net/ipv4/ipmr.c

   1 /*
   2  *      IP multicast routing support for mrouted 3.6/3.8
   3  *
   4  *              (c) 1995 Alan Cox, <alan@lxorguk.ukuu.org.uk>
   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/capability.h>
  32 #include <linux/errno.h>
  33 #include <linux/timer.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/sock.h>
  56 #include <net/icmp.h>
  57 #include <net/udp.h>
  58 #include <net/raw.h>
  59 #include <linux/notifier.h>
  60 #include <linux/if_arp.h>
  61 #include <linux/netfilter_ipv4.h>
  62 #include <linux/compat.h>
  63 #include <linux/export.h>
  64 #include <net/ip_tunnels.h>
  65 #include <net/checksum.h>
  66 #include <net/netlink.h>
  67 #include <net/fib_rules.h>
  68 #include <linux/netconf.h>
  69 #include <net/nexthop.h>
  70
  71 #include <linux/nospec.h>
  72
  73 struct ipmr_rule {
  74         struct fib_rule         common;
  75 };
  76
  77 struct ipmr_result {
  78         struct mr_table         *mrt;
  79 };
  80
  81 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  82  * Note that the changes are semaphored via rtnl_lock.
  83  */
  84
  85 static DEFINE_RWLOCK(mrt_lock);
  86
  87 /* Multicast router control variables */
  88
  89 /* Special spinlock for queue of unresolved entries */
  90 static DEFINE_SPINLOCK(mfc_unres_lock);
  91
  92 /* We return to original Alan's scheme. Hash table of resolved
  93  * entries is changed only in process context and protected
  94  * with weak lock mrt_lock. Queue of unresolved entries is protected
  95  * with strong spinlock mfc_unres_lock.
  96  *
  97  * In this case data path is free of exclusive locks at all.
  98  */
  99
 100 static struct kmem_cache *mrt_cachep __read_mostly;
 101
 102 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 103 static void ipmr_free_table(struct mr_table *mrt);
 104
 105 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 106                           struct net_device *dev, struct sk_buff *skb,
 107                           struct mfc_cache *cache, int local);
 108 static int ipmr_cache_report(struct mr_table *mrt,
 109                              struct sk_buff *pkt, vifi_t vifi, int assert);
 110 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 111                               struct mfc_cache *c, struct rtmsg *rtm);
 112 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 113                                  int cmd);
 114 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt);
 115 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 116 static void ipmr_expire_process(unsigned long arg);
 117
 118 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 119 #define ipmr_for_each_table(mrt, net) \
 120         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 121
 122 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 123 {
 124         struct mr_table *mrt;
 125
 126         ipmr_for_each_table(mrt, net) {
 127                 if (mrt->id == id)
 128                         return mrt;
 129         }
 130         return NULL;
 131 }
 132
 133 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 134                            struct mr_table **mrt)
 135 {
 136         int err;
 137         struct ipmr_result res;
 138         struct fib_lookup_arg arg = {
 139                 .result = &res,
 140                 .flags = FIB_LOOKUP_NOREF,
 141         };
 142
 143         /* update flow if oif or iif point to device enslaved to l3mdev */
 144         l3mdev_update_flow(net, flowi4_to_flowi(flp4));
 145
 146         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 147                                flowi4_to_flowi(flp4), 0, &arg);
 148         if (err < 0)
 149                 return err;
 150         *mrt = res.mrt;
 151         return 0;
 152 }
 153
 154 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 155                             int flags, struct fib_lookup_arg *arg)
 156 {
 157         struct ipmr_result *res = arg->result;
 158         struct mr_table *mrt;
 159
 160         switch (rule->action) {
 161         case FR_ACT_TO_TBL:
 162                 break;
 163         case FR_ACT_UNREACHABLE:
 164                 return -ENETUNREACH;
 165         case FR_ACT_PROHIBIT:
 166                 return -EACCES;
 167         case FR_ACT_BLACKHOLE:
 168         default:
 169                 return -EINVAL;
 170         }
 171
 172         arg->table = fib_rule_get_table(rule, arg);
 173
 174         mrt = ipmr_get_table(rule->fr_net, arg->table);
 175         if (!mrt)
 176                 return -EAGAIN;
 177         res->mrt = mrt;
 178         return 0;
 179 }
 180
 181 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 182 {
 183         return 1;
 184 }
 185
 186 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 187         FRA_GENERIC_POLICY,
 188 };
 189
 190 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 191                                struct fib_rule_hdr *frh, struct nlattr **tb)
 192 {
 193         return 0;
 194 }
 195
 196 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 197                              struct nlattr **tb)
 198 {
 199         return 1;
 200 }
 201
 202 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 203                           struct fib_rule_hdr *frh)
 204 {
 205         frh->dst_len = 0;
 206         frh->src_len = 0;
 207         frh->tos     = 0;
 208         return 0;
 209 }
 210
 211 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 212         .family         = RTNL_FAMILY_IPMR,
 213         .rule_size      = sizeof(struct ipmr_rule),
 214         .addr_size      = sizeof(u32),
 215         .action         = ipmr_rule_action,
 216         .match          = ipmr_rule_match,
 217         .configure      = ipmr_rule_configure,
 218         .compare        = ipmr_rule_compare,
 219         .fill           = ipmr_rule_fill,
 220         .nlgroup        = RTNLGRP_IPV4_RULE,
 221         .policy         = ipmr_rule_policy,
 222         .owner          = THIS_MODULE,
 223 };
 224
 225 static int __net_init ipmr_rules_init(struct net *net)
 226 {
 227         struct fib_rules_ops *ops;
 228         struct mr_table *mrt;
 229         int err;
 230
 231         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 232         if (IS_ERR(ops))
 233                 return PTR_ERR(ops);
 234
 235         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 236
 237         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 238         if (IS_ERR(mrt)) {
 239                 err = PTR_ERR(mrt);
 240                 goto err1;
 241         }
 242
 243         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 244         if (err < 0)
 245                 goto err2;
 246
 247         net->ipv4.mr_rules_ops = ops;
 248         return 0;
 249
 250 err2:
 251         rtnl_lock();
 252         ipmr_free_table(mrt);
 253         rtnl_unlock();
 254 err1:
 255         fib_rules_unregister(ops);
 256         return err;
 257 }
 258
 259 static void __net_exit ipmr_rules_exit(struct net *net)
 260 {
 261         struct mr_table *mrt, *next;
 262
 263         rtnl_lock();
 264         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 265                 list_del(&mrt->list);
 266                 ipmr_free_table(mrt);
 267         }
 268         fib_rules_unregister(net->ipv4.mr_rules_ops);
 269         rtnl_unlock();
 270 }
 271 #else
 272 #define ipmr_for_each_table(mrt, net) \
 273         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 274
 275 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 276 {
 277         return net->ipv4.mrt;
 278 }
 279
 280 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 281                            struct mr_table **mrt)
 282 {
 283         *mrt = net->ipv4.mrt;
 284         return 0;
 285 }
 286
 287 static int __net_init ipmr_rules_init(struct net *net)
 288 {
 289         struct mr_table *mrt;
 290
 291         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 292         if (IS_ERR(mrt))
 293                 return PTR_ERR(mrt);
 294         net->ipv4.mrt = mrt;
 295         return 0;
 296 }
 297
 298 static void __net_exit ipmr_rules_exit(struct net *net)
 299 {
 300         rtnl_lock();
 301         ipmr_free_table(net->ipv4.mrt);
 302         net->ipv4.mrt = NULL;
 303         rtnl_unlock();
 304 }
 305 #endif
 306
 307 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 308                                 const void *ptr)
 309 {
 310         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 311         struct mfc_cache *c = (struct mfc_cache *)ptr;
 312
 313         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 314                cmparg->mfc_origin != c->mfc_origin;
 315 }
 316
 317 static const struct rhashtable_params ipmr_rht_params = {
 318         .head_offset = offsetof(struct mfc_cache, mnode),
 319         .key_offset = offsetof(struct mfc_cache, cmparg),
 320         .key_len = sizeof(struct mfc_cache_cmp_arg),
 321         .nelem_hint = 3,
 322         .locks_mul = 1,
 323         .obj_cmpfn = ipmr_hash_cmp,
 324         .automatic_shrinking = true,
 325 };
 326
 327 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 328 {
 329         struct mr_table *mrt;
 330         int err;
 331
 332         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 333         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 334                 return ERR_PTR(-EINVAL);
 335
 336         mrt = ipmr_get_table(net, id);
 337         if (mrt)
 338                 return mrt;
 339
 340         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 341         if (!mrt)
 342                 return ERR_PTR(-ENOMEM);
 343         write_pnet(&mrt->net, net);
 344         mrt->id = id;
 345
 346         err = rhltable_init(&mrt->mfc_hash, &ipmr_rht_params);
 347         if (err) {
 348                 kfree(mrt);
 349                 return ERR_PTR(err);
 350         }
 351         INIT_LIST_HEAD(&mrt->mfc_cache_list);
 352         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 353
 354         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 355                     (unsigned long)mrt);
 356
 357         mrt->mroute_reg_vif_num = -1;
 358 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 359         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 360 #endif
 361         return mrt;
 362 }
 363
 364 static void ipmr_free_table(struct mr_table *mrt)
 365 {
 366         del_timer_sync(&mrt->ipmr_expire_timer);
 367         mroute_clean_tables(mrt, true);
 368         rhltable_destroy(&mrt->mfc_hash);
 369         kfree(mrt);
 370 }
 371
 372 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 373
 374 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 375 {
 376         struct net *net = dev_net(dev);
 377
 378         dev_close(dev);
 379
 380         dev = __dev_get_by_name(net, "tunl0");
 381         if (dev) {
 382                 const struct net_device_ops *ops = dev->netdev_ops;
 383                 struct ifreq ifr;
 384                 struct ip_tunnel_parm p;
 385
 386                 memset(&p, 0, sizeof(p));
 387                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 388                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 389                 p.iph.version = 4;
 390                 p.iph.ihl = 5;
 391                 p.iph.protocol = IPPROTO_IPIP;
 392                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 393                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 394
 395                 if (ops->ndo_do_ioctl) {
 396                         mm_segment_t oldfs = get_fs();
 397
 398                         set_fs(KERNEL_DS);
 399                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 400                         set_fs(oldfs);
 401                 }
 402         }
 403 }
 404
 405 /* Initialize ipmr pimreg/tunnel in_device */
 406 static bool ipmr_init_vif_indev(const struct net_device *dev)
 407 {
 408         struct in_device *in_dev;
 409
 410         ASSERT_RTNL();
 411
 412         in_dev = __in_dev_get_rtnl(dev);
 413         if (!in_dev)
 414                 return false;
 415         ipv4_devconf_setall(in_dev);
 416         neigh_parms_data_state_setall(in_dev->arp_parms);
 417         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 418
 419         return true;
 420 }
 421
 422 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 423 {
 424         struct net_device  *dev;
 425
 426         dev = __dev_get_by_name(net, "tunl0");
 427
 428         if (dev) {
 429                 const struct net_device_ops *ops = dev->netdev_ops;
 430                 int err;
 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                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 448                         set_fs(oldfs);
 449                 } else {
 450                         err = -EOPNOTSUPP;
 451                 }
 452                 dev = NULL;
 453
 454                 if (err == 0 &&
 455                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 456                         dev->flags |= IFF_MULTICAST;
 457                         if (!ipmr_init_vif_indev(dev))
 458                                 goto failure;
 459                         if (dev_open(dev))
 460                                 goto failure;
 461                         dev_hold(dev);
 462                 }
 463         }
 464         return dev;
 465
 466 failure:
 467         unregister_netdevice(dev);
 468         return NULL;
 469 }
 470
 471 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 472 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 473 {
 474         struct net *net = dev_net(dev);
 475         struct mr_table *mrt;
 476         struct flowi4 fl4 = {
 477                 .flowi4_oif     = dev->ifindex,
 478                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 479                 .flowi4_mark    = skb->mark,
 480         };
 481         int err;
