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         del_timer_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.tx_bytes += skb->len;
 510         dev->stats.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                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1008                 }
1009         }
1010 }
1011
1012 /* Bounce a cache query up to mrouted and netlink.
1013  *
1014  * Called under rcu_read_lock().
1015  */
1016 static int ipmr_cache_report(const struct mr_table *mrt,
1017                              struct sk_buff *pkt, vifi_t vifi, int assert)
1018 {
1019         const int ihl = ip_hdrlen(pkt);
1020         struct sock *mroute_sk;
1021         struct igmphdr *igmp;
1022         struct igmpmsg *msg;
1023         struct sk_buff *skb;
1024         int ret;
1025
1026         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE)
1027                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1028         else
1029                 skb = alloc_skb(128, GFP_ATOMIC);
1030
1031         if (!skb)
1032                 return -ENOBUFS;
1033
1034         if (assert == IGMPMSG_WHOLEPKT || assert == IGMPMSG_WRVIFWHOLE) {
1035                 /* Ugly, but we have no choice with this interface.
1036                  * Duplicate old header, fix ihl, length etc.
1037                  * And all this only to mangle msg->im_msgtype and
1038                  * to set msg->im_mbz to "mbz" :-)
1039                  */
1040                 skb_push(skb, sizeof(struct iphdr));
1041                 skb_reset_network_header(skb);
1042                 skb_reset_transport_header(skb);
1043                 msg = (struct igmpmsg *)skb_network_header(skb);
1044                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1045                 msg->im_msgtype = assert;
1046                 msg->im_mbz = 0;
1047                 if (assert == IGMPMSG_WRVIFWHOLE) {
1048                         msg->im_vif = vifi;
1049                         msg->im_vif_hi = vifi >> 8;
1050                 } else {
1051                         /* Pairs with WRITE_ONCE() in vif_add() and vif_delete() */
1052                         int vif_num = READ_ONCE(mrt->mroute_reg_vif_num);
1053
1054                         msg->im_vif = vif_num;
1055                         msg->im_vif_hi = vif_num >> 8;
1056                 }
1057                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1058                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1059                                              sizeof(struct iphdr));
1060         } else {
1061                 /* Copy the IP header */
1062                 skb_set_network_header(skb, skb->len);
1063                 skb_put(skb, ihl);
1064                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1065                 /* Flag to the kernel this is a route add */
1066                 ip_hdr(skb)->protocol = 0;
1067                 msg = (struct igmpmsg *)skb_network_header(skb);
1068                 msg->im_vif = vifi;
1069                 msg->im_vif_hi = vifi >> 8;
1070                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1071                 /* Add our header */
1072                 igmp = skb_put(skb, sizeof(struct igmphdr));
1073                 igmp->type = assert;
1074                 msg->im_msgtype = assert;
1075                 igmp->code = 0;
1076                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1077                 skb->transport_header = skb->network_header;
1078         }
1079
1080         mroute_sk = rcu_dereference(mrt->mroute_sk);
1081         if (!mroute_sk) {
1082                 kfree_skb(skb);
1083                 return -EINVAL;
1084         }
1085
1086         igmpmsg_netlink_event(mrt, skb);
1087
1088         /* Deliver to mrouted */
1089         ret = sock_queue_rcv_skb(mroute_sk, skb);
1090
1091         if (ret < 0) {
1092                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1093                 kfree_skb(skb);
1094         }
1095
1096         return ret;
1097 }
1098
1099 /* Queue a packet for resolution. It gets locked cache entry! */
1100 /* Called under rcu_read_lock() */
1101 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1102                                  struct sk_buff *skb, struct net_device *dev)
1103 {
1104         const struct iphdr *iph = ip_hdr(skb);
1105         struct mfc_cache *c;
1106         bool found = false;
1107         int err;
1108
1109         spin_lock_bh(&mfc_unres_lock);
1110         list_for_each_entry(c, &mrt->mfc_unres_queue, _c.list) {
1111                 if (c->mfc_mcastgrp == iph->daddr &&
1112                     c->mfc_origin == iph->saddr) {
1113                         found = true;
1114                         break;
1115                 }
1116         }
1117
1118         if (!found) {
1119                 /* Create a new entry if allowable */
1120                 c = ipmr_cache_alloc_unres();
1121                 if (!c) {
1122                         spin_unlock_bh(&mfc_unres_lock);
1123
1124                         kfree_skb(skb);
1125                         return -ENOBUFS;
1126                 }
1127
1128                 /* Fill in the new cache entry */
1129                 c->_c.mfc_parent = -1;
1130                 c->mfc_origin   = iph->saddr;
1131                 c->mfc_mcastgrp = iph->daddr;
1132
1133                 /* Reflect first query at mrouted. */
1134                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1135
1136                 if (err < 0) {
1137                         /* If the report failed throw the cache entry
1138                            out - Brad Parker
1139                          */
1140                         spin_unlock_bh(&mfc_unres_lock);
1141
1142                         ipmr_cache_free(c);
1143                         kfree_skb(skb);
1144                         return err;
1145                 }
1146
1147                 atomic_inc(&mrt->cache_resolve_queue_len);
1148                 list_add(&c->_c.list, &mrt->mfc_unres_queue);
1149                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1150
1151                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1152                         mod_timer(&mrt->ipmr_expire_timer,
1153                                   c->_c.mfc_un.unres.expires);
1154         }
1155
1156         /* See if we can append the packet */
1157         if (c->_c.mfc_un.unres.unresolved.qlen > 3) {
1158                 kfree_skb(skb);
1159                 err = -ENOBUFS;
1160         } else {
1161                 if (dev) {
1162                         skb->dev = dev;
1163                         skb->skb_iif = dev->ifindex;
1164                 }
1165                 skb_queue_tail(&c->_c.mfc_un.unres.unresolved, skb);
1166                 err = 0;
1167         }
1168
1169         spin_unlock_bh(&mfc_unres_lock);
1170         return err;
1171 }
1172
1173 /* MFC cache manipulation by user space mroute daemon */
1174
1175 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1176 {
1177         struct net *net = read_pnet(&mrt->net);
1178         struct mfc_cache *c;
1179
1180         /* The entries are added/deleted only under RTNL */
1181         rcu_read_lock();
1182         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1183                                    mfc->mfcc_mcastgrp.s_addr, parent);
1184         rcu_read_unlock();
1185         if (!c)
1186                 return -ENOENT;
1187         rhltable_remove(&mrt->mfc_hash, &c->_c.mnode, ipmr_rht_params);
1188         list_del_rcu(&c->_c.list);
1189         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, c, mrt->id);
1190         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1191         mr_cache_put(&c->_c);
1192
1193         return 0;
1194 }
1195
1196 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1197                         struct mfcctl *mfc, int mrtsock, int parent)
1198 {
1199         struct mfc_cache *uc, *c;
1200         struct mr_mfc *_uc;
1201         bool found;
1202         int ret;
1203
1204         if (mfc->mfcc_parent >= MAXVIFS)
1205                 return -ENFILE;
1206
1207         /* The entries are added/deleted only under RTNL */
1208         rcu_read_lock();
1209         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1210                                    mfc->mfcc_mcastgrp.s_addr, parent);
1211         rcu_read_unlock();
1212         if (c) {
1213                 spin_lock(&mrt_lock);
1214                 c->_c.mfc_parent = mfc->mfcc_parent;
1215                 ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1216                 if (!mrtsock)
1217                         c->_c.mfc_flags |= MFC_STATIC;
1218                 spin_unlock(&mrt_lock);
1219                 call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, c,
1220                                               mrt->id);
1221                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1222                 return 0;
1223         }
1224
1225         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1226             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1227                 return -EINVAL;
1228
1229         c = ipmr_cache_alloc();
1230         if (!c)
1231                 return -ENOMEM;
1232
1233         c->mfc_origin = mfc->mfcc_origin.s_addr;
1234         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1235         c->_c.mfc_parent = mfc->mfcc_parent;
1236         ipmr_update_thresholds(mrt, &c->_c, mfc->mfcc_ttls);
1237         if (!mrtsock)
1238                 c->_c.mfc_flags |= MFC_STATIC;
1239
1240         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->_c.mnode,
1241                                   ipmr_rht_params);
1242         if (ret) {
1243                 pr_err("ipmr: rhtable insert error %d\n", ret);
1244                 ipmr_cache_free(c);
1245                 return ret;
1246         }
1247         list_add_tail_rcu(&c->_c.list, &mrt->mfc_cache_list);
1248         /* Check to see if we resolved a queued list. If so we
1249          * need to send on the frames and tidy up.
1250          */
1251         found = false;
1252         spin_lock_bh(&mfc_unres_lock);
1253         list_for_each_entry(_uc, &mrt->mfc_unres_queue, list) {
1254                 uc = (struct mfc_cache *)_uc;
1255                 if (uc->mfc_origin == c->mfc_origin &&
1256                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1257                         list_del(&_uc->list);
1258                         atomic_dec(&mrt->cache_resolve_queue_len);
1259                         found = true;
1260                         break;
1261                 }
1262         }
1263         if (list_empty(&mrt->mfc_unres_queue))
1264                 del_timer(&mrt->ipmr_expire_timer);
1265         spin_unlock_bh(&mfc_unres_lock);
1266
1267         if (found) {
1268                 ipmr_cache_resolve(net, mrt, uc, c);
1269                 ipmr_cache_free(uc);
1270         }
1271         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, c, mrt->id);
1272         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1273         return 0;
1274 }
1275
1276 /* Close the multicast socket, and clear the vif tables etc */
1277 static void mroute_clean_tables(struct mr_table *mrt, int flags)
1278 {
1279         struct net *net = read_pnet(&mrt->net);
1280         struct mr_mfc *c, *tmp;
1281         struct mfc_cache *cache;
1282         LIST_HEAD(list);
1283         int i;
1284
1285         /* Shut down all active vif entries */
1286         if (flags & (MRT_FLUSH_VIFS | MRT_FLUSH_VIFS_STATIC)) {
1287                 for (i = 0; i < mrt->maxvif; i++) {
1288                         if (((mrt->vif_table[i].flags & VIFF_STATIC) &&
1289                              !(flags & MRT_FLUSH_VIFS_STATIC)) ||
1290                             (!(mrt->vif_table[i].flags & VIFF_STATIC) && !(flags & MRT_FLUSH_VIFS)))
1291                                 continue;
1292                         vif_delete(mrt, i, 0, &list);
1293                 }
1294                 unregister_netdevice_many(&list);
1295         }
1296
1297         /* Wipe the cache */
1298         if (flags & (MRT_FLUSH_MFC | MRT_FLUSH_MFC_STATIC)) {
1299                 list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1300                         if (((c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC_STATIC)) ||
1301                             (!(c->mfc_flags & MFC_STATIC) && !(flags & MRT_FLUSH_MFC)))
1302                                 continue;
1303                         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1304                         list_del_rcu(&c->list);
1305                         cache = (struct mfc_cache *)c;
1306                         call_ipmr_mfc_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, cache,
1307                                                       mrt->id);
1308                         mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1309                         mr_cache_put(c);
1310                 }
1311         }
1312
1313         if (flags & MRT_FLUSH_MFC) {
1314                 if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1315                         spin_lock_bh(&mfc_unres_lock);
1316                         list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1317                                 list_del(&c->list);
1318                                 cache = (struct mfc_cache *)c;
1319                                 mroute_netlink_event(mrt, cache, RTM_DELROUTE);
1320                                 ipmr_destroy_unres(mrt, cache);
1321                         }
1322                         spin_unlock_bh(&mfc_unres_lock);
1323                 }
1324         }
1325 }
1326
1327 /* called from ip_ra_control(), before an RCU grace period,
1328  * we don't need to call synchronize_rcu() here
1329  */
1330 static void mrtsock_destruct(struct sock *sk)
1331 {
1332         struct net *net = sock_net(sk);
1333         struct mr_table *mrt;
1334
1335         rtnl_lock();
1336         ipmr_for_each_table(mrt, net) {
1337                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1338                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1339                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1340                                                     NETCONFA_MC_FORWARDING,
1341                                                     NETCONFA_IFINDEX_ALL,
1342                                                     net->ipv4.devconf_all);
1343                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1344                         mroute_clean_tables(mrt, MRT_FLUSH_VIFS | MRT_FLUSH_MFC);
1345                 }
1346         }
1347         rtnl_unlock();
1348 }
1349
1350 /* Socket options and virtual interface manipulation. The whole
1351  * virtual interface system is a complete heap, but unfortunately
1352  * that's how BSD mrouted happens to think. Maybe one day with a proper
1353  * MOSPF/PIM router set up we can clean this up.
1354  */
1355
1356 int ip_mroute_setsockopt(struct sock *sk, int optname, sockptr_t optval,
1357                          unsigned int optlen)
1358 {
1359         struct net *net = sock_net(sk);
1360         int val, ret = 0, parent = 0;
1361         struct mr_table *mrt;
1362         struct vifctl vif;
1363         struct mfcctl mfc;
1364         bool do_wrvifwhole;
1365         u32 uval;
1366
1367         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1368         rtnl_lock();
1369         if (sk->sk_type != SOCK_RAW ||
1370             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1371                 ret = -EOPNOTSUPP;
1372                 goto out_unlock;
1373         }
1374
1375         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1376         if (!mrt) {
1377                 ret = -ENOENT;
1378                 goto out_unlock;
1379         }
1380         if (optname != MRT_INIT) {
1381                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1382                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1383                         ret = -EACCES;
1384                         goto out_unlock;
1385                 }
1386         }
1387
1388         switch (optname) {
1389         case MRT_INIT:
1390                 if (optlen != sizeof(int)) {
1391                         ret = -EINVAL;
1392                         break;
1393                 }
1394                 if (rtnl_dereference(mrt->mroute_sk)) {
1395                         ret = -EADDRINUSE;
1396                         break;
1397                 }
1398
1399                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1400                 if (ret == 0) {
1401                         rcu_assign_pointer(mrt->mroute_sk, sk);
1402                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1403                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1404                                                     NETCONFA_MC_FORWARDING,
1405                                                     NETCONFA_IFINDEX_ALL,
1406                                                     net->ipv4.devconf_all);
1407                 }
1408                 break;
1409         case MRT_DONE:
1410                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1411                         ret = -EACCES;
1412                 } else {
1413                         /* We need to unlock here because mrtsock_destruct takes
1414                          * care of rtnl itself and we can't change that due to
1415                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1416                          */
1417                         rtnl_unlock();
1418                         ret = ip_ra_control(sk, 0, NULL);
1419                         goto out;
1420                 }
1421                 break;
1422         case MRT_ADD_VIF:
1423         case MRT_DEL_VIF:
1424                 if (optlen != sizeof(vif)) {
1425                         ret = -EINVAL;
1426                         break;
1427                 }
1428                 if (copy_from_sockptr(&vif, optval, sizeof(vif))) {
1429                         ret = -EFAULT;
1430                         break;
1431                 }
1432                 if (vif.vifc_vifi >= MAXVIFS) {
1433                         ret = -ENFILE;
1434                         break;
1435                 }
1436                 if (optname == MRT_ADD_VIF) {
1437                         ret = vif_add(net, mrt, &vif,
1438                                       sk == rtnl_dereference(mrt->mroute_sk));
1439                 } else {
1440                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1441                 }
1442                 break;
1443         /* Manipulate the forwarding caches. These live
1444          * in a sort of kernel/user symbiosis.
1445          */
1446         case MRT_ADD_MFC:
1447         case MRT_DEL_MFC:
1448                 parent = -1;
1449                 fallthrough;
1450         case MRT_ADD_MFC_PROXY:
1451         case MRT_DEL_MFC_PROXY:
1452                 if (optlen != sizeof(mfc)) {
1453                         ret = -EINVAL;
1454                         break;
1455                 }
1456                 if (copy_from_sockptr(&mfc, optval, sizeof(mfc))) {
1457                         ret = -EFAULT;
1458                         break;
1459                 }
1460                 if (parent == 0)
1461                         parent = mfc.mfcc_parent;
1462                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1463                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1464                 else
1465                         ret = ipmr_mfc_add(net, mrt, &mfc,
1466                                            sk == rtnl_dereference(mrt->mroute_sk),
1467                                            parent);
1468                 break;
1469         case MRT_FLUSH:
1470                 if (optlen != sizeof(val)) {
1471                         ret = -EINVAL;
1472                         break;
1473                 }
1474                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1475                         ret = -EFAULT;
1476                         break;
1477                 }
1478                 mroute_clean_tables(mrt, val);
1479                 break;
1480         /* Control PIM assert. */
1481         case MRT_ASSERT:
1482                 if (optlen != sizeof(val)) {
1483                         ret = -EINVAL;
1484                         break;
1485                 }
1486                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1487                         ret = -EFAULT;
1488                         break;
1489                 }
1490                 mrt->mroute_do_assert = val;
1491                 break;
1492         case MRT_PIM:
1493                 if (!ipmr_pimsm_enabled()) {
1494                         ret = -ENOPROTOOPT;
1495                         break;
1496                 }
1497                 if (optlen != sizeof(val)) {
1498                         ret = -EINVAL;
1499                         break;
1500                 }
1501                 if (copy_from_sockptr(&val, optval, sizeof(val))) {
1502                         ret = -EFAULT;
1503                         break;
1504                 }
1505
1506                 do_wrvifwhole = (val == IGMPMSG_WRVIFWHOLE);
1507                 val = !!val;
1508                 if (val != mrt->mroute_do_pim) {
1509                         mrt->mroute_do_pim = val;
1510                         mrt->mroute_do_assert = val;
1511                         mrt->mroute_do_wrvifwhole = do_wrvifwhole;
1512                 }
1513                 break;
1514         case MRT_TABLE:
1515                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1516                         ret = -ENOPROTOOPT;
1517                         break;
1518                 }
1519                 if (optlen != sizeof(uval)) {
1520                         ret = -EINVAL;
1521                         break;
1522                 }
1523                 if (copy_from_sockptr(&uval, optval, sizeof(uval))) {
1524                         ret = -EFAULT;
1525                         break;
1526                 }
1527
1528                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1529                         ret = -EBUSY;
1530                 } else {
1531                         mrt = ipmr_new_table(net, uval);
1532                         if (IS_ERR(mrt))
1533                                 ret = PTR_ERR(mrt);
1534                         else
1535                                 raw_sk(sk)->ipmr_table = uval;
1536                 }
1537                 break;
1538         /* Spurious command, or MRT_VERSION which you cannot set. */
1539         default:
1540                 ret = -ENOPROTOOPT;
1541         }
1542 out_unlock:
1543         rtnl_unlock();
1544 out:
1545         return ret;
1546 }
1547
1548 /* Getsock opt support for the multicast routing system. */
1549 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1550 {
1551         int olr;
1552         int val;
1553         struct net *net = sock_net(sk);
1554         struct mr_table *mrt;
1555
1556         if (sk->sk_type != SOCK_RAW ||
1557             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1558                 return -EOPNOTSUPP;
1559
1560         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1561         if (!mrt)
1562                 return -ENOENT;
1563
1564         switch (optname) {
1565         case MRT_VERSION:
1566                 val = 0x0305;
1567                 break;
1568         case MRT_PIM:
1569                 if (!ipmr_pimsm_enabled())
1570                         return -ENOPROTOOPT;
1571                 val = mrt->mroute_do_pim;
1572                 break;
1573         case MRT_ASSERT:
1574                 val = mrt->mroute_do_assert;
1575                 break;
1576         default:
1577                 return -ENOPROTOOPT;
1578         }
1579
1580         if (get_user(olr, optlen))
1581                 return -EFAULT;
1582         olr = min_t(unsigned int, olr, sizeof(int));
1583         if (olr < 0)
1584                 return -EINVAL;
1585         if (put_user(olr, optlen))
1586                 return -EFAULT;
1587         if (copy_to_user(optval, &val, olr))
1588                 return -EFAULT;
1589         return 0;
1590 }
1591
1592 /* The IP multicast ioctl support routines. */
1593 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1594 {
1595         struct sioc_sg_req sr;
1596         struct sioc_vif_req vr;
1597         struct vif_device *vif;
1598         struct mfc_cache *c;
1599         struct net *net = sock_net(sk);
1600         struct mr_table *mrt;
1601
1602         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1603         if (!mrt)
1604                 return -ENOENT;
1605
1606         switch (cmd) {
1607         case SIOCGETVIFCNT:
1608                 if (copy_from_user(&vr, arg, sizeof(vr)))
1609                         return -EFAULT;
1610                 if (vr.vifi >= mrt->maxvif)
1611                         return -EINVAL;
1612                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1613                 rcu_read_lock();
1614                 vif = &mrt->vif_table[vr.vifi];
1615                 if (VIF_EXISTS(mrt, vr.vifi)) {
1616                         vr.icount = READ_ONCE(vif->pkt_in);
1617                         vr.ocount = READ_ONCE(vif->pkt_out);
1618                         vr.ibytes = READ_ONCE(vif->bytes_in);
1619                         vr.obytes = READ_ONCE(vif->bytes_out);
1620                         rcu_read_unlock();
1621
1622                         if (copy_to_user(arg, &vr, sizeof(vr)))
1623                                 return -EFAULT;
1624                         return 0;
1625                 }
1626                 rcu_read_unlock();
1627                 return -EADDRNOTAVAIL;
1628         case SIOCGETSGCNT:
1629                 if (copy_from_user(&sr, arg, sizeof(sr)))
1630                         return -EFAULT;
1631
1632                 rcu_read_lock();
1633                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1634                 if (c) {
1635                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1636                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1637                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1638                         rcu_read_unlock();
1639
1640                         if (copy_to_user(arg, &sr, sizeof(sr)))
1641                                 return -EFAULT;
1642                         return 0;
1643                 }
1644                 rcu_read_unlock();
1645                 return -EADDRNOTAVAIL;
1646         default:
1647                 return -ENOIOCTLCMD;
1648         }
1649 }
1650
1651 #ifdef CONFIG_COMPAT
1652 struct compat_sioc_sg_req {
1653         struct in_addr src;
1654         struct in_addr grp;
1655         compat_ulong_t pktcnt;
1656         compat_ulong_t bytecnt;
1657         compat_ulong_t wrong_if;
1658 };
1659
1660 struct compat_sioc_vif_req {
1661         vifi_t  vifi;           /* Which iface */
1662         compat_ulong_t icount;
1663         compat_ulong_t ocount;
1664         compat_ulong_t ibytes;
1665         compat_ulong_t obytes;
1666 };
1667
1668 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1669 {
1670         struct compat_sioc_sg_req sr;
1671         struct compat_sioc_vif_req vr;
1672         struct vif_device *vif;
1673         struct mfc_cache *c;
1674         struct net *net = sock_net(sk);
1675         struct mr_table *mrt;
1676
1677         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1678         if (!mrt)
1679                 return -ENOENT;
1680
1681         switch (cmd) {
1682         case SIOCGETVIFCNT:
1683                 if (copy_from_user(&vr, arg, sizeof(vr)))
1684                         return -EFAULT;
1685                 if (vr.vifi >= mrt->maxvif)
1686                         return -EINVAL;
1687                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1688                 rcu_read_lock();
1689                 vif = &mrt->vif_table[vr.vifi];
1690                 if (VIF_EXISTS(mrt, vr.vifi)) {
1691                         vr.icount = READ_ONCE(vif->pkt_in);
1692                         vr.ocount = READ_ONCE(vif->pkt_out);
1693                         vr.ibytes = READ_ONCE(vif->bytes_in);
1694                         vr.obytes = READ_ONCE(vif->bytes_out);
1695                         rcu_read_unlock();
1696
1697                         if (copy_to_user(arg, &vr, sizeof(vr)))
1698                                 return -EFAULT;
1699                         return 0;
1700                 }
1701                 rcu_read_unlock();
1702                 return -EADDRNOTAVAIL;
1703         case SIOCGETSGCNT:
1704                 if (copy_from_user(&sr, arg, sizeof(sr)))
1705                         return -EFAULT;
1706
1707                 rcu_read_lock();
1708                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1709                 if (c) {
1710                         sr.pktcnt = c->_c.mfc_un.res.pkt;
1711                         sr.bytecnt = c->_c.mfc_un.res.bytes;
1712                         sr.wrong_if = c->_c.mfc_un.res.wrong_if;
1713                         rcu_read_unlock();
1714
1715                         if (copy_to_user(arg, &sr, sizeof(sr)))
1716                                 return -EFAULT;
1717                         return 0;
1718                 }
1719                 rcu_read_unlock();
1720                 return -EADDRNOTAVAIL;
1721         default:
1722                 return -ENOIOCTLCMD;
1723         }
1724 }
1725 #endif
1726
1727 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1728 {
1729         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1730         struct net *net = dev_net(dev);
1731         struct mr_table *mrt;
1732         struct vif_device *v;
1733         int ct;
1734
1735         if (event != NETDEV_UNREGISTER)
1736                 return NOTIFY_DONE;
1737
1738         ipmr_for_each_table(mrt, net) {
1739                 v = &mrt->vif_table[0];
1740                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1741                         if (rcu_access_pointer(v->dev) == dev)
1742                                 vif_delete(mrt, ct, 1, NULL);
1743                 }
1744         }
1745         return NOTIFY_DONE;
1746 }
1747
1748 static struct notifier_block ip_mr_notifier = {
1749         .notifier_call = ipmr_device_event,
1750 };
1751
1752 /* Encapsulate a packet by attaching a valid IPIP header to it.
1753  * This avoids tunnel drivers and other mess and gives us the speed so
1754  * important for multicast video.
1755  */
1756 static void ip_encap(struct net *net, struct sk_buff *skb,
1757                      __be32 saddr, __be32 daddr)
1758 {
1759         struct iphdr *iph;
1760         const struct iphdr *old_iph = ip_hdr(skb);
1761
1762         skb_push(skb, sizeof(struct iphdr));
1763         skb->transport_header = skb->network_header;
1764         skb_reset_network_header(skb);
1765         iph = ip_hdr(skb);
1766
1767         iph->version    =       4;
1768         iph->tos        =       old_iph->tos;
1769         iph->ttl        =       old_iph->ttl;
1770         iph->frag_off   =       0;
1771         iph->daddr      =       daddr;
1772         iph->saddr      =       saddr;
1773         iph->protocol   =       IPPROTO_IPIP;
1774         iph->ihl        =       5;
1775         iph->tot_len    =       htons(skb->len);
1776         ip_select_ident(net, skb, NULL);
1777         ip_send_check(iph);
1778
1779         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1780         nf_reset_ct(skb);
1781 }
1782
1783 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1784                                       struct sk_buff *skb)
1785 {
1786         struct ip_options *opt = &(IPCB(skb)->opt);
1787
1788         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1789         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1790
1791         if (unlikely(opt->optlen))
1792                 ip_forward_options(skb);
1793
1794         return dst_output(net, sk, skb);
1795 }
1796
1797 #ifdef CONFIG_NET_SWITCHDEV
1798 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1799                                    int in_vifi, int out_vifi)
1800 {
1801         struct vif_device *out_vif = &mrt->vif_table[out_vifi];
1802         struct vif_device *in_vif = &mrt->vif_table[in_vifi];
1803
1804         if (!skb->offload_l3_fwd_mark)
1805                 return false;
1806         if (!out_vif->dev_parent_id.id_len || !in_vif->dev_parent_id.id_len)
1807                 return false;
1808         return netdev_phys_item_id_same(&out_vif->dev_parent_id,
1809                                         &in_vif->dev_parent_id);
1810 }
1811 #else
1812 static bool ipmr_forward_offloaded(struct sk_buff *skb, struct mr_table *mrt,
1813                                    int in_vifi, int out_vifi)
1814 {
1815         return false;
1816 }
1817 #endif
1818
1819 /* Processing handlers for ipmr_forward, under rcu_read_lock() */
1820
1821 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1822                             int in_vifi, struct sk_buff *skb, int vifi)
1823 {
1824         const struct iphdr *iph = ip_hdr(skb);
1825         struct vif_device *vif = &mrt->vif_table[vifi];
1826         struct net_device *vif_dev;
1827         struct net_device *dev;
1828         struct rtable *rt;
1829         struct flowi4 fl4;
1830         int    encap = 0;
1831
1832         vif_dev = vif_dev_read(vif);
1833         if (!vif_dev)
1834                 goto out_free;
1835
1836         if (vif->flags & VIFF_REGISTER) {
1837                 WRITE_ONCE(vif->pkt_out, vif->pkt_out + 1);
1838                 WRITE_ONCE(vif->bytes_out, vif->bytes_out + skb->len);
1839                 vif_dev->stats.tx_bytes += skb->len;
1840                 vif_dev->stats.tx_packets++;
1841                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1842                 goto out_free;
1843         }
1844
1845         if (ipmr_forward_offloaded(skb, mrt, in_vifi, vifi))
1846                 goto out_free;
1847
1848         if (vif->flags & VIFF_TUNNEL) {
1849                 rt = ip_route_output_ports(net, &fl4, NULL,
1850                                            vif->remote, vif->local,
1851                                            0, 0,
1852                                            IPPROTO_IPIP,
1853                                            RT_TOS(iph->tos), vif->link);
1854                 if (IS_ERR(rt))
1855                         goto out_free;
1856                 encap = sizeof(struct iphdr);
1857         } else {
1858                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1859                                            0, 0,
1860                                            IPPROTO_IPIP,
1861                                            RT_TOS(iph->tos), vif->link);
1862                 if (IS_ERR(rt))
1863                         goto out_free;
1864         }
1865
1866         dev = rt->dst.dev;
1867
1868         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1869                 /* Do not fragment multicasts. Alas, IPv4 does not
1870                  * allow to send ICMP, so that packets will disappear
1871                  * to blackhole.
1872                  */
1873                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1874                 ip_rt_put(rt);
1875                 goto out_free;
1876         }
1877
1878         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1879
1880         if (skb_cow(skb, encap)) {
1881                 ip_rt_put(rt);
1882                 goto out_free;
1883         }
1884
1885         WRITE_ONCE(vif->pkt_out, vif->pkt_out + 1);
1886         WRITE_ONCE(vif->bytes_out, vif->bytes_out + skb->len);
1887
1888         skb_dst_drop(skb);
1889         skb_dst_set(skb, &rt->dst);
1890         ip_decrease_ttl(ip_hdr(skb));
1891
1892         /* FIXME: forward and output firewalls used to be called here.
1893          * What do we do with netfilter? -- RR
1894          */
1895         if (vif->flags & VIFF_TUNNEL) {
1896                 ip_encap(net, skb, vif->local, vif->remote);
1897                 /* FIXME: extra output firewall step used to be here. --RR */
1898                 vif_dev->stats.tx_packets++;
1899                 vif_dev->stats.tx_bytes += skb->len;
1900         }
1901
1902         IPCB(skb)->flags |= IPSKB_FORWARDED;
1903
1904         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1905          * not only before forwarding, but after forwarding on all output
1906          * interfaces. It is clear, if mrouter runs a multicasting
1907          * program, it should receive packets not depending to what interface
1908          * program is joined.
1909          * If we will not make it, the program will have to join on all
1910          * interfaces. On the other hand, multihoming host (or router, but
1911          * not mrouter) cannot join to more than one interface - it will
1912          * result in receiving multiple packets.
1913          */
1914         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1915                 net, NULL, skb, skb->dev, dev,
1916                 ipmr_forward_finish);
1917         return;
1918
1919 out_free:
1920         kfree_skb(skb);
1921 }
1922
1923 /* Called with mrt_lock or rcu_read_lock() */
1924 static int ipmr_find_vif(const struct mr_table *mrt, struct net_device *dev)
1925 {
1926         int ct;
1927         /* Pairs with WRITE_ONCE() in vif_delete()/vif_add() */
1928         for (ct = READ_ONCE(mrt->maxvif) - 1; ct >= 0; ct--) {
1929                 if (rcu_access_pointer(mrt->vif_table[ct].dev) == dev)
1930                         break;
1931         }
1932         return ct;
1933 }
1934
1935 /* "local" means that we should preserve one skb (for local delivery) */
1936 /* Called uner rcu_read_lock() */
1937 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1938                           struct net_device *dev, struct sk_buff *skb,
1939                           struct mfc_cache *c, int local)
1940 {
1941         int true_vifi = ipmr_find_vif(mrt, dev);
1942         int psend = -1;
1943         int vif, ct;
1944
1945         vif = c->_c.mfc_parent;
1946         c->_c.mfc_un.res.pkt++;
1947         c->_c.mfc_un.res.bytes += skb->len;
1948         c->_c.mfc_un.res.lastuse = jiffies;
1949
1950         if (c->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1951                 struct mfc_cache *cache_proxy;
1952
1953                 /* For an (*,G) entry, we only check that the incoming
1954                  * interface is part of the static tree.
1955                  */
1956                 cache_proxy = mr_mfc_find_any_parent(mrt, vif);
1957                 if (cache_proxy &&
1958                     cache_proxy->_c.mfc_un.res.ttls[true_vifi] < 255)
1959                         goto forward;
1960         }
1961
1962         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1963         if (rcu_access_pointer(mrt->vif_table[vif].dev) != dev) {
1964                 if (rt_is_output_route(skb_rtable(skb))) {
1965                         /* It is our own packet, looped back.
1966                          * Very complicated situation...
1967                          *
1968                          * The best workaround until routing daemons will be
1969                          * fixed is not to redistribute packet, if it was
1970                          * send through wrong interface. It means, that
1971                          * multicast applications WILL NOT work for
1972                          * (S,G), which have default multicast route pointing
1973                          * to wrong oif. In any case, it is not a good
1974                          * idea to use multicasting applications on router.
1975                          */
1976                         goto dont_forward;
1977                 }
1978
1979                 c->_c.mfc_un.res.wrong_if++;
1980
1981                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1982                     /* pimsm uses asserts, when switching from RPT to SPT,
1983                      * so that we cannot check that packet arrived on an oif.
1984                      * It is bad, but otherwise we would need to move pretty
1985                      * large chunk of pimd to kernel. Ough... --ANK
1986                      */
1987                     (mrt->mroute_do_pim ||
1988                      c->_c.mfc_un.res.ttls[true_vifi] < 255) &&
1989                     time_after(jiffies,
1990                                c->_c.mfc_un.res.last_assert +
1991                                MFC_ASSERT_THRESH)) {
1992                         c->_c.mfc_un.res.last_assert = jiffies;
1993                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1994                         if (mrt->mroute_do_wrvifwhole)
1995                                 ipmr_cache_report(mrt, skb, true_vifi,
1996                                                   IGMPMSG_WRVIFWHOLE);
1997                 }
1998                 goto dont_forward;
1999         }
2000
2001 forward:
2002         WRITE_ONCE(mrt->vif_table[vif].pkt_in,
2003                    mrt->vif_table[vif].pkt_in + 1);
2004         WRITE_ONCE(mrt->vif_table[vif].bytes_in,
2005                    mrt->vif_table[vif].bytes_in + skb->len);
2006
2007         /* Forward the frame */
2008         if (c->mfc_origin == htonl(INADDR_ANY) &&
2009             c->mfc_mcastgrp == htonl(INADDR_ANY)) {
2010                 if (true_vifi >= 0 &&
2011                     true_vifi != c->_c.mfc_parent &&
2012                     ip_hdr(skb)->ttl >
2013                                 c->_c.mfc_un.res.ttls[c->_c.mfc_parent]) {
2014                         /* It's an (*,*) entry and the packet is not coming from
2015                          * the upstream: forward the packet to the upstream
2016                          * only.
2017                          */
2018                         psend = c->_c.mfc_parent;
2019                         goto last_forward;
2020                 }
2021                 goto dont_forward;
2022         }
2023         for (ct = c->_c.mfc_un.res.maxvif - 1;
2024              ct >= c->_c.mfc_un.res.minvif; ct--) {
2025                 /* For (*,G) entry, don't forward to the incoming interface */
2026                 if ((c->mfc_origin != htonl(INADDR_ANY) ||
2027                      ct != true_vifi) &&
2028                     ip_hdr(skb)->ttl > c->_c.mfc_un.res.ttls[ct]) {
2029                         if (psend != -1) {
2030                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2031
2032                                 if (skb2)
2033                                         ipmr_queue_xmit(net, mrt, true_vifi,
2034                                                         skb2, psend);
2035                         }
2036                         psend = ct;
2037                 }
2038         }
2039 last_forward:
2040         if (psend != -1) {
2041                 if (local) {
2042                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2043
2044                         if (skb2)
2045                                 ipmr_queue_xmit(net, mrt, true_vifi, skb2,
2046                                                 psend);
2047                 } else {
2048                         ipmr_queue_xmit(net, mrt, true_vifi, skb, psend);
2049                         return;
2050                 }
2051         }
2052
2053 dont_forward:
2054         if (!local)
2055                 kfree_skb(skb);
2056 }
2057
2058 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
2059 {
2060         struct rtable *rt = skb_rtable(skb);
2061         struct iphdr *iph = ip_hdr(skb);
2062         struct flowi4 fl4 = {
2063                 .daddr = iph->daddr,
2064                 .saddr = iph->saddr,
2065                 .flowi4_tos = RT_TOS(iph->tos),
2066                 .flowi4_oif = (rt_is_output_route(rt) ?
2067                                skb->dev->ifindex : 0),
2068                 .flowi4_iif = (rt_is_output_route(rt) ?
2069                                LOOPBACK_IFINDEX :
2070                                skb->dev->ifindex),
2071                 .flowi4_mark = skb->mark,
2072         };
2073         struct mr_table *mrt;
2074         int err;
2075
2076         err = ipmr_fib_lookup(net, &fl4, &mrt);
2077         if (err)
2078                 return ERR_PTR(err);
2079         return mrt;
2080 }
2081
2082 /* Multicast packets for forwarding arrive here
2083  * Called with rcu_read_lock();
2084  */
2085 int ip_mr_input(struct sk_buff *skb)
2086 {
2087         struct mfc_cache *cache;
2088         struct net *net = dev_net(skb->dev);
2089         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
2090         struct mr_table *mrt;
2091         struct net_device *dev;
2092
2093         /* skb->dev passed in is the loX master dev for vrfs.
2094          * As there are no vifs associated with loopback devices,
2095          * get the proper interface that does have a vif associated with it.
2096          */
2097         dev = skb->dev;
2098         if (netif_is_l3_master(skb->dev)) {
2099                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2100                 if (!dev) {
2101                         kfree_skb(skb);
2102                         return -ENODEV;
2103                 }
2104         }
2105
2106         /* Packet is looped back after forward, it should not be
2107          * forwarded second time, but still can be delivered locally.
2108          */
2109         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2110                 goto dont_forward;
2111
2112         mrt = ipmr_rt_fib_lookup(net, skb);
2113         if (IS_ERR(mrt)) {
2114                 kfree_skb(skb);
2115                 return PTR_ERR(mrt);
2116         }
2117         if (!local) {
2118                 if (IPCB(skb)->opt.router_alert) {
2119                         if (ip_call_ra_chain(skb))
2120                                 return 0;
2121                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2122                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2123                          * Cisco IOS <= 11.2(8)) do not put router alert
2124                          * option to IGMP packets destined to routable
2125                          * groups. It is very bad, because it means
2126                          * that we can forward NO IGMP messages.
2127                          */
2128                         struct sock *mroute_sk;
2129
2130                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2131                         if (mroute_sk) {
2132                                 nf_reset_ct(skb);
2133                                 raw_rcv(mroute_sk, skb);
2134                                 return 0;
2135                         }
2136                 }
2137         }
2138
2139         /* already under rcu_read_lock() */
2140         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2141         if (!cache) {
2142                 int vif = ipmr_find_vif(mrt, dev);
2143
2144                 if (vif >= 0)
2145                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2146                                                     vif);
2147         }
2148
2149         /* No usable cache entry */
2150         if (!cache) {
2151                 int vif;
2152
2153                 if (local) {
2154                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2155                         ip_local_deliver(skb);
2156                         if (!skb2)
2157                                 return -ENOBUFS;
2158                         skb = skb2;
2159                 }
2160
2161                 vif = ipmr_find_vif(mrt, dev);
2162                 if (vif >= 0)
2163                         return ipmr_cache_unresolved(mrt, vif, skb, dev);
2164                 kfree_skb(skb);
2165                 return -ENODEV;
2166         }
2167
2168         ip_mr_forward(net, mrt, dev, skb, cache, local);
2169
2170         if (local)
2171                 return ip_local_deliver(skb);
2172
2173         return 0;
2174
2175 dont_forward:
2176         if (local)
2177                 return ip_local_deliver(skb);
2178         kfree_skb(skb);
2179         return 0;
2180 }
2181
2182 #ifdef CONFIG_IP_PIMSM_V1
2183 /* Handle IGMP messages of PIMv1 */
2184 int pim_rcv_v1(struct sk_buff *skb)
2185 {
2186         struct igmphdr *pim;
2187         struct net *net = dev_net(skb->dev);
2188         struct mr_table *mrt;
2189
2190         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2191                 goto drop;
2192
2193         pim = igmp_hdr(skb);
2194
2195         mrt = ipmr_rt_fib_lookup(net, skb);
2196         if (IS_ERR(mrt))
2197                 goto drop;
2198         if (!mrt->mroute_do_pim ||
2199             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2200                 goto drop;
2201
2202         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2203 drop:
2204                 kfree_skb(skb);
2205         }
2206         return 0;
2207 }
2208 #endif
2209
2210 #ifdef CONFIG_IP_PIMSM_V2
2211 static int pim_rcv(struct sk_buff *skb)
2212 {
2213         struct pimreghdr *pim;
2214         struct net *net = dev_net(skb->dev);
2215         struct mr_table *mrt;
2216
2217         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2218                 goto drop;
2219
2220         pim = (struct pimreghdr *)skb_transport_header(skb);
2221         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2222             (pim->flags & PIM_NULL_REGISTER) ||
2223             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2224              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2225                 goto drop;
2226
2227         mrt = ipmr_rt_fib_lookup(net, skb);
2228         if (IS_ERR(mrt))
2229                 goto drop;
2230         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2231 drop:
2232                 kfree_skb(skb);
2233         }
2234         return 0;
2235 }
2236 #endif
2237
2238 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2239                    __be32 saddr, __be32 daddr,
2240                    struct rtmsg *rtm, u32 portid)
2241 {
2242         struct mfc_cache *cache;
2243         struct mr_table *mrt;
2244         int err;
2245
2246         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2247         if (!mrt)
2248                 return -ENOENT;
2249
2250         rcu_read_lock();
2251         cache = ipmr_cache_find(mrt, saddr, daddr);
2252         if (!cache && skb->dev) {
2253                 int vif = ipmr_find_vif(mrt, skb->dev);
2254
2255                 if (vif >= 0)
2256                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2257         }
2258         if (!cache) {
2259                 struct sk_buff *skb2;
2260                 struct iphdr *iph;
2261                 struct net_device *dev;
2262                 int vif = -1;
2263
2264                 dev = skb->dev;
2265                 if (dev)
2266                         vif = ipmr_find_vif(mrt, dev);
2267                 if (vif < 0) {
2268                         rcu_read_unlock();
2269                         return -ENODEV;
2270                 }
2271
2272                 skb2 = skb_realloc_headroom(skb, sizeof(struct iphdr));
2273                 if (!skb2) {
2274                         rcu_read_unlock();
2275                         return -ENOMEM;
2276                 }
2277
2278                 NETLINK_CB(skb2).portid = portid;
2279                 skb_push(skb2, sizeof(struct iphdr));
2280                 skb_reset_network_header(skb2);
2281                 iph = ip_hdr(skb2);
2282                 iph->ihl = sizeof(struct iphdr) >> 2;
2283                 iph->saddr = saddr;
2284                 iph->daddr = daddr;
2285                 iph->version = 0;
2286                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2287                 rcu_read_unlock();
2288                 return err;
2289         }
2290
2291         err = mr_fill_mroute(mrt, skb, &cache->_c, rtm);
2292         rcu_read_unlock();
2293         return err;
2294 }
2295
2296 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2297                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2298                             int flags)
2299 {
2300         struct nlmsghdr *nlh;
2301         struct rtmsg *rtm;
2302         int err;
2303
2304         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2305         if (!nlh)
2306                 return -EMSGSIZE;
2307
2308         rtm = nlmsg_data(nlh);
2309         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2310         rtm->rtm_dst_len  = 32;
2311         rtm->rtm_src_len  = 32;
2312         rtm->rtm_tos      = 0;
2313         rtm->rtm_table    = mrt->id;
2314         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2315                 goto nla_put_failure;
2316         rtm->rtm_type     = RTN_MULTICAST;
2317         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2318         if (c->_c.mfc_flags & MFC_STATIC)
2319                 rtm->rtm_protocol = RTPROT_STATIC;
2320         else
2321                 rtm->rtm_protocol = RTPROT_MROUTED;
2322         rtm->rtm_flags    = 0;
2323
2324         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2325             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2326                 goto nla_put_failure;
2327         err = mr_fill_mroute(mrt, skb, &c->_c, rtm);
2328         /* do not break the dump if cache is unresolved */
2329         if (err < 0 && err != -ENOENT)
2330                 goto nla_put_failure;
2331
2332         nlmsg_end(skb, nlh);
2333         return 0;
2334
2335 nla_put_failure:
2336         nlmsg_cancel(skb, nlh);
2337         return -EMSGSIZE;
2338 }
2339
2340 static int _ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2341                              u32 portid, u32 seq, struct mr_mfc *c, int cmd,
2342                              int flags)
2343 {
2344         return ipmr_fill_mroute(mrt, skb, portid, seq, (struct mfc_cache *)c,
2345                                 cmd, flags);
2346 }
2347
2348 static size_t mroute_msgsize(bool unresolved, int maxvif)
2349 {
2350         size_t len =
2351                 NLMSG_ALIGN(sizeof(struct rtmsg))
2352                 + nla_total_size(4)     /* RTA_TABLE */
2353                 + nla_total_size(4)     /* RTA_SRC */
2354                 + nla_total_size(4)     /* RTA_DST */
2355                 ;
2356
2357         if (!unresolved)
2358                 len = len
2359                       + nla_total_size(4)       /* RTA_IIF */
2360                       + nla_total_size(0)       /* RTA_MULTIPATH */
2361                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2362                                                 /* RTA_MFC_STATS */
2363                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2364                 ;
2365
2366         return len;
2367 }
2368
2369 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2370                                  int cmd)
2371 {
2372         struct net *net = read_pnet(&mrt->net);
2373         struct sk_buff *skb;
2374         int err = -ENOBUFS;
2375
2376         skb = nlmsg_new(mroute_msgsize(mfc->_c.mfc_parent >= MAXVIFS,
2377                                        mrt->maxvif),
2378                         GFP_ATOMIC);
2379         if (!skb)
2380                 goto errout;
2381
2382         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2383         if (err < 0)
2384                 goto errout;
2385
2386         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2387         return;
2388
2389 errout:
2390         kfree_skb(skb);
2391         if (err < 0)
2392                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2393 }
2394
2395 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2396 {
2397         size_t len =
2398                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2399                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2400                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2401                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2402                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2403                 + nla_total_size(4)     /* IPMRA_CREPORT_TABLE */
2404                                         /* IPMRA_CREPORT_PKT */
2405                 + nla_total_size(payloadlen)
2406                 ;
2407
2408         return len;
2409 }
2410
2411 static void igmpmsg_netlink_event(const struct mr_table *mrt, struct sk_buff *pkt)
2412 {
2413         struct net *net = read_pnet(&mrt->net);
2414         struct nlmsghdr *nlh;
2415         struct rtgenmsg *rtgenm;
2416         struct igmpmsg *msg;
2417         struct sk_buff *skb;
2418         struct nlattr *nla;
2419         int payloadlen;
2420
2421         payloadlen = pkt->len - sizeof(struct igmpmsg);
2422         msg = (struct igmpmsg *)skb_network_header(pkt);
2423
2424         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2425         if (!skb)
2426                 goto errout;
2427
2428         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2429                         sizeof(struct rtgenmsg), 0);
2430         if (!nlh)
2431                 goto errout;
2432         rtgenm = nlmsg_data(nlh);
2433         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2434         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2435             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif | (msg->im_vif_hi << 8)) ||
2436             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2437                             msg->im_src.s_addr) ||
2438             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2439                             msg->im_dst.s_addr) ||
2440             nla_put_u32(skb, IPMRA_CREPORT_TABLE, mrt->id))
2441                 goto nla_put_failure;
2442
2443         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2444         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2445                                   nla_data(nla), payloadlen))
2446                 goto nla_put_failure;
2447
2448         nlmsg_end(skb, nlh);
2449
2450         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2451         return;
2452
2453 nla_put_failure:
2454         nlmsg_cancel(skb, nlh);
2455 errout:
2456         kfree_skb(skb);
2457         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2458 }
2459
2460 static int ipmr_rtm_valid_getroute_req(struct sk_buff *skb,
2461                                        const struct nlmsghdr *nlh,
2462                                        struct nlattr **tb,
2463                                        struct netlink_ext_ack *extack)
2464 {
2465         struct rtmsg *rtm;
2466         int i, err;
2467
2468         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
2469                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for multicast route get request");
2470                 return -EINVAL;
2471         }
2472
2473         if (!netlink_strict_get_check(skb))
2474                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
2475                                               rtm_ipv4_policy, extack);
2476
2477         rtm = nlmsg_data(nlh);
2478         if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) ||
2479             (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) ||
2480             rtm->rtm_tos || rtm->rtm_table || rtm->rtm_protocol ||
2481             rtm->rtm_scope || rtm->rtm_type || rtm->rtm_flags) {
2482                 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for multicast route get request");
2483                 return -EINVAL;
2484         }
2485
2486         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
2487                                             rtm_ipv4_policy, extack);
2488         if (err)
2489                 return err;
2490
2491         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
2492             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
2493                 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4");
2494                 return -EINVAL;
2495         }
2496
2497         for (i = 0; i <= RTA_MAX; i++) {
2498                 if (!tb[i])
2499                         continue;
2500
2501                 switch (i) {
2502                 case RTA_SRC:
2503                 case RTA_DST:
2504                 case RTA_TABLE:
2505                         break;
2506                 default:
2507                         NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in multicast route get request");
2508                         return -EINVAL;
2509                 }
2510         }
2511
2512         return 0;
2513 }
2514
2515 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2516                              struct netlink_ext_ack *extack)
2517 {
2518         struct net *net = sock_net(in_skb->sk);
2519         struct nlattr *tb[RTA_MAX + 1];
2520         struct sk_buff *skb = NULL;
2521         struct mfc_cache *cache;
2522         struct mr_table *mrt;
2523         __be32 src, grp;
2524         u32 tableid;
2525         int err;
2526
2527         err = ipmr_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
2528         if (err < 0)
2529                 goto errout;
2530
2531         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2532         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2533         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2534
2535         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2536         if (!mrt) {
2537                 err = -ENOENT;
2538                 goto errout_free;
2539         }
2540
2541         /* entries are added/deleted only under RTNL */
2542         rcu_read_lock();
2543         cache = ipmr_cache_find(mrt, src, grp);
2544         rcu_read_unlock();
2545         if (!cache) {
2546                 err = -ENOENT;
2547                 goto errout_free;
2548         }
2549
2550         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2551         if (!skb) {
2552                 err = -ENOBUFS;
2553                 goto errout_free;
2554         }
2555
2556         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2557                                nlh->nlmsg_seq, cache,
2558                                RTM_NEWROUTE, 0);
2559         if (err < 0)
2560                 goto errout_free;
2561
2562         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2563
2564 errout:
2565         return err;
2566
2567 errout_free:
2568         kfree_skb(skb);
2569         goto errout;
2570 }
2571
2572 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2573 {
2574         struct fib_dump_filter filter = {};
2575         int err;
2576
2577         if (cb->strict_check) {
2578                 err = ip_valid_fib_dump_req(sock_net(skb->sk), cb->nlh,
2579                                             &filter, cb);
2580                 if (err < 0)
2581                         return err;
2582         }
2583
2584         if (filter.table_id) {
2585                 struct mr_table *mrt;
2586
2587                 mrt = ipmr_get_table(sock_net(skb->sk), filter.table_id);
2588                 if (!mrt) {
2589                         if (rtnl_msg_family(cb->nlh) != RTNL_FAMILY_IPMR)
2590                                 return skb->len;
2591
2592                         NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
2593                         return -ENOENT;
2594                 }
2595                 err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
2596                                     &mfc_unres_lock, &filter);
2597                 return skb->len ? : err;
2598         }
2599
2600         return mr_rtm_dumproute(skb, cb, ipmr_mr_table_iter,
2601                                 _ipmr_fill_mroute, &mfc_unres_lock, &filter);
2602 }
2603
2604 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2605         [RTA_SRC]       = { .type = NLA_U32 },
2606         [RTA_DST]       = { .type = NLA_U32 },
2607         [RTA_IIF]       = { .type = NLA_U32 },
2608         [RTA_TABLE]     = { .type = NLA_U32 },
2609         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2610 };
2611
2612 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2613 {
2614         switch (rtm_protocol) {
2615         case RTPROT_STATIC:
2616         case RTPROT_MROUTED:
2617                 return true;
2618         }
2619         return false;
2620 }
2621
2622 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2623 {
2624         struct rtnexthop *rtnh = nla_data(nla);
2625         int remaining = nla_len(nla), vifi = 0;
2626
2627         while (rtnh_ok(rtnh, remaining)) {
2628                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2629                 if (++vifi == MAXVIFS)
2630                         break;
2631                 rtnh = rtnh_next(rtnh, &remaining);
2632         }
2633
2634         return remaining > 0 ? -EINVAL : vifi;
2635 }
2636
2637 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2638 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2639                             struct mfcctl *mfcc, int *mrtsock,
2640                             struct mr_table **mrtret,
2641                             struct netlink_ext_ack *extack)
2642 {
2643         struct net_device *dev = NULL;
2644         u32 tblid = RT_TABLE_DEFAULT;
2645         struct mr_table *mrt;
2646         struct nlattr *attr;
2647         struct rtmsg *rtm;
2648         int ret, rem;
2649
2650         ret = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
2651                                         rtm_ipmr_policy, extack);
2652         if (ret < 0)
2653                 goto out;
2654         rtm = nlmsg_data(nlh);
2655
2656         ret = -EINVAL;
2657         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2658             rtm->rtm_type != RTN_MULTICAST ||
2659             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2660             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2661                 goto out;
2662
2663         memset(mfcc, 0, sizeof(*mfcc));
2664         mfcc->mfcc_parent = -1;
2665         ret = 0;
2666         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2667                 switch (nla_type(attr)) {
2668                 case RTA_SRC:
2669                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2670                         break;
2671                 case RTA_DST:
2672                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2673                         break;
2674                 case RTA_IIF:
2675                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2676                         if (!dev) {
2677                                 ret = -ENODEV;
2678                                 goto out;
2679                         }
2680                         break;
2681                 case RTA_MULTIPATH:
2682                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2683                                 ret = -EINVAL;
2684                                 goto out;
2685                         }
2686                         break;
2687                 case RTA_PREFSRC:
2688                         ret = 1;
2689                         break;
2690                 case RTA_TABLE:
2691                         tblid = nla_get_u32(attr);
2692                         break;
2693                 }
2694         }
2695         mrt = ipmr_get_table(net, tblid);
2696         if (!mrt) {
2697                 ret = -ENOENT;
2698                 goto out;
2699         }
2700         *mrtret = mrt;
2701         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2702         if (dev)
2703                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2704
2705 out:
2706         return ret;
2707 }
2708
2709 /* takes care of both newroute and delroute */
2710 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2711                           struct netlink_ext_ack *extack)
2712 {
2713         struct net *net = sock_net(skb->sk);
2714         int ret, mrtsock, parent;
2715         struct mr_table *tbl;
2716         struct mfcctl mfcc;
2717
2718         mrtsock = 0;
2719         tbl = NULL;
2720         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2721         if (ret < 0)
2722                 return ret;
2723
2724         parent = ret ? mfcc.mfcc_parent : -1;
2725         if (nlh->nlmsg_type == RTM_NEWROUTE)
2726                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2727         else
2728                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2729 }
2730
2731 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2732 {
2733         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2734
2735         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2736             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2737             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2738                         mrt->mroute_reg_vif_num) ||
2739             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2740                        mrt->mroute_do_assert) ||
2741             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim) ||
2742             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_WRVIFWHOLE,
2743                        mrt->mroute_do_wrvifwhole))
2744                 return false;
2745
2746         return true;
2747 }
2748
2749 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2750 {
2751         struct net_device *vif_dev;
2752         struct nlattr *vif_nest;
2753         struct vif_device *vif;
2754
2755         vif = &mrt->vif_table[vifid];
2756         vif_dev = rtnl_dereference(vif->dev);
2757         /* if the VIF doesn't exist just continue */
2758         if (!vif_dev)
2759                 return true;
2760
2761         vif_nest = nla_nest_start_noflag(skb, IPMRA_VIF);
2762         if (!vif_nest)
2763                 return false;
2764
2765         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif_dev->ifindex) ||
2766             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2767             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2768             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2769                               IPMRA_VIFA_PAD) ||
2770             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2771                               IPMRA_VIFA_PAD) ||
2772             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2773                               IPMRA_VIFA_PAD) ||
2774             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2775                               IPMRA_VIFA_PAD) ||
2776             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2777             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2778                 nla_nest_cancel(skb, vif_nest);
2779                 return false;
2780         }
2781         nla_nest_end(skb, vif_nest);
2782
2783         return true;
2784 }
2785
2786 static int ipmr_valid_dumplink(const struct nlmsghdr *nlh,
2787                                struct netlink_ext_ack *extack)
2788 {
2789         struct ifinfomsg *ifm;
2790
2791         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ifm))) {
2792                 NL_SET_ERR_MSG(extack, "ipv4: Invalid header for ipmr link dump");
2793                 return -EINVAL;
2794         }
2795
2796         if (nlmsg_attrlen(nlh, sizeof(*ifm))) {
2797                 NL_SET_ERR_MSG(extack, "Invalid data after header in ipmr link dump");
2798                 return -EINVAL;
2799         }
2800
2801         ifm = nlmsg_data(nlh);
2802         if (ifm->__ifi_pad || ifm->ifi_type || ifm->ifi_flags ||
2803             ifm->ifi_change || ifm->ifi_index) {
2804                 NL_SET_ERR_MSG(extack, "Invalid values in header for ipmr link dump request");
2805                 return -EINVAL;
2806         }
2807
2808         return 0;
2809 }
2810
2811 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2812 {
2813         struct net *net = sock_net(skb->sk);
2814         struct nlmsghdr *nlh = NULL;
2815         unsigned int t = 0, s_t;
2816         unsigned int e = 0, s_e;
2817         struct mr_table *mrt;
2818
2819         if (cb->strict_check) {
2820                 int err = ipmr_valid_dumplink(cb->nlh, cb->extack);
2821
2822                 if (err < 0)
2823                         return err;
2824         }
2825
2826         s_t = cb->args[0];
2827         s_e = cb->args[1];
2828
2829         ipmr_for_each_table(mrt, net) {
2830                 struct nlattr *vifs, *af;
2831                 struct ifinfomsg *hdr;
2832                 u32 i;
2833
2834                 if (t < s_t)
2835                         goto skip_table;
2836                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2837                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2838                                 sizeof(*hdr), NLM_F_MULTI);
2839                 if (!nlh)
2840                         break;
2841
2842                 hdr = nlmsg_data(nlh);
2843                 memset(hdr, 0, sizeof(*hdr));
2844                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2845
2846                 af = nla_nest_start_noflag(skb, IFLA_AF_SPEC);
2847                 if (!af) {
2848                         nlmsg_cancel(skb, nlh);
2849                         goto out;
2850                 }
2851
2852                 if (!ipmr_fill_table(mrt, skb)) {
2853                         nlmsg_cancel(skb, nlh);
2854                         goto out;
2855                 }
2856
2857                 vifs = nla_nest_start_noflag(skb, IPMRA_TABLE_VIFS);
2858                 if (!vifs) {
2859                         nla_nest_end(skb, af);
2860                         nlmsg_end(skb, nlh);
2861                         goto out;
2862                 }
2863                 for (i = 0; i < mrt->maxvif; i++) {
2864                         if (e < s_e)
2865                                 goto skip_entry;
2866                         if (!ipmr_fill_vif(mrt, i, skb)) {
2867                                 nla_nest_end(skb, vifs);
2868                                 nla_nest_end(skb, af);
2869                                 nlmsg_end(skb, nlh);
2870                                 goto out;
2871                         }
2872 skip_entry:
2873                         e++;
2874                 }
2875                 s_e = 0;
2876                 e = 0;
2877                 nla_nest_end(skb, vifs);
2878                 nla_nest_end(skb, af);
2879                 nlmsg_end(skb, nlh);
2880 skip_table:
2881                 t++;
2882         }
2883
2884 out:
2885         cb->args[1] = e;
2886         cb->args[0] = t;
2887
2888         return skb->len;
2889 }
2890
2891 #ifdef CONFIG_PROC_FS
2892 /* The /proc interfaces to multicast routing :
2893  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2894  */
2895
2896 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2897         __acquires(RCU)
2898 {
2899         struct mr_vif_iter *iter = seq->private;
2900         struct net *net = seq_file_net(seq);
2901         struct mr_table *mrt;
2902
2903         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2904         if (!mrt)
2905                 return ERR_PTR(-ENOENT);
2906
2907         iter->mrt = mrt;
2908
2909         rcu_read_lock();
2910         return mr_vif_seq_start(seq, pos);
2911 }
2912
2913 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2914         __releases(RCU)
2915 {
2916         rcu_read_unlock();
2917 }
2918
2919 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2920 {
2921         struct mr_vif_iter *iter = seq->private;
2922         struct mr_table *mrt = iter->mrt;
2923
2924         if (v == SEQ_START_TOKEN) {
2925                 seq_puts(seq,
2926                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2927         } else {
2928                 const struct vif_device *vif = v;
2929                 const struct net_device *vif_dev;
2930                 const char *name;
2931
2932                 vif_dev = vif_dev_read(vif);
2933                 name = vif_dev ? vif_dev->name : "none";
2934                 seq_printf(seq,
2935                            "%2td %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2936                            vif - mrt->vif_table,
2937                            name, vif->bytes_in, vif->pkt_in,
2938                            vif->bytes_out, vif->pkt_out,
2939                            vif->flags, vif->local, vif->remote);
2940         }
2941         return 0;
2942 }
2943
2944 static const struct seq_operations ipmr_vif_seq_ops = {
2945         .start = ipmr_vif_seq_start,
2946         .next  = mr_vif_seq_next,
2947         .stop  = ipmr_vif_seq_stop,
2948         .show  = ipmr_vif_seq_show,
2949 };
2950
2951 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2952 {
2953         struct net *net = seq_file_net(seq);
2954         struct mr_table *mrt;
2955
2956         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2957         if (!mrt)
2958                 return ERR_PTR(-ENOENT);
2959
2960         return mr_mfc_seq_start(seq, pos, mrt, &mfc_unres_lock);
2961 }
2962
2963 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2964 {
2965         int n;
2966
2967         if (v == SEQ_START_TOKEN) {
2968                 seq_puts(seq,
2969                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2970         } else {
2971                 const struct mfc_cache *mfc = v;
2972                 const struct mr_mfc_iter *it = seq->private;
2973                 const struct mr_table *mrt = it->mrt;
2974
2975                 seq_printf(seq, "%08X %08X %-3hd",
2976                            (__force u32) mfc->mfc_mcastgrp,
2977                            (__force u32) mfc->mfc_origin,
2978                            mfc->_c.mfc_parent);
2979
2980                 if (it->cache != &mrt->mfc_unres_queue) {
2981                         seq_printf(seq, " %8lu %8lu %8lu",
2982                                    mfc->_c.mfc_un.res.pkt,
2983                                    mfc->_c.mfc_un.res.bytes,
2984                                    mfc->_c.mfc_un.res.wrong_if);
2985                         for (n = mfc->_c.mfc_un.res.minvif;
2986                              n < mfc->_c.mfc_un.res.maxvif; n++) {
2987                                 if (VIF_EXISTS(mrt, n) &&
2988                                     mfc->_c.mfc_un.res.ttls[n] < 255)
2989                                         seq_printf(seq,
2990                                            " %2d:%-3d",
2991                                            n, mfc->_c.mfc_un.res.ttls[n]);
2992                         }
2993                 } else {
2994                         /* unresolved mfc_caches don't contain
2995                          * pkt, bytes and wrong_if values
2996                          */
2997                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2998                 }
2999                 seq_putc(seq, '\n');
3000         }
3001         return 0;
3002 }
3003
3004 static const struct seq_operations ipmr_mfc_seq_ops = {
3005         .start = ipmr_mfc_seq_start,
3006         .next  = mr_mfc_seq_next,
3007         .stop  = mr_mfc_seq_stop,
3008         .show  = ipmr_mfc_seq_show,
3009 };
3010 #endif
3011
3012 #ifdef CONFIG_IP_PIMSM_V2
3013 static const struct net_protocol pim_protocol = {
3014         .handler        =       pim_rcv,
3015 };
3016 #endif
3017
3018 static unsigned int ipmr_seq_read(struct net *net)
3019 {
3020         ASSERT_RTNL();
3021
3022         return net->ipv4.ipmr_seq + ipmr_rules_seq_read(net);
3023 }
3024
3025 static int ipmr_dump(struct net *net, struct notifier_block *nb,
3026                      struct netlink_ext_ack *extack)
3027 {
3028         return mr_dump(net, nb, RTNL_FAMILY_IPMR, ipmr_rules_dump,
3029                        ipmr_mr_table_iter, extack);
3030 }
3031
3032 static const struct fib_notifier_ops ipmr_notifier_ops_template = {
3033         .family         = RTNL_FAMILY_IPMR,
3034         .fib_seq_read   = ipmr_seq_read,
3035         .fib_dump       = ipmr_dump,
3036         .owner          = THIS_MODULE,
3037 };
3038
3039 static int __net_init ipmr_notifier_init(struct net *net)
3040 {
3041         struct fib_notifier_ops *ops;
3042
3043         net->ipv4.ipmr_seq = 0;
3044
3045         ops = fib_notifier_ops_register(&ipmr_notifier_ops_template, net);
3046         if (IS_ERR(ops))
3047                 return PTR_ERR(ops);
3048         net->ipv4.ipmr_notifier_ops = ops;
3049
3050         return 0;
3051 }
3052
3053 static void __net_exit ipmr_notifier_exit(struct net *net)
3054 {
3055         fib_notifier_ops_unregister(net->ipv4.ipmr_notifier_ops);
3056         net->ipv4.ipmr_notifier_ops = NULL;
3057 }
3058
3059 /* Setup for IP multicast routing */
3060 static int __net_init ipmr_net_init(struct net *net)
3061 {
3062         int err;
3063
3064         err = ipmr_notifier_init(net);
3065         if (err)
3066                 goto ipmr_notifier_fail;
3067
3068         err = ipmr_rules_init(net);
3069         if (err < 0)
3070                 goto ipmr_rules_fail;
3071
3072 #ifdef CONFIG_PROC_FS
3073         err = -ENOMEM;
3074         if (!proc_create_net("ip_mr_vif", 0, net->proc_net, &ipmr_vif_seq_ops,
3075                         sizeof(struct mr_vif_iter)))
3076                 goto proc_vif_fail;
3077         if (!proc_create_net("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_seq_ops,
3078                         sizeof(struct mr_mfc_iter)))
3079                 goto proc_cache_fail;
3080 #endif
3081         return 0;
3082
3083 #ifdef CONFIG_PROC_FS
3084 proc_cache_fail:
3085         remove_proc_entry("ip_mr_vif", net->proc_net);
3086 proc_vif_fail:
3087         rtnl_lock();
3088         ipmr_rules_exit(net);
3089         rtnl_unlock();
3090 #endif
3091 ipmr_rules_fail:
3092         ipmr_notifier_exit(net);
3093 ipmr_notifier_fail:
3094         return err;
3095 }
3096
3097 static void __net_exit ipmr_net_exit(struct net *net)
3098 {
3099 #ifdef CONFIG_PROC_FS
3100         remove_proc_entry("ip_mr_cache", net->proc_net);
3101         remove_proc_entry("ip_mr_vif", net->proc_net);
3102 #endif
3103         ipmr_notifier_exit(net);
3104 }
3105
3106 static void __net_exit ipmr_net_exit_batch(struct list_head *net_list)
3107 {
3108         struct net *net;
3109
3110         rtnl_lock();
3111         list_for_each_entry(net, net_list, exit_list)
3112                 ipmr_rules_exit(net);
3113         rtnl_unlock();
3114 }
3115
3116 static struct pernet_operations ipmr_net_ops = {
3117         .init = ipmr_net_init,
3118         .exit = ipmr_net_exit,
3119         .exit_batch = ipmr_net_exit_batch,
3120 };
3121
3122 int __init ip_mr_init(void)
3123 {
3124         int err;
3125
3126         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3127                                        sizeof(struct mfc_cache),
3128                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3129                                        NULL);
3130
3131         err = register_pernet_subsys(&ipmr_net_ops);
3132         if (err)
3133                 goto reg_pernet_fail;
3134
3135         err = register_netdevice_notifier(&ip_mr_notifier);
3136         if (err)
3137                 goto reg_notif_fail;
3138 #ifdef CONFIG_IP_PIMSM_V2
3139         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3140                 pr_err("%s: can't add PIM protocol\n", __func__);
3141                 err = -EAGAIN;
3142                 goto add_proto_fail;
3143         }
3144 #endif
3145         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3146                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3147         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3148                       ipmr_rtm_route, NULL, 0);
3149         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3150                       ipmr_rtm_route, NULL, 0);
3151
3152         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3153                       NULL, ipmr_rtm_dumplink, 0);
3154         return 0;
3155
3156 #ifdef CONFIG_IP_PIMSM_V2
3157 add_proto_fail:
3158         unregister_netdevice_notifier(&ip_mr_notifier);
3159 #endif
3160 reg_notif_fail:
3161         unregister_pernet_subsys(&ipmr_net_ops);
3162 reg_pernet_fail:
3163         kmem_cache_destroy(mrt_cachep);
3164         return err;
3165 }