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         ipmr_free_table(mrt);
 252 err1:
 253         fib_rules_unregister(ops);
 254         return err;
 255 }
 256
 257 static void __net_exit ipmr_rules_exit(struct net *net)
 258 {
 259         struct mr_table *mrt, *next;
 260
 261         rtnl_lock();
 262         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 263                 list_del(&mrt->list);
 264                 ipmr_free_table(mrt);
 265         }
 266         fib_rules_unregister(net->ipv4.mr_rules_ops);
 267         rtnl_unlock();
 268 }
 269 #else
 270 #define ipmr_for_each_table(mrt, net) \
 271         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 272
 273 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 274 {
 275         return net->ipv4.mrt;
 276 }
 277
 278 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 279                            struct mr_table **mrt)
 280 {
 281         *mrt = net->ipv4.mrt;
 282         return 0;
 283 }
 284
 285 static int __net_init ipmr_rules_init(struct net *net)
 286 {
 287         struct mr_table *mrt;
 288
 289         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 290         if (IS_ERR(mrt))
 291                 return PTR_ERR(mrt);
 292         net->ipv4.mrt = mrt;
 293         return 0;
 294 }
 295
 296 static void __net_exit ipmr_rules_exit(struct net *net)
 297 {
 298         rtnl_lock();
 299         ipmr_free_table(net->ipv4.mrt);
 300         net->ipv4.mrt = NULL;
 301         rtnl_unlock();
 302 }
 303 #endif
 304
 305 static inline int ipmr_hash_cmp(struct rhashtable_compare_arg *arg,
 306                                 const void *ptr)
 307 {
 308         const struct mfc_cache_cmp_arg *cmparg = arg->key;
 309         struct mfc_cache *c = (struct mfc_cache *)ptr;
 310
 311         return cmparg->mfc_mcastgrp != c->mfc_mcastgrp ||
 312                cmparg->mfc_origin != c->mfc_origin;
 313 }
 314
 315 static const struct rhashtable_params ipmr_rht_params = {
 316         .head_offset = offsetof(struct mfc_cache, mnode),
 317         .key_offset = offsetof(struct mfc_cache, cmparg),
 318         .key_len = sizeof(struct mfc_cache_cmp_arg),
 319         .nelem_hint = 3,
 320         .locks_mul = 1,
 321         .obj_cmpfn = ipmr_hash_cmp,
 322         .automatic_shrinking = true,
 323 };
 324
 325 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 326 {
 327         struct mr_table *mrt;
 328         int err;
 329
 330         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 331         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 332                 return ERR_PTR(-EINVAL);
 333
 334         mrt = ipmr_get_table(net, id);
 335         if (mrt)
 336                 return mrt;
 337
 338         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 339         if (!mrt)
 340                 return ERR_PTR(-ENOMEM);
 341         write_pnet(&mrt->net, net);
 342         mrt->id = id;
 343
 344         err = rhltable_init(&mrt->mfc_hash, &ipmr_rht_params);
 345         if (err) {
 346                 kfree(mrt);
 347                 return ERR_PTR(err);
 348         }
 349         INIT_LIST_HEAD(&mrt->mfc_cache_list);
 350         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 351
 352         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 353                     (unsigned long)mrt);
 354
 355         mrt->mroute_reg_vif_num = -1;
 356 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 357         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 358 #endif
 359         return mrt;
 360 }
 361
 362 static void ipmr_free_table(struct mr_table *mrt)
 363 {
 364         del_timer_sync(&mrt->ipmr_expire_timer);
 365         mroute_clean_tables(mrt, true);
 366         rhltable_destroy(&mrt->mfc_hash);
 367         kfree(mrt);
 368 }
 369
 370 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 371
 372 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 373 {
 374         struct net *net = dev_net(dev);
 375
 376         dev_close(dev);
 377
 378         dev = __dev_get_by_name(net, "tunl0");
 379         if (dev) {
 380                 const struct net_device_ops *ops = dev->netdev_ops;
 381                 struct ifreq ifr;
 382                 struct ip_tunnel_parm p;
 383
 384                 memset(&p, 0, sizeof(p));
 385                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 386                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 387                 p.iph.version = 4;
 388                 p.iph.ihl = 5;
 389                 p.iph.protocol = IPPROTO_IPIP;
 390                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 391                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 392
 393                 if (ops->ndo_do_ioctl) {
 394                         mm_segment_t oldfs = get_fs();
 395
 396                         set_fs(KERNEL_DS);
 397                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 398                         set_fs(oldfs);
 399                 }
 400         }
 401 }
 402
 403 /* Initialize ipmr pimreg/tunnel in_device */
 404 static bool ipmr_init_vif_indev(const struct net_device *dev)
 405 {
 406         struct in_device *in_dev;
 407
 408         ASSERT_RTNL();
 409
 410         in_dev = __in_dev_get_rtnl(dev);
 411         if (!in_dev)
 412                 return false;
 413         ipv4_devconf_setall(in_dev);
 414         neigh_parms_data_state_setall(in_dev->arp_parms);
 415         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 416
 417         return true;
 418 }
 419
 420 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 421 {
 422         struct net_device  *dev;
 423
 424         dev = __dev_get_by_name(net, "tunl0");
 425
 426         if (dev) {
 427                 const struct net_device_ops *ops = dev->netdev_ops;
 428                 int err;
 429                 struct ifreq ifr;
 430                 struct ip_tunnel_parm p;
 431
 432                 memset(&p, 0, sizeof(p));
 433                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 434                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 435                 p.iph.version = 4;
 436                 p.iph.ihl = 5;
 437                 p.iph.protocol = IPPROTO_IPIP;
 438                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 439                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 440
 441                 if (ops->ndo_do_ioctl) {
 442                         mm_segment_t oldfs = get_fs();
 443
 444                         set_fs(KERNEL_DS);
 445                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 446                         set_fs(oldfs);
 447                 } else {
 448                         err = -EOPNOTSUPP;
 449                 }
 450                 dev = NULL;
 451
 452                 if (err == 0 &&