 482
 483         err = ipmr_fib_lookup(net, &fl4, &mrt);
 484         if (err < 0) {
 485                 kfree_skb(skb);
 486                 return err;
 487         }
 488
 489         read_lock(&mrt_lock);
 490         dev->stats.tx_bytes += skb->len;
 491         dev->stats.tx_packets++;
 492         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 493         read_unlock(&mrt_lock);
 494         kfree_skb(skb);
 495         return NETDEV_TX_OK;
 496 }
 497
 498 static int reg_vif_get_iflink(const struct net_device *dev)
 499 {
 500         return 0;
 501 }
 502
 503 static const struct net_device_ops reg_vif_netdev_ops = {
 504         .ndo_start_xmit = reg_vif_xmit,
 505         .ndo_get_iflink = reg_vif_get_iflink,
 506 };
 507
 508 static void reg_vif_setup(struct net_device *dev)
 509 {
 510         dev->type               = ARPHRD_PIMREG;
 511         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 512         dev->flags              = IFF_NOARP;
 513         dev->netdev_ops         = &reg_vif_netdev_ops;
 514         dev->needs_free_netdev  = true;
 515         dev->features           |= NETIF_F_NETNS_LOCAL;
 516 }
 517
 518 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 519 {
 520         struct net_device *dev;
 521         char name[IFNAMSIZ];
 522
 523         if (mrt->id == RT_TABLE_DEFAULT)
 524                 sprintf(name, "pimreg");
 525         else
 526                 sprintf(name, "pimreg%u", mrt->id);
 527
 528         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 529
 530         if (!dev)
 531                 return NULL;
 532
 533         dev_net_set(dev, net);
 534
 535         if (register_netdevice(dev)) {
 536                 free_netdev(dev);
 537                 return NULL;
 538         }
 539
 540         if (!ipmr_init_vif_indev(dev))
 541                 goto failure;
 542         if (dev_open(dev))
 543                 goto failure;
 544
 545         dev_hold(dev);
 546
 547         return dev;
 548
 549 failure:
 550         unregister_netdevice(dev);
 551         return NULL;
 552 }
 553
 554 /* called with rcu_read_lock() */
 555 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 556                      unsigned int pimlen)
 557 {
 558         struct net_device *reg_dev = NULL;
 559         struct iphdr *encap;
 560
 561         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 562         /* Check that:
 563          * a. packet is really sent to a multicast group
 564          * b. packet is not a NULL-REGISTER
 565          * c. packet is not truncated
 566          */
 567         if (!ipv4_is_multicast(encap->daddr) ||
 568             encap->tot_len == 0 ||
 569             ntohs(encap->tot_len) + pimlen > skb->len)
 570                 return 1;
 571
 572         read_lock(&mrt_lock);
 573         if (mrt->mroute_reg_vif_num >= 0)
 574                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 575         read_unlock(&mrt_lock);
 576
 577         if (!reg_dev)
 578                 return 1;
 579
 580         skb->mac_header = skb->network_header;
 581         skb_pull(skb, (u8 *)encap - skb->data);
 582         skb_reset_network_header(skb);
 583         skb->protocol = htons(ETH_P_IP);
 584         skb->ip_summed = CHECKSUM_NONE;
 585
 586         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 587
 588         netif_rx(skb);
 589
 590         return NET_RX_SUCCESS;
 591 }
 592 #else
 593 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 594 {
 595         return NULL;
 596 }
 597 #endif
 598
 599 /**
 600  *      vif_delete - Delete a VIF entry
 601  *      @notify: Set to 1, if the caller is a notifier_call
 602  */
 603 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 604                       struct list_head *head)
 605 {
 606         struct vif_device *v;
 607         struct net_device *dev;
 608         struct in_device *in_dev;
 609
 610         if (vifi < 0 || vifi >= mrt->maxvif)
 611                 return -EADDRNOTAVAIL;
 612
 613         v = &mrt->vif_table[vifi];
 614
 615         write_lock_bh(&mrt_lock);
 616         dev = v->dev;
 617         v->dev = NULL;
 618
 619         if (!dev) {
 620                 write_unlock_bh(&mrt_lock);
 621                 return -EADDRNOTAVAIL;
 622         }
 623
 624         if (vifi == mrt->mroute_reg_vif_num)
 625                 mrt->mroute_reg_vif_num = -1;
 626
 627         if (vifi + 1 == mrt->maxvif) {
 628                 int tmp;
 629
 630                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 631                         if (VIF_EXISTS(mrt, tmp))
 632                                 break;
 633                 }
 634                 mrt->maxvif = tmp+1;
 635         }
 636
 637         write_unlock_bh(&mrt_lock);
 638
 639         dev_set_allmulti(dev, -1);
 640
 641         in_dev = __in_dev_get_rtnl(dev);
 642         if (in_dev) {
 643                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 644                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 645                                             NETCONFA_MC_FORWARDING,
 646                                             dev->ifindex, &in_dev->cnf);
 647                 ip_rt_multicast_event(in_dev);
 648         }
 649
 650         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 651                 unregister_netdevice_queue(dev, head);
 652
 653         dev_put(dev);
 654         return 0;
 655 }
 656
 657 static void ipmr_cache_free_rcu(struct rcu_head *head)
 658 {
 659         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 660
 661         kmem_cache_free(mrt_cachep, c);
 662 }
 663
 664 static inline void ipmr_cache_free(struct mfc_cache *c)
 665 {
 666         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 667 }
 668
 669 /* Destroy an unresolved cache entry, killing queued skbs
 670  * and reporting error to netlink readers.
 671  */
 672 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 673 {
 674         struct net *net = read_pnet(&mrt->net);
 675         struct sk_buff *skb;
 676         struct nlmsgerr *e;
 677
 678         atomic_dec(&mrt->cache_resolve_queue_len);
 679
 680         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 681                 if (ip_hdr(skb)->version == 0) {
 682                         struct nlmsghdr *nlh = skb_pull(skb,
 683                                                         sizeof(struct iphdr));
 684                         nlh->nlmsg_type = NLMSG_ERROR;
 685                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 686                         skb_trim(skb, nlh->nlmsg_len);
 687                         e = nlmsg_data(nlh);
 688                         e->error = -ETIMEDOUT;
 689                         memset(&e->msg, 0, sizeof(e->msg));
 690
 691                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 692                 } else {
 693                         kfree_skb(skb);
 694                 }
 695         }
 696
 697         ipmr_cache_free(c);
 698 }
 699
 700 /* Timer process for the unresolved queue. */
 701 static void ipmr_expire_process(unsigned long arg)
 702 {
 703         struct mr_table *mrt = (struct mr_table *)arg;
 704         unsigned long now;
 705         unsigned long expires;
 706         struct mfc_cache *c, *next;
 707
 708         if (!spin_trylock(&mfc_unres_lock)) {
 709                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 710                 return;
 711         }
 712
 713         if (list_empty(&mrt->mfc_unres_queue))
 714                 goto out;
 715
 716         now = jiffies;
 717         expires = 10*HZ;
 718
 719         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 720                 if (time_after(c->mfc_un.unres.expires, now)) {
 721                         unsigned long interval = c->mfc_un.unres.expires - now;
 722                         if (interval < expires)
 723                                 expires = interval;
 724                         continue;
 725                 }
 726
 727                 list_del(&c->list);
 728                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 729                 ipmr_destroy_unres(mrt, c);
 730         }
 731
 732         if (!list_empty(&mrt->mfc_unres_queue))
 733                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 734
 735 out:
 736         spin_unlock(&mfc_unres_lock);
 737 }
 738
 739 /* Fill oifs list. It is called under write locked mrt_lock. */
 740 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 741                                    unsigned char *ttls)
 742 {
 743         int vifi;
 744
 745         cache->mfc_un.res.minvif = MAXVIFS;
 746         cache->mfc_un.res.maxvif = 0;
 747         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 748
 749         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 750                 if (VIF_EXISTS(mrt, vifi) &&
 751                     ttls[vifi] && ttls[vifi] < 255) {
 752                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 753                         if (cache->mfc_un.res.minvif > vifi)
 754                                 cache->mfc_un.res.minvif = vifi;
 755                         if (cache->mfc_un.res.maxvif <= vifi)
 756                                 cache->mfc_un.res.maxvif = vifi + 1;
 757                 }
 758         }
 759         cache->mfc_un.res.lastuse = jiffies;
 760 }
 761
 762 static int vif_add(struct net *net, struct mr_table *mrt,
 763                    struct vifctl *vifc, int mrtsock)
 764 {
 765         int vifi = vifc->vifc_vifi;
 766         struct vif_device *v = &mrt->vif_table[vifi];
 767         struct net_device *dev;
 768         struct in_device *in_dev;
 769         int err;
 770
 771         /* Is vif busy ? */
 772         if (VIF_EXISTS(mrt, vifi))
 773                 return -EADDRINUSE;
 774
 775         switch (vifc->vifc_flags) {
 776         case VIFF_REGISTER:
 777                 if (!ipmr_pimsm_enabled())
 778                         return -EINVAL;
 779                 /* Special Purpose VIF in PIM
 780                  * All the packets will be sent to the daemon
 781                  */
 782                 if (mrt->mroute_reg_vif_num >= 0)
 783                         return -EADDRINUSE;
 784                 dev = ipmr_reg_vif(net, mrt);
 785                 if (!dev)
 786                         return -ENOBUFS;
 787                 err = dev_set_allmulti(dev, 1);
 788                 if (err) {
 789                         unregister_netdevice(dev);
 790                         dev_put(dev);
 791                         return err;
 792                 }
 793                 break;
 794         case VIFF_TUNNEL:
 795                 dev = ipmr_new_tunnel(net, vifc);
 796                 if (!dev)
 797                         return -ENOBUFS;
 798                 err = dev_set_allmulti(dev, 1);
 799                 if (err) {
 800                         ipmr_del_tunnel(dev, vifc);
 801                         dev_put(dev);
 802                         return err;
 803                 }
 804                 break;
 805         case VIFF_USE_IFINDEX:
 806         case 0:
 807                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 808                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 809                         if (dev && !__in_dev_get_rtnl(dev)) {
 810                                 dev_put(dev);
 811                                 return -EADDRNOTAVAIL;
 812                         }
 813                 } else {
 814                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 815                 }
 816                 if (!dev)
 817                         return -EADDRNOTAVAIL;
 818                 err = dev_set_allmulti(dev, 1);
 819                 if (err) {
 820                         dev_put(dev);
 821                         return err;
 822                 }
 823                 break;
 824         default:
 825                 return -EINVAL;
 826         }
 827
 828         in_dev = __in_dev_get_rtnl(dev);
 829         if (!in_dev) {
 830                 dev_put(dev);
 831                 return -EADDRNOTAVAIL;
 832         }
 833         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 834         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 835                                     dev->ifindex, &in_dev->cnf);
 836         ip_rt_multicast_event(in_dev);
 837
 838         /* Fill in the VIF structures */
 839
 840         v->rate_limit = vifc->vifc_rate_limit;
 841         v->local = vifc->vifc_lcl_addr.s_addr;
 842         v->remote = vifc->vifc_rmt_addr.s_addr;
 843         v->flags = vifc->vifc_flags;
 844         if (!mrtsock)
 845                 v->flags |= VIFF_STATIC;
 846         v->threshold = vifc->vifc_threshold;
 847         v->bytes_in = 0;
 848         v->bytes_out = 0;
 849         v->pkt_in = 0;
 850         v->pkt_out = 0;
 851         v->link = dev->ifindex;
 852         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 853                 v->link = dev_get_iflink(dev);
 854
 855         /* And finish update writing critical data */
 856         write_lock_bh(&mrt_lock);
 857         v->dev = dev;
 858         if (v->flags & VIFF_REGISTER)
 859                 mrt->mroute_reg_vif_num = vifi;
 860         if (vifi+1 > mrt->maxvif)
 861                 mrt->maxvif = vifi+1;
 862         write_unlock_bh(&mrt_lock);
 863         return 0;
 864 }
 865
 866 /* called with rcu_read_lock() */
 867 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 868                                          __be32 origin,
 869                                          __be32 mcastgrp)
 870 {
 871         struct mfc_cache_cmp_arg arg = {
 872                         .mfc_mcastgrp = mcastgrp,
 873                         .mfc_origin = origin
 874         };
 875         struct rhlist_head *tmp, *list;
 876         struct mfc_cache *c;
 877
