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, <[email protected]>
   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_SPINLOCK(mrt_lock);
  81
  82 static struct net_device *vif_dev_read(const struct vif_device *vif)
  83 {
  84         return rcu_dereference(vif->dev);
  85 }
  86
  87 /* Multicast router control variables */
  88
  89 /* Special spinlock for queue of unresolved entries */
  90 static DEFINE_SPINLOCK(mfc_unres_lock);
  91
  92 /* We return to original Alan's scheme. Hash table of resolved
  93  * entries is changed only in process context and protected
  94  * with weak lock mrt_lock. Queue of unresolved entries is protected
  95  * with strong spinlock mfc_unres_lock.
  96  *
  97  * In this case data path is free of exclusive locks at all.
  98  */
  99
 100 static struct kmem_cache *mrt_cachep __ro_after_init;
 101
 102 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 103 static void ipmr_free_table(struct mr_table *mrt);
 104
 105 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 106                           struct net_device *dev, struct sk_buff *skb,
 107                           struct mfc_cache *cache, int local);
 108 static int ipmr_cache_report(const struct mr_table *mrt,
 109                              struct sk_buff *pkt, vifi_t vifi, int assert);
 110 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 111                                  int cmd);
 112 static void igmpmsg_netlink_event(const struct mr_table *mrt, struct sk_buff *pkt);
 113 static void mroute_clean_tables(struct mr_table *mrt, int flags);
 114 static void ipmr_expire_process(struct timer_list *t);
 115
 116 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 117 #define ipmr_for_each_table(mrt, net)                                   \
 118         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list,        \
 119                                 lockdep_rtnl_is_held() ||               \
 120                                 list_empty(&net->ipv4.mr_tables))
 121
 122 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 123                                            struct mr_table *mrt)
 124 {
 125         struct mr_table *ret;
 126
 127         if (!mrt)
 128                 ret = list_entry_rcu(net->ipv4.mr_tables.next,
 129                                      struct mr_table, list);
 130         else
 131                 ret = list_entry_rcu(mrt->list.next,
 132                                      struct mr_table, list);
 133
 134         if (&ret->list == &net->ipv4.mr_tables)
 135                 return NULL;
 136         return ret;
 137 }
 138
 139 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 140 {
 141         struct mr_table *mrt;
 142
 143         ipmr_for_each_table(mrt, net) {
 144                 if (mrt->id == id)
 145                         return mrt;
 146         }
 147         return NULL;
 148 }
 149
 150 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 151                            struct mr_table **mrt)
 152 {
 153         int err;
 154         struct ipmr_result res;
 155         struct fib_lookup_arg arg = {
 156                 .result = &res,
 157                 .flags = FIB_LOOKUP_NOREF,
 158         };
 159
 160         /* update flow if oif or iif point to device enslaved to l3mdev */
 161         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 162
 163         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 164                                flowi4_to_flowi(flp4), 0, &arg);
 165         if (err < 0)
 166                 return err;
 167         *mrt = res.mrt;
 168         return 0;
 169 }
 170
 171 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 172                             int flags, struct fib_lookup_arg *arg)
 173 {
 174         struct ipmr_result *res = arg->result;
 175         struct mr_table *mrt;
 176
 177         switch (rule->action) {
 178         case FR_ACT_TO_TBL:
 179                 break;
 180         case FR_ACT_UNREACHABLE:
 181                 return -ENETUNREACH;
 182         case FR_ACT_PROHIBIT:
 183                 return -EACCES;
 184         case FR_ACT_BLACKHOLE:
 185         default:
 186                 return -EINVAL;
 187         }
 188
 189         arg->table = fib_rule_get_table(rule, arg);
 190
 191         mrt = ipmr_get_table(rule->fr_net, arg->table);
 192         if (!mrt)
 193                 return -EAGAIN;
 194         res->mrt = mrt;
 195         return 0;
 196 }
 197
 198 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 199 {
 200         return 1;
 201 }
 202
 203 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 204                                struct fib_rule_hdr *frh, struct nlattr **tb,
 205                                struct netlink_ext_ack *extack)
 206 {
 207         return 0;
 208 }
 209
 210 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 211                              struct nlattr **tb)
 212 {
 213         return 1;
 214 }
 215
 216 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 217                           struct fib_rule_hdr *frh)
 218 {
 219         frh->dst_len = 0;
 220         frh->src_len = 0;
 221         frh->tos     = 0;
 222         return 0;
 223 }
 224
 225 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 226         .family         = RTNL_FAMILY_IPMR,
 227         .rule_size      = sizeof(struct ipmr_rule),
 228         .addr_size      = sizeof(u32),
 229         .action         = ipmr_rule_action,
 230         .match          = ipmr_rule_match,
 231         .configure      = ipmr_rule_configure,
 232         .compare        = ipmr_rule_compare,
 233         .fill           = ipmr_rule_fill,
 234         .nlgroup        = RTNLGRP_IPV4_RULE,
 235         .owner          = THIS_MODULE,
 236 };
 237
 238 static int __net_init ipmr_rules_init(struct net *net)
 239 {
 240         struct fib_rules_ops *ops;
 241         struct mr_table *mrt;
 242         int err;
 243
 244         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 245         if (IS_ERR(ops))
 246                 return PTR_ERR(ops);
 247
 248         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 249
 250         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 251         if (IS_ERR(mrt)) {
 252                 err = PTR_ERR(mrt);
 253                 goto err1;
 254         }
 255
 256         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 257         if (err < 0)
 258                 goto err2;
 259
 260         net->ipv4.mr_rules_ops = ops;
 261         return 0;
 262
 263 err2:
 264         rtnl_lock();
 265         ipmr_free_table(mrt);
 266         rtnl_unlock();
 267 err1:
 268         fib_rules_unregister(ops);
 269         return err;
 270 }
 271
 272 static void __net_exit ipmr_rules_exit(struct net *net)
 273 {
 274         struct mr_table *mrt, *next;
 275
 276         ASSERT_RTNL();
 277         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 278                 list_del(&mrt->list);
 279                 ipmr_free_table(mrt);
 280         }
 281         fib_rules_unregister(net->ipv4.mr_rules_ops);
 282 }
 283
 284 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 285                            struct netlink_ext_ack *extack)
 286 {
 287         return fib_rules_dump(net, nb, RTNL_FAMILY_IPMR, extack);
 288 }
 289
 290 static unsigned int ipmr_rules_seq_read(struct net *net)
 291 {
 292         return fib_rules_seq_read(net, RTNL_FAMILY_IPMR);
 293 }
 294
 295 bool ipmr_rule_default(const struct fib_rule *rule)
 296 {
 297         return fib_rule_matchall(rule) && rule->table == RT_TABLE_DEFAULT;
 298 }
 299 EXPORT_SYMBOL(ipmr_rule_default);
 300 #else
 301 #define ipmr_for_each_table(mrt, net) \
 302         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 303
 304 static struct mr_table *ipmr_mr_table_iter(struct net *net,
 305                                            struct mr_table *mrt)
 306 {
 307         if (!mrt)
 308                 return net->ipv4.mrt;
 309         return NULL;
 310 }
 311
 312 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 313 {
 314         return net->ipv4.mrt;
 315 }
 316
 317 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 318                            struct mr_table **mrt)
 319 {
 320         *mrt = net->ipv4.mrt;
 321         return 0;
 322 }
 323
 324 static int __net_init ipmr_rules_init(struct net *net)
 325 {
 326         struct mr_table *mrt;
 327
 328         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 329         if (IS_ERR(mrt))
 330                 return PTR_ERR(mrt);
 331         net->ipv4.mrt = mrt;
 332         return 0;
 333 }
 334
 335 static void __net_exit ipmr_rules_exit(struct net *net)
 336 {
 337         ASSERT_RTNL();
 338         ipmr_free_table(net->ipv4.mrt);
 339         net->ipv4.mrt = NULL;
 340 }
 341
 342 static int ipmr_rules_dump(struct net *net, struct notifier_block *nb,
 343                            struct netlink_ext_ack *extack)
 344 {
 345         return 0;
 346 }
 347
 348 static unsigned int ipmr_rules_seq_read(struct net *net)
 349 {
 350         return 0;
 351 }
 352
 353 bool ipmr_rule_default(const struct fib_rule *rule)
 354 {
 355         return true;
 356 }
 357 EXPORT_SYMBOL(ipmr_rule_default);
 358 #endif
 359
 360 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 361                                 const void *ptr)
 362 {
 363         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 364         const struct mfc_cache *c = ptr;
 365
 366         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 367                cmparg->mfc_origin != c->mfc_origin;
 368 }
 369
 370 static const struct rhashtable_params ipmr_rht_params = {
 371         .head_offset = offsetof(struct mr_mfc, mnode),
 372         .key_offset = offsetof(struct mfc_cache, cmparg),
 373         .key_len = sizeof(struct mfc_cache_cmp_arg),
 374         .nelem_hint = 3,
 375         .obj_cmpfn = ipmr_hash_cmp,
 376         .automatic_shrinking = true,
 377 };
 378
 379 static void ipmr_new_table_set(struct mr_table *mrt,
 380                                struct net *net)
 381 {
 382 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 383         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 384 #endif
 385 }
 386
 387 static struct mfc_cache_cmp_arg ipmr_mr_table_ops_cmparg_any = {
 388         .mfc_mcastgrp = htonl(INADDR_ANY),
 389         .mfc_origin = htonl(INADDR_ANY),
 390 };
 391
 392 static struct mr_table_ops ipmr_mr_table_ops = {
 393         .rht_params = &ipmr_rht_params,
 394         .cmparg_any = &ipmr_mr_table_ops_cmparg_any,
 395 };
 396
 397 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 398 {
 399         struct mr_table *mrt;
 400
 401         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 402         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 403                 return ERR_PTR(-EINVAL);
 404
 405         mrt = ipmr_get_table(net, id);
 406         if (mrt)
 407                 return mrt;
 408
 409         return mr_table_alloc(net, id, &ipmr_mr_table_ops,
 410                               ipmr_expire_process, ipmr_new_table_set);
 411 }
 412
 413 static void ipmr_free_table(struct mr_table *mrt)
 414 {
 415         timer_shutdown_sync(&mrt->ipmr_expire_timer);
 416         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC |
 417                                  MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC);
 418         rhltable_destroy(&mrt->mfc_hash);
 419         kfree(mrt);
 420 }
 421
 422 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 423
 424 /* Initialize ipmr pimreg/tunnel in_device */
 425 static bool ipmr_init_vif_indev(const struct net_device *dev)
 426 {
 427         struct in_device *in_dev;
 428
 429         ASSERT_RTNL();
 430
 431         in_dev = __in_dev_get_rtnl(dev);
 432         if (!in_dev)
 433                 return false;
 434         ipv4_devconf_setall(in_dev);
 435         neigh_parms_data_state_setall(in_dev->arp_parms);
 436         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 437
 438         return true;
 439 }
 440
 441 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 442 {
 443         struct net_device *tunnel_dev, *new_dev;
 444         struct ip_tunnel_parm p = { };
 445         int err;
 446
 447         tunnel_dev = __dev_get_by_name(net, "tunl0");
 448         if (!tunnel_dev)
 449                 goto out;
 450
 451         p.iph.daddr = v->vifc_rmt_addr.s_addr;
 452         p.iph.saddr = v->vifc_lcl_addr.s_addr;
 453         p.iph.version = 4;
 454         p.iph.ihl = 5;
 455         p.iph.protocol = IPPROTO_IPIP;
 456         sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 457
 458         if (!tunnel_dev->netdev_ops->ndo_tunnel_ctl)
 459                 goto out;
 460         err = tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 461                         SIOCADDTUNNEL);
 462         if (err)
 463                 goto out;
 464
 465         new_dev = __dev_get_by_name(net, p.name);
 466         if (!new_dev)
 467                 goto out;
 468
 469         new_dev->flags |= IFF_MULTICAST;
 470         if (!ipmr_init_vif_indev(new_dev))
 471                 goto out_unregister;
 472         if (dev_open(new_dev, NULL))
 473                 goto out_unregister;
 474         dev_hold(new_dev);
 475         err = dev_set_allmulti(new_dev, 1);
 476         if (err) {
 477                 dev_close(new_dev);
 478                 tunnel_dev->netdev_ops->ndo_tunnel_ctl(tunnel_dev, &p,
 479                                 SIOCDELTUNNEL);
 480                 dev_put(new_dev);
 481                 new_dev = ERR_PTR(err);
 482         }
 483         return new_dev;
 484
 485 out_unregister:
 486         unregister_netdevice(new_dev);
 487 out:
 488         return ERR_PTR(-ENOBUFS);
 489 }
 490
 491 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 492 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 493 {
 494         struct net *net = dev_net(dev);
 495         struct mr_table *mrt;
 496         struct flowi4 fl4 = {
 497                 .flowi4_oif     = dev->ifindex,
 498                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 499                 .flowi4_mark    = skb->mark,
 500         };
 501         int err;
 502
 503         err = ipmr_fib_lookup(net, &fl4, &mrt);
 504         if (err < 0) {
 505                 kfree_skb(skb);
 506                 return err;
 507         }
 508
 509         DEV_STATS_ADD(dev, tx_bytes, skb->len);