 453                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 454                         dev->flags |= IFF_MULTICAST;
 455                         if (!ipmr_init_vif_indev(dev))
 456                                 goto failure;
 457                         if (dev_open(dev))
 458                                 goto failure;
 459                         dev_hold(dev);
 460                 }
 461         }
 462         return dev;
 463
 464 failure:
 465         unregister_netdevice(dev);
 466         return NULL;
 467 }
 468
 469 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 470 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 471 {
 472         struct net *net = dev_net(dev);
 473         struct mr_table *mrt;
 474         struct flowi4 fl4 = {
 475                 .flowi4_oif     = dev->ifindex,
 476                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 477                 .flowi4_mark    = skb->mark,
 478         };
 479         int err;
 480
 481         err = ipmr_fib_lookup(net, &fl4, &mrt);
 482         if (err < 0) {
 483                 kfree_skb(skb);
 484                 return err;
 485         }
 486
 487         read_lock(&mrt_lock);
 488         dev->stats.tx_bytes += skb->len;
 489         dev->stats.tx_packets++;
 490         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 491         read_unlock(&mrt_lock);
 492         kfree_skb(skb);
 493         return NETDEV_TX_OK;
 494 }
 495
 496 static int reg_vif_get_iflink(const struct net_device *dev)
 497 {
 498         return 0;
 499 }
 500
 501 static const struct net_device_ops reg_vif_netdev_ops = {
 502         .ndo_start_xmit = reg_vif_xmit,
 503         .ndo_get_iflink = reg_vif_get_iflink,
 504 };
 505
 506 static void reg_vif_setup(struct net_device *dev)
 507 {
 508         dev->type               = ARPHRD_PIMREG;
 509         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 510         dev->flags              = IFF_NOARP;
 511         dev->netdev_ops         = &reg_vif_netdev_ops;
 512         dev->needs_free_netdev  = true;
 513         dev->features           |= NETIF_F_NETNS_LOCAL;
 514 }
 515
 516 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 517 {
 518         struct net_device *dev;
 519         char name[IFNAMSIZ];
 520
 521         if (mrt->id == RT_TABLE_DEFAULT)
 522                 sprintf(name, "pimreg");
 523         else
 524                 sprintf(name, "pimreg%u", mrt->id);
 525
 526         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 527
 528         if (!dev)
 529                 return NULL;
 530
 531         dev_net_set(dev, net);
 532
 533         if (register_netdevice(dev)) {
 534                 free_netdev(dev);
 535                 return NULL;
 536         }
 537
 538         if (!ipmr_init_vif_indev(dev))
 539                 goto failure;
 540         if (dev_open(dev))
 541                 goto failure;
 542
 543         dev_hold(dev);
 544
 545         return dev;
 546
 547 failure:
 548         unregister_netdevice(dev);
 549         return NULL;
 550 }
 551
 552 /* called with rcu_read_lock() */
 553 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 554                      unsigned int pimlen)
 555 {
 556         struct net_device *reg_dev = NULL;
 557         struct iphdr *encap;
 558
 559         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 560         /* Check that:
 561          * a. packet is really sent to a multicast group
 562          * b. packet is not a NULL-REGISTER
 563          * c. packet is not truncated
 564          */
 565         if (!ipv4_is_multicast(encap->daddr) ||
 566             encap->tot_len == 0 ||
 567             ntohs(encap->tot_len) + pimlen > skb->len)
 568                 return 1;
 569
 570         read_lock(&mrt_lock);
 571         if (mrt->mroute_reg_vif_num >= 0)
 572                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 573         read_unlock(&mrt_lock);
 574
 575         if (!reg_dev)
 576                 return 1;
 577
 578         skb->mac_header = skb->network_header;
 579         skb_pull(skb, (u8 *)encap - skb->data);
 580         skb_reset_network_header(skb);
 581         skb->protocol = htons(ETH_P_IP);
 582         skb->ip_summed = CHECKSUM_NONE;
 583
 584         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 585
 586         netif_rx(skb);
 587
 588         return NET_RX_SUCCESS;
 589 }
 590 #else
 591 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 592 {
 593         return NULL;
 594 }
 595 #endif
 596
 597 /**
 598  *      vif_delete - Delete a VIF entry
 599  *      @notify: Set to 1, if the caller is a notifier_call
 600  */
 601 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 602                       struct list_head *head)
 603 {
 604         struct vif_device *v;
 605         struct net_device *dev;
 606         struct in_device *in_dev;
 607
 608         if (vifi < 0 || vifi >= mrt->maxvif)
 609                 return -EADDRNOTAVAIL;
 610
 611         v = &mrt->vif_table[vifi];
 612
 613         write_lock_bh(&mrt_lock);
 614         dev = v->dev;
 615         v->dev = NULL;
 616
 617         if (!dev) {
 618                 write_unlock_bh(&mrt_lock);
 619                 return -EADDRNOTAVAIL;
 620         }
 621
 622         if (vifi == mrt->mroute_reg_vif_num)
 623                 mrt->mroute_reg_vif_num = -1;
 624
 625         if (vifi + 1 == mrt->maxvif) {
 626                 int tmp;
 627
 628                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 629                         if (VIF_EXISTS(mrt, tmp))
 630                                 break;
 631                 }
 632                 mrt->maxvif = tmp+1;
 633         }
 634
 635         write_unlock_bh(&mrt_lock);
 636
 637         dev_set_allmulti(dev, -1);
 638
 639         in_dev = __in_dev_get_rtnl(dev);
 640         if (in_dev) {
 641                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 642                 inet_netconf_notify_devconf(dev_net(dev), RTM_NEWNETCONF,
 643                                             NETCONFA_MC_FORWARDING,
 644                                             dev->ifindex, &in_dev->cnf);
 645                 ip_rt_multicast_event(in_dev);
 646         }
 647
 648         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 649                 unregister_netdevice_queue(dev, head);
 650
 651         dev_put(dev);
 652         return 0;
 653 }
 654
 655 static void ipmr_cache_free_rcu(struct rcu_head *head)
 656 {
 657         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 658
 659         kmem_cache_free(mrt_cachep, c);
 660 }
 661
 662 static inline void ipmr_cache_free(struct mfc_cache *c)
 663 {
 664         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 665 }
 666
 667 /* Destroy an unresolved cache entry, killing queued skbs
 668  * and reporting error to netlink readers.