 878         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 879         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 880                 return c;
 881
 882         return NULL;
 883 }
 884
 885 /* Look for a (*,*,oif) entry */
 886 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 887                                                     int vifi)
 888 {
 889         struct mfc_cache_cmp_arg arg = {
 890                         .mfc_mcastgrp = htonl(INADDR_ANY),
 891                         .mfc_origin = htonl(INADDR_ANY)
 892         };
 893         struct rhlist_head *tmp, *list;
 894         struct mfc_cache *c;
 895
 896         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 897         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 898                 if (c->mfc_un.res.ttls[vifi] < 255)
 899                         return c;
 900
 901         return NULL;
 902 }
 903
 904 /* Look for a (*,G) entry */
 905 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 906                                              __be32 mcastgrp, int vifi)
 907 {
 908         struct mfc_cache_cmp_arg arg = {
 909                         .mfc_mcastgrp = mcastgrp,
 910                         .mfc_origin = htonl(INADDR_ANY)
 911         };
 912         struct rhlist_head *tmp, *list;
 913         struct mfc_cache *c, *proxy;
 914
 915         if (mcastgrp == htonl(INADDR_ANY))
 916                 goto skip;
 917
 918         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 919         rhl_for_each_entry_rcu(c, tmp, list, mnode) {
 920                 if (c->mfc_un.res.ttls[vifi] < 255)
 921                         return c;
 922
 923                 /* It's ok if the vifi is part of the static tree */
 924                 proxy = ipmr_cache_find_any_parent(mrt, c->mfc_parent);
 925                 if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 926                         return c;
 927         }
 928
 929 skip:
 930         return ipmr_cache_find_any_parent(mrt, vifi);
 931 }
 932
 933 /* Look for a (S,G,iif) entry if parent != -1 */
 934 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 935                                                 __be32 origin, __be32 mcastgrp,
 936                                                 int parent)
 937 {
 938         struct mfc_cache_cmp_arg arg = {
 939                         .mfc_mcastgrp = mcastgrp,
 940                         .mfc_origin = origin,
 941         };
 942         struct rhlist_head *tmp, *list;
 943         struct mfc_cache *c;
 944
 945         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 946         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 947                 if (parent == -1 || parent == c->mfc_parent)
 948                         return c;
 949
 950         return NULL;
 951 }
 952
 953 /* Allocate a multicast cache entry */
 954 static struct mfc_cache *ipmr_cache_alloc(void)
 955 {
 956         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 957
 958         if (c) {
 959                 c->mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 960                 c->mfc_un.res.minvif = MAXVIFS;
 961         }
 962         return c;
 963 }
 964
 965 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 966 {
 967         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 968
 969         if (c) {
 970                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 971                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 972         }
 973         return c;
 974 }
 975
 976 /* A cache entry has gone into a resolved state from queued */
 977 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 978                                struct mfc_cache *uc, struct mfc_cache *c)
 979 {
 980         struct sk_buff *skb;
 981         struct nlmsgerr *e;
 982
 983         /* Play the pending entries through our router */
 984         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 985                 if (ip_hdr(skb)->version == 0) {
 986                         struct nlmsghdr *nlh = skb_pull(skb,
 987                                                         sizeof(struct iphdr));
 988
 989                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 990                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 991                                                  (u8 *)nlh;
 992                         } else {
 993                                 nlh->nlmsg_type = NLMSG_ERROR;
 994                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 995                                 skb_trim(skb, nlh->nlmsg_len);
 996                                 e = nlmsg_data(nlh);
 997                                 e->error = -EMSGSIZE;
 998                                 memset(&e->msg, 0, sizeof(e->msg));
 999                         }
1000
1001                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1002                 } else {
1003                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1004                 }
1005         }
1006 }
1007
1008 /* Bounce a cache query up to mrouted and netlink.
1009  *
1010  * Called under mrt_lock.
1011  */
1012 static int ipmr_cache_report(struct mr_table *mrt,
1013                              struct sk_buff *pkt, vifi_t vifi, int assert)
1014 {
1015         const int ihl = ip_hdrlen(pkt);
1016         struct sock *mroute_sk;
1017         struct igmphdr *igmp;
1018         struct igmpmsg *msg;
1019         struct sk_buff *skb;
1020         int ret;
1021
1022         if (assert == IGMPMSG_WHOLEPKT)
1023                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1024         else
1025                 skb = alloc_skb(128, GFP_ATOMIC);
1026
1027         if (!skb)
1028                 return -ENOBUFS;
1029
1030         if (assert == IGMPMSG_WHOLEPKT) {
1031                 /* Ugly, but we have no choice with this interface.
1032                  * Duplicate old header, fix ihl, length etc.
1033                  * And all this only to mangle msg->im_msgtype and
1034                  * to set msg->im_mbz to "mbz" :-)
1035                  */
1036                 skb_push(skb, sizeof(struct iphdr));
1037                 skb_reset_network_header(skb);
1038                 skb_reset_transport_header(skb);
1039                 msg = (struct igmpmsg *)skb_network_header(skb);
1040                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1041                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
1042                 msg->im_mbz = 0;
1043                 msg->im_vif = mrt->mroute_reg_vif_num;
1044                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1045                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1046                                              sizeof(struct iphdr));
1047         } else {
1048                 /* Copy the IP header */
1049                 skb_set_network_header(skb, skb->len);
1050                 skb_put(skb, ihl);
1051                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1052                 /* Flag to the kernel this is a route add */
1053                 ip_hdr(skb)->protocol = 0;
1054                 msg = (struct igmpmsg *)skb_network_header(skb);
1055                 msg->im_vif = vifi;
1056                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1057                 /* Add our header */
1058                 igmp = skb_put(skb, sizeof(struct igmphdr));
1059                 igmp->type = assert;
1060                 msg->im_msgtype = assert;
1061                 igmp->code = 0;
1062                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1063                 skb->transport_header = skb->network_header;
1064         }
1065
1066         rcu_read_lock();
1067         mroute_sk = rcu_dereference(mrt->mroute_sk);
1068         if (!mroute_sk) {
1069                 rcu_read_unlock();
1070                 kfree_skb(skb);
1071                 return -EINVAL;
1072         }
1073
1074         igmpmsg_netlink_event(mrt, skb);
1075
1076         /* Deliver to mrouted */
1077         ret = sock_queue_rcv_skb(mroute_sk, skb);
1078         rcu_read_unlock();
1079         if (ret < 0) {
1080                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1081                 kfree_skb(skb);
1082         }
1083
1084         return ret;
1085 }
1086
1087 /* Queue a packet for resolution. It gets locked cache entry! */
1088 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1089                                  struct sk_buff *skb, struct net_device *dev)
1090 {
1091         const struct iphdr *iph = ip_hdr(skb);
1092         struct mfc_cache *c;
1093         bool found = false;
1094         int err;
1095
1096         spin_lock_bh(&mfc_unres_lock);
1097         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1098                 if (c->mfc_mcastgrp == iph->daddr &&
1099                     c->mfc_origin == iph->saddr) {
1100                         found = true;
1101                         break;
1102                 }
1103         }
1104
1105         if (!found) {
1106                 /* Create a new entry if allowable */
1107                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1108                     (c = ipmr_cache_alloc_unres()) == NULL) {
1109                         spin_unlock_bh(&mfc_unres_lock);
1110
1111                         kfree_skb(skb);
1112                         return -ENOBUFS;
1113                 }
1114
1115                 /* Fill in the new cache entry */
1116                 c->mfc_parent   = -1;
1117                 c->mfc_origin   = iph->saddr;
1118                 c->mfc_mcastgrp = iph->daddr;
1119
1120                 /* Reflect first query at mrouted. */
1121                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1122                 if (err < 0) {
1123                         /* If the report failed throw the cache entry
1124                            out - Brad Parker
1125                          */
1126                         spin_unlock_bh(&mfc_unres_lock);
1127
1128                         ipmr_cache_free(c);
1129                         kfree_skb(skb);
1130                         return err;
1131                 }
1132
1133                 atomic_inc(&mrt->cache_resolve_queue_len);
1134                 list_add(&c->list, &mrt->mfc_unres_queue);
1135                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1136
1137                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1138                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1139         }
1140
1141         /* See if we can append the packet */
1142         if (c->mfc_un.unres.unresolved.qlen > 3) {
1143                 kfree_skb(skb);
1144                 err = -ENOBUFS;
1145         } else {
1146                 if (dev) {
1147                         skb->dev = dev;
1148                         skb->skb_iif = dev->ifindex;
1149                 }
1150                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1151                 err = 0;
1152         }
1153
1154         spin_unlock_bh(&mfc_unres_lock);
1155         return err;
1156 }
1157
1158 /* MFC cache manipulation by user space mroute daemon */
1159
1160 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1161 {
1162         struct mfc_cache *c;
1163
1164         /* The entries are added/deleted only under RTNL */
1165         rcu_read_lock();
1166         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1167                                    mfc->mfcc_mcastgrp.s_addr, parent);
1168         rcu_read_unlock();
1169         if (!c)
1170                 return -ENOENT;
1171         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1172         list_del_rcu(&c->list);
1173         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1174         ipmr_cache_free(c);
1175
1176         return 0;
1177 }
1178
1179 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1180                         struct mfcctl *mfc, int mrtsock, int parent)
1181 {
1182         struct mfc_cache *uc, *c;
1183         bool found;
1184         int ret;
1185
1186         if (mfc->mfcc_parent >= MAXVIFS)
1187                 return -ENFILE;
1188
1189         /* The entries are added/deleted only under RTNL */
1190         rcu_read_lock();
1191         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1192                                    mfc->mfcc_mcastgrp.s_addr, parent);
1193         rcu_read_unlock();
1194         if (c) {
1195                 write_lock_bh(&mrt_lock);
1196                 c->mfc_parent = mfc->mfcc_parent;
1197                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1198                 if (!mrtsock)
1199                         c->mfc_flags |= MFC_STATIC;
1200                 write_unlock_bh(&mrt_lock);
1201                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1202                 return 0;
1203         }
1204
1205         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1206             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1207                 return -EINVAL;
1208
1209         c = ipmr_cache_alloc();
1210         if (!c)
1211                 return -ENOMEM;
1212
1213         c->mfc_origin = mfc->mfcc_origin.s_addr;
1214         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1215         c->mfc_parent = mfc->mfcc_parent;
1216         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1217         if (!mrtsock)
1218                 c->mfc_flags |= MFC_STATIC;
1219
1220         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->mnode,
1221                                   ipmr_rht_params);
1222         if (ret) {
1223                 pr_err("ipmr: rhtable insert error %d\n", ret);
1224                 ipmr_cache_free(c);
1225                 return ret;
1226         }
1227         list_add_tail_rcu(&c->list, &mrt->mfc_cache_list);
1228         /* Check to see if we resolved a queued list. If so we
1229          * need to send on the frames and tidy up.