 510         DEV_STATS_INC(dev, tx_packets);
 511         rcu_read_lock();
 512
 513         /* Pairs with WRITE_ONCE() in vif_add() and vif_delete() */
 514         ipmr_cache_report(mrt, skb, READ_ONCE(mrt->mroute_reg_vif_num),
 515                           IGMPMSG_WHOLEPKT);
 516
 517         rcu_read_unlock();
 518         kfree_skb(skb);
 519         return NETDEV_TX_OK;
 520 }
 521
 522 static int reg_vif_get_iflink(const struct net_device *dev)
 523 {
 524         return 0;
 525 }
 526
 527 static const struct net_device_ops reg_vif_netdev_ops = {
 528         .ndo_start_xmit = reg_vif_xmit,
 529         .ndo_get_iflink = reg_vif_get_iflink,
 530 };
 531
 532 static void reg_vif_setup(struct net_device *dev)
 533 {
 534         dev->type               = ARPHRD_PIMREG;
 535         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 536         dev->flags              = IFF_NOARP;
 537         dev->netdev_ops         = &reg_vif_netdev_ops;
 538         dev->needs_free_netdev  = true;
 539         dev->features           |= NETIF_F_NETNS_LOCAL;
 540 }
 541
 542 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 543 {
 544         struct net_device *dev;
 545         char name[IFNAMSIZ];
 546
 547         if (mrt->id == RT_TABLE_DEFAULT)
 548                 sprintf(name, "pimreg");
 549         else
 550                 sprintf(name, "pimreg%u", mrt->id);
 551
 552         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 553
 554         if (!dev)
 555                 return NULL;
 556
 557         dev_net_set(dev, net);
 558
 559         if (register_netdevice(dev)) {
 560                 free_netdev(dev);
 561                 return NULL;
 562         }
 563
 564         if (!ipmr_init_vif_indev(dev))
 565                 goto failure;
 566         if (dev_open(dev, NULL))
 567                 goto failure;
 568
 569         dev_hold(dev);
 570
 571         return dev;
 572
 573 failure:
 574         unregister_netdevice(dev);
 575         return NULL;
 576 }
 577
 578 /* called with rcu_read_lock() */
 579 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 580                      unsigned int pimlen)
 581 {
 582         struct net_device *reg_dev = NULL;
 583         struct iphdr *encap;
 584         int vif_num;
 585
 586         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 587         /* Check that:
 588          * a. packet is really sent to a multicast group
 589          * b. packet is not a NULL-REGISTER
 590          * c. packet is not truncated
 591          */
 592         if (!ipv4_is_multicast(encap->daddr) ||
 593             encap->tot_len == 0 ||
 594             ntohs(encap->tot_len) + pimlen > skb->len)
 595                 return 1;
 596
 597         /* Pairs with WRITE_ONCE() in vif_add()/vid_delete() */
 598         vif_num = READ_ONCE(mrt->mroute_reg_vif_num);
 599         if (vif_num >= 0)
 600                 reg_dev = vif_dev_read(&mrt->vif_table[vif_num]);
 601         if (!reg_dev)
 602                 return 1;
 603
 604         skb->mac_header = skb->network_header;
 605         skb_pull(skb, (u8 *)encap - skb->data);
 606         skb_reset_network_header(skb);
 607         skb->protocol = htons(ETH_P_IP);
 608         skb->ip_summed = CHECKSUM_NONE;
 609
 610         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 611
 612         netif_rx(skb);
 613
 614         return NET_RX_SUCCESS;
 615 }
 616 #else
 617 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 618 {
 619         return NULL;
 620 }
 621 #endif
 622
 623 static int call_ipmr_vif_entry_notifiers(struct net *net,
 624                                          enum fib_event_type event_type,
 625                                          struct vif_device *vif,
 626                                          struct net_device *vif_dev,
 627                                          vifi_t vif_index, u32 tb_id)
 628 {
 629         return mr_call_vif_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 630                                      vif, vif_dev, vif_index, tb_id,
 631                                      &net->ipv4.ipmr_seq);
 632 }
 633
 634 static int call_ipmr_mfc_entry_notifiers(struct net *net,
 635                                          enum fib_event_type event_type,
 636                                          struct mfc_cache *mfc, u32 tb_id)
 637 {
 638         return mr_call_mfc_notifiers(net, RTNL_FAMILY_IPMR, event_type,
 639                                      &mfc->_c, tb_id, &net->ipv4.ipmr_seq);
 640 }
 641
 642 /**
 643  *      vif_delete - Delete a VIF entry
 644  *      @mrt: Table to delete from
 645  *      @vifi: VIF identifier to delete
 646  *      @notify: Set to 1, if the caller is a notifier_call
 647  *      @head: if unregistering the VIF, place it on this queue
 648  */
 649 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 650                       struct list_head *head)
 651 {
 652         struct net *net = read_pnet(&mrt->net);
 653         struct vif_device *v;
 654         struct net_device *dev;
 655         struct in_device *in_dev;
 656
 657         if (vifi < 0 || vifi >= mrt->maxvif)
 658                 return -EADDRNOTAVAIL;
 659
 660         v = &mrt->vif_table[vifi];
 661
 662         dev = rtnl_dereference(v->dev);
 663         if (!dev)
 664                 return -EADDRNOTAVAIL;
 665
 666         spin_lock(&mrt_lock);
 667         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_DEL, v, dev,
 668                                       vifi, mrt->id);
 669         RCU_INIT_POINTER(v->dev, NULL);
 670
 671         if (vifi == mrt->mroute_reg_vif_num) {
 672                 /* Pairs with READ_ONCE() in ipmr_cache_report() and reg_vif_xmit() */
 673                 WRITE_ONCE(mrt->mroute_reg_vif_num, -1);
 674         }
 675         if (vifi + 1 == mrt->maxvif) {
 676                 int tmp;
 677
 678                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 679                         if (VIF_EXISTS(mrt, tmp))
 680                                 break;
 681                 }
 682                 WRITE_ONCE(mrt->maxvif, tmp + 1);
 683         }
 684
 685         spin_unlock(&mrt_lock);
 686
 687         dev_set_allmulti(dev, -1);
 688
 689         in_dev = __in_dev_get_rtnl(dev);
 690         if (in_dev) {
 691                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 692                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 693                                             NETCONFA_MC_FORWARDING,
 694                                             dev->ifindex, &in_dev->cnf);
 695                 ip_rt_multicast_event(in_dev);
 696         }
 697
 698         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 699                 unregister_netdevice_queue(dev, head);
 700
 701         netdev_put(dev, &v->dev_tracker);
 702         return 0;
 703 }
 704
 705 static void ipmr_cache_free_rcu(struct rcu_head *head)
 706 {
 707         struct mr_mfc *c = container_of(head, struct mr_mfc, rcu);
 708
 709         kmem_cache_free(mrt_cachep, (struct mfc_cache *)c);
 710 }
 711
 712 static void ipmr_cache_free(struct mfc_cache *c)
 713 {
 714         call_rcu(&c->_c.rcu, ipmr_cache_free_rcu);
 715 }
 716
 717 /* Destroy an unresolved cache entry, killing queued skbs
 718  * and reporting error to netlink readers.
 719  */
 720 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 721 {
 722         struct net *net = read_pnet(&mrt->net);
 723         struct sk_buff *skb;
 724         struct nlmsgerr *e;
 725
 726         atomic_dec(&mrt->cache_resolve_queue_len);
 727
 728         while ((skb = skb_dequeue(&c->_c.mfc_un.unres.unresolved))) {
 729                 if (ip_hdr(skb)->version == 0) {
 730                         struct nlmsghdr *nlh = skb_pull(skb,
 731                                                         sizeof(struct iphdr));
 732                         nlh->nlmsg_type = NLMSG_ERROR;
 733                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 734                         skb_trim(skb, nlh->nlmsg_len);
 735                         e = nlmsg_data(nlh);
 736                         e->error = -ETIMEDOUT;
 737                         memset(&e->msg, 0, sizeof(e->msg));
 738
 739                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 740                 } else {
 741                         kfree_skb(skb);
 742                 }
 743         }
 744
 745         ipmr_cache_free(c);
 746 }
 747
 748 /* Timer process for the unresolved queue. */
 749 static void ipmr_expire_process(struct timer_list *t)
 750 {
 751         struct mr_table *mrt = from_timer(mrt, t, ipmr_expire_timer);
 752         struct mr_mfc *c, *next;
 753         unsigned long expires;
 754         unsigned long now;
 755
 756         if (!spin_trylock(&mfc_unres_lock)) {
 757                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 758                 return;
 759         }
 760
 761         if (list_empty(&mrt->mfc_unres_queue))
 762                 goto out;
 763
 764         now = jiffies;
 765         expires = 10*HZ;
 766
 767         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 768                 if (time_after(c->mfc_un.unres.expires, now)) {
 769                         unsigned long interval = c->mfc_un.unres.expires - now;
 770                         if (interval < expires)
 771                                 expires = interval;
 772                         continue;
 773                 }
 774
 775                 list_del(&c->list);
 776                 mroute_netlink_event(mrt, (struct mfc_cache *)c, RTM_DELROUTE);
 777                 ipmr_destroy_unres(mrt, (struct mfc_cache *)c);
 778         }
 779
 780         if (!list_empty(&mrt->mfc_unres_queue))
 781                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 782
 783 out:
 784         spin_unlock(&mfc_unres_lock);
 785 }
 786
 787 /* Fill oifs list. It is called under locked mrt_lock. */
 788 static void ipmr_update_thresholds(struct mr_table *mrt, struct mr_mfc *cache,
 789                                    unsigned char *ttls)
 790 {
 791         int vifi;
 792
 793         cache->mfc_un.res.minvif = MAXVIFS;
 794         cache->mfc_un.res.maxvif = 0;
 795         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 796
 797         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 798                 if (VIF_EXISTS(mrt, vifi) &&
 799                     ttls[vifi] && ttls[vifi] < 255) {
 800                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 801                         if (cache->mfc_un.res.minvif > vifi)
 802                                 cache->mfc_un.res.minvif = vifi;
 803                         if (cache->mfc_un.res.maxvif <= vifi)
 804                                 cache->mfc_un.res.maxvif = vifi + 1;
 805                 }
 806         }
 807         cache->mfc_un.res.lastuse = jiffies;
 808 }
 809
 810 static int vif_add(struct net *net, struct mr_table *mrt,
 811                    struct vifctl *vifc, int mrtsock)
 812 {
 813         struct netdev_phys_item_id ppid = { };
 814         int vifi = vifc->vifc_vifi;
 815         struct vif_device *v = &mrt->vif_table[vifi];
 816         struct net_device *dev;
 817         struct in_device *in_dev;
 818         int err;
 819
 820         /* Is vif busy ? */
 821         if (VIF_EXISTS(mrt, vifi))
 822                 return -EADDRINUSE;
 823
 824         switch (vifc->vifc_flags) {
 825         case VIFF_REGISTER:
 826                 if (!ipmr_pimsm_enabled())
 827                         return -EINVAL;
 828                 /* Special Purpose VIF in PIM
 829                  * All the packets will be sent to the daemon
 830                  */
 831                 if (mrt->mroute_reg_vif_num >= 0)
 832                         return -EADDRINUSE;
 833                 dev = ipmr_reg_vif(net, mrt);
 834                 if (!dev)
 835                         return -ENOBUFS;
 836                 err = dev_set_allmulti(dev, 1);
 837                 if (err) {
 838                         unregister_netdevice(dev);
 839                         dev_put(dev);
 840                         return err;
 841                 }
 842                 break;
 843         case VIFF_TUNNEL:
 844                 dev = ipmr_new_tunnel(net, vifc);
 845                 if (IS_ERR(dev))
 846                         return PTR_ERR(dev);
 847                 break;
 848         case VIFF_USE_IFINDEX:
 849         case 0:
 850                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 851                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 852                         if (dev && !__in_dev_get_rtnl(dev)) {
 853                                 dev_put(dev);
 854                                 return -EADDRNOTAVAIL;
 855                         }
 856                 } else {
 857                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 858                 }
 859                 if (!dev)
 860                         return -EADDRNOTAVAIL;
 861                 err = dev_set_allmulti(dev, 1);
 862                 if (err) {
 863                         dev_put(dev);
 864                         return err;
 865                 }
 866                 break;
 867         default:
 868                 return -EINVAL;
 869         }
 870
 871         in_dev = __in_dev_get_rtnl(dev);
 872         if (!in_dev) {
 873                 dev_put(dev);
 874                 return -EADDRNOTAVAIL;
 875         }
 876         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 877         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 878                                     dev->ifindex, &in_dev->cnf);
 879         ip_rt_multicast_event(in_dev);
 880
 881         /* Fill in the VIF structures */
 882         vif_device_init(v, dev, vifc->vifc_rate_limit,
 883                         vifc->vifc_threshold,
 884                         vifc->vifc_flags | (!mrtsock ? VIFF_STATIC : 0),
 885                         (VIFF_TUNNEL | VIFF_REGISTER));
 886
 887         err = dev_get_port_parent_id(dev, &ppid, true);
 888         if (err == 0) {
 889                 memcpy(v->dev_parent_id.id, ppid.id, ppid.id_len);
 890                 v->dev_parent_id.id_len = ppid.id_len;
 891         } else {
 892                 v->dev_parent_id.id_len = 0;
 893         }
 894
 895         v->local = vifc->vifc_lcl_addr.s_addr;
 896         v->remote = vifc->vifc_rmt_addr.s_addr;
 897