 669  */
 670 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 671 {
 672         struct net *net = read_pnet(&mrt->net);
 673         struct sk_buff *skb;
 674         struct nlmsgerr *e;
 675
 676         atomic_dec(&mrt->cache_resolve_queue_len);
 677
 678         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 679                 if (ip_hdr(skb)->version == 0) {
 680                         struct nlmsghdr *nlh = skb_pull(skb,
 681                                                         sizeof(struct iphdr));
 682                         nlh->nlmsg_type = NLMSG_ERROR;
 683                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 684                         skb_trim(skb, nlh->nlmsg_len);
 685                         e = nlmsg_data(nlh);
 686                         e->error = -ETIMEDOUT;
 687                         memset(&e->msg, 0, sizeof(e->msg));
 688
 689                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 690                 } else {
 691                         kfree_skb(skb);
 692                 }
 693         }
 694
 695         ipmr_cache_free(c);
 696 }
 697
 698 /* Timer process for the unresolved queue. */
 699 static void ipmr_expire_process(unsigned long arg)
 700 {
 701         struct mr_table *mrt = (struct mr_table *)arg;
 702         unsigned long now;
 703         unsigned long expires;
 704         struct mfc_cache *c, *next;
 705
 706         if (!spin_trylock(&mfc_unres_lock)) {
 707                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 708                 return;
 709         }
 710
 711         if (list_empty(&mrt->mfc_unres_queue))
 712                 goto out;
 713
 714         now = jiffies;
 715         expires = 10*HZ;
 716
 717         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 718                 if (time_after(c->mfc_un.unres.expires, now)) {
 719                         unsigned long interval = c->mfc_un.unres.expires - now;
 720                         if (interval < expires)
 721                                 expires = interval;
 722                         continue;
 723                 }
 724
 725                 list_del(&c->list);
 726                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 727                 ipmr_destroy_unres(mrt, c);
 728         }
 729
 730         if (!list_empty(&mrt->mfc_unres_queue))
 731                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 732
 733 out:
 734         spin_unlock(&mfc_unres_lock);
 735 }
 736
 737 /* Fill oifs list. It is called under write locked mrt_lock. */
 738 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 739                                    unsigned char *ttls)
 740 {
 741         int vifi;
 742
 743         cache->mfc_un.res.minvif = MAXVIFS;
 744         cache->mfc_un.res.maxvif = 0;
 745         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 746
 747         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 748                 if (VIF_EXISTS(mrt, vifi) &&
 749                     ttls[vifi] && ttls[vifi] < 255) {
 750                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 751                         if (cache->mfc_un.res.minvif > vifi)
 752                                 cache->mfc_un.res.minvif = vifi;
 753                         if (cache->mfc_un.res.maxvif <= vifi)
 754                                 cache->mfc_un.res.maxvif = vifi + 1;
 755                 }
 756         }
 757         cache->mfc_un.res.lastuse = jiffies;
 758 }
 759
 760 static int vif_add(struct net *net, struct mr_table *mrt,
 761                    struct vifctl *vifc, int mrtsock)
 762 {
 763         int vifi = vifc->vifc_vifi;
 764         struct vif_device *v = &mrt->vif_table[vifi];
 765         struct net_device *dev;
 766         struct in_device *in_dev;
 767         int err;
 768
 769         /* Is vif busy ? */
 770         if (VIF_EXISTS(mrt, vifi))
 771                 return -EADDRINUSE;
 772
 773         switch (vifc->vifc_flags) {
 774         case VIFF_REGISTER:
 775                 if (!ipmr_pimsm_enabled())
 776                         return -EINVAL;
 777                 /* Special Purpose VIF in PIM
 778                  * All the packets will be sent to the daemon
 779                  */
 780                 if (mrt->mroute_reg_vif_num >= 0)
 781                         return -EADDRINUSE;
 782                 dev = ipmr_reg_vif(net, mrt);
 783                 if (!dev)
 784                         return -ENOBUFS;
 785                 err = dev_set_allmulti(dev, 1);
 786                 if (err) {
 787                         unregister_netdevice(dev);
 788                         dev_put(dev);
 789                         return err;
 790                 }
 791                 break;
 792         case VIFF_TUNNEL:
 793                 dev = ipmr_new_tunnel(net, vifc);
 794                 if (!dev)
 795                         return -ENOBUFS;
 796                 err = dev_set_allmulti(dev, 1);
 797                 if (err) {
 798                         ipmr_del_tunnel(dev, vifc);
 799                         dev_put(dev);
 800                         return err;
 801                 }
 802                 break;
 803         case VIFF_USE_IFINDEX:
 804         case 0:
 805                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 806                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 807                         if (dev && !__in_dev_get_rtnl(dev)) {
 808                                 dev_put(dev);
 809                                 return -EADDRNOTAVAIL;
 810                         }
 811                 } else {
 812                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 813                 }
 814                 if (!dev)
 815                         return -EADDRNOTAVAIL;
 816                 err = dev_set_allmulti(dev, 1);
 817                 if (err) {
 818                         dev_put(dev);
 819                         return err;
 820                 }
 821                 break;
 822         default:
 823                 return -EINVAL;
 824         }
 825
 826         in_dev = __in_dev_get_rtnl(dev);
 827         if (!in_dev) {
 828                 dev_put(dev);
 829                 return -EADDRNOTAVAIL;
 830         }
 831         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 832         inet_netconf_notify_devconf(net, RTM_NEWNETCONF, NETCONFA_MC_FORWARDING,
 833                                     dev->ifindex, &in_dev->cnf);
 834         ip_rt_multicast_event(in_dev);
 835
 836         /* Fill in the VIF structures */
 837
 838         v->rate_limit = vifc->vifc_rate_limit;
 839         v->local = vifc->vifc_lcl_addr.s_addr;
 840         v->remote = vifc->vifc_rmt_addr.s_addr;
 841         v->flags = vifc->vifc_flags;
 842         if (!mrtsock)
 843                 v->flags |= VIFF_STATIC;
 844         v->threshold = vifc->vifc_threshold;
 845         v->bytes_in = 0;
 846         v->bytes_out = 0;
 847         v->pkt_in = 0;
 848         v->pkt_out = 0;
 849         v->link = dev->ifindex;
 850         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 851                 v->link = dev_get_iflink(dev);
 852
 853         /* And finish update writing critical data */
 854         write_lock_bh(&mrt_lock);
 855         v->dev = dev;
 856         if (v->flags & VIFF_REGISTER)
 857                 mrt->mroute_reg_vif_num = vifi;
 858         if (vifi+1 > mrt->maxvif)
 859                 mrt->maxvif = vifi+1;
 860         write_unlock_bh(&mrt_lock);
 861         return 0;
 862 }
 863
 864 /* called with rcu_read_lock() */
 865 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 866                                          __be32 origin,
 867                                          __be32 mcastgrp)
 868 {
 869         struct mfc_cache_cmp_arg arg = {
 870                         .mfc_mcastgrp = mcastgrp,
 871                         .mfc_origin = origin
 872         };
 873         struct rhlist_head *tmp, *list;
 874         struct mfc_cache *c;
 875
 876         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 877         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 878                 return c;
 879
 880         return NULL;
 881 }
 882
 883 /* Look for a (*,*,oif) entry */
 884 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 885                                                     int vifi)
 886 {
 887         struct mfc_cache_cmp_arg arg = {
 888                         .mfc_mcastgrp = htonl(INADDR_ANY),
 889                         .mfc_origin = htonl(INADDR_ANY)