1230          */
1231         found = false;
1232         spin_lock_bh(&mfc_unres_lock);
1233         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1234                 if (uc->mfc_origin == c->mfc_origin &&
1235                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1236                         list_del(&uc->list);
1237                         atomic_dec(&mrt->cache_resolve_queue_len);
1238                         found = true;
1239                         break;
1240                 }
1241         }
1242         if (list_empty(&mrt->mfc_unres_queue))
1243                 del_timer(&mrt->ipmr_expire_timer);
1244         spin_unlock_bh(&mfc_unres_lock);
1245
1246         if (found) {
1247                 ipmr_cache_resolve(net, mrt, uc, c);
1248                 ipmr_cache_free(uc);
1249         }
1250         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1251         return 0;
1252 }
1253
1254 /* Close the multicast socket, and clear the vif tables etc */
1255 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1256 {
1257         struct mfc_cache *c, *tmp;
1258         LIST_HEAD(list);
1259         int i;
1260
1261         /* Shut down all active vif entries */
1262         for (i = 0; i < mrt->maxvif; i++) {
1263                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1264                         continue;
1265                 vif_delete(mrt, i, 0, &list);
1266         }
1267         unregister_netdevice_many(&list);
1268
1269         /* Wipe the cache */
1270         list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1271                 if (!all && (c->mfc_flags & MFC_STATIC))
1272                         continue;
1273                 rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1274                 list_del_rcu(&c->list);
1275                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
1276                 ipmr_cache_free(c);
1277         }
1278
1279         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1280                 spin_lock_bh(&mfc_unres_lock);
1281                 list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1282                         list_del(&c->list);
1283                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1284                         ipmr_destroy_unres(mrt, c);
1285                 }
1286                 spin_unlock_bh(&mfc_unres_lock);
1287         }
1288 }
1289
1290 /* called from ip_ra_control(), before an RCU grace period,
1291  * we dont need to call synchronize_rcu() here
1292  */
1293 static void mrtsock_destruct(struct sock *sk)
1294 {
1295         struct net *net = sock_net(sk);
1296         struct mr_table *mrt;
1297
1298         ASSERT_RTNL();
1299         ipmr_for_each_table(mrt, net) {
1300                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1301                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1302                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1303                                                     NETCONFA_MC_FORWARDING,
1304                                                     NETCONFA_IFINDEX_ALL,
1305                                                     net->ipv4.devconf_all);
1306                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1307                         mroute_clean_tables(mrt, false);
1308                 }
1309         }
1310 }
1311
1312 /* Socket options and virtual interface manipulation. The whole
1313  * virtual interface system is a complete heap, but unfortunately
1314  * that's how BSD mrouted happens to think. Maybe one day with a proper
1315  * MOSPF/PIM router set up we can clean this up.
1316  */
1317
1318 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1319                          unsigned int optlen)
1320 {
1321         struct net *net = sock_net(sk);
1322         int val, ret = 0, parent = 0;
1323         struct mr_table *mrt;
1324         struct vifctl vif;
1325         struct mfcctl mfc;
1326         u32 uval;
1327
1328         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1329         rtnl_lock();
1330         if (sk->sk_type != SOCK_RAW ||
1331             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1332                 ret = -EOPNOTSUPP;
1333                 goto out_unlock;
1334         }
1335
1336         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1337         if (!mrt) {
1338                 ret = -ENOENT;
1339                 goto out_unlock;
1340         }
1341         if (optname != MRT_INIT) {
1342                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1343                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1344                         ret = -EACCES;
1345                         goto out_unlock;
1346                 }
1347         }
1348
1349         switch (optname) {
1350         case MRT_INIT:
1351                 if (optlen != sizeof(int)) {
1352                         ret = -EINVAL;
1353                         break;
1354                 }
1355                 if (rtnl_dereference(mrt->mroute_sk)) {
1356                         ret = -EADDRINUSE;
1357                         break;
1358                 }
1359
1360                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1361                 if (ret == 0) {
1362                         rcu_assign_pointer(mrt->mroute_sk, sk);
1363                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1364                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1365                                                     NETCONFA_MC_FORWARDING,
1366                                                     NETCONFA_IFINDEX_ALL,
1367                                                     net->ipv4.devconf_all);
1368                 }
1369                 break;
1370         case MRT_DONE:
1371                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1372                         ret = -EACCES;
1373                 } else {
1374                         ret = ip_ra_control(sk, 0, NULL);
1375                         goto out_unlock;
1376                 }
1377                 break;
1378         case MRT_ADD_VIF:
1379         case MRT_DEL_VIF:
1380                 if (optlen != sizeof(vif)) {
1381                         ret = -EINVAL;
1382                         break;
1383                 }
1384                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1385                         ret = -EFAULT;
1386                         break;
1387                 }
1388                 if (vif.vifc_vifi >= MAXVIFS) {
1389                         ret = -ENFILE;
1390                         break;
1391                 }
1392                 if (optname == MRT_ADD_VIF) {
1393                         ret = vif_add(net, mrt, &vif,
1394                                       sk == rtnl_dereference(mrt->mroute_sk));
1395                 } else {
1396                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1397                 }
1398                 break;
1399         /* Manipulate the forwarding caches. These live
1400          * in a sort of kernel/user symbiosis.
1401          */
1402         case MRT_ADD_MFC:
1403         case MRT_DEL_MFC:
1404                 parent = -1;
1405         case MRT_ADD_MFC_PROXY:
1406         case MRT_DEL_MFC_PROXY:
1407                 if (optlen != sizeof(mfc)) {
1408                         ret = -EINVAL;
1409                         break;
1410                 }
1411                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1412                         ret = -EFAULT;
1413                         break;
1414                 }
1415                 if (parent == 0)
1416                         parent = mfc.mfcc_parent;
1417                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1418                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1419                 else
1420                         ret = ipmr_mfc_add(net, mrt, &mfc,
1421                                            sk == rtnl_dereference(mrt->mroute_sk),
1422                                            parent);
1423                 break;
1424         /* Control PIM assert. */
1425         case MRT_ASSERT:
1426                 if (optlen != sizeof(val)) {
1427                         ret = -EINVAL;
1428                         break;
1429                 }
1430                 if (get_user(val, (int __user *)optval)) {
1431                         ret = -EFAULT;
1432                         break;
1433                 }
1434                 mrt->mroute_do_assert = val;
1435                 break;
1436         case MRT_PIM:
1437                 if (!ipmr_pimsm_enabled()) {
1438                         ret = -ENOPROTOOPT;
1439                         break;
1440                 }
1441                 if (optlen != sizeof(val)) {
1442                         ret = -EINVAL;
1443                         break;
1444                 }
1445                 if (get_user(val, (int __user *)optval)) {
1446                         ret = -EFAULT;
1447                         break;
1448                 }
1449
1450                 val = !!val;
1451                 if (val != mrt->mroute_do_pim) {
1452                         mrt->mroute_do_pim = val;
1453                         mrt->mroute_do_assert = val;
1454                 }
1455                 break;
1456         case MRT_TABLE:
1457                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1458                         ret = -ENOPROTOOPT;
1459                         break;
1460                 }
1461                 if (optlen != sizeof(uval)) {
1462                         ret = -EINVAL;
1463                         break;
1464                 }
1465                 if (get_user(uval, (u32 __user *)optval)) {
1466                         ret = -EFAULT;
1467                         break;
1468                 }
1469
1470                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1471                         ret = -EBUSY;
1472                 } else {
1473                         mrt = ipmr_new_table(net, uval);
1474                         if (IS_ERR(mrt))
1475                                 ret = PTR_ERR(mrt);
1476                         else
1477                                 raw_sk(sk)->ipmr_table = uval;
1478                 }
1479                 break;
1480         /* Spurious command, or MRT_VERSION which you cannot set. */
1481         default:
1482                 ret = -ENOPROTOOPT;
1483         }
1484 out_unlock:
1485         rtnl_unlock();
1486         return ret;
1487 }
1488
1489 /* Getsock opt support for the multicast routing system. */
1490 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1491 {
1492         int olr;
1493         int val;
1494         struct net *net = sock_net(sk);
1495         struct mr_table *mrt;
1496
1497         if (sk->sk_type != SOCK_RAW ||
1498             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1499                 return -EOPNOTSUPP;
1500
1501         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1502         if (!mrt)
1503                 return -ENOENT;
1504
1505         switch (optname) {
1506         case MRT_VERSION:
1507                 val = 0x0305;
1508                 break;
1509         case MRT_PIM:
1510                 if (!ipmr_pimsm_enabled())
1511                         return -ENOPROTOOPT;
1512                 val = mrt->mroute_do_pim;
1513                 break;
1514         case MRT_ASSERT:
1515                 val = mrt->mroute_do_assert;
1516                 break;
1517         default:
1518                 return -ENOPROTOOPT;
1519         }
1520
1521         if (get_user(olr, optlen))
1522                 return -EFAULT;
1523         olr = min_t(unsigned int, olr, sizeof(int));
1524         if (olr < 0)
1525                 return -EINVAL;
1526         if (put_user(olr, optlen))
1527                 return -EFAULT;
1528         if (copy_to_user(optval, &val, olr))
1529                 return -EFAULT;
1530         return 0;
1531 }
1532
1533 /* The IP multicast ioctl support routines. */
1534 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1535 {
1536         struct sioc_sg_req sr;
1537         struct sioc_vif_req vr;
1538         struct vif_device *vif;
1539         struct mfc_cache *c;
1540         struct net *net = sock_net(sk);
1541         struct mr_table *mrt;
1542
1543         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1544         if (!mrt)
1545                 return -ENOENT;
1546
1547         switch (cmd) {
1548         case SIOCGETVIFCNT:
1549                 if (copy_from_user(&vr, arg, sizeof(vr)))
1550                         return -EFAULT;
1551                 if (vr.vifi >= mrt->maxvif)
1552                         return -EINVAL;
1553                 read_lock(&mrt_lock);
1554                 vif = &mrt->vif_table[vr.vifi];
1555                 if (VIF_EXISTS(mrt, vr.vifi)) {
1556                         vr.icount = vif->pkt_in;
1557                         vr.ocount = vif->pkt_out;
1558                         vr.ibytes = vif->bytes_in;
1559                         vr.obytes = vif->bytes_out;
1560                         read_unlock(&mrt_lock);
1561
1562                         if (copy_to_user(arg, &vr, sizeof(vr)))
1563                                 return -EFAULT;
1564                         return 0;
1565                 }
1566                 read_unlock(&mrt_lock);
1567                 return -EADDRNOTAVAIL;
1568         case SIOCGETSGCNT:
1569                 if (copy_from_user(&sr, arg, sizeof(sr)))
1570                         return -EFAULT;
1571
1572                 rcu_read_lock();
1573                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1574                 if (c) {
1575                         sr.pktcnt = c->mfc_un.res.pkt;
1576                         sr.bytecnt = c->mfc_un.res.bytes;
1577                         sr.wrong_if = c->mfc_un.res.wrong_if;
1578                         rcu_read_unlock();
1579
1580                         if (copy_to_user(arg, &sr, sizeof(sr)))
1581                                 return -EFAULT;
1582                         return 0;
1583                 }
1584                 rcu_read_unlock();
1585                 return -EADDRNOTAVAIL;
1586         default:
1587                 return -ENOIOCTLCMD;
1588         }
1589 }
1590
1591 #ifdef CONFIG_COMPAT
1592 struct compat_sioc_sg_req {
1593         struct in_addr src;
1594         struct in_addr grp;
1595         compat_ulong_t pktcnt;
1596         compat_ulong_t bytecnt;
1597         compat_ulong_t wrong_if;
1598 };
1599
1600 struct compat_sioc_vif_req {
1601         vifi_t  vifi;           /* Which iface */
1602         compat_ulong_t icount;
1603         compat_ulong_t ocount;
1604         compat_ulong_t ibytes;
1605         compat_ulong_t obytes;
1606 };
1607
1608 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1609 {
1610         struct compat_sioc_sg_req sr;
1611         struct compat_sioc_vif_req vr;
1612         struct vif_device *vif;
1613         struct mfc_cache *c;
1614         struct net *net = sock_net(sk);
1615         struct mr_table *mrt;
1616
1617         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1618         if (!mrt)
1619                 return -ENOENT;
1620
1621         switch (cmd) {
1622         case SIOCGETVIFCNT:
1623                 if (copy_from_user(&vr, arg, sizeof(vr)))
1624                         return -EFAULT;
1625                 if (vr.vifi >= mrt->maxvif)
1626                         return -EINVAL;
1627                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1628                 read_lock(&mrt_lock);
1629                 vif = &mrt->vif_table[vr.vifi];
1630                 if (VIF_EXISTS(mrt, vr.vifi)) {
1631                         vr.icount = vif->pkt_in;
1632                         vr.ocount = vif->pkt_out;
1633                         vr.ibytes = vif->bytes_in;
1634                         vr.obytes = vif->bytes_out;
1635                         read_unlock(&mrt_lock);
1636
1637                         if (copy_to_user(arg, &vr, sizeof(vr)))
1638                                 return -EFAULT;
1639                         return 0;
1640                 }
1641                 read_unlock(&mrt_lock);
1642                 return -EADDRNOTAVAIL;
1643         case SIOCGETSGCNT:
1644                 if (copy_from_user(&sr, arg, sizeof(sr)))
1645                         return -EFAULT;
1646
1647                 rcu_read_lock();
1648                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1649                 if (c) {
1650                         sr.pktcnt = c->mfc_un.res.pkt;
1651                         sr.bytecnt = c->mfc_un.res.bytes;
1652                         sr.wrong_if = c->mfc_un.res.wrong_if;
1653                         rcu_read_unlock();
1654
1655                         if (copy_to_user(arg, &sr, sizeof(sr)))
1656                                 return -EFAULT;
1657                         return 0;
1658                 }
1659                 rcu_read_unlock();
1660                 return -EADDRNOTAVAIL;
1661         default:
1662                 return -ENOIOCTLCMD;
1663         }
1664 }
1665 #endif
1666
1667 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1668 {
1669         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1670         struct net *net = dev_net(dev);
1671         struct mr_table *mrt;
1672         struct vif_device *v;
1673         int ct;
1674
1675         if (event != NETDEV_UNREGISTER)
1676                 return NOTIFY_DONE;
1677
1678         ipmr_for_each_table(mrt, net) {
1679                 v = &mrt->vif_table[0];
1680                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1681                         if (v->dev == dev)
1682                                 vif_delete(mrt, ct, 1, NULL);
1683                 }
1684         }
1685         return NOTIFY_DONE;
1686 }
1687
1688 static struct notifier_block ip_mr_notifier = {
1689         .notifier_call = ipmr_device_event,
1690 };
1691
1692 /* Encapsulate a packet by attaching a valid IPIP header to it.
1693  * This avoids tunnel drivers and other mess and gives us the speed so
1694  * important for multicast video.
1695  */
1696 static void ip_encap(struct net *net, struct sk_buff *skb,
1697                      __be32 saddr, __be32 daddr)
1698 {
1699         struct iphdr *iph;
1700         const struct iphdr *old_iph = ip_hdr(skb);
1701
1702         skb_push(skb, sizeof(struct iphdr));
1703         skb->transport_header = skb->network_header;
1704         skb_reset_network_header(skb);
1705         iph = ip_hdr(skb);
1706
1707         iph->version    =       4;
1708         iph->tos        =       old_iph->tos;
1709         iph->ttl        =       old_iph->ttl;
1710         iph->frag_off   =       0;
1711         iph->daddr      =       daddr;
1712         iph->saddr      =       saddr;
1713         iph->protocol   =       IPPROTO_IPIP;
1714         iph->ihl        =       5;
1715         iph->tot_len    =       htons(skb->len);
1716         ip_select_ident(net, skb, NULL);
1717         ip_send_check(iph);
1718
1719         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1720         nf_reset(skb);
1721 }
1722
1723 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1724                                       struct sk_buff *skb)
1725 {
1726         struct ip_options *opt = &(IPCB(skb)->opt);
1727
1728         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1729         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1730
1731         if (unlikely(opt->optlen))
1732                 ip_forward_options(skb);
1733
1734         return dst_output(net, sk, skb);
1735 }
1736
1737 /* Processing handlers for ipmr_forward */
1738
1739 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1740                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1741 {
1742         const struct iphdr *iph = ip_hdr(skb);
1743         struct vif_device *vif = &mrt->vif_table[vifi];
1744         struct net_device *dev;
1745         struct rtable *rt;
1746         struct flowi4 fl4;
1747         int    encap = 0;
1748
1749         if (!vif->dev)
1750                 goto out_free;
1751
1752         if (vif->flags & VIFF_REGISTER) {
1753                 vif->pkt_out++;
1754                 vif->bytes_out += skb->len;
1755                 vif->dev->stats.tx_bytes += skb->len;
1756                 vif->dev->stats.tx_packets++;
1757                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1758                 goto out_free;
1759         }
1760
1761         if (vif->flags & VIFF_TUNNEL) {
1762                 rt = ip_route_output_ports(net, &fl4, NULL,
1763                                            vif->remote, vif->local,
1764                                            0, 0,
1765                                            IPPROTO_IPIP,
1766                                            RT_TOS(iph->tos), vif->link);
1767                 if (IS_ERR(rt))
1768                         goto out_free;
1769                 encap = sizeof(struct iphdr);
1770         } else {
1771                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1772                                            0, 0,
1773                                            IPPROTO_IPIP,
1774                                            RT_TOS(iph->tos), vif->link);
1775                 if (IS_ERR(rt))
1776                         goto out_free;
1777         }
1778
1779         dev = rt->dst.dev;
1780
1781         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1782                 /* Do not fragment multicasts. Alas, IPv4 does not
1783                  * allow to send ICMP, so that packets will disappear
1784                  * to blackhole.
1785                  */
1786                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1787                 ip_rt_put(rt);
1788                 goto out_free;
1789         }
1790
1791         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1792
1793         if (skb_cow(skb, encap)) {
1794                 ip_rt_put(rt);
1795                 goto out_free;
1796         }
1797
1798         vif->pkt_out++;
1799         vif->bytes_out += skb->len;
1800
1801         skb_dst_drop(skb);
1802         skb_dst_set(skb, &rt->dst);
1803         ip_decrease_ttl(ip_hdr(skb));
1804
1805         /* FIXME: forward and output firewalls used to be called here.
1806          * What do we do with netfilter? -- RR
1807          */
1808         if (vif->flags & VIFF_TUNNEL) {
1809                 ip_encap(net, skb, vif->local, vif->remote);
1810                 /* FIXME: extra output firewall step used to be here. --RR */
1811                 vif->dev->stats.tx_packets++;
1812                 vif->dev->stats.tx_bytes += skb->len;
1813         }
1814
1815         IPCB(skb)->flags |= IPSKB_FORWARDED;
1816
1817         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1818          * not only before forwarding, but after forwarding on all output
1819          * interfaces. It is clear, if mrouter runs a multicasting
1820          * program, it should receive packets not depending to what interface
1821          * program is joined.
1822          * If we will not make it, the program will have to join on all
1823          * interfaces. On the other hand, multihoming host (or router, but
1824          * not mrouter) cannot join to more than one interface - it will
1825          * result in receiving multiple packets.
1826          */
1827         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1828                 net, NULL, skb, skb->dev, dev,
1829                 ipmr_forward_finish);
1830         return;
1831
1832 out_free:
1833         kfree_skb(skb);
1834 }
1835
1836 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1837 {
1838         int ct;
1839
1840         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1841                 if (mrt->vif_table[ct].dev == dev)
1842                         break;
1843         }
1844         return ct;
1845 }
1846
1847 /* "local" means that we should preserve one skb (for local delivery) */
1848 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1849                           struct net_device *dev, struct sk_buff *skb,
1850                           struct mfc_cache *cache, int local)
1851 {
1852         int true_vifi = ipmr_find_vif(mrt, dev);
1853         int psend = -1;
1854         int vif, ct;
1855
1856         vif = cache->mfc_parent;
1857         cache->mfc_un.res.pkt++;
1858         cache->mfc_un.res.bytes += skb->len;
1859         cache->mfc_un.res.lastuse = jiffies;
1860
1861         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1862                 struct mfc_cache *cache_proxy;
1863
1864                 /* For an (*,G) entry, we only check that the incomming
1865                  * interface is part of the static tree.
1866                  */
1867                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1868                 if (cache_proxy &&
1869                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1870                         goto forward;
1871         }
1872
1873         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1874         if (mrt->vif_table[vif].dev != dev) {
1875                 if (rt_is_output_route(skb_rtable(skb))) {
1876                         /* It is our own packet, looped back.
1877                          * Very complicated situation...
1878                          *
1879                          * The best workaround until routing daemons will be
1880                          * fixed is not to redistribute packet, if it was
1881                          * send through wrong interface. It means, that
1882                          * multicast applications WILL NOT work for
1883                          * (S,G), which have default multicast route pointing
1884                          * to wrong oif. In any case, it is not a good
1885                          * idea to use multicasting applications on router.
1886                          */
1887                         goto dont_forward;
1888                 }
1889
1890                 cache->mfc_un.res.wrong_if++;
1891
1892                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1893                     /* pimsm uses asserts, when switching from RPT to SPT,
1894                      * so that we cannot check that packet arrived on an oif.
1895                      * It is bad, but otherwise we would need to move pretty
1896                      * large chunk of pimd to kernel. Ough... --ANK
1897                      */
1898                     (mrt->mroute_do_pim ||
1899                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1900                     time_after(jiffies,
1901                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1902                         cache->mfc_un.res.last_assert = jiffies;
1903                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1904                 }
1905                 goto dont_forward;
1906         }
1907
1908 forward:
1909         mrt->vif_table[vif].pkt_in++;
1910         mrt->vif_table[vif].bytes_in += skb->len;
1911
1912         /* Forward the frame */
1913         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1914             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1915                 if (true_vifi >= 0 &&
1916                     true_vifi != cache->mfc_parent &&
1917                     ip_hdr(skb)->ttl >
1918                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1919                         /* It's an (*,*) entry and the packet is not coming from
1920                          * the upstream: forward the packet to the upstream
1921                          * only.
1922                          */
1923                         psend = cache->mfc_parent;
1924                         goto last_forward;
1925                 }
1926                 goto dont_forward;
1927         }
1928         for (ct = cache->mfc_un.res.maxvif - 1;
1929              ct >= cache->mfc_un.res.minvif; ct--) {
1930                 /* For (*,G) entry, don't forward to the incoming interface */
1931                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1932                      ct != true_vifi) &&
1933                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1934                         if (psend != -1) {
1935                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1936
1937                                 if (skb2)
1938                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1939                                                         psend);
1940                         }
1941                         psend = ct;
1942                 }
1943         }
1944 last_forward:
1945         if (psend != -1) {
1946                 if (local) {
1947                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1948
1949                         if (skb2)
1950                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1951                 } else {
1952                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1953                         return;
1954                 }
1955         }
1956
1957 dont_forward:
1958         if (!local)
1959                 kfree_skb(skb);
1960 }
1961
1962 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1963 {
1964         struct rtable *rt = skb_rtable(skb);
1965         struct iphdr *iph = ip_hdr(skb);
1966         struct flowi4 fl4 = {
1967                 .daddr = iph->daddr,
1968                 .saddr = iph->saddr,
1969                 .flowi4_tos = RT_TOS(iph->tos),
1970                 .flowi4_oif = (rt_is_output_route(rt) ?
1971                                skb->dev->ifindex : 0),
1972                 .flowi4_iif = (rt_is_output_route(rt) ?
1973                                LOOPBACK_IFINDEX :
1974                                skb->dev->ifindex),
1975                 .flowi4_mark = skb->mark,
1976         };
1977         struct mr_table *mrt;
1978         int err;
1979
1980         err = ipmr_fib_lookup(net, &fl4, &mrt);
1981         if (err)
1982                 return ERR_PTR(err);
1983         return mrt;
1984 }
1985
1986 /* Multicast packets for forwarding arrive here
1987  * Called with rcu_read_lock();
1988  */
1989 int ip_mr_input(struct sk_buff *skb)
1990 {
1991         struct mfc_cache *cache;
1992         struct net *net = dev_net(skb->dev);
1993         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1994         struct mr_table *mrt;
1995         struct net_device *dev;
1996
1997         /* skb->dev passed in is the loX master dev for vrfs.
1998          * As there are no vifs associated with loopback devices,
1999          * get the proper interface that does have a vif associated with it.
2000          */
2001         dev = skb->dev;
2002         if (netif_is_l3_master(skb->dev)) {
2003                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2004                 if (!dev) {
2005                         kfree_skb(skb);
2006                         return -ENODEV;
2007                 }
2008         }
2009
2010         /* Packet is looped back after forward, it should not be
2011          * forwarded second time, but still can be delivered locally.
2012          */
2013         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2014                 goto dont_forward;
2015
2016         mrt = ipmr_rt_fib_lookup(net, skb);
2017         if (IS_ERR(mrt)) {
2018                 kfree_skb(skb);
2019                 return PTR_ERR(mrt);
2020         }
2021         if (!local) {
2022                 if (IPCB(skb)->opt.router_alert) {
2023                         if (ip_call_ra_chain(skb))
2024                                 return 0;
2025                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2026                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2027                          * Cisco IOS <= 11.2(8)) do not put router alert
2028                          * option to IGMP packets destined to routable
2029                          * groups. It is very bad, because it means
2030                          * that we can forward NO IGMP messages.
2031                          */
2032                         struct sock *mroute_sk;
2033
2034                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2035                         if (mroute_sk) {
2036                                 nf_reset(skb);
2037                                 raw_rcv(mroute_sk, skb);
2038                                 return 0;
2039                         }
2040                     }
2041         }
2042
2043         /* already under rcu_read_lock() */
2044         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2045         if (!cache) {
2046                 int vif = ipmr_find_vif(mrt, dev);
2047
2048                 if (vif >= 0)
2049                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2050                                                     vif);
2051         }
2052
2053         /* No usable cache entry */
2054         if (!cache) {
2055                 int vif;
2056
2057                 if (local) {
2058                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2059                         ip_local_deliver(skb);
2060                         if (!skb2)
2061                                 return -ENOBUFS;
2062                         skb = skb2;
2063                 }
2064
2065                 read_lock(&mrt_lock);
2066                 vif = ipmr_find_vif(mrt, dev);
2067                 if (vif >= 0) {
2068                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2069                         read_unlock(&mrt_lock);
2070
2071                         return err2;
2072                 }
2073                 read_unlock(&mrt_lock);
2074                 kfree_skb(skb);
2075                 return -ENODEV;
2076         }
2077
2078         read_lock(&mrt_lock);
2079         ip_mr_forward(net, mrt, dev, skb, cache, local);
2080         read_unlock(&mrt_lock);
2081
2082         if (local)
2083                 return ip_local_deliver(skb);
2084
2085         return 0;
2086
2087 dont_forward:
2088         if (local)
2089                 return ip_local_deliver(skb);
2090         kfree_skb(skb);
2091         return 0;
2092 }
2093
2094 #ifdef CONFIG_IP_PIMSM_V1
2095 /* Handle IGMP messages of PIMv1 */
2096 int pim_rcv_v1(struct sk_buff *skb)
2097 {
2098         struct igmphdr *pim;
2099         struct net *net = dev_net(skb->dev);
2100         struct mr_table *mrt;
2101
2102         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2103                 goto drop;
2104
2105         pim = igmp_hdr(skb);
2106
2107         mrt = ipmr_rt_fib_lookup(net, skb);
2108         if (IS_ERR(mrt))
2109                 goto drop;
2110         if (!mrt->mroute_do_pim ||
2111             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2112                 goto drop;
2113
2114         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2115 drop:
2116                 kfree_skb(skb);
2117         }
2118         return 0;
2119 }
2120 #endif
2121
2122 #ifdef CONFIG_IP_PIMSM_V2
2123 static int pim_rcv(struct sk_buff *skb)
2124 {
2125         struct pimreghdr *pim;
2126         struct net *net = dev_net(skb->dev);
2127         struct mr_table *mrt;
2128
2129         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2130                 goto drop;
2131
2132         pim = (struct pimreghdr *)skb_transport_header(skb);
2133         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2134             (pim->flags & PIM_NULL_REGISTER) ||
2135             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2136              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2137                 goto drop;
2138
2139         mrt = ipmr_rt_fib_lookup(net, skb);
2140         if (IS_ERR(mrt))
2141                 goto drop;
2142         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2143 drop:
2144                 kfree_skb(skb);
2145         }
2146         return 0;
2147 }
2148 #endif
2149
2150 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2151                               struct mfc_cache *c, struct rtmsg *rtm)
2152 {
2153         struct rta_mfc_stats mfcs;
2154         struct nlattr *mp_attr;
2155         struct rtnexthop *nhp;
2156         unsigned long lastuse;
2157         int ct;
2158
2159         /* If cache is unresolved, don't try to parse IIF and OIF */
2160         if (c->mfc_parent >= MAXVIFS) {
2161                 rtm->rtm_flags |= RTNH_F_UNRESOLVED;
2162                 return -ENOENT;
2163         }
2164
2165         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2166             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2167                 return -EMSGSIZE;
2168
2169         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2170                 return -EMSGSIZE;
2171
2172         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2173                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2174                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2175                                 nla_nest_cancel(skb, mp_attr);
2176                                 return -EMSGSIZE;
2177                         }
2178
2179                         nhp->rtnh_flags = 0;
2180                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2181                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2182                         nhp->rtnh_len = sizeof(*nhp);
2183                 }
2184         }
2185
2186         nla_nest_end(skb, mp_attr);
2187
2188         lastuse = READ_ONCE(c->mfc_un.res.lastuse);
2189         lastuse = time_after_eq(jiffies, lastuse) ? jiffies - lastuse : 0;
2190
2191         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2192         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2193         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2194         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) ||
2195             nla_put_u64_64bit(skb, RTA_EXPIRES, jiffies_to_clock_t(lastuse),
2196                               RTA_PAD))
2197                 return -EMSGSIZE;
2198
2199         rtm->rtm_type = RTN_MULTICAST;
2200         return 1;
2201 }
2202
2203 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2204                    __be32 saddr, __be32 daddr,
2205                    struct rtmsg *rtm, u32 portid)
2206 {
2207         struct mfc_cache *cache;
2208         struct mr_table *mrt;
2209         int err;
2210
2211         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2212         if (!mrt)
2213                 return -ENOENT;
2214
2215         rcu_read_lock();
2216         cache = ipmr_cache_find(mrt, saddr, daddr);
2217         if (!cache && skb->dev) {
2218                 int vif = ipmr_find_vif(mrt, skb->dev);
2219
2220                 if (vif >= 0)
2221                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2222         }
2223         if (!cache) {
2224                 struct sk_buff *skb2;
2225                 struct iphdr *iph;
2226                 struct net_device *dev;
2227                 int vif = -1;
2228
2229                 dev = skb->dev;
2230                 read_lock(&mrt_lock);
2231                 if (dev)
2232                         vif = ipmr_find_vif(mrt, dev);
2233                 if (vif < 0) {
2234                         read_unlock(&mrt_lock);
2235                         rcu_read_unlock();
2236                         return -ENODEV;
2237                 }
2238                 skb2 = skb_clone(skb, GFP_ATOMIC);
2239                 if (!skb2) {
2240                         read_unlock(&mrt_lock);
2241                         rcu_read_unlock();
2242                         return -ENOMEM;
2243                 }
2244
2245                 NETLINK_CB(skb2).portid = portid;
2246                 skb_push(skb2, sizeof(struct iphdr));
2247                 skb_reset_network_header(skb2);
2248                 iph = ip_hdr(skb2);
2249                 iph->ihl = sizeof(struct iphdr) >> 2;
2250                 iph->saddr = saddr;
2251                 iph->daddr = daddr;
2252                 iph->version = 0;
2253                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2254                 read_unlock(&mrt_lock);
2255                 rcu_read_unlock();
2256                 return err;
2257         }
2258
2259         read_lock(&mrt_lock);
2260         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2261         read_unlock(&mrt_lock);
2262         rcu_read_unlock();
2263         return err;
2264 }
2265
2266 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2267                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2268                             int flags)
2269 {
2270         struct nlmsghdr *nlh;
2271         struct rtmsg *rtm;
2272         int err;
2273
2274         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2275         if (!nlh)
2276                 return -EMSGSIZE;
2277
2278         rtm = nlmsg_data(nlh);
2279         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2280         rtm->rtm_dst_len  = 32;
2281         rtm->rtm_src_len  = 32;
2282         rtm->rtm_tos      = 0;
2283         rtm->rtm_table    = mrt->id;
2284         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2285                 goto nla_put_failure;
2286         rtm->rtm_type     = RTN_MULTICAST;
2287         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2288         if (c->mfc_flags & MFC_STATIC)
2289                 rtm->rtm_protocol = RTPROT_STATIC;
2290         else
2291                 rtm->rtm_protocol = RTPROT_MROUTED;
2292         rtm->rtm_flags    = 0;
2293
2294         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2295             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2296                 goto nla_put_failure;
2297         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2298         /* do not break the dump if cache is unresolved */
2299         if (err < 0 && err != -ENOENT)
2300                 goto nla_put_failure;
2301
2302         nlmsg_end(skb, nlh);
2303         return 0;
2304
2305 nla_put_failure:
2306         nlmsg_cancel(skb, nlh);
2307         return -EMSGSIZE;
2308 }
2309
2310 static size_t mroute_msgsize(bool unresolved, int maxvif)
2311 {
2312         size_t len =
2313                 NLMSG_ALIGN(sizeof(struct rtmsg))
2314                 + nla_total_size(4)     /* RTA_TABLE */
2315                 + nla_total_size(4)     /* RTA_SRC */
2316                 + nla_total_size(4)     /* RTA_DST */
2317                 ;
2318
2319         if (!unresolved)
2320                 len = len
2321                       + nla_total_size(4)       /* RTA_IIF */
2322                       + nla_total_size(0)       /* RTA_MULTIPATH */
2323                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2324                                                 /* RTA_MFC_STATS */
2325                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2326                 ;
2327
2328         return len;
2329 }
2330
2331 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2332                                  int cmd)
2333 {
2334         struct net *net = read_pnet(&mrt->net);
2335         struct sk_buff *skb;
2336         int err = -ENOBUFS;
2337
2338         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2339                         GFP_ATOMIC);
2340         if (!skb)
2341                 goto errout;
2342
2343         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2344         if (err < 0)
2345                 goto errout;
2346
2347         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2348         return;
2349
2350 errout:
2351         kfree_skb(skb);
2352         if (err < 0)
2353                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2354 }
2355
2356 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2357 {
2358         size_t len =
2359                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2360                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2361                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2362                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2363                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2364                                         /* IPMRA_CREPORT_PKT */
2365                 + nla_total_size(payloadlen)
2366                 ;
2367
2368         return len;
2369 }
2370
2371 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2372 {
2373         struct net *net = read_pnet(&mrt->net);
2374         struct nlmsghdr *nlh;
2375         struct rtgenmsg *rtgenm;
2376         struct igmpmsg *msg;
2377         struct sk_buff *skb;
2378         struct nlattr *nla;
2379         int payloadlen;
2380
2381         payloadlen = pkt->len - sizeof(struct igmpmsg);
2382         msg = (struct igmpmsg *)skb_network_header(pkt);
2383
2384         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2385         if (!skb)
2386                 goto errout;
2387
2388         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2389                         sizeof(struct rtgenmsg), 0);
2390         if (!nlh)
2391                 goto errout;
2392         rtgenm = nlmsg_data(nlh);
2393         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2394         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2395             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif) ||
2396             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2397                             msg->im_src.s_addr) ||
2398             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2399                             msg->im_dst.s_addr))
2400                 goto nla_put_failure;
2401
2402         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2403         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2404                                   nla_data(nla), payloadlen))
2405                 goto nla_put_failure;
2406
2407         nlmsg_end(skb, nlh);
2408
2409         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2410         return;
2411
2412 nla_put_failure:
2413         nlmsg_cancel(skb, nlh);
2414 errout:
2415         kfree_skb(skb);
2416         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2417 }
2418
2419 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2420                              struct netlink_ext_ack *extack)
2421 {
2422         struct net *net = sock_net(in_skb->sk);
2423         struct nlattr *tb[RTA_MAX + 1];
2424         struct sk_buff *skb = NULL;
2425         struct mfc_cache *cache;
2426         struct mr_table *mrt;
2427         struct rtmsg *rtm;
2428         __be32 src, grp;
2429         u32 tableid;
2430         int err;
2431
2432         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2433                           rtm_ipv4_policy, extack);
2434         if (err < 0)
2435                 goto errout;
2436
2437         rtm = nlmsg_data(nlh);
2438
2439         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2440         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2441         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2442
2443         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2444         if (!mrt) {
2445                 err = -ENOENT;
2446                 goto errout_free;
2447         }
2448
2449         /* entries are added/deleted only under RTNL */
2450         rcu_read_lock();
2451         cache = ipmr_cache_find(mrt, src, grp);
2452         rcu_read_unlock();
2453         if (!cache) {
2454                 err = -ENOENT;
2455                 goto errout_free;
2456         }
2457
2458         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2459         if (!skb) {
2460                 err = -ENOBUFS;
2461                 goto errout_free;
2462         }
2463
2464         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2465                                nlh->nlmsg_seq, cache,
2466                                RTM_NEWROUTE, 0);
2467         if (err < 0)
2468                 goto errout_free;
2469
2470         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2471
2472 errout:
2473         return err;
2474
2475 errout_free:
2476         kfree_skb(skb);
2477         goto errout;
2478 }
2479
2480 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2481 {
2482         struct net *net = sock_net(skb->sk);
2483         struct mr_table *mrt;
2484         struct mfc_cache *mfc;
2485         unsigned int t = 0, s_t;
2486         unsigned int e = 0, s_e;
2487
2488         s_t = cb->args[0];
2489         s_e = cb->args[1];
2490
2491         rcu_read_lock();
2492         ipmr_for_each_table(mrt, net) {
2493                 if (t < s_t)
2494                         goto next_table;
2495                 list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list) {
2496                         if (e < s_e)
2497                                 goto next_entry;
2498                         if (ipmr_fill_mroute(mrt, skb,
2499                                              NETLINK_CB(cb->skb).portid,
2500                                              cb->nlh->nlmsg_seq,
2501                                              mfc, RTM_NEWROUTE,
2502                                              NLM_F_MULTI) < 0)
2503                                 goto done;
2504 next_entry:
2505                         e++;
2506                 }
2507
2508                 spin_lock_bh(&mfc_unres_lock);
2509                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2510                         if (e < s_e)
2511                                 goto next_entry2;
2512                         if (ipmr_fill_mroute(mrt, skb,
2513                                              NETLINK_CB(cb->skb).portid,
2514                                              cb->nlh->nlmsg_seq,
2515                                              mfc, RTM_NEWROUTE,
2516                                              NLM_F_MULTI) < 0) {
2517                                 spin_unlock_bh(&mfc_unres_lock);
2518                                 goto done;
2519                         }
2520 next_entry2:
2521                         e++;
2522                 }
2523                 spin_unlock_bh(&mfc_unres_lock);
2524                 e = 0;
2525                 s_e = 0;
2526 next_table:
2527                 t++;
2528         }
2529 done:
2530         rcu_read_unlock();
2531
2532         cb->args[1] = e;
2533         cb->args[0] = t;
2534
2535         return skb->len;
2536 }
2537
2538 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2539         [RTA_SRC]       = { .type = NLA_U32 },
2540         [RTA_DST]       = { .type = NLA_U32 },
2541         [RTA_IIF]       = { .type = NLA_U32 },
2542         [RTA_TABLE]     = { .type = NLA_U32 },
2543         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2544 };
2545
2546 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2547 {
2548         switch (rtm_protocol) {
2549         case RTPROT_STATIC:
2550         case RTPROT_MROUTED:
2551                 return true;
2552         }
2553         return false;
2554 }
2555
2556 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2557 {
2558         struct rtnexthop *rtnh = nla_data(nla);
2559         int remaining = nla_len(nla), vifi = 0;
2560
2561         while (rtnh_ok(rtnh, remaining)) {
2562                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2563                 if (++vifi == MAXVIFS)
2564                         break;
2565                 rtnh = rtnh_next(rtnh, &remaining);
2566         }
2567
2568         return remaining > 0 ? -EINVAL : vifi;
2569 }
2570
2571 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2572 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2573                             struct mfcctl *mfcc, int *mrtsock,
2574                             struct mr_table **mrtret,
2575                             struct netlink_ext_ack *extack)
2576 {
2577         struct net_device *dev = NULL;
2578         u32 tblid = RT_TABLE_DEFAULT;
2579         struct mr_table *mrt;
2580         struct nlattr *attr;
2581         struct rtmsg *rtm;
2582         int ret, rem;
2583
2584         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy,
2585                              extack);
2586         if (ret < 0)
2587                 goto out;
2588         rtm = nlmsg_data(nlh);
2589
2590         ret = -EINVAL;
2591         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2592             rtm->rtm_type != RTN_MULTICAST ||
2593             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2594             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2595                 goto out;
2596
2597         memset(mfcc, 0, sizeof(*mfcc));
2598         mfcc->mfcc_parent = -1;
2599         ret = 0;
2600         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2601                 switch (nla_type(attr)) {
2602                 case RTA_SRC:
2603                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2604                         break;
2605                 case RTA_DST:
2606                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2607                         break;
2608                 case RTA_IIF:
2609                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2610                         if (!dev) {
2611                                 ret = -ENODEV;
2612                                 goto out;
2613                         }
2614                         break;
2615                 case RTA_MULTIPATH:
2616                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2617                                 ret = -EINVAL;
2618                                 goto out;
2619                         }
2620                         break;
2621                 case RTA_PREFSRC:
2622                         ret = 1;
2623                         break;
2624                 case RTA_TABLE:
2625                         tblid = nla_get_u32(attr);
2626                         break;
2627                 }
2628         }
2629         mrt = ipmr_get_table(net, tblid);
2630         if (!mrt) {
2631                 ret = -ENOENT;
2632                 goto out;
2633         }
2634         *mrtret = mrt;
2635         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2636         if (dev)
2637                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2638
2639 out:
2640         return ret;
2641 }
2642
2643 /* takes care of both newroute and delroute */
2644 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2645                           struct netlink_ext_ack *extack)
2646 {
2647         struct net *net = sock_net(skb->sk);
2648         int ret, mrtsock, parent;
2649         struct mr_table *tbl;
2650         struct mfcctl mfcc;
2651
2652         mrtsock = 0;
2653         tbl = NULL;
2654         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2655         if (ret < 0)
2656                 return ret;
2657
2658         parent = ret ? mfcc.mfcc_parent : -1;
2659         if (nlh->nlmsg_type == RTM_NEWROUTE)
2660                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2661         else
2662                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2663 }
2664
2665 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2666 {
2667         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2668
2669         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2670             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2671             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2672                         mrt->mroute_reg_vif_num) ||
2673             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2674                        mrt->mroute_do_assert) ||
2675             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim))
2676                 return false;
2677
2678         return true;
2679 }
2680
2681 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2682 {
2683         struct nlattr *vif_nest;
2684         struct vif_device *vif;
2685
2686         /* if the VIF doesn't exist just continue */
2687         if (!VIF_EXISTS(mrt, vifid))
2688                 return true;
2689
2690         vif = &mrt->vif_table[vifid];
2691         vif_nest = nla_nest_start(skb, IPMRA_VIF);
2692         if (!vif_nest)
2693                 return false;
2694         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2695             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2696             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2697             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2698                               IPMRA_VIFA_PAD) ||
2699             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2700                               IPMRA_VIFA_PAD) ||
2701             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2702                               IPMRA_VIFA_PAD) ||
2703             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2704                               IPMRA_VIFA_PAD) ||
2705             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2706             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2707                 nla_nest_cancel(skb, vif_nest);
2708                 return false;
2709         }
2710         nla_nest_end(skb, vif_nest);
2711
2712         return true;
2713 }
2714
2715 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2716 {
2717         struct net *net = sock_net(skb->sk);
2718         struct nlmsghdr *nlh = NULL;
2719         unsigned int t = 0, s_t;
2720         unsigned int e = 0, s_e;
2721         struct mr_table *mrt;
2722
2723         s_t = cb->args[0];
2724         s_e = cb->args[1];
2725
2726         ipmr_for_each_table(mrt, net) {
2727                 struct nlattr *vifs, *af;
2728                 struct ifinfomsg *hdr;
2729                 u32 i;
2730
2731                 if (t < s_t)
2732                         goto skip_table;
2733                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2734                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2735                                 sizeof(*hdr), NLM_F_MULTI);
2736                 if (!nlh)
2737                         break;
2738
2739                 hdr = nlmsg_data(nlh);
2740                 memset(hdr, 0, sizeof(*hdr));
2741                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2742
2743                 af = nla_nest_start(skb, IFLA_AF_SPEC);
2744                 if (!af) {
2745                         nlmsg_cancel(skb, nlh);
2746                         goto out;
2747                 }
2748
2749                 if (!ipmr_fill_table(mrt, skb)) {
2750                         nlmsg_cancel(skb, nlh);
2751                         goto out;
2752                 }
2753
2754                 vifs = nla_nest_start(skb, IPMRA_TABLE_VIFS);
2755                 if (!vifs) {
2756                         nla_nest_end(skb, af);
2757                         nlmsg_end(skb, nlh);
2758                         goto out;
2759                 }
2760                 for (i = 0; i < mrt->maxvif; i++) {
2761                         if (e < s_e)
2762                                 goto skip_entry;
2763                         if (!ipmr_fill_vif(mrt, i, skb)) {
2764                                 nla_nest_end(skb, vifs);
2765                                 nla_nest_end(skb, af);
2766                                 nlmsg_end(skb, nlh);
2767                                 goto out;
2768                         }
2769 skip_entry:
2770                         e++;
2771                 }
2772                 s_e = 0;
2773                 e = 0;
2774                 nla_nest_end(skb, vifs);
2775                 nla_nest_end(skb, af);
2776                 nlmsg_end(skb, nlh);
2777 skip_table:
2778                 t++;
2779         }
2780
2781 out:
2782         cb->args[1] = e;
2783         cb->args[0] = t;
2784
2785         return skb->len;
2786 }
2787
2788 #ifdef CONFIG_PROC_FS
2789 /* The /proc interfaces to multicast routing :
2790  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2791  */
2792 struct ipmr_vif_iter {
2793         struct seq_net_private p;
2794         struct mr_table *mrt;
2795         int ct;
2796 };
2797
2798 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2799                                            struct ipmr_vif_iter *iter,
2800                                            loff_t pos)
2801 {
2802         struct mr_table *mrt = iter->mrt;
2803
2804         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2805                 if (!VIF_EXISTS(mrt, iter->ct))
2806                         continue;
2807                 if (pos-- == 0)
2808                         return &mrt->vif_table[iter->ct];
2809         }
2810         return NULL;
2811 }
2812
2813 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2814         __acquires(mrt_lock)
2815 {
2816         struct ipmr_vif_iter *iter = seq->private;
2817         struct net *net = seq_file_net(seq);
2818         struct mr_table *mrt;
2819
2820         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2821         if (!mrt)
2822                 return ERR_PTR(-ENOENT);
2823
2824         iter->mrt = mrt;
2825
2826         read_lock(&mrt_lock);
2827         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2828                 : SEQ_START_TOKEN;
2829 }
2830
2831 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2832 {
2833         struct ipmr_vif_iter *iter = seq->private;
2834         struct net *net = seq_file_net(seq);
2835         struct mr_table *mrt = iter->mrt;
2836
2837         ++*pos;
2838         if (v == SEQ_START_TOKEN)
2839                 return ipmr_vif_seq_idx(net, iter, 0);
2840
2841         while (++iter->ct < mrt->maxvif) {
2842                 if (!VIF_EXISTS(mrt, iter->ct))
2843                         continue;
2844                 return &mrt->vif_table[iter->ct];
2845         }
2846         return NULL;
2847 }
2848
2849 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2850         __releases(mrt_lock)
2851 {
2852         read_unlock(&mrt_lock);
2853 }
2854
2855 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2856 {
2857         struct ipmr_vif_iter *iter = seq->private;
2858         struct mr_table *mrt = iter->mrt;
2859
2860         if (v == SEQ_START_TOKEN) {
2861                 seq_puts(seq,
2862                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2863         } else {
2864                 const struct vif_device *vif = v;
2865                 const char *name =  vif->dev ? vif->dev->name : "none";
2866
2867                 seq_printf(seq,
2868                            "%2zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2869                            vif - mrt->vif_table,
2870                            name, vif->bytes_in, vif->pkt_in,
2871                            vif->bytes_out, vif->pkt_out,
2872                            vif->flags, vif->local, vif->remote);
2873         }
2874         return 0;
2875 }
2876
2877 static const struct seq_operations ipmr_vif_seq_ops = {
2878         .start = ipmr_vif_seq_start,
2879         .next  = ipmr_vif_seq_next,
2880         .stop  = ipmr_vif_seq_stop,
2881         .show  = ipmr_vif_seq_show,
2882 };
2883
2884 static int ipmr_vif_open(struct inode *inode, struct file *file)
2885 {
2886         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2887                             sizeof(struct ipmr_vif_iter));
2888 }
2889
2890 static const struct file_operations ipmr_vif_fops = {
2891         .owner   = THIS_MODULE,
2892         .open    = ipmr_vif_open,
2893         .read    = seq_read,
2894         .llseek  = seq_lseek,
2895         .release = seq_release_net,
2896 };
2897
2898 struct ipmr_mfc_iter {
2899         struct seq_net_private p;
2900         struct mr_table *mrt;
2901         struct list_head *cache;
2902 };
2903
2904 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2905                                           struct ipmr_mfc_iter *it, loff_t pos)
2906 {
2907         struct mr_table *mrt = it->mrt;
2908         struct mfc_cache *mfc;
2909
2910         rcu_read_lock();
2911         it->cache = &mrt->mfc_cache_list;
2912         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list)
2913                 if (pos-- == 0)
2914                         return mfc;
2915         rcu_read_unlock();
2916
2917         spin_lock_bh(&mfc_unres_lock);
2918         it->cache = &mrt->mfc_unres_queue;
2919         list_for_each_entry(mfc, it->cache, list)
2920                 if (pos-- == 0)
2921                         return mfc;
2922         spin_unlock_bh(&mfc_unres_lock);
2923
2924         it->cache = NULL;
2925         return NULL;
2926 }
2927
2928
2929 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2930 {
2931         struct ipmr_mfc_iter *it = seq->private;
2932         struct net *net = seq_file_net(seq);
2933         struct mr_table *mrt;
2934
2935         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2936         if (!mrt)
2937                 return ERR_PTR(-ENOENT);
2938
2939         it->mrt = mrt;
2940         it->cache = NULL;
2941         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2942                 : SEQ_START_TOKEN;
2943 }
2944
2945 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2946 {
2947         struct ipmr_mfc_iter *it = seq->private;
2948         struct net *net = seq_file_net(seq);
2949         struct mr_table *mrt = it->mrt;
2950         struct mfc_cache *mfc = v;
2951
2952         ++*pos;
2953
2954         if (v == SEQ_START_TOKEN)
2955                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2956
2957         if (mfc->list.next != it->cache)
2958                 return list_entry(mfc->list.next, struct mfc_cache, list);
2959
2960         if (it->cache == &mrt->mfc_unres_queue)
2961                 goto end_of_list;
2962
2963         /* exhausted cache_array, show unresolved */
2964         rcu_read_unlock();
2965         it->cache = &mrt->mfc_unres_queue;
2966
2967         spin_lock_bh(&mfc_unres_lock);
2968         if (!list_empty(it->cache))
2969                 return list_first_entry(it->cache, struct mfc_cache, list);
2970
2971 end_of_list:
2972         spin_unlock_bh(&mfc_unres_lock);
2973         it->cache = NULL;
2974
2975         return NULL;
2976 }
2977
2978 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2979 {
2980         struct ipmr_mfc_iter *it = seq->private;
2981         struct mr_table *mrt = it->mrt;
2982
2983         if (it->cache == &mrt->mfc_unres_queue)
2984                 spin_unlock_bh(&mfc_unres_lock);
2985         else if (it->cache == &mrt->mfc_cache_list)
2986                 rcu_read_unlock();
2987 }
2988
2989 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2990 {
2991         int n;
2992
2993         if (v == SEQ_START_TOKEN) {
2994                 seq_puts(seq,
2995                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2996         } else {
2997                 const struct mfc_cache *mfc = v;
2998                 const struct ipmr_mfc_iter *it = seq->private;
2999                 const struct mr_table *mrt = it->mrt;
3000
3001                 seq_printf(seq, "%08X %08X %-3hd",
3002                            (__force u32) mfc->mfc_mcastgrp,
3003                            (__force u32) mfc->mfc_origin,
3004                            mfc->mfc_parent);
3005
3006                 if (it->cache != &mrt->mfc_unres_queue) {
3007                         seq_printf(seq, " %8lu %8lu %8lu",
3008                                    mfc->mfc_un.res.pkt,
3009                                    mfc->mfc_un.res.bytes,
3010                                    mfc->mfc_un.res.wrong_if);
3011                         for (n = mfc->mfc_un.res.minvif;
3012                              n < mfc->mfc_un.res.maxvif; n++) {
3013                                 if (VIF_EXISTS(mrt, n) &&
3014                                     mfc->mfc_un.res.ttls[n] < 255)
3015                                         seq_printf(seq,
3016                                            " %2d:%-3d",
3017                                            n, mfc->mfc_un.res.ttls[n]);
3018                         }
3019                 } else {
3020                         /* unresolved mfc_caches don't contain
3021                          * pkt, bytes and wrong_if values
3022                          */
3023                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
3024                 }
3025                 seq_putc(seq, '\n');
3026         }
3027         return 0;
3028 }
3029
3030 static const struct seq_operations ipmr_mfc_seq_ops = {
3031         .start = ipmr_mfc_seq_start,
3032         .next  = ipmr_mfc_seq_next,
3033         .stop  = ipmr_mfc_seq_stop,
3034         .show  = ipmr_mfc_seq_show,
3035 };
3036
3037 static int ipmr_mfc_open(struct inode *inode, struct file *file)
3038 {
3039         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
3040                             sizeof(struct ipmr_mfc_iter));
3041 }
3042
3043 static const struct file_operations ipmr_mfc_fops = {
3044         .owner   = THIS_MODULE,
3045         .open    = ipmr_mfc_open,
3046         .read    = seq_read,
3047         .llseek  = seq_lseek,
3048         .release = seq_release_net,
3049 };
3050 #endif
3051
3052 #ifdef CONFIG_IP_PIMSM_V2
3053 static const struct net_protocol pim_protocol = {
3054         .handler        =       pim_rcv,
3055         .netns_ok       =       1,
3056 };
3057 #endif
3058
3059 /* Setup for IP multicast routing */
3060 static int __net_init ipmr_net_init(struct net *net)
3061 {
3062         int err;
3063
3064         err = ipmr_rules_init(net);
3065         if (err < 0)
3066                 goto fail;
3067
3068 #ifdef CONFIG_PROC_FS
3069         err = -ENOMEM;
3070         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
3071                 goto proc_vif_fail;
3072         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
3073                 goto proc_cache_fail;
3074 #endif
3075         return 0;
3076
3077 #ifdef CONFIG_PROC_FS
3078 proc_cache_fail:
3079         remove_proc_entry("ip_mr_vif", net->proc_net);
3080 proc_vif_fail:
3081         ipmr_rules_exit(net);
3082 #endif
3083 fail:
3084         return err;
3085 }
3086
3087 static void __net_exit ipmr_net_exit(struct net *net)
3088 {
3089 #ifdef CONFIG_PROC_FS
3090         remove_proc_entry("ip_mr_cache", net->proc_net);
3091         remove_proc_entry("ip_mr_vif", net->proc_net);
3092 #endif
3093         ipmr_rules_exit(net);
3094 }
3095
3096 static struct pernet_operations ipmr_net_ops = {
3097         .init = ipmr_net_init,
3098         .exit = ipmr_net_exit,
3099 };
3100
3101 int __init ip_mr_init(void)
3102 {
3103         int err;
3104
3105         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3106                                        sizeof(struct mfc_cache),
3107                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3108                                        NULL);
3109
3110         err = register_pernet_subsys(&ipmr_net_ops);
3111         if (err)
3112                 goto reg_pernet_fail;
3113
3114         err = register_netdevice_notifier(&ip_mr_notifier);
3115         if (err)
3116                 goto reg_notif_fail;
3117 #ifdef CONFIG_IP_PIMSM_V2
3118         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3119                 pr_err("%s: can't add PIM protocol\n", __func__);
3120                 err = -EAGAIN;
3121                 goto add_proto_fail;
3122         }
3123 #endif
3124         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3125                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3126         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3127                       ipmr_rtm_route, NULL, 0);
3128         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3129                       ipmr_rtm_route, NULL, 0);
3130
3131         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3132                       NULL, ipmr_rtm_dumplink, 0);
3133         return 0;
3134
3135 #ifdef CONFIG_IP_PIMSM_V2
3136 add_proto_fail:
3137         unregister_netdevice_notifier(&ip_mr_notifier);
3138 #endif
3139 reg_notif_fail:
3140         unregister_pernet_subsys(&ipmr_net_ops);
3141 reg_pernet_fail:
3142         kmem_cache_destroy(mrt_cachep);
3143         return err;
3144 }