 898         /* And finish update writing critical data */
 899         spin_lock(&mrt_lock);
 900         rcu_assign_pointer(v->dev, dev);
 901         netdev_tracker_alloc(dev, &v->dev_tracker, GFP_ATOMIC);
 902         if (v->flags & VIFF_REGISTER) {
 903                 /* Pairs with READ_ONCE() in ipmr_cache_report() and reg_vif_xmit() */
 904                 WRITE_ONCE(mrt->mroute_reg_vif_num, vifi);
 905         }
 906         if (vifi+1 > mrt->maxvif)
 907                 WRITE_ONCE(mrt->maxvif, vifi + 1);
 908         spin_unlock(&mrt_lock);
 909         call_ipmr_vif_entry_notifiers(net, FIB_EVENT_VIF_ADD, v, dev,
 910                                       vifi, mrt->id);
 911         return 0;
 912 }
 913
 914 /* called with rcu_read_lock() */
 915 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 916                                          __be32 origin,
 917                                          __be32 mcastgrp)
 918 {
 919         struct mfc_cache_cmp_arg arg = {
 920                         .mfc_mcastgrp = mcastgrp,
 921                         .mfc_origin = origin
 922         };
 923
 924         return mr_mfc_find(mrt, &arg);
 925 }
 926
 927 /* Look for a (*,G) entry */
 928 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 929                                              __be32 mcastgrp, int vifi)
 930 {
 931         struct mfc_cache_cmp_arg arg = {
 932                         .mfc_mcastgrp = mcastgrp,
 933                         .mfc_origin = htonl(INADDR_ANY)
 934         };
 935
 936         if (mcastgrp == htonl(INADDR_ANY))
 937                 return mr_mfc_find_any_parent(mrt, vifi);
 938         return mr_mfc_find_any(mrt, vifi, &arg);
 939 }
 940
 941 /* Look for a (S,G,iif) entry if parent != -1 */
 942 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 943                                                 __be32 origin, __be32 mcastgrp,
 944                                                 int parent)
 945 {
 946         struct mfc_cache_cmp_arg arg = {
 947                         .mfc_mcastgrp = mcastgrp,
 948                         .mfc_origin = origin,
 949         };
 950
 951         return mr_mfc_find_parent(mrt, &arg, parent);
 952 }
 953
 954 /* Allocate a multicast cache entry */
 955 static struct mfc_cache *ipmr_cache_alloc(void)
 956 {
 957         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 958
 959         if (c) {
 960                 c->_c.mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 961                 c->_c.mfc_un.res.minvif = MAXVIFS;
 962                 c->_c.free = ipmr_cache_free_rcu;
 963                 refcount_set(&c->_c.mfc_un.res.refcount, 1);
 964         }
 965         return c;
 966 }
 967
 968 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 969 {
 970         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 971
 972         if (c) {
 973                 skb_queue_head_init(&c->_c.mfc_un.unres.unresolved);
 974                 c->_c.mfc_un.unres.expires = jiffies + 10 * HZ;
 975         }
 976         return c;
 977 }
 978
 979 /* A cache entry has gone into a resolved state from queued */
 980 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 981                                struct mfc_cache *uc, struct mfc_cache *c)
 982 {
 983         struct sk_buff *skb;
 984         struct nlmsgerr *e;
 985
 986         /* Play the pending entries through our router */
 987         while ((skb = __skb_dequeue(&uc->_c.mfc_un.unres.unresolved))) {
 988                 if (ip_hdr(skb)->version == 0) {
 989                         struct nlmsghdr *nlh = skb_pull(skb,
 990                                                         sizeof(struct iphdr));
 991
 992                         if (mr_fill_mroute(mrt, skb, &c->_c,
 993                                            nlmsg_data(nlh)) > 0) {
 994                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 995                                                  (u8 *)nlh;
 996                         } else {
 997                                 nlh->nlmsg_type = NLMSG_ERROR;
 998                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 999                                 skb_trim(skb, nlh->nlmsg_len);
1000                                 e = nlmsg_data(nlh);
1001                                 e->error = -EMSGSIZE;
1002                                 memset(&e->msg, 0, sizeof(e->msg));
1003                         }
1004
1005                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1006                 } else {
1007                         rcu_read_lock();
1008                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1009                         rcu_read_unlock();
1010                 }
1011         }
1012 }
1013
1014 /* Bounce a cache query up to mrouted and netlink.
1015  *
1016  * Called under rcu_read_lock().
1017  */
1018 static int ipmr_cache_report(const struct mr_table *mrt,
1019                              struct sk_buff *pkt, vifi_t vifi, int assert)
1020 {
1021         const int ihl = ip_hdrlen(pkt);
1022         struct sock *mroute_sk;
1023         struct igmphdr *igmp;
1024         struct igmpmsg *msg;
1025         struct sk_buff *skb;
1026         int ret;
1027
1028         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1029                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1030         else
1031                 skb = alloc_skb(128, GFP_ATOMIC);
1032
1033         if (!skb)
1034                 return -ENOBUFS;
1035
1036         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1037                 /* Ugly, but we have no choice with this interface.
1038                  * Duplicate old header, fix ihl, length etc.
1039                  * And all this only to mangle msg->im_msgtype and
1040                  * to set msg->im_mbz to "mbz" :-)
1041                  */
1042                 skb_push(skb, sizeof(struct iphdr));
1043                 skb_reset_network_header(skb);
1044                 skb_reset_transport_header(skb);
1045                 msg = (struct igmpmsg *)skb_network_header(skb);
1046                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1047                 msg->im_msgtype = assert;
1048                 msg->im_mbz = 0;
1049                 if (assert == IGMPMSG_WRVIFWHOLE) {
1050                         msg->im_vif = vifi;
1051                         msg->im_vif_hi = vifi >> 8;
1052                 } else {
1053                         /* Pairs with WRITE_ONCE() in vif_add() and vif_delete() */
1054                         int vif_num = READ_ONCE(mrt->mroute_reg_vif_num);
1055
1056                         msg->im_vif = vif_num;
1057                         msg->im_vif_hi = vif_num >> 8;
1058                 }
1059                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1060                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1061                                              sizeof(struct iphdr));
1062         } else {
1063                 /* Copy the IP header */
1064                 skb_set_network_header(skb, skb->len);
1065                 skb_put(skb, ihl);
1066                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1067                 /* Flag to the kernel this is a route add */
1068                 ip_hdr(skb)->protocol = 0;
1069                 msg = (struct igmpmsg *)skb_network_header(skb);
1070                 msg->im_vif = vifi;
1071                 msg->im_vif_hi = vifi >> 8;
1072                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1073                 /* Add our header */
1074                 igmp = skb_put(skb, sizeof(struct igmphdr));
1075                 igmp->type = assert;
1076                 msg->im_msgtype = assert;
1077                 igmp->code = 0;
1078                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1079                 skb->transport_header = skb->network_header;
1080         }
1081
1082         mroute_sk = rcu_dereference(mrt->mroute_sk);
1083         if (!mroute_sk) {
1084                 kfree_skb(skb);
1085                 return -EINVAL;
1086         }
1087
1088         igmpmsg_netlink_event(mrt, skb);
1089
1090         /* Deliver to mrouted */
1091         ret = sock_queue_rcv_skb(mroute_sk, skb);
1092
1093         if (ret < 0) {
1094                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1095                 kfree_skb(skb);
1096         }
1097
1098         return ret;
1099 }
1100
1101 /* Queue a packet for resolution. It gets locked cache entry! */
1102 /* Called under rcu_read_lock() */
1103 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1104                                  struct sk_buff *skb, struct net_device *dev)
1105 {
1106         const struct iphdr *iph = ip_hdr(skb);
1107         struct mfc_cache *c;
1108         bool found = false;
1109         int err;
1110
1111         spin_lock_bh(&mfc_unres_lock);
1112         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1113                 if (c->mfc_mcastgrp == iph->daddr &&
1114                     c->mfc_origin == iph->saddr) {
1115                         found = true;
1116                         break;
1117                 }
1118         }
1119
1120         if (!found) {
1121                 /* Create a new entry if allowable */
1122                 c = ipmr_cache_alloc_unres();
1123                 if (!c) {
1124                         spin_unlock_bh(&mfc_unres_lock);
1125
1126                         kfree_skb(skb);
1127                         return -ENOBUFS;
1128                 }
1129
1130                 /* Fill in the new cache entry */
1131                 c->_c.mfc_parent = -1;
1132                 c->mfc_origin   = iph->saddr;
1133                 c->mfc_mcastgrp = iph->daddr;
1134
1135                 /* Reflect first query at mrouted. */
1136                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1137
1138                 if (err < 0) {
1139                         /* If the report failed throw the cache entry
1140                            out - Brad Parker
1141                          */
1142                         spin_unlock_bh(&mfc_unres_lock);
1143
1144                         ipmr_cache_free(c);
1145                         kfree_skb(skb);
1146                         return err;
1147                 }
1148
1149                 atomic_inc(&mrt->cache_resolve_queue_len);
1150                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1151                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1152
1153                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1154                         mod_timer(&mrt->ipmr_expire_timer,
1155                                   c->_c.mfc_un.unres.expires);
1156         }
1157
1158         /* See if we can append the packet */
1159         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1160                 kfree_skb(skb);
1161                 err = -ENOBUFS;
1162         } else {
1163                 if (dev) {
1164                         skb->dev = dev;
1165                         skb->skb_iif = dev->ifindex;
1166                 }
1167                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1168                 err = 0;
1169         }
1170
1171         spin_unlock_bh(&mfc_unres_lock);
1172         return err;
1173 }
1174
1175 /* MFC cache manipulation by user space mroute daemon */
1176
1177 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1178 {
1179         struct net *net = read_pnet(&mrt->net);
1180         struct mfc_cache *c;
1181
1182         /* The entries are added/deleted only under RTNL */
1183         rcu_read_lock();
1184         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1185                                    mfc->mfcc_mcastgrp.s_addr, parent);
1186         rcu_read_unlock();
1187         if (!c)
1188                 return -ENOENT;
1189         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1190         list_del_rcu(&c->_c.list);
1191         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1192         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1193         mr_cache_put(&c->_c);
1194
1195         return 0;
1196 }
1197
1198 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1199                         struct mfcctl *mfc, int mrtsock, int parent)
1200 {
1201         struct mfc_cache *uc, *c;
1202         struct mr_mfc *_uc;
1203         bool found;
1204         int ret;
1205
1206         if (mfc->mfcc_parent >= MAXVIFS)
1207                 return -ENFILE;
1208
1209         /* The entries are added/deleted only under RTNL */
1210         rcu_read_lock();
1211         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1212                                    mfc->mfcc_mcastgrp.s_addr, parent);
1213         rcu_read_unlock();
1214         if (c) {
1215                 spin_lock(&mrt_lock);
1216                 c->_c.mfc_parent = mfc->mfcc_parent;
1217                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1218                 if (!mrtsock)
1219                         c->_c.mfc_flags |= MFC_STATIC;
1220                 spin_unlock(&mrt_lock);
1221                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1222                                               mrt->id);
1223                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1224                 return 0;
1225         }
1226
1227         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1228             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1229                 return -EINVAL;
1230
1231         c = ipmr_cache_alloc();
1232         if (!c)
1233                 return -ENOMEM;
1234
1235         c->mfc_origin = mfc->mfcc_origin.s_addr;
1236         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1237         c->_c.mfc_parent = mfc->mfcc_parent;
1238         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1239         if (!mrtsock)
1240                 c->_c.mfc_flags |= MFC_STATIC;
1241
1242         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1243                                   ipmr_rht_params);
1244         if (ret) {
1245                 pr_err("ipmr: rhtable insert error %d\n", ret);
1246                 ipmr_cache_free(c);
1247                 return ret;
1248         }
1249         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1250         /* Check to see if we resolved a queued list. If so we
1251          * need to send on the frames and tidy up.