 890         };
 891         struct rhlist_head *tmp, *list;
 892         struct mfc_cache *c;
 893
 894         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 895         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 896                 if (c->mfc_un.res.ttls[vifi] < 255)
 897                         return c;
 898
 899         return NULL;
 900 }
 901
 902 /* Look for a (*,G) entry */
 903 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 904                                              __be32 mcastgrp, int vifi)
 905 {
 906         struct mfc_cache_cmp_arg arg = {
 907                         .mfc_mcastgrp = mcastgrp,
 908                         .mfc_origin = htonl(INADDR_ANY)
 909         };
 910         struct rhlist_head *tmp, *list;
 911         struct mfc_cache *c, *proxy;
 912
 913         if (mcastgrp == htonl(INADDR_ANY))
 914                 goto skip;
 915
 916         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 917         rhl_for_each_entry_rcu(c, tmp, list, mnode) {
 918                 if (c->mfc_un.res.ttls[vifi] < 255)
 919                         return c;
 920
 921                 /* It's ok if the vifi is part of the static tree */
 922                 proxy = ipmr_cache_find_any_parent(mrt, c->mfc_parent);
 923                 if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 924                         return c;
 925         }
 926
 927 skip:
 928         return ipmr_cache_find_any_parent(mrt, vifi);
 929 }
 930
 931 /* Look for a (S,G,iif) entry if parent != -1 */
 932 static struct mfc_cache *ipmr_cache_find_parent(struct mr_table *mrt,
 933                                                 __be32 origin, __be32 mcastgrp,
 934                                                 int parent)
 935 {
 936         struct mfc_cache_cmp_arg arg = {
 937                         .mfc_mcastgrp = mcastgrp,
 938                         .mfc_origin = origin,
 939         };
 940         struct rhlist_head *tmp, *list;
 941         struct mfc_cache *c;
 942
 943         list = rhltable_lookup(&mrt->mfc_hash, &arg, ipmr_rht_params);
 944         rhl_for_each_entry_rcu(c, tmp, list, mnode)
 945                 if (parent == -1 || parent == c->mfc_parent)
 946                         return c;
 947
 948         return NULL;
 949 }
 950
 951 /* Allocate a multicast cache entry */
 952 static struct mfc_cache *ipmr_cache_alloc(void)
 953 {
 954         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 955
 956         if (c) {
 957                 c->mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 958                 c->mfc_un.res.minvif = MAXVIFS;
 959         }
 960         return c;
 961 }
 962
 963 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 964 {
 965         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 966
 967         if (c) {
 968                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 969                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 970         }
 971         return c;
 972 }
 973
 974 /* A cache entry has gone into a resolved state from queued */
 975 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 976                                struct mfc_cache *uc, struct mfc_cache *c)
 977 {
 978         struct sk_buff *skb;
 979         struct nlmsgerr *e;
 980
 981         /* Play the pending entries through our router */
 982         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 983                 if (ip_hdr(skb)->version == 0) {
 984                         struct nlmsghdr *nlh = skb_pull(skb,
 985                                                         sizeof(struct iphdr));
 986
 987                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 988                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 989                                                  (u8 *)nlh;
 990                         } else {
 991                                 nlh->nlmsg_type = NLMSG_ERROR;
 992                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 993                                 skb_trim(skb, nlh->nlmsg_len);
 994                                 e = nlmsg_data(nlh);
 995                                 e->error = -EMSGSIZE;
 996                                 memset(&e->msg, 0, sizeof(e->msg));
 997                         }
 998
 999                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
1000                 } else {
1001                         ip_mr_forward(net, mrt, skb->dev, skb, c, 0);
1002                 }
1003         }
1004 }
1005
1006 /* Bounce a cache query up to mrouted and netlink.
1007  *
1008  * Called under mrt_lock.
1009  */
1010 static int ipmr_cache_report(struct mr_table *mrt,
1011                              struct sk_buff *pkt, vifi_t vifi, int assert)
1012 {
1013         const int ihl = ip_hdrlen(pkt);
1014         struct sock *mroute_sk;
1015         struct igmphdr *igmp;
1016         struct igmpmsg *msg;
1017         struct sk_buff *skb;
1018         int ret;
1019
1020         if (assert == IGMPMSG_WHOLEPKT)
1021                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
1022         else
1023                 skb = alloc_skb(128, GFP_ATOMIC);
1024
1025         if (!skb)
1026                 return -ENOBUFS;
1027
1028         if (assert == IGMPMSG_WHOLEPKT) {
1029                 /* Ugly, but we have no choice with this interface.
1030                  * Duplicate old header, fix ihl, length etc.
1031                  * And all this only to mangle msg->im_msgtype and
1032                  * to set msg->im_mbz to "mbz" :-)
1033                  */
1034                 skb_push(skb, sizeof(struct iphdr));
1035                 skb_reset_network_header(skb);
1036                 skb_reset_transport_header(skb);
1037                 msg = (struct igmpmsg *)skb_network_header(skb);
1038                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
1039                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
1040                 msg->im_mbz = 0;
1041                 msg->im_vif = mrt->mroute_reg_vif_num;
1042                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
1043                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
1044                                              sizeof(struct iphdr));
1045         } else {
1046                 /* Copy the IP header */
1047                 skb_set_network_header(skb, skb->len);
1048                 skb_put(skb, ihl);
1049                 skb_copy_to_linear_data(skb, pkt->data, ihl);
1050                 /* Flag to the kernel this is a route add */
1051                 ip_hdr(skb)->protocol = 0;
1052                 msg = (struct igmpmsg *)skb_network_header(skb);
1053                 msg->im_vif = vifi;
1054                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
1055                 /* Add our header */
1056                 igmp = skb_put(skb, sizeof(struct igmphdr));
1057                 igmp->type = assert;
1058                 msg->im_msgtype = assert;
1059                 igmp->code = 0;
1060                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
1061                 skb->transport_header = skb->network_header;
1062         }
1063
1064         rcu_read_lock();
1065         mroute_sk = rcu_dereference(mrt->mroute_sk);
1066         if (!mroute_sk) {
1067                 rcu_read_unlock();
1068                 kfree_skb(skb);
1069                 return -EINVAL;
1070         }
1071
1072         igmpmsg_netlink_event(mrt, skb);
1073
1074         /* Deliver to mrouted */
1075         ret = sock_queue_rcv_skb(mroute_sk, skb);
1076         rcu_read_unlock();
1077         if (ret < 0) {
1078                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1079                 kfree_skb(skb);
1080         }
1081
1082         return ret;
1083 }
1084
1085 /* Queue a packet for resolution. It gets locked cache entry! */
1086 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1087                                  struct sk_buff *skb, struct net_device *dev)
1088 {
1089         const struct iphdr *iph = ip_hdr(skb);
1090         struct mfc_cache *c;
1091         bool found = false;
1092         int err;
1093
1094         spin_lock_bh(&mfc_unres_lock);
1095         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1096                 if (c->mfc_mcastgrp == iph->daddr &&
1097                     c->mfc_origin == iph->saddr) {
1098                         found = true;
1099                         break;
1100                 }
1101         }
1102
1103         if (!found) {
