_mfc_seq_next,
2996         .stop  = mr_mfc_seq_stop,
2997         .show  = ipmr_mfc_seq_show,
2998 };
2999 #endif
3000
3001 #ifdef CONFIG_IP_PIMSM_V2
3002 static const struct net_protocol pim_protocol = {
3003         .handler        =       pim_rcv,
3004 };
3005 #endif
3006
3007 static unsigned int ipmr_seq_read(struct net *net)
3008 {
3009         ASSERT_RTNL();
3010
3011         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3012 }
3013
3014 static int ipmr_dump(struct net *net, struct notifier_block *nb,
3015                      struct netlink_ext_ack *extack)
3016 {
3017         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
3018                        ipmr_mr_table_iter, &mrt_lock, extack);
3019 }
3020
3021 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3022         .family         = RTNL_FAMILY_IPMR,
3023         .fib_seq_read   = ipmr_seq_read,
3024         .fib_dump       = ipmr_dump,
3025         .owner          = THIS_MODULE,
3026 };
3027
3028 static int __net_init ipmr_notifier_init(struct net *net)
3029 {
3030         struct fib_notifier_ops *ops;
3031
3032         net->ipv4.ipmr_seq = 0;
3033
3034         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3035         if (IS_ERR(ops))
3036                 return PTR_ERR(ops);
3037         net->ipv4.ipmr_notifier_ops = ops;
3038
3039         return 0;
3040 }
3041
3042 static void __net_exit ipmr_notifier_exit(struct net *net)
3043 {
3044         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3045         net->ipv4.ipmr_notifier_ops = NULL;
3046 }
3047
3048 /* Setup for IP multicast routing */
3049 static int __net_init ipmr_net_init(struct net *net)
3050 {
3051         int err;
3052
3053         err = ipmr_notifier_init(net);
3054         if (err)
3055                 goto ipmr_notifier_fail;
3056
3057         err = ipmr_rules_init(net);
3058         if (err < 0)
3059                 goto ipmr_rules_fail;
3060
3061 #ifdef CONFIG_PROC_FS
3062         err = -ENOMEM;
3063         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
3064                         sizeof(struct mr_vif_iter)))
3065                 goto proc_vif_fail;
3066         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
3067                         sizeof(struct mr_mfc_iter)))
3068                 goto proc_cache_fail;
3069 #endif
3070         return 0;
3071
3072 #ifdef CONFIG_PROC_FS
3073 proc_cache_fail:
3074         remove_proc_entry("ip_mr_vif", net->proc_net);
3075 proc_vif_fail:
3076         rtnl_lock();
3077         ipmr_rules_exit(net);
3078         rtnl_unlock();
3079 #endif
3080 ipmr_rules_fail:
3081         ipmr_notifier_exit(net);
3082 ipmr_notifier_fail:
3083         return err;
3084 }
3085
3086 static void __net_exit ipmr_net_exit(struct net *net)
3087 {
3088 #ifdef CONFIG_PROC_FS
3089         remove_proc_entry("ip_mr_cache", net->proc_net);
3090         remove_proc_entry("ip_mr_vif", net->proc_net);
3091 #endif
3092         ipmr_notifier_exit(net);
3093 }
3094
3095 static void __net_exit ipmr_net_exit_batch(struct list_head *net_list)
3096 {
3097         struct net *net;
3098
3099         rtnl_lock();
3100         list_for_each_entry(net, net_list, exit_list)
3101                 ipmr_rules_exit(net);
3102         rtnl_unlock();
3103 }
3104
3105 static struct pernet_operations ipmr_net_ops = {
3106         .init = ipmr_net_init,
3107         .exit = ipmr_net_exit,
3108         .exit_batch = ipmr_net_exit_batch,
3109 };
3110
3111 int __init ip_mr_init(void)
3112 {
3113         int err;
3114
3115         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3116                                        sizeof(struct mfc_cache),
3117                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3118                                        NULL);
3119
3120         err = register_pernet_subsys(&ipmr_net_ops);
3121         if (err)
3122                 goto reg_pernet_fail;
3123
3124         err = register_netdevice_notifier(&ip_mr_notifier);
3125         if (err)
3126                 goto reg_notif_fail;
3127 #ifdef CONFIG_IP_PIMSM_V2
3128         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3129                 pr_err("%s: can't add PIM protocol\n", __func__);
3130                 err = -EAGAIN;
3131                 goto add_proto_fail;
3132         }
3133 #endif
3134         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3135                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3136         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3137                       ipmr_rtm_route, NULL, 0);
3138         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3139                       ipmr_rtm_route, NULL, 0);
3140
3141         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3142                       NULL, ipmr_rtm_dumplink, 0);
3143         return 0;
3144
3145 #ifdef CONFIG_IP_PIMSM_V2
3146 add_proto_fail:
3147         unregister_netdevice_notifier(&ip_mr_notifier);
3148 #endif
3149 reg_notif_fail:
3150         unregister_pernet_subsys(&ipmr_net_ops);
3151 reg_pernet_fail:
3152         kmem_cache_destroy(mrt_cachep);
3153         return err;
3154 }