1252          */
1253         found = false;
1254         spin_lock_bh(&mfc_unres_lock);
1255         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1256                 uc = (struct mfc_cache *)_uc;
1257                 if (uc->mfc_origin == c->mfc_origin &&
1258                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1259                         list_del(&_uc->list);
1260                         atomic_dec(&mrt->cache_resolve_queue_len);
1261                         found = true;
1262                         break;
1263                 }
1264         }
1265         if (list_empty(&mrt->mfc_unres_queue))
1266                 del_timer(&mrt->ipmr_expire_timer);
1267         spin_unlock_bh(&mfc_unres_lock);
1268
1269         if (found) {
1270                 ipmr_cache_resolve(net, mrt, uc, c);
1271                 ipmr_cache_free(uc);
1272         }
1273         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1274         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1275         return 0;
1276 }
1277
1278 /* Close the multicast socket, and clear the vif tables etc */
1279 static void mroute_clean_tables(struct mr_table *mrt, int flags)
1280 {
1281         struct net *net = read_pnet(&mrt->net);
1282         struct mr_mfc *c, *tmp;
1283         struct mfc_cache *cache;
1284         LIST_HEAD(list);
1285         int i;
1286
1287         /* Shut down all active vif entries */
1288         if (flags & (MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC)) {
1289                 for (i = 0; i < mrt->maxvif; i++) {
1290                         if (((mrt->vif_table[i].flags & VIFF_STATIC) &&
1291                              !(flags & MRT_FLUSH_VIFS_STATIC)) ||
1292                             (!(mrt->vif_table[i].flags & VIFF_STATIC) && !(flags & MRT_FLUSH_VIFS)))
1293                                 continue;
1294                         vif_delete(mrt, i, 0, &list);
1295                 }
1296                 unregister_netdevice_many(&list);
1297         }
1298
1299         /* Wipe the cache */
1300         if (flags & (MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC)) {
1301                 list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1302                         if (((c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC_STATIC)) ||
1303                             (!(c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC)))
1304                                 continue;
1305                         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1306                         list_del_rcu(&c->list);
1307                         cache = (struct mfc_cache *)c;
1308                         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1309                                                       mrt->id);
1310                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1311                         mr_cache_put(c);
1312                 }
1313         }
1314
1315         if (flags & MRT_FLUSH_MFC) {
1316                 if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1317                         spin_lock_bh(&mfc_unres_lock);
1318                         list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1319                                 list_del(&c->list);
1320                                 cache = (struct mfc_cache *)c;
1321                                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1322                                 ipmr_destroy_unres(mrt, cache);
1323                         }
1324                         spin_unlock_bh(&mfc_unres_lock);
1325                 }
1326         }
1327 }
1328
1329 /* called from ip_ra_control(), before an RCU grace period,
1330  * we don't need to call synchronize_rcu() here
1331  */
1332 static void mrtsock_destruct(struct sock *sk)
1333 {
1334         struct net *net = sock_net(sk);
1335         struct mr_table *mrt;
1336
1337         rtnl_lock();
1338         ipmr_for_each_table(mrt, net) {
1339                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1340                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1341                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1342                                                     NETCONFA_MC_FORWARDING,
1343                                                     NETCONFA_IFINDEX_ALL,
1344                                                     net->ipv4.devconf_all);
1345                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1346                         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_MFC);
1347                 }
1348         }
1349         rtnl_unlock();
1350 }
1351
1352 /* Socket options and virtual interface manipulation. The whole
1353  * virtual interface system is a complete heap, but unfortunately
1354  * that's how BSD mrouted happens to think. Maybe one day with a proper
1355  * MOSPF/PIM router set up we can clean this up.
1356  */
1357
1358 int ip_mroute_setsockopt(struct sock *sk, int optname, sockptr_t optval,
1359                          unsigned int optlen)
1360 {
1361         struct net *net = sock_net(sk);
1362         int val, ret = 0, parent = 0;
1363         struct mr_table *mrt;
1364         struct vifctl vif;
1365         struct mfcctl mfc;
1366         bool do_wrvifwhole;
1367         u32 uval;
1368
1369         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1370         rtnl_lock();
1371         if (sk->sk_type != SOCK_RAW ||
1372             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1373                 ret = -EOPNOTSUPP;
1374                 goto out_unlock;
1375         }
1376
1377         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1378         if (!mrt) {
1379                 ret = -ENOENT;
1380                 goto out_unlock;
1381         }
1382         if (optname != MRT_INIT) {
1383                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1384                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1385                         ret = -EACCES;
1386                         goto out_unlock;
1387                 }
1388         }
1389
1390         switch (optname) {
1391         case MRT_INIT:
1392                 if (optlen != sizeof(int)) {
1393                         ret = -EINVAL;
1394                         break;
1395                 }
1396                 if (rtnl_dereference(mrt->mroute_sk)) {
1397                         ret = -EADDRINUSE;
1398                         break;
1399                 }
1400
1401                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1402                 if (ret == 0) {
1403                         rcu_assign_pointer(mrt->mroute_sk, sk);
1404                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1405                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1406                                                     NETCONFA_MC_FORWARDING,
1407                                                     NETCONFA_IFINDEX_ALL,
1408                                                     net->ipv4.devconf_all);
1409                 }
1410                 break;
1411         case MRT_DONE:
1412                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1413                         ret = -EACCES;
1414                 } else {
1415                         /* We need to unlock here because mrtsock_destruct takes
1416                          * care of rtnl itself and we can't change that due to
1417                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1418                          */
1419                         rtnl_unlock();
1420                         ret = ip_ra_control(sk, 0, NULL);
1421                         goto out;
1422                 }
1423                 break;
1424         case MRT_ADD_VIF:
1425         case MRT_DEL_VIF:
1426                 if (optlen != sizeof(vif)) {
1427                         ret = -EINVAL;
1428                         break;
1429                 }
1430                 if (copy_from_sockptr(&vif, optval, sizeof(vif))) {
1431                         ret = -EFAULT;
1432                         break;
1433                 }
1434                 if (vif.vifc_vifi >= MAXVIFS) {
1435                         ret = -ENFILE;
1436                         break;
1437                 }
1438                 if (optname == MRT_ADD_VIF) {
1439                         ret = vif_add(net, mrt, &vif,
1440                                       sk == rtnl_dereference(mrt->mroute_sk));
1441                 } else {
1442                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1443                 }
1444                 break;
1445         /* Manipulate the forwarding caches. These live
1446          * in a sort of kernel/user symbiosis.
1447          */
1448         case MRT_ADD_MFC:
1449         case MRT_DEL_MFC:
1450                 parent = -1;
1451                 fallthrough;
1452         case MRT_ADD_MFC_PROXY:
1453         case MRT_DEL_MFC_PROXY:
1454                 if (optlen != sizeof(mfc)) {
1455                         ret = -EINVAL;
1456                         break;
1457                 }
1458                 if (copy_from_sockptr(&mfc, optval, sizeof(mfc))) {
1459                         ret = -EFAULT;
1460                         break;
1461                 }
1462                 if (parent == 0)
1463                         parent = mfc.mfcc_parent;
1464                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1465                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1466                 else
1467                         ret = ipmr_mfc_add(net, mrt, &mfc,
1468                                            sk == rtnl_dereference(mrt->mroute_sk),
1469                                            parent);
1470                 break;
1471         case MRT_FLUSH:
1472                 if (optlen != sizeof(val)) {
1473                         ret = -EINVAL;
1474                         break;
1475                 }
1476                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1477                         ret = -EFAULT;
1478                         break;
1479                 }
1480                 mroute_clean_tables(mrt, val);
1481                 break;
1482         /* Control PIM assert. */
1483         case MRT_ASSERT:
1484                 if (optlen != sizeof(val)) {
1485                         ret = -EINVAL;
1486                         break;
1487                 }
1488                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1489                         ret = -EFAULT;
1490                         break;
1491                 }
1492                 mrt->mroute_do_assert = val;
1493                 break;
1494         case MRT_PIM:
1495                 if (!ipmr_pimsm_enabled()) {
1496                         ret = -ENOPROTOOPT;
1497                         break;
1498                 }
1499                 if (optlen != sizeof(val)) {
1500                         ret = -EINVAL;
1501                         break;
1502                 }
1503                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1504                         ret = -EFAULT;
1505                         break;
1506                 }
1507
1508                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1509                 val = !!val;
1510                 if (val != mrt->mroute_do_pim) {
1511                         mrt->mroute_do_pim = val;
1512                         mrt->mroute_do_assert = val;
1513                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1514                 }
1515                 break;
1516         case MRT_TABLE:
1517                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1518                         ret = -ENOPROTOOPT;
1519                         break;
1520                 }
1521                 if (optlen != sizeof(uval)) {
1522                         ret = -EINVAL;
1523                         break;
1524                 }
1525                 if (copy_from_sockptr(&uval, optval, sizeof(uval))) {
1526                         ret = -EFAULT;
1527                         break;
1528                 }
1529
1530                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1531                         ret = -EBUSY;
1532                 } else {
1533                         mrt = ipmr_new_table(net, uval);
1534                         if (IS_ERR(mrt))
1535                                 ret = PTR_ERR(mrt);
1536                         else
1537                                 raw_sk(sk)->ipmr_table = uval;
1538                 }
1539                 break;
1540         /* Spurious command, or MRT_VERSION which you cannot set. */
1541         default:
1542                 ret = -ENOPROTOOPT;
1543         }
1544 out_unlock:
1545         rtnl_unlock();
1546 out:
1547         return ret;
1548 }
1549
1550 /* Getsock opt support for the multicast routing system. */
1551 int ip_mroute_getsockopt(struct sock *sk, int optname, sockptr_t optval,
1552                          sockptr_t optlen)
1553 {
1554         int olr;
1555         int val;
1556         struct net *net = sock_net(sk);
1557         struct mr_table *mrt;
1558
1559         if (sk->sk_type != SOCK_RAW ||
1560             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1561                 return -EOPNOTSUPP;
1562
1563         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1564         if (!mrt)
1565                 return -ENOENT;
1566
1567         switch (optname) {
1568         case MRT_VERSION:
1569                 val = 0x0305;
1570                 break;
1571         case MRT_PIM:
1572                 if (!ipmr_pimsm_enabled())
1573                         return -ENOPROTOOPT;
1574                 val = mrt->mroute_do_pim;
1575                 break;
1576         case MRT_ASSERT:
1577                 val = mrt->mroute_do_assert;
1578                 break;
1579         default:
1580                 return -ENOPROTOOPT;
1581         }
1582
1583         if (copy_from_sockptr(&olr, optlen, sizeof(int)))
1584                 return -EFAULT;
1585         olr = min_t(unsigned int, olr, sizeof(int));
1586         if (olr < 0)
1587                 return -EINVAL;
1588         if (copy_to_sockptr(optlen, &olr, sizeof(int)))
1589                 return -EFAULT;
1590         if (copy_to_sockptr(optval, &val, olr))
1591                 return -EFAULT;
1592         return 0;
1593 }
1594
1595 /* The IP multicast ioctl support routines. */
1596 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1597 {
1598         struct sioc_sg_req sr;
1599         struct sioc_vif_req vr;
1600         struct vif_device *vif;
1601         struct mfc_cache *c;
1602         struct net *net = sock_net(sk);
1603         struct mr_table *mrt;
1604
1605         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1606         if (!mrt)
1607                 return -ENOENT;
1608
1609         switch (cmd) {
1610         case SIOCGETVIFCNT:
1611                 if (copy_from_user(&vr, arg, sizeof(vr)))
1612                         return -EFAULT;
1613                 if (vr.vifi >= mrt->maxvif)
1614                         return -EINVAL;
1615                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1616                 rcu_read_lock();
1617                 vif = &mrt->vif_table[vr.vifi];
1618                 if (VIF_EXISTS(mrt, vr.vifi)) {
1619                         vr.icount = READ_ONCE(vif->pkt_in);
1620                         vr.ocount = READ_ONCE(vif->pkt_out);
1621                         vr.ibytes = READ_ONCE(vif->bytes_in);
1622                         vr.obytes = READ_ONCE(vif->bytes_out);
1623                         rcu_read_unlock();
1624
1625                         if (copy_to_user(arg, &vr, sizeof(vr)))
1626                                 return -EFAULT;
1627                         return 0;
1628                 }
1629                 rcu_read_unlock();
1630                 return -EADDRNOTAVAIL;
1631         case SIOCGETSGCNT:
1632                 if (copy_from_user(&sr, arg, sizeof(sr)))
1633                         return -EFAULT;
1634
1635                 rcu_read_lock();
1636                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1637                 if (c) {
1638                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1639                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1640                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1641                         rcu_read_unlock();
1642
1643                         if (copy_to_user(arg, &sr, sizeof(sr)))
1644                                 return -EFAULT;
1645                         return 0;
1646                 }
1647                 rcu_read_unlock();
1648                 return -EADDRNOTAVAIL;
1649         default:
1650                 return -ENOIOCTLCMD;
1651         }
1652 }
1653
1654 #ifdef CONFIG_COMPAT
1655 struct compat_sioc_sg_req {
1656         struct in_addr src;
1657         struct in_addr grp;
1658         compat_ulong_t pktcnt;
1659         compat_ulong_t bytecnt;
1660         compat_ulong_t wrong_if;
1661 };
1662
1663 struct compat_sioc_vif_req {
1664         vifi_t  vifi;           /* Which iface */
1665         compat_ulong_t icount;
1666         compat_ulong_t ocount;
1667         compat_ulong_t ibytes;
1668         compat_ulong_t obytes;
1669 };
1670
1671 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1672 {
1673         struct compat_sioc_sg_req sr;
1674         struct compat_sioc_vif_req vr;
1675         struct vif_device *vif;
1676         struct mfc_cache *c;
1677         struct net *net = sock_net(sk);
1678         struct mr_table *mrt;
1679
1680         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1681         if (!mrt)
1682                 return -ENOENT;
1683
1684         switch (cmd) {
1685         case SIOCGETVIFCNT:
1686                 if (copy_from_user(&vr, arg, sizeof(vr)))
1687                         return -EFAULT;
1688                 if (vr.vifi >= mrt->maxvif)
1689                         return -EINVAL;
1690                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1691                 rcu_read_lock();
1692                 vif = &mrt->vif_table[vr.vifi];
1693                 if (VIF_EXISTS(mrt, vr.vifi)) {
1694                         vr.icount = READ_ONCE(vif->pkt_in);
1695                         vr.ocount = READ_ONCE(vif->pkt_out);
1696                         vr.ibytes = READ_ONCE(vif->bytes_in);
1697                         vr.obytes = READ_ONCE(vif->bytes_out);
1698                         rcu_read_unlock();
1699
1700                         if (copy_to_user(arg, &vr, sizeof(vr)))
1701                                 return -EFAULT;
1702                         return 0;
1703                 }
1704                 rcu_read_unlock();
1705                 return -EADDRNOTAVAIL;
1706         case SIOCGETSGCNT:
1707                 if (copy_from_user(&sr, arg, sizeof(sr)))
1708                         return -EFAULT;
1709
1710                 rcu_read_lock();
1711                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1712                 if (c) {
1713                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1714                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1715                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1716                         rcu_read_unlock();
1717
1718                         if (copy_to_user(arg, &sr, sizeof(sr)))
1719                                 return -EFAULT;
1720                         return 0;
1721                 }
1722                 rcu_read_unlock();
1723                 return -EADDRNOTAVAIL;
1724         default:
1725                 return -ENOIOCTLCMD;
1726         }
1727 }
1728 #endif
1729
1730 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1731 {
1732         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1733         struct net *net = dev_net(dev);
1734         struct mr_table *mrt;
1735         struct vif_device *v;
1736         int ct;
1737
1738         if (event != NETDEV_UNREGISTER)
1739                 return NOTIFY_DONE;
1740
1741         ipmr_for_each_table(mrt, net) {
1742                 v = &mrt->vif_table[0];
1743                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1744                         if (rcu_access_pointer(v->dev) == dev)
1745                                 vif_delete(mrt, ct, 1, NULL);
1746                 }
1747         }
1748         return NOTIFY_DONE;
1749 }
1750
1751 static struct notifier_block ip_mr_notifier = {
1752         .notifier_call = ipmr_device_event,
1753 };
1754
1755 /* Encapsulate a packet by attaching a valid IPIP header to it.
1756  * This avoids tunnel drivers and other mess and gives us the speed so
1757  * important for multicast video.
1758  */
1759 static void ip_encap(struct net *net, struct sk_buff *skb,
1760                      __be32 saddr, __be32 daddr)
1761 {
1762         struct iphdr *iph;
1763         const struct iphdr *old_iph = ip_hdr(skb);
1764
1765         skb_push(skb, sizeof(struct iphdr));
1766         skb->transport_header = skb->network_header;
1767         skb_reset_network_header(skb);
1768         iph = ip_hdr(skb);
1769
1770         iph->version    =       4;
1771         iph->tos        =       old_iph->tos;
1772         iph->ttl        =       old_iph->ttl;
1773         iph->frag_off   =       0;
1774         iph->daddr      =       daddr;
1775         iph->saddr      =       saddr;
1776         iph->protocol   =       IPPROTO_IPIP;
1777         iph->ihl        =       5;
1778         iph->tot_len    =       htons(skb->len);
1779         ip_select_ident(net, skb, NULL);
1780         ip_send_check(iph);
1781
1782         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1783         nf_reset_ct(skb);
1784 }
1785
1786 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1787                                       struct sk_buff *skb)
1788 {
1789         struct ip_options *opt = &(IPCB(skb)->opt);
1790
1791         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1792         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1793
1794         if (unlikely(opt->optlen))
1795                 ip_forward_options(skb);
1796
1797         return dst_output(net, sk, skb);
1798 }
1799
1800 #ifdef CONFIG_NET_SWITCHDEV
1801 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1802                                    int in_vifi, int out_vifi)
1803 {
1804         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1805         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1806
1807         if (!skb->offload_l3_fwd_mark)
1808                 return false;
1809         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1810                 return false;
1811         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1812                                         &in_vif->dev_parent_id);
1813 }
1814 #else
1815 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1816                                    int in_vifi, int out_vifi)
1817 {
1818         return false;
1819 }
1820 #endif
1821
1822 /* Processing handlers for ipmr_forward, under rcu_read_lock() */
1823
1824 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1825                             int in_vifi, struct sk_buff *skb, int vifi)
1826 {
1827         const struct iphdr *iph = ip_hdr(skb);
1828         struct vif_device *vif = &mrt->vif_table[vifi];
1829         struct net_device *vif_dev;
1830         struct net_device *dev;
1831         struct rtable *rt;
1832         struct flowi4 fl4;
1833         int    encap = 0;
1834
1835         vif_dev = vif_dev_read(vif);
1836         if (!vif_dev)
1837                 goto out_free;
1838
1839         if (vif->flags & VIFF_REGISTER) {
1840                 WRITE_ONCE(vif->pkt_out, vif->pkt_out + 1);
1841                 WRITE_ONCE(vif->bytes_out, vif->bytes_out + skb->len);
1842                 DEV_STATS_ADD(vif_dev, tx_bytes, skb->len);
1843                 DEV_STATS_INC(vif_dev, tx_packets);
1844                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1845                 goto out_free;
1846         }
1847
1848         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1849                 goto out_free;
1850
1851         if (vif->flags & VIFF_TUNNEL) {
1852                 rt = ip_route_output_ports(net, &fl4, NULL,
1853                                            vif->remote, vif->local,
1854                                            0, 0,
1855                                            IPPROTO_IPIP,
1856                                            RT_TOS(iph->tos), vif->link);
1857                 if (IS_ERR(rt))
1858                         goto out_free;
1859                 encap = sizeof(struct iphdr);
1860         } else {
1861                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1862                                            0, 0,
1863                                            IPPROTO_IPIP,
1864                                            RT_TOS(iph->tos), vif->link);
1865                 if (IS_ERR(rt))
1866                         goto out_free;
1867         }
1868
1869         dev = rt->dst.dev;
1870
1871         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1872                 /* Do not fragment multicasts. Alas, IPv4 does not
1873                  * allow to send ICMP, so that packets will disappear
1874                  * to blackhole.
1875                  */
1876                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1877                 ip_rt_put(rt);
1878                 goto out_free;
1879         }
1880
1881         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1882
1883         if (skb_cow(skb, encap)) {
1884                 ip_rt_put(rt);
1885                 goto out_free;
1886         }
1887
1888         WRITE_ONCE(vif->pkt_out, vif->pkt_out + 1);
1889         WRITE_ONCE(vif->bytes_out, vif->bytes_out + skb->len);
1890
1891         skb_dst_drop(skb);
1892         skb_dst_set(skb, &rt->dst);
1893         ip_decrease_ttl(ip_hdr(skb));
1894
1895         /* FIXME: forward and output firewalls used to be called here.
1896          * What do we do with netfilter? -- RR
1897          */
1898         if (vif->flags & VIFF_TUNNEL) {
1899                 ip_encap(net, skb, vif->local, vif->remote);
1900                 /* FIXME: extra output firewall step used to be here. --RR */
1901                 DEV_STATS_INC(vif_dev, tx_packets);
1902                 DEV_STATS_ADD(vif_dev, tx_bytes, skb->len);
1903         }
1904
1905         IPCB(skb)->flags |= IPSKB_FORWARDED;
1906
1907         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1908          * not only before forwarding, but after forwarding on all output
1909          * interfaces. It is clear, if mrouter runs a multicasting
1910          * program, it should receive packets not depending to what interface
1911          * program is joined.
1912          * If we will not make it, the program will have to join on all
1913          * interfaces. On the other hand, multihoming host (or router, but
1914          * not mrouter) cannot join to more than one interface - it will
1915          * result in receiving multiple packets.
1916          */
1917         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1918                 net, NULL, skb, skb->dev, dev,
1919                 ipmr_forward_finish);
1920         return;
1921
1922 out_free:
1923         kfree_skb(skb);
1924 }
1925
1926 /* Called with mrt_lock or rcu_read_lock() */
1927 static int ipmr_find_vif(const struct mr_table *mrt, struct net_device *dev)
1928 {
1929         int ct;
1930         /* Pairs with WRITE_ONCE() in vif_delete()/vif_add() */
1931         for (ct = READ_ONCE(mrt->maxvif) - 1; ct >= 0; ct--) {
1932                 if (rcu_access_pointer(mrt->vif_table[ct].dev) == dev)
1933                         break;
1934         }
1935         return ct;
1936 }
1937
1938 /* "local" means that we should preserve one skb (for local delivery) */
1939 /* Called uner rcu_read_lock() */
1940 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1941                           struct net_device *dev, struct sk_buff *skb,
1942                           struct mfc_cache *c, int local)
1943 {
1944         int true_vifi = ipmr_find_vif(mrt, dev);
1945         int psend = -1;
1946         int vif, ct;
1947
1948         vif = c->_c.mfc_parent;
1949         c->_c.mfc_un.res.pkt++;
1950         c->_c.mfc_un.res.bytes += skb->len;
1951         c->_c.mfc_un.res.lastuse = jiffies;
1952
1953         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1954                 struct mfc_cache *cache_proxy;
1955
1956                 /* For an (*,G) entry, we only check that the incoming
1957                  * interface is part of the static tree.
1958                  */
1959                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1960                 if (cache_proxy &&
1961                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1962                         goto forward;
1963         }
1964
1965         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1966         if (rcu_access_pointer(mrt->vif_table[vif].dev) != dev) {
1967                 if (rt_is_output_route(skb_rtable(skb))) {
1968                         /* It is our own packet, looped back.
1969                          * Very complicated situation...
1970                          *
1971                          * The best workaround until routing daemons will be
1972                          * fixed is not to redistribute packet, if it was
1973                          * send through wrong interface. It means, that
1974                          * multicast applications WILL NOT work for
1975                          * (S,G), which have default multicast route pointing
1976                          * to wrong oif. In any case, it is not a good
1977                          * idea to use multicasting applications on router.
1978                          */
1979                         goto dont_forward;
1980                 }
1981
1982                 c->_c.mfc_un.res.wrong_if++;
1983
1984                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1985                     /* pimsm uses asserts, when switching from RPT to SPT,
1986                      * so that we cannot check that packet arrived on an oif.
1987                      * It is bad, but otherwise we would need to move pretty
1988                      * large chunk of pimd to kernel. Ough... --ANK
1989                      */
1990                     (mrt->mroute_do_pim ||
1991                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1992                     time_after(jiffies,
1993                                c->_c.mfc_un.res.last_assert +
1994                                MFC_ASSERT_THRESH)) {
1995                         c->_c.mfc_un.res.last_assert = jiffies;
1996                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1997                         if (mrt->mroute_do_wrvifwhole)
1998                                 ipmr_cache_report(mrt, skb, true_vifi,
1999                                                   IGMPMSG_WRVIFWHOLE);
2000                 }
2001                 goto dont_forward;
2002         }
2003
2004 forward:
2005         WRITE_ONCE(mrt->vif_table[vif].pkt_in,
2006                    mrt->vif_table[vif].pkt_in + 1);
2007         WRITE_ONCE(mrt->vif_table[vif].bytes_in,
2008                    mrt->vif_table[vif].bytes_in + skb->len);
2009
2010         /* Forward the frame */
2011         if (c->mfc_origin == htonl(INADDR_ANY) &&
2012             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
2013                 if (true_vifi >= 0 &&
2014                     true_vifi != c->_c.mfc_parent &&
2015                     ip_hdr(skb)->ttl >
2016                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2017                         /* It's an (*,*) entry and the packet is not coming from
2018                          * the upstream: forward the packet to the upstream
2019                          * only.
2020                          */
2021                         psend = c->_c.mfc_parent;
2022                         goto last_forward;
2023                 }
2024                 goto dont_forward;
2025         }
2026         for (ct = c->_c.mfc_un.res.maxvif - 1;
2027              ct >= c->_c.mfc_un.res.minvif; ct--) {
2028                 /* For (*,G) entry, don't forward to the incoming interface */
2029                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2030                      ct != true_vifi) &&
2031                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2032                         if (psend != -1) {
2033                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2034
2035                                 if (skb2)
2036                                         ipmr_queue_xmit(net, mrt, true_vifi,
2037                                                         skb2, psend);
2038                         }
2039                         psend = ct;
2040                 }
2041         }
2042 last_forward:
2043         if (psend != -1) {
2044                 if (local) {
2045                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2046
2047                         if (skb2)
2048                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2049                                                 psend);
2050                 } else {
2051                         ipmr_queue_xmit(net, mrt, true_vifi, skb, psend);
2052                         return;
2053                 }
2054         }
2055
2056 dont_forward:
2057         if (!local)
2058                 kfree_skb(skb);
2059 }
2060
2061 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2062 {
2063         struct rtable *rt = skb_rtable(skb);
2064         struct iphdr *iph = ip_hdr(skb);
2065         struct flowi4 fl4 = {
2066                 .daddr = iph->daddr,
2067                 .saddr = iph->saddr,
2068                 .flowi4_tos = RT_TOS(iph->tos),
2069                 .flowi4_oif = (rt_is_output_route(rt) ?
2070                                skb->dev->ifindex : 0),
2071                 .flowi4_iif = (rt_is_output_route(rt) ?
2072                                LOOPBACK_IFINDEX :
2073                                skb->dev->ifindex),
2074                 .flowi4_mark = skb->mark,
2075         };
2076         struct mr_table *mrt;
2077         int err;
2078
2079         err = ipmr_fib_lookup(net, &fl4, &mrt);
2080         if (err)
2081                 return ERR_PTR(err);
2082         return mrt;
2083 }
2084
2085 /* Multicast packets for forwarding arrive here
2086  * Called with rcu_read_lock();
2087  */
2088 int ip_mr_input(struct sk_buff *skb)
2089 {
2090         struct mfc_cache *cache;
2091         struct net *net = dev_net(skb->dev);
2092         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2093         struct mr_table *mrt;
2094         struct net_device *dev;
2095
2096         /* skb->dev passed in is the loX master dev for vrfs.
2097          * As there are no vifs associated with loopback devices,
2098          * get the proper interface that does have a vif associated with it.
2099          */
2100         dev = skb->dev;
2101         if (netif_is_l3_master(skb->dev)) {
2102                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2103                 if (!dev) {
2104                         kfree_skb(skb);
2105                         return -ENODEV;
2106                 }
2107         }
2108
2109         /* Packet is looped back after forward, it should not be
2110          * forwarded second time, but still can be delivered locally.
2111          */
2112         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2113                 goto dont_forward;
2114
2115         mrt = ipmr_rt_fib_lookup(net, skb);
2116         if (IS_ERR(mrt)) {
2117                 kfree_skb(skb);
2118                 return PTR_ERR(mrt);
2119         }
2120         if (!local) {
2121                 if (IPCB(skb)->opt.router_alert) {
2122                         if (ip_call_ra_chain(skb))
2123                                 return 0;
2124                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2125                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2126                          * Cisco IOS <= 11.2(8)) do not put router alert
2127                          * option to IGMP packets destined to routable
2128                          * groups. It is very bad, because it means
2129                          * that we can forward NO IGMP messages.
2130                          */
2131                         struct sock *mroute_sk;
2132
2133                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2134                         if (mroute_sk) {
2135                                 nf_reset_ct(skb);
2136                                 raw_rcv(mroute_sk, skb);
2137                                 return 0;
2138                         }
2139                 }
2140         }
2141
2142         /* already under rcu_read_lock() */
2143         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2144         if (!cache) {
2145                 int vif = ipmr_find_vif(mrt, dev);
2146
2147                 if (vif >= 0)
2148                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2149                                                     vif);
2150         }
2151
2152         /* No usable cache entry */
2153         if (!cache) {
2154                 int vif;
2155
2156                 if (local) {
2157                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2158                         ip_local_deliver(skb);
2159                         if (!skb2)
2160                                 return -ENOBUFS;
2161                         skb = skb2;
2162                 }
2163
2164                 vif = ipmr_find_vif(mrt, dev);
2165                 if (vif >= 0)
2166                         return ipmr_cache_unresolved(mrt, vif, skb, dev);
2167                 kfree_skb(skb);
2168                 return -ENODEV;
2169         }
2170
2171         ip_mr_forward(net, mrt, dev, skb, cache, local);
2172
2173         if (local)
2174                 return ip_local_deliver(skb);
2175
2176         return 0;
2177
2178 dont_forward:
2179         if (local)
2180                 return ip_local_deliver(skb);
2181         kfree_skb(skb);
2182         return 0;
2183 }
2184
2185 #ifdef CONFIG_IP_PIMSM_V1
2186 /* Handle IGMP messages of PIMv1 */
2187 int pim_rcv_v1(struct sk_buff *skb)
2188 {
2189         struct igmphdr *pim;
2190         struct net *net = dev_net(skb->dev);
2191         struct mr_table *mrt;
2192
2193         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2194                 goto drop;
2195
2196         pim = igmp_hdr(skb);
2197
2198         mrt = ipmr_rt_fib_lookup(net, skb);
2199         if (IS_ERR(mrt))
2200                 goto drop;
2201         if (!mrt->mroute_do_pim ||
2202             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2203                 goto drop;
2204
2205         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2206 drop:
2207                 kfree_skb(skb);
2208         }
2209         return 0;
2210 }
2211 #endif
2212
2213 #ifdef CONFIG_IP_PIMSM_V2
2214 static int pim_rcv(struct sk_buff *skb)
2215 {
2216         struct pimreghdr *pim;
2217         struct net *net = dev_net(skb->dev);
2218         struct mr_table *mrt;
2219
2220         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2221                 goto drop;
2222
2223         pim = (struct pimreghdr *)skb_transport_header(skb);
2224         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2225             (pim->flags & PIM_NULL_REGISTER) ||
2226             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2227              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2228                 goto drop;
2229
2230         mrt = ipmr_rt_fib_lookup(net, skb);
2231         if (IS_ERR(mrt))
2232                 goto drop;
2233         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2234 drop:
2235                 kfree_skb(skb);
2236         }
2237         return 0;
2238 }
2239 #endif
2240
2241 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2242                    __be32 saddr, __be32 daddr,
2243                    struct rtmsg *rtm, u32 portid)
2244 {
2245         struct mfc_cache *cache;
2246         struct mr_table *mrt;
2247         int err;
2248
2249         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2250         if (!mrt)
2251                 return -ENOENT;
2252
2253         rcu_read_lock();
2254         cache = ipmr_cache_find(mrt, saddr, daddr);
2255         if (!cache && skb->dev) {
2256                 int vif = ipmr_find_vif(mrt, skb->dev);
2257
2258                 if (vif >= 0)
2259                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2260         }
2261         if (!cache) {
2262                 struct sk_buff *skb2;
2263                 struct iphdr *iph;
2264                 struct net_device *dev;
2265                 int vif = -1;
2266
2267                 dev = skb->dev;
2268                 if (dev)
2269                         vif = ipmr_find_vif(mrt, dev);
2270                 if (vif < 0) {
2271                         rcu_read_unlock();
2272                         return -ENODEV;
2273                 }
2274
2275                 skb2 = skb_realloc_headroom(skb, sizeof(struct iphdr));
2276                 if (!skb2) {
2277                         rcu_read_unlock();
2278                         return -ENOMEM;
2279                 }
2280
2281                 NETLINK_CB(skb2).portid = portid;
2282                 skb_push(skb2, sizeof(struct iphdr));
2283                 skb_reset_network_header(skb2);
2284                 iph = ip_hdr(skb2);
2285                 iph->ihl = sizeof(struct iphdr) >> 2;
2286                 iph->saddr = saddr;
2287                 iph->daddr = daddr;
2288                 iph->version = 0;
2289                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2290                 rcu_read_unlock();
2291                 return err;
2292         }
2293
2294         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2295         rcu_read_unlock();
2296         return err;
2297 }
2298
2299 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2300                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2301                             int flags)
2302 {
2303         struct nlmsghdr *nlh;
2304         struct rtmsg *rtm;
2305         int err;
2306
2307         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2308         if (!nlh)
2309                 return -EMSGSIZE;
2310
2311         rtm = nlmsg_data(nlh);
2312         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2313         rtm->rtm_dst_len  = 32;
2314         rtm->rtm_src_len  = 32;
2315         rtm->rtm_tos      = 0;
2316         rtm->rtm_table    = mrt->id;
2317         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2318                 goto nla_put_failure;
2319         rtm->rtm_type     = RTN_MULTICAST;
2320         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2321         if (c->_c.mfc_flags & MFC_STATIC)
2322                 rtm->rtm_protocol = RTPROT_STATIC;
2323         else
2324                 rtm->rtm_protocol = RTPROT_MROUTED;
2325         rtm->rtm_flags    = 0;
2326
2327         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2328             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2329                 goto nla_put_failure;
2330         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2331         /* do not break the dump if cache is unresolved */
2332         if (err < 0 && err != -ENOENT)
2333                 goto nla_put_failure;
2334
2335         nlmsg_end(skb, nlh);
2336         return 0;
2337
2338 nla_put_failure:
2339         nlmsg_cancel(skb, nlh);
2340         return -EMSGSIZE;
2341 }
2342
2343 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2344                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2345                              int flags)
2346 {
2347         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2348                                 cmd, flags);
2349 }
2350
2351 static size_t mroute_msgsize(bool unresolved, int maxvif)
2352 {
2353         size_t len =
2354                 NLMSG_ALIGN(sizeof(struct rtmsg))
2355                 + nla_total_size(4)     /* RTA_TABLE */
2356                 + nla_total_size(4)     /* RTA_SRC */
2357                 + nla_total_size(4)     /* RTA_DST */
2358                 ;
2359
2360         if (!unresolved)
2361                 len = len
2362                       + nla_total_size(4)       /* RTA_IIF */
2363                       + nla_total_size(0)       /* RTA_MULTIPATH */
2364                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2365                                                 /* RTA_MFC_STATS */
2366                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2367                 ;
2368
2369         return len;
2370 }
2371
2372 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2373                                  int cmd)
2374 {
2375         struct net *net = read_pnet(&mrt->net);
2376         struct sk_buff *skb;
2377         int err = -ENOBUFS;
2378
2379         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2380                                        mrt->maxvif),
2381                         GFP_ATOMIC);
2382         if (!skb)
2383                 goto errout;
2384
2385         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2386         if (err < 0)
2387                 goto errout;
2388
2389         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2390         return;
2391
2392 errout:
2393         kfree_skb(skb);
2394         if (err < 0)
2395                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2396 }
2397
2398 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2399 {
2400         size_t len =
2401                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2402                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2403                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2404                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2405                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2406                 + nla_total_size(4)     /* IPMRA_CREPORT_TABLE */
2407                                         /* IPMRA_CREPORT_PKT */
2408                 + nla_total_size(payloadlen)
2409                 ;
2410
2411         return len;
2412 }
2413
2414 static void igmpmsg_netlink_event(const struct mr_table *mrt, struct sk_buff *pkt)
2415 {
2416         struct net *net = read_pnet(&mrt->net);
2417         struct nlmsghdr *nlh;
2418         struct rtgenmsg *rtgenm;
2419         struct igmpmsg *msg;
2420         struct sk_buff *skb;
2421         struct nlattr *nla;
2422         int payloadlen;
2423
2424         payloadlen = pkt->len - sizeof(struct igmpmsg);
2425         msg = (struct igmpmsg *)skb_network_header(pkt);
2426
2427         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2428         if (!skb)
2429                 goto errout;
2430
2431         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2432                         sizeof(struct rtgenmsg), 0);
2433         if (!nlh)
2434                 goto errout;
2435         rtgenm = nlmsg_data(nlh);
2436         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2437         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2438             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif | (msg->im_vif_hi << 8)) ||
2439             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2440                             msg->im_src.s_addr) ||
2441             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2442                             msg->im_dst.s_addr) ||
2443             nla_put_u32(skb, IPMRA_CREPORT_TABLE, mrt->id))
2444                 goto nla_put_failure;
2445
2446         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2447         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2448                                   nla_data(nla), payloadlen))
2449                 goto nla_put_failure;
2450
2451         nlmsg_end(skb, nlh);
2452
2453         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2454         return;
2455
2456 nla_put_failure:
2457         nlmsg_cancel(skb, nlh);
2458 errout:
2459         kfree_skb(skb);
2460         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2461 }
2462
2463 static int ipmr_rtm_valid_getroute_req(struct sk_buff *skb,
2464                                        const struct nlmsghdr *nlh,
2465                                        struct nlattr **tb,
2466                                        struct netlink_ext_ack *extack)
2467 {
2468         struct rtmsg *rtm;
2469         int i, err;
2470
2471         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2472                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for multicast route get request");
2473                 return -EINVAL;
2474         }
2475
2476         if (!netlink_strict_get_check(skb))
2477                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2478                                               rtm_ipv4_policy, extack);
2479
2480         rtm = nlmsg_data(nlh);
2481         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2482             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2483             rtm->rtm_tos || rtm->rtm_table || rtm->rtm_protocol ||
2484             rtm->rtm_scope || rtm->rtm_type || rtm->rtm_flags) {
2485                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for multicast route get request");
2486                 return -EINVAL;
2487         }
2488
2489         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2490                                             rtm_ipv4_policy, extack);
2491         if (err)
2492                 return err;
2493
2494         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2495             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2496                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2497                 return -EINVAL;
2498         }
2499
2500         for (i = 0; i <= RTA_MAX; i++) {
2501                 if (!tb[i])
2502                         continue;
2503
2504                 switch (i) {
2505                 case RTA_SRC:
2506                 case RTA_DST:
2507                 case RTA_TABLE:
2508                         break;
2509                 default:
2510                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in multicast route get request");
2511                         return -EINVAL;
2512                 }
2513         }
2514
2515         return 0;
2516 }
2517
2518 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2519                              struct netlink_ext_ack *extack)
2520 {
2521         struct net *net = sock_net(in_skb->sk);
2522         struct nlattr *tb[RTA_MAX + 1];
2523         struct sk_buff *skb = NULL;
2524         struct mfc_cache *cache;
2525         struct mr_table *mrt;
2526         __be32 src, grp;
2527         u32 tableid;
2528         int err;
2529
2530         err = ipmr_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2531         if (err < 0)
2532                 goto errout;
2533
2534         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2535         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2536         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2537
2538         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2539         if (!mrt) {
2540                 err = -ENOENT;
2541                 goto errout_free;
2542         }
2543
2544         /* entries are added/deleted only under RTNL */
2545         rcu_read_lock();
2546         cache = ipmr_cache_find(mrt, src, grp);
2547         rcu_read_unlock();
2548         if (!cache) {
2549                 err = -ENOENT;
2550                 goto errout_free;
2551         }
2552
2553         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2554         if (!skb) {
2555                 err = -ENOBUFS;
2556                 goto errout_free;
2557         }
2558
2559         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2560                                nlh->nlmsg_seq, cache,
2561                                RTM_NEWROUTE, 0);
2562         if (err < 0)
2563                 goto errout_free;
2564
2565         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2566
2567 errout:
2568         return err;
2569
2570 errout_free:
2571         kfree_skb(skb);
2572         goto errout;
2573 }
2574
2575 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2576 {
2577         struct fib_dump_filter filter = {};
2578         int err;
2579
2580         if (cb->strict_check) {
2581                 err = ip_valid_fib_dump_req(sock_net(skb->sk), cb->nlh,
2582                                             &filter, cb);
2583                 if (err < 0)
2584                         return err;
2585         }
2586
2587         if (filter.table_id) {
2588                 struct mr_table *mrt;
2589
2590                 mrt = ipmr_get_table(sock_net(skb->sk), filter.table_id);
2591                 if (!mrt) {
2592                         if (rtnl_msg_family(cb->nlh) != RTNL_FAMILY_IPMR)
2593                                 return skb->len;
2594
2595                         NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
2596                         return -ENOENT;
2597                 }
2598                 err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
2599                                     &mfc_unres_lock, &filter);
2600                 return skb->len ? : err;
2601         }
2602
2603         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2604                                 _ipmr_fill_mroute, &mfc_unres_lock, &filter);
2605 }
2606
2607 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2608         [RTA_SRC]       = { .type = NLA_U32 },
2609         [RTA_DST]       = { .type = NLA_U32 },
2610         [RTA_IIF]       = { .type = NLA_U32 },
2611         [RTA_TABLE]     = { .type = NLA_U32 },
2612         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2613 };
2614
2615 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2616 {
2617         switch (rtm_protocol) {
2618         case RTPROT_STATIC:
2619         case RTPROT_MROUTED:
2620                 return true;
2621         }
2622         return false;
2623 }
2624
2625 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2626 {
2627         struct rtnexthop *rtnh = nla_data(nla);
2628         int remaining = nla_len(nla), vifi = 0;
2629
2630         while (rtnh_ok(rtnh, remaining)) {
2631                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2632                 if (++vifi == MAXVIFS)
2633                         break;
2634                 rtnh = rtnh_next(rtnh, &remaining);
2635         }
2636
2637         return remaining > 0 ? -EINVAL : vifi;
2638 }
2639
2640 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2641 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2642                             struct mfcctl *mfcc, int *mrtsock,
2643                             struct mr_table **mrtret,
2644                             struct netlink_ext_ack *extack)
2645 {
2646         struct net_device *dev = NULL;
2647         u32 tblid = RT_TABLE_DEFAULT;
2648         struct mr_table *mrt;
2649         struct nlattr *attr;
2650         struct rtmsg *rtm;
2651         int ret, rem;
2652
2653         ret = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
2654                                         rtm_ipmr_policy, extack);
2655         if (ret < 0)
2656                 goto out;
2657         rtm = nlmsg_data(nlh);
2658
2659         ret = -EINVAL;
2660         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2661             rtm->rtm_type != RTN_MULTICAST ||
2662             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2663             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2664                 goto out;
2665
2666         memset(mfcc, 0, sizeof(*mfcc));
2667         mfcc->mfcc_parent = -1;
2668         ret = 0;
2669         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2670                 switch (nla_type(attr)) {
2671                 case RTA_SRC:
2672                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2673                         break;
2674                 case RTA_DST:
2675                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2676                         break;
2677                 case RTA_IIF:
2678                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2679                         if (!dev) {
2680                                 ret = -ENODEV;
2681                                 goto out;
2682                         }
2683                         break;
2684                 case RTA_MULTIPATH:
2685                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2686                                 ret = -EINVAL;
2687                                 goto out;
2688                         }
2689                         break;
2690                 case RTA_PREFSRC:
2691                         ret = 1;
2692                         break;
2693                 case RTA_TABLE:
2694                         tblid = nla_get_u32(attr);
2695                         break;
2696                 }
2697         }
2698         mrt = ipmr_get_table(net, tblid);
2699         if (!mrt) {
2700                 ret = -ENOENT;
2701                 goto out;
2702         }
2703         *mrtret = mrt;
2704         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2705         if (dev)
2706                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2707
2708 out:
2709         return ret;
2710 }
2711
2712 /* takes care of both newroute and delroute */
2713 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2714                           struct netlink_ext_ack *extack)
2715 {
2716         struct net *net = sock_net(skb->sk);
2717         int ret, mrtsock, parent;
2718         struct mr_table *tbl;
2719         struct mfcctl mfcc;
2720
2721         mrtsock = 0;
2722         tbl = NULL;
2723         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2724         if (ret < 0)
2725                 return ret;
2726
2727         parent = ret ? mfcc.mfcc_parent : -1;
2728         if (nlh->nlmsg_type == RTM_NEWROUTE)
2729                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2730         else
2731                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2732 }
2733
2734 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2735 {
2736         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2737
2738         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2739             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2740             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2741                         mrt->mroute_reg_vif_num) ||
2742             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2743                        mrt->mroute_do_assert) ||
2744             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2745             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2746                        mrt->mroute_do_wrvifwhole))
2747                 return false;
2748
2749         return true;
2750 }
2751
2752 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2753 {
2754         struct net_device *vif_dev;
2755         struct nlattr *vif_nest;
2756         struct vif_device *vif;
2757
2758         vif = &mrt->vif_table[vifid];
2759         vif_dev = rtnl_dereference(vif->dev);
2760         /* if the VIF doesn't exist just continue */
2761         if (!vif_dev)
2762                 return true;
2763
2764         vif_nest = nla_nest_start_noflag(skb, IPMRA_VIF);
2765         if (!vif_nest)
2766                 return false;
2767
2768         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif_dev->ifindex) ||
2769             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2770             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2771             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2772                               IPMRA_VIFA_PAD) ||
2773             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2774                               IPMRA_VIFA_PAD) ||
2775             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2776                               IPMRA_VIFA_PAD) ||
2777             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2778                               IPMRA_VIFA_PAD) ||
2779             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2780             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2781                 nla_nest_cancel(skb, vif_nest);
2782                 return false;
2783         }
2784         nla_nest_end(skb, vif_nest);
2785
2786         return true;
2787 }
2788
2789 static int ipmr_valid_dumplink(const struct nlmsghdr *nlh,
2790                                struct netlink_ext_ack *extack)
2791 {
2792         struct ifinfomsg *ifm;
2793
2794         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ifm))) {
2795                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for ipmr link dump");
2796                 return -EINVAL;
2797         }
2798
2799         if (nlmsg_attrlen(nlh, sizeof(*ifm))) {
2800                 NL_SET_ERR_MSG(extack, "Invalid data after header in ipmr link dump");
2801                 return -EINVAL;
2802         }
2803
2804         ifm = nlmsg_data(nlh);
2805         if (ifm->__ifi_pad || ifm->ifi_type || ifm->ifi_flags ||
2806             ifm->ifi_change || ifm->ifi_index) {
2807                 NL_SET_ERR_MSG(extack, "Invalid values in header for ipmr link dump request");
2808                 return -EINVAL;
2809         }
2810
2811         return 0;
2812 }
2813
2814 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2815 {
2816         struct net *net = sock_net(skb->sk);
2817         struct nlmsghdr *nlh = NULL;
2818         unsigned int t = 0, s_t;
2819         unsigned int e = 0, s_e;
2820         struct mr_table *mrt;
2821
2822         if (cb->strict_check) {
2823                 int err = ipmr_valid_dumplink(cb->nlh, cb->extack);
2824
2825                 if (err < 0)
2826                         return err;
2827         }
2828
2829         s_t = cb->args[0];
2830         s_e = cb->args[1];
2831
2832         ipmr_for_each_table(mrt, net) {
2833                 struct nlattr *vifs, *af;
2834                 struct ifinfomsg *hdr;
2835                 u32 i;
2836
2837                 if (t < s_t)
2838                         goto skip_table;
2839                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2840                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2841                                 sizeof(*hdr), NLM_F_MULTI);
2842                 if (!nlh)
2843                         break;
2844
2845                 hdr = nlmsg_data(nlh);
2846                 memset(hdr, 0, sizeof(*hdr));
2847                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2848
2849                 af = nla_nest_start_noflag(skb, IFLA_AF_SPEC);
2850                 if (!af) {
2851                         nlmsg_cancel(skb, nlh);
2852                         goto out;
2853                 }
2854
2855                 if (!ipmr_fill_table(mrt, skb)) {
2856                         nlmsg_cancel(skb, nlh);
2857                         goto out;
2858                 }
2859
2860                 vifs = nla_nest_start_noflag(skb, IPMRA_TABLE_VIFS);
2861                 if (!vifs) {
2862                         nla_nest_end(skb, af);
2863                         nlmsg_end(skb, nlh);
2864                         goto out;
2865                 }
2866                 for (i = 0; i < mrt->maxvif; i++) {
2867                         if (e < s_e)
2868                                 goto skip_entry;
2869                         if (!ipmr_fill_vif(mrt, i, skb)) {
2870                                 nla_nest_end(skb, vifs);
2871                                 nla_nest_end(skb, af);
2872                                 nlmsg_end(skb, nlh);
2873                                 goto out;
2874                         }
2875 skip_entry:
2876                         e++;
2877                 }
2878                 s_e = 0;
2879                 e = 0;
2880                 nla_nest_end(skb, vifs);
2881                 nla_nest_end(skb, af);
2882                 nlmsg_end(skb, nlh);
2883 skip_table:
2884                 t++;
2885         }
2886
2887 out:
2888         cb->args[1] = e;
2889         cb->args[0] = t;
2890
2891         return skb->len;
2892 }
2893
2894 #ifdef CONFIG_PROC_FS
2895 /* The /proc interfaces to multicast routing :
2896  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2897  */
2898
2899 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2900         __acquires(RCU)
2901 {
2902         struct mr_vif_iter *iter = seq->private;
2903         struct net *net = seq_file_net(seq);
2904         struct mr_table *mrt;
2905
2906         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2907         if (!mrt)
2908                 return ERR_PTR(-ENOENT);
2909
2910         iter->mrt = mrt;
2911
2912         rcu_read_lock();
2913         return mr_vif_seq_start(seq, pos);
2914 }
2915
2916 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2917         __releases(RCU)
2918 {
2919         rcu_read_unlock();
2920 }
2921
2922 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2923 {
2924         struct mr_vif_iter *iter = seq->private;
2925         struct mr_table *mrt = iter->mrt;
2926
2927         if (v == SEQ_START_TOKEN) {
2928                 seq_puts(seq,
2929                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2930         } else {
2931                 const struct vif_device *vif = v;
2932                 const struct net_device *vif_dev;
2933                 const char *name;
2934
2935                 vif_dev = vif_dev_read(vif);
2936                 name = vif_dev ? vif_dev->name : "none";
2937                 seq_printf(seq,
2938                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2939                            vif - mrt->vif_table,
2940                            name, vif->bytes_in, vif->pkt_in,
2941                            vif->bytes_out, vif->pkt_out,
2942                            vif->flags, vif->local, vif->remote);
2943         }
2944         return 0;
2945 }
2946
2947 static const struct seq_operations ipmr_vif_seq_ops = {
2948         .start = ipmr_vif_seq_start,
2949         .next  = mr_vif_seq_next,
2950         .stop  = ipmr_vif_seq_stop,
2951         .show  = ipmr_vif_seq_show,
2952 };
2953
2954 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2955 {
2956         struct net *net = seq_file_net(seq);
2957         struct mr_table *mrt;
2958
2959         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2960         if (!mrt)
2961                 return ERR_PTR(-ENOENT);
2962
2963         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2964 }
2965
2966 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2967 {
2968         int n;
2969
2970         if (v == SEQ_START_TOKEN) {
2971                 seq_puts(seq,
2972                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2973         } else {
2974                 const struct mfc_cache *mfc = v;
2975                 const struct mr_mfc_iter *it = seq->private;
2976                 const struct mr_table *mrt = it->mrt;
2977
2978                 seq_printf(seq, "%08X %08X %-3hd",
2979                            (__force u32) mfc->mfc_mcastgrp,
2980                            (__force u32) mfc->mfc_origin,
2981                            mfc->_c.mfc_parent);
2982
2983                 if (it->cache != &mrt->mfc_unres_queue) {
2984                         seq_printf(seq, " %8lu %8lu %8lu",
2985                                    mfc->_c.mfc_un.res.pkt,
2986                                    mfc->_c.mfc_un.res.bytes,
2987                                    mfc->_c.mfc_un.res.wrong_if);
2988                         for (n = mfc->_c.mfc_un.res.minvif;
2989                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2990                                 if (VIF_EXISTS(mrt, n) &&
2991                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2992                                         seq_printf(seq,
2993                                            " %2d:%-3d",
2994                                            n, mfc->_c.mfc_un.res.ttls[n]);
2995                         }
2996                 } else {
2997                         /* unresolved mfc_caches don't contain
2998                          * pkt, bytes and wrong_if values
2999                          */
3000                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
3001                 }
3002                 seq_putc(seq, '\n');
3003         }
3004         return 0;
3005 }
3006
3007 static const struct seq_operations ipmr_mfc_seq_ops = {
3008         .start = ipmr_mfc_seq_start,
3009         .next  = mr_mfc_seq_next,
3010         .stop  = mr_mfc_seq_stop,
3011         .show  = ipmr_mfc_seq_show,
3012 };
3013 #endif
3014
3015 #ifdef CONFIG_IP_PIMSM_V2
3016 static const struct net_protocol pim_protocol = {
3017         .handler        =       pim_rcv,
3018 };
3019 #endif
3020
3021 static unsigned int ipmr_seq_read(struct net *net)
3022 {
3023         ASSERT_RTNL();
3024
3025         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3026 }
3027
3028 static int ipmr_dump(struct net *net, struct notifier_block *nb,
3029                      struct netlink_ext_ack *extack)
3030 {
3031         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
3032                        ipmr_mr_table_iter, extack);
3033 }
3034
3035 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3036         .family         = RTNL_FAMILY_IPMR,
3037         .fib_seq_read   = ipmr_seq_read,
3038         .fib_dump       = ipmr_dump,
3039         .owner          = THIS_MODULE,
3040 };
3041
3042 static int __net_init ipmr_notifier_init(struct net *net)
3043 {
3044         struct fib_notifier_ops *ops;
3045
3046         net->ipv4.ipmr_seq = 0;
3047
3048         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3049         if (IS_ERR(ops))
3050                 return PTR_ERR(ops);
3051         net->ipv4.ipmr_notifier_ops = ops;
3052
3053         return 0;
3054 }
3055
3056 static void __net_exit ipmr_notifier_exit(struct net *net)
3057 {
3058         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3059         net->ipv4.ipmr_notifier_ops = NULL;
3060 }
3061
3062 /* Setup for IP multicast routing */
3063 static int __net_init ipmr_net_init(struct net *net)
3064 {
3065         int err;
3066
3067         err = ipmr_notifier_init(net);
3068         if (err)
3069                 goto ipmr_notifier_fail;
3070
3071         err = ipmr_rules_init(net);
3072         if (err < 0)
3073                 goto ipmr_rules_fail;
3074
3075 #ifdef CONFIG_PROC_FS
3076         err = -ENOMEM;
3077         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
3078                         sizeof(struct mr_vif_iter)))
3079                 goto proc_vif_fail;
3080         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
3081                         sizeof(struct mr_mfc_iter)))
3082                 goto proc_cache_fail;
3083 #endif
3084         return 0;
3085
3086 #ifdef CONFIG_PROC_FS
3087 proc_cache_fail:
3088         remove_proc_entry("ip_mr_vif", net->proc_net);
3089 proc_vif_fail:
3090         rtnl_lock();
3091         ipmr_rules_exit(net);
3092         rtnl_unlock();
3093 #endif
3094 ipmr_rules_fail:
3095         ipmr_notifier_exit(net);
3096 ipmr_notifier_fail:
3097         return err;
3098 }
3099
3100 static void __net_exit ipmr_net_exit(struct net *net)
3101 {
3102 #ifdef CONFIG_PROC_FS
3103         remove_proc_entry("ip_mr_cache", net->proc_net);
3104         remove_proc_entry("ip_mr_vif", net->proc_net);
3105 #endif
3106         ipmr_notifier_exit(net);
3107 }
3108
3109 static void __net_exit ipmr_net_exit_batch(struct list_head *net_list)
3110 {
3111         struct net *net;
3112
3113         rtnl_lock();
3114         list_for_each_entry(net, net_list, exit_list)
3115                 ipmr_rules_exit(net);
3116         rtnl_unlock();
3117 }
3118
3119 static struct pernet_operations ipmr_net_ops = {
3120         .init = ipmr_net_init,
3121         .exit = ipmr_net_exit,
3122         .exit_batch = ipmr_net_exit_batch,
3123 };
3124
3125 int __init ip_mr_init(void)
3126 {
3127         int err;
3128
3129         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3130                                        sizeof(struct mfc_cache),
3131                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3132                                        NULL);
3133
3134         err = register_pernet_subsys(&ipmr_net_ops);
3135         if (err)
3136                 goto reg_pernet_fail;
3137
3138         err = register_netdevice_notifier(&ip_mr_notifier);
3139         if (err)
3140                 goto reg_notif_fail;
3141 #ifdef CONFIG_IP_PIMSM_V2
3142         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3143                 pr_err("%s: can't add PIM protocol\n", __func__);
3144                 err = -EAGAIN;
3145                 goto add_proto_fail;
3146         }
3147 #endif
3148         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3149                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3150         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3151                       ipmr_rtm_route, NULL, 0);
3152         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3153                       ipmr_rtm_route, NULL, 0);
3154
3155         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3156                       NULL, ipmr_rtm_dumplink, 0);
3157         return 0;
3158
3159 #ifdef CONFIG_IP_PIMSM_V2
3160 add_proto_fail:
3161         unregister_netdevice_notifier(&ip_mr_notifier);
3162 #endif
3163 reg_notif_fail:
3164         unregister_pernet_subsys(&ipmr_net_ops);
3165 reg_pernet_fail:
3166         kmem_cache_destroy(mrt_cachep);
3167         return err;
3168 }