1104                 /* Create a new entry if allowable */
1105                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1106                     (c = ipmr_cache_alloc_unres()) == NULL) {
1107                         spin_unlock_bh(&mfc_unres_lock);
1108
1109                         kfree_skb(skb);
1110                         return -ENOBUFS;
1111                 }
1112
1113                 /* Fill in the new cache entry */
1114                 c->mfc_parent   = -1;
1115                 c->mfc_origin   = iph->saddr;
1116                 c->mfc_mcastgrp = iph->daddr;
1117
1118                 /* Reflect first query at mrouted. */
1119                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1120                 if (err < 0) {
1121                         /* If the report failed throw the cache entry
1122                            out - Brad Parker
1123                          */
1124                         spin_unlock_bh(&mfc_unres_lock);
1125
1126                         ipmr_cache_free(c);
1127                         kfree_skb(skb);
1128                         return err;
1129                 }
1130
1131                 atomic_inc(&mrt->cache_resolve_queue_len);
1132                 list_add(&c->list, &mrt->mfc_unres_queue);
1133                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1134
1135                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1136                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1137         }
1138
1139         /* See if we can append the packet */
1140         if (c->mfc_un.unres.unresolved.qlen > 3) {
1141                 kfree_skb(skb);
1142                 err = -ENOBUFS;
1143         } else {
1144                 if (dev) {
1145                         skb->dev = dev;
1146                         skb->skb_iif = dev->ifindex;
1147                 }
1148                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1149                 err = 0;
1150         }
1151
1152         spin_unlock_bh(&mfc_unres_lock);
1153         return err;
1154 }
1155
1156 /* MFC cache manipulation by user space mroute daemon */
1157
1158 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1159 {
1160         struct mfc_cache *c;
1161
1162         /* The entries are added/deleted only under RTNL */
1163         rcu_read_lock();
1164         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1165                                    mfc->mfcc_mcastgrp.s_addr, parent);
1166         rcu_read_unlock();
1167         if (!c)
1168                 return -ENOENT;
1169         rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1170         list_del_rcu(&c->list);
1171         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1172         ipmr_cache_free(c);
1173
1174         return 0;
1175 }
1176
1177 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1178                         struct mfcctl *mfc, int mrtsock, int parent)
1179 {
1180         struct mfc_cache *uc, *c;
1181         bool found;
1182         int ret;
1183
1184         if (mfc->mfcc_parent >= MAXVIFS)
1185                 return -ENFILE;
1186
1187         /* The entries are added/deleted only under RTNL */
1188         rcu_read_lock();
1189         c = ipmr_cache_find_parent(mrt, mfc->mfcc_origin.s_addr,
1190                                    mfc->mfcc_mcastgrp.s_addr, parent);
1191         rcu_read_unlock();
1192         if (c) {
1193                 write_lock_bh(&mrt_lock);
1194                 c->mfc_parent = mfc->mfcc_parent;
1195                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1196                 if (!mrtsock)
1197                         c->mfc_flags |= MFC_STATIC;
1198                 write_unlock_bh(&mrt_lock);
1199                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1200                 return 0;
1201         }
1202
1203         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1204             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1205                 return -EINVAL;
1206
1207         c = ipmr_cache_alloc();
1208         if (!c)
1209                 return -ENOMEM;
1210
1211         c->mfc_origin = mfc->mfcc_origin.s_addr;
1212         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1213         c->mfc_parent = mfc->mfcc_parent;
1214         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1215         if (!mrtsock)
1216                 c->mfc_flags |= MFC_STATIC;
1217
1218         ret = rhltable_insert_key(&mrt->mfc_hash, &c->cmparg, &c->mnode,
1219                                   ipmr_rht_params);
1220         if (ret) {
1221                 pr_err("ipmr: rhtable insert error %d\n", ret);
1222                 ipmr_cache_free(c);
1223                 return ret;
1224         }
1225         list_add_tail_rcu(&c->list, &mrt->mfc_cache_list);
1226         /* Check to see if we resolved a queued list. If so we
1227          * need to send on the frames and tidy up.
1228          */
1229         found = false;
1230         spin_lock_bh(&mfc_unres_lock);
1231         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1232                 if (uc->mfc_origin == c->mfc_origin &&
1233                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1234                         list_del(&uc->list);
1235                         atomic_dec(&mrt->cache_resolve_queue_len);
1236                         found = true;
1237                         break;
1238                 }
1239         }
1240         if (list_empty(&mrt->mfc_unres_queue))
1241                 del_timer(&mrt->ipmr_expire_timer);
1242         spin_unlock_bh(&mfc_unres_lock);
1243
1244         if (found) {
1245                 ipmr_cache_resolve(net, mrt, uc, c);
1246                 ipmr_cache_free(uc);
1247         }
1248         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1249         return 0;
1250 }
1251
1252 /* Close the multicast socket, and clear the vif tables etc */
1253 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1254 {
1255         struct mfc_cache *c, *tmp;
1256         LIST_HEAD(list);
1257         int i;
1258
1259         /* Shut down all active vif entries */
1260         for (i = 0; i < mrt->maxvif; i++) {
1261                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1262                         continue;
1263                 vif_delete(mrt, i, 0, &list);
1264         }
1265         unregister_netdevice_many(&list);
1266
1267         /* Wipe the cache */
1268         list_for_each_entry_safe(c, tmp, &mrt->mfc_cache_list, list) {
1269                 if (!all && (c->mfc_flags & MFC_STATIC))
1270                         continue;
1271                 rhltable_remove(&mrt->mfc_hash, &c->mnode, ipmr_rht_params);
1272                 list_del_rcu(&c->list);
1273                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
1274                 ipmr_cache_free(c);
1275         }
1276
1277         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1278                 spin_lock_bh(&mfc_unres_lock);
1279                 list_for_each_entry_safe(c, tmp, &mrt->mfc_unres_queue, list) {
1280                         list_del(&c->list);
1281                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1282                         ipmr_destroy_unres(mrt, c);
1283                 }
1284                 spin_unlock_bh(&mfc_unres_lock);
1285         }
1286 }
1287
1288 /* called from ip_ra_control(), before an RCU grace period,
1289  * we dont need to call synchronize_rcu() here
1290  */
1291 static void mrtsock_destruct(struct sock *sk)
1292 {
1293         struct net *net = sock_net(sk);
1294         struct mr_table *mrt;
1295
1296         ASSERT_RTNL();
1297         ipmr_for_each_table(mrt, net) {
1298                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1299                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1300                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1301                                                     NETCONFA_MC_FORWARDING,
1302                                                     NETCONFA_IFINDEX_ALL,
1303                                                     net->ipv4.devconf_all);
1304                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1305                         mroute_clean_tables(mrt, false);
1306                 }
1307         }
1308 }
1309
1310 /* Socket options and virtual interface manipulation. The whole
1311  * virtual interface system is a complete heap, but unfortunately
1312  * that's how BSD mrouted happens to think. Maybe one day with a proper
1313  * MOSPF/PIM router set up we can clean this up.
1314  */
1315
1316 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1317                          unsigned int optlen)
1318 {
1319         struct net *net = sock_net(sk);
1320         int val, ret = 0, parent = 0;
1321         struct mr_table *mrt;
1322         struct vifctl vif;
1323         struct mfcctl mfc;
1324         u32 uval;
1325
1326         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1327         rtnl_lock();
1328         if (sk->sk_type != SOCK_RAW ||
1329             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1330                 ret = -EOPNOTSUPP;
1331                 goto out_unlock;
1332         }
1333
1334         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1335         if (!mrt) {
1336                 ret = -ENOENT;
1337                 goto out_unlock;
1338         }
1339         if (optname != MRT_INIT) {
1340                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1341                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1342                         ret = -EACCES;
1343                         goto out_unlock;
1344                 }
1345         }
1346
1347         switch (optname) {
1348         case MRT_INIT:
1349                 if (optlen != sizeof(int)) {
1350                         ret = -EINVAL;
1351                         break;
1352                 }
1353                 if (rtnl_dereference(mrt->mroute_sk)) {
1354                         ret = -EADDRINUSE;
1355                         break;
1356                 }
1357
1358                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1359                 if (ret == 0) {
1360                         rcu_assign_pointer(mrt->mroute_sk, sk);
1361                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1362                         inet_netconf_notify_devconf(net, RTM_NEWNETCONF,
1363                                                     NETCONFA_MC_FORWARDING,
1364                                                     NETCONFA_IFINDEX_ALL,
1365                                                     net->ipv4.devconf_all);
1366                 }
1367                 break;
1368         case MRT_DONE:
1369                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1370                         ret = -EACCES;
1371                 } else {
1372                         ret = ip_ra_control(sk, 0, NULL);
1373                         goto out_unlock;
1374                 }
1375                 break;
1376         case MRT_ADD_VIF:
1377         case MRT_DEL_VIF:
1378                 if (optlen != sizeof(vif)) {
1379                         ret = -EINVAL;
1380                         break;
1381                 }
1382                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1383                         ret = -EFAULT;
1384                         break;
1385                 }
1386                 if (vif.vifc_vifi >= MAXVIFS) {
1387                         ret = -ENFILE;
1388                         break;
1389                 }
1390                 if (optname == MRT_ADD_VIF) {
1391                         ret = vif_add(net, mrt, &vif,
1392                                       sk == rtnl_dereference(mrt->mroute_sk));
1393                 } else {
1394                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1395                 }
1396                 break;
1397         /* Manipulate the forwarding caches. These live
1398          * in a sort of kernel/user symbiosis.
1399          */
1400         case MRT_ADD_MFC:
1401         case MRT_DEL_MFC:
1402                 parent = -1;
1403         case MRT_ADD_MFC_PROXY:
1404         case MRT_DEL_MFC_PROXY:
1405                 if (optlen != sizeof(mfc)) {
1406                         ret = -EINVAL;
1407                         break;
1408                 }
1409                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1410                         ret = -EFAULT;
1411                         break;
1412                 }
1413                 if (parent == 0)
1414                         parent = mfc.mfcc_parent;
1415                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1416                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1417                 else
1418                         ret = ipmr_mfc_add(net, mrt, &mfc,
1419                                            sk == rtnl_dereference(mrt->mroute_sk),
1420                                            parent);
1421                 break;
1422         /* Control PIM assert. */
1423         case MRT_ASSERT:
1424                 if (optlen != sizeof(val)) {
1425                         ret = -EINVAL;
1426                         break;
1427                 }
1428                 if (get_user(val, (int __user *)optval)) {
1429                         ret = -EFAULT;
1430                         break;
1431                 }
1432                 mrt->mroute_do_assert = val;
1433                 break;
1434         case MRT_PIM:
1435                 if (!ipmr_pimsm_enabled()) {
1436                         ret = -ENOPROTOOPT;
1437                         break;
1438                 }
1439                 if (optlen != sizeof(val)) {
1440                         ret = -EINVAL;
1441                         break;
1442                 }
1443                 if (get_user(val, (int __user *)optval)) {
1444                         ret = -EFAULT;
1445                         break;
1446                 }
1447
1448                 val = !!val;
1449                 if (val != mrt->mroute_do_pim) {
1450                         mrt->mroute_do_pim = val;
1451                         mrt->mroute_do_assert = val;
1452                 }
1453                 break;
1454         case MRT_TABLE:
1455                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1456                         ret = -ENOPROTOOPT;
1457                         break;
1458                 }
1459                 if (optlen != sizeof(uval)) {
1460                         ret = -EINVAL;
1461                         break;
1462                 }
1463                 if (get_user(uval, (u32 __user *)optval)) {
1464                         ret = -EFAULT;
1465                         break;
1466                 }
1467
1468                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1469                         ret = -EBUSY;
1470                 } else {
1471                         mrt = ipmr_new_table(net, uval);
1472                         if (IS_ERR(mrt))
1473                                 ret = PTR_ERR(mrt);
1474                         else
1475                                 raw_sk(sk)->ipmr_table = uval;
1476                 }
1477                 break;
1478         /* Spurious command, or MRT_VERSION which you cannot set. */
1479         default:
1480                 ret = -ENOPROTOOPT;
1481         }
1482 out_unlock:
1483         rtnl_unlock();
1484         return ret;
1485 }
1486
1487 /* Getsock opt support for the multicast routing system. */
1488 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1489 {
1490         int olr;
1491         int val;
1492         struct net *net = sock_net(sk);
1493         struct mr_table *mrt;
1494
1495         if (sk->sk_type != SOCK_RAW ||
1496             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1497                 return -EOPNOTSUPP;
1498
1499         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1500         if (!mrt)
1501                 return -ENOENT;
1502
1503         switch (optname) {
1504         case MRT_VERSION:
1505                 val = 0x0305;
1506                 break;
1507         case MRT_PIM:
1508                 if (!ipmr_pimsm_enabled())
1509                         return -ENOPROTOOPT;
1510                 val = mrt->mroute_do_pim;
1511                 break;
1512         case MRT_ASSERT:
1513                 val = mrt->mroute_do_assert;
1514                 break;
1515         default:
1516                 return -ENOPROTOOPT;
1517         }
1518
1519         if (get_user(olr, optlen))
1520                 return -EFAULT;
1521         olr = min_t(unsigned int, olr, sizeof(int));
1522         if (olr < 0)
1523                 return -EINVAL;
1524         if (put_user(olr, optlen))
1525                 return -EFAULT;
1526         if (copy_to_user(optval, &val, olr))
1527                 return -EFAULT;
1528         return 0;
1529 }
1530
1531 /* The IP multicast ioctl support routines. */
1532 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1533 {
1534         struct sioc_sg_req sr;
1535         struct sioc_vif_req vr;
1536         struct vif_device *vif;
1537         struct mfc_cache *c;
1538         struct net *net = sock_net(sk);
1539         struct mr_table *mrt;
1540
1541         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1542         if (!mrt)
1543                 return -ENOENT;
1544
1545         switch (cmd) {
1546         case SIOCGETVIFCNT:
1547                 if (copy_from_user(&vr, arg, sizeof(vr)))
1548                         return -EFAULT;
1549                 if (vr.vifi >= mrt->maxvif)
1550                         return -EINVAL;
1551                 read_lock(&mrt_lock);
1552                 vif = &mrt->vif_table[vr.vifi];
1553                 if (VIF_EXISTS(mrt, vr.vifi)) {
1554                         vr.icount = vif->pkt_in;
1555                         vr.ocount = vif->pkt_out;
1556                         vr.ibytes = vif->bytes_in;
1557                         vr.obytes = vif->bytes_out;
1558                         read_unlock(&mrt_lock);
1559
1560                         if (copy_to_user(arg, &vr, sizeof(vr)))
1561                                 return -EFAULT;
1562                         return 0;
1563                 }
1564                 read_unlock(&mrt_lock);
1565                 return -EADDRNOTAVAIL;
1566         case SIOCGETSGCNT:
1567                 if (copy_from_user(&sr, arg, sizeof(sr)))
1568                         return -EFAULT;
1569
1570                 rcu_read_lock();
1571                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1572                 if (c) {
1573                         sr.pktcnt = c->mfc_un.res.pkt;
1574                         sr.bytecnt = c->mfc_un.res.bytes;
1575                         sr.wrong_if = c->mfc_un.res.wrong_if;
1576                         rcu_read_unlock();
1577
1578                         if (copy_to_user(arg, &sr, sizeof(sr)))
1579                                 return -EFAULT;
1580                         return 0;
1581                 }
1582                 rcu_read_unlock();
1583                 return -EADDRNOTAVAIL;
1584         default:
1585                 return -ENOIOCTLCMD;
1586         }
1587 }
1588
1589 #ifdef CONFIG_COMPAT
1590 struct compat_sioc_sg_req {
1591         struct in_addr src;
1592         struct in_addr grp;
1593         compat_ulong_t pktcnt;
1594         compat_ulong_t bytecnt;
1595         compat_ulong_t wrong_if;
1596 };
1597
1598 struct compat_sioc_vif_req {
1599         vifi_t  vifi;           /* Which iface */
1600         compat_ulong_t icount;
1601         compat_ulong_t ocount;
1602         compat_ulong_t ibytes;
1603         compat_ulong_t obytes;
1604 };
1605
1606 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1607 {
1608         struct compat_sioc_sg_req sr;
1609         struct compat_sioc_vif_req vr;
1610         struct vif_device *vif;
1611         struct mfc_cache *c;
1612         struct net *net = sock_net(sk);
1613         struct mr_table *mrt;
1614
1615         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1616         if (!mrt)
1617                 return -ENOENT;
1618
1619         switch (cmd) {
1620         case SIOCGETVIFCNT:
1621                 if (copy_from_user(&vr, arg, sizeof(vr)))
1622                         return -EFAULT;
1623                 if (vr.vifi >= mrt->maxvif)
1624                         return -EINVAL;
1625                 vr.vifi = array_index_nospec(vr.vifi, mrt->maxvif);
1626                 read_lock(&mrt_lock);
1627                 vif = &mrt->vif_table[vr.vifi];
1628                 if (VIF_EXISTS(mrt, vr.vifi)) {
1629                         vr.icount = vif->pkt_in;
1630                         vr.ocount = vif->pkt_out;
1631                         vr.ibytes = vif->bytes_in;
1632                         vr.obytes = vif->bytes_out;
1633                         read_unlock(&mrt_lock);
1634
1635                         if (copy_to_user(arg, &vr, sizeof(vr)))
1636                                 return -EFAULT;
1637                         return 0;
1638                 }
1639                 read_unlock(&mrt_lock);
1640                 return -EADDRNOTAVAIL;
1641         case SIOCGETSGCNT:
1642                 if (copy_from_user(&sr, arg, sizeof(sr)))
1643                         return -EFAULT;
1644
1645                 rcu_read_lock();
1646                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1647                 if (c) {
1648                         sr.pktcnt = c->mfc_un.res.pkt;
1649                         sr.bytecnt = c->mfc_un.res.bytes;
1650                         sr.wrong_if = c->mfc_un.res.wrong_if;
1651                         rcu_read_unlock();
1652
1653                         if (copy_to_user(arg, &sr, sizeof(sr)))
1654                                 return -EFAULT;
1655                         return 0;
1656                 }
1657                 rcu_read_unlock();
1658                 return -EADDRNOTAVAIL;
1659         default:
1660                 return -ENOIOCTLCMD;
1661         }
1662 }
1663 #endif
1664
1665 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1666 {
1667         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1668         struct net *net = dev_net(dev);
1669         struct mr_table *mrt;
1670         struct vif_device *v;
1671         int ct;
1672
1673         if (event != NETDEV_UNREGISTER)
1674                 return NOTIFY_DONE;
1675
1676         ipmr_for_each_table(mrt, net) {
1677                 v = &mrt->vif_table[0];
1678                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1679                         if (v->dev == dev)
1680                                 vif_delete(mrt, ct, 1, NULL);
1681                 }
1682         }
1683         return NOTIFY_DONE;
1684 }
1685
1686 static struct notifier_block ip_mr_notifier = {
1687         .notifier_call = ipmr_device_event,
1688 };
1689
1690 /* Encapsulate a packet by attaching a valid IPIP header to it.
1691  * This avoids tunnel drivers and other mess and gives us the speed so
1692  * important for multicast video.
1693  */
1694 static void ip_encap(struct net *net, struct sk_buff *skb,
1695                      __be32 saddr, __be32 daddr)
1696 {
1697         struct iphdr *iph;
1698         const struct iphdr *old_iph = ip_hdr(skb);
1699
1700         skb_push(skb, sizeof(struct iphdr));
1701         skb->transport_header = skb->network_header;
1702         skb_reset_network_header(skb);
1703         iph = ip_hdr(skb);
1704
1705         iph->version    =       4;
1706         iph->tos        =       old_iph->tos;
1707         iph->ttl        =       old_iph->ttl;
1708         iph->frag_off   =       0;
1709         iph->daddr      =       daddr;
1710         iph->saddr      =       saddr;
1711         iph->protocol   =       IPPROTO_IPIP;
1712         iph->ihl        =       5;
1713         iph->tot_len    =       htons(skb->len);
1714         ip_select_ident(net, skb, NULL);
1715         ip_send_check(iph);
1716
1717         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1718         nf_reset(skb);
1719 }
1720
1721 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1722                                       struct sk_buff *skb)
1723 {
1724         struct ip_options *opt = &(IPCB(skb)->opt);
1725
1726         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1727         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1728
1729         if (unlikely(opt->optlen))
1730                 ip_forward_options(skb);
1731
1732         return dst_output(net, sk, skb);
1733 }
1734
1735 /* Processing handlers for ipmr_forward */
1736
1737 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1738                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1739 {
1740         const struct iphdr *iph = ip_hdr(skb);
1741         struct vif_device *vif = &mrt->vif_table[vifi];
1742         struct net_device *dev;
1743         struct rtable *rt;
1744         struct flowi4 fl4;
1745         int    encap = 0;
1746
1747         if (!vif->dev)
1748                 goto out_free;
1749
1750         if (vif->flags & VIFF_REGISTER) {
1751                 vif->pkt_out++;
1752                 vif->bytes_out += skb->len;
1753                 vif->dev->stats.tx_bytes += skb->len;
1754                 vif->dev->stats.tx_packets++;
1755                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1756                 goto out_free;
1757         }
1758
1759         if (vif->flags & VIFF_TUNNEL) {
1760                 rt = ip_route_output_ports(net, &fl4, NULL,
1761                                            vif->remote, vif->local,
1762                                            0, 0,
1763                                            IPPROTO_IPIP,
1764                                            RT_TOS(iph->tos), vif->link);
1765                 if (IS_ERR(rt))
1766                         goto out_free;
1767                 encap = sizeof(struct iphdr);
1768         } else {
1769                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1770                                            0, 0,
1771                                            IPPROTO_IPIP,
1772                                            RT_TOS(iph->tos), vif->link);
1773                 if (IS_ERR(rt))
1774                         goto out_free;
1775         }
1776
1777         dev = rt->dst.dev;
1778
1779         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1780                 /* Do not fragment multicasts. Alas, IPv4 does not
1781                  * allow to send ICMP, so that packets will disappear
1782                  * to blackhole.
1783                  */
1784                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1785                 ip_rt_put(rt);
1786                 goto out_free;
1787         }
1788
1789         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1790
1791         if (skb_cow(skb, encap)) {
1792                 ip_rt_put(rt);
1793                 goto out_free;
1794         }
1795
1796         vif->pkt_out++;
1797         vif->bytes_out += skb->len;
1798
1799         skb_dst_drop(skb);
1800         skb_dst_set(skb, &rt->dst);
1801         ip_decrease_ttl(ip_hdr(skb));
1802
1803         /* FIXME: forward and output firewalls used to be called here.
1804          * What do we do with netfilter? -- RR
1805          */
1806         if (vif->flags & VIFF_TUNNEL) {
1807                 ip_encap(net, skb, vif->local, vif->remote);
1808                 /* FIXME: extra output firewall step used to be here. --RR */
1809                 vif->dev->stats.tx_packets++;
1810                 vif->dev->stats.tx_bytes += skb->len;
1811         }
1812
1813         IPCB(skb)->flags |= IPSKB_FORWARDED;
1814
1815         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1816          * not only before forwarding, but after forwarding on all output
1817          * interfaces. It is clear, if mrouter runs a multicasting
1818          * program, it should receive packets not depending to what interface
1819          * program is joined.
1820          * If we will not make it, the program will have to join on all
1821          * interfaces. On the other hand, multihoming host (or router, but
1822          * not mrouter) cannot join to more than one interface - it will
1823          * result in receiving multiple packets.
1824          */
1825         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1826                 net, NULL, skb, skb->dev, dev,
1827                 ipmr_forward_finish);
1828         return;
1829
1830 out_free:
1831         kfree_skb(skb);
1832 }
1833
1834 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1835 {
1836         int ct;
1837
1838         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1839                 if (mrt->vif_table[ct].dev == dev)
1840                         break;
1841         }
1842         return ct;
1843 }
1844
1845 /* "local" means that we should preserve one skb (for local delivery) */
1846 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1847                           struct net_device *dev, struct sk_buff *skb,
1848                           struct mfc_cache *cache, int local)
1849 {
1850         int true_vifi = ipmr_find_vif(mrt, dev);
1851         int psend = -1;
1852         int vif, ct;
1853
1854         vif = cache->mfc_parent;
1855         cache->mfc_un.res.pkt++;
1856         cache->mfc_un.res.bytes += skb->len;
1857         cache->mfc_un.res.lastuse = jiffies;
1858
1859         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1860                 struct mfc_cache *cache_proxy;
1861
1862                 /* For an (*,G) entry, we only check that the incomming
1863                  * interface is part of the static tree.
1864                  */
1865                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1866                 if (cache_proxy &&
1867                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1868                         goto forward;
1869         }
1870
1871         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1872         if (mrt->vif_table[vif].dev != dev) {
1873                 if (rt_is_output_route(skb_rtable(skb))) {
1874                         /* It is our own packet, looped back.
1875                          * Very complicated situation...
1876                          *
1877                          * The best workaround until routing daemons will be
1878                          * fixed is not to redistribute packet, if it was
1879                          * send through wrong interface. It means, that
1880                          * multicast applications WILL NOT work for
1881                          * (S,G), which have default multicast route pointing
1882                          * to wrong oif. In any case, it is not a good
1883                          * idea to use multicasting applications on router.
1884                          */
1885                         goto dont_forward;
1886                 }
1887
1888                 cache->mfc_un.res.wrong_if++;
1889
1890                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1891                     /* pimsm uses asserts, when switching from RPT to SPT,
1892                      * so that we cannot check that packet arrived on an oif.
1893                      * It is bad, but otherwise we would need to move pretty
1894                      * large chunk of pimd to kernel. Ough... --ANK
1895                      */
1896                     (mrt->mroute_do_pim ||
1897                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1898                     time_after(jiffies,
1899                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1900                         cache->mfc_un.res.last_assert = jiffies;
1901                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1902                 }
1903                 goto dont_forward;
1904         }
1905
1906 forward:
1907         mrt->vif_table[vif].pkt_in++;
1908         mrt->vif_table[vif].bytes_in += skb->len;
1909
1910         /* Forward the frame */
1911         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1912             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1913                 if (true_vifi >= 0 &&
1914                     true_vifi != cache->mfc_parent &&
1915                     ip_hdr(skb)->ttl >
1916                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1917                         /* It's an (*,*) entry and the packet is not coming from
1918                          * the upstream: forward the packet to the upstream
1919                          * only.
1920                          */
1921                         psend = cache->mfc_parent;
1922                         goto last_forward;
1923                 }
1924                 goto dont_forward;
1925         }
1926         for (ct = cache->mfc_un.res.maxvif - 1;
1927              ct >= cache->mfc_un.res.minvif; ct--) {
1928                 /* For (*,G) entry, don't forward to the incoming interface */
1929                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1930                      ct != true_vifi) &&
1931                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1932                         if (psend != -1) {
1933                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1934
1935                                 if (skb2)
1936                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1937                                                         psend);
1938                         }
1939                         psend = ct;
1940                 }
1941         }
1942 last_forward:
1943         if (psend != -1) {
1944                 if (local) {
1945                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1946
1947                         if (skb2)
1948                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1949                 } else {
1950                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1951                         return;
1952                 }
1953         }
1954
1955 dont_forward:
1956         if (!local)
1957                 kfree_skb(skb);
1958 }
1959
1960 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1961 {
1962         struct rtable *rt = skb_rtