net/ipv4/ipmr.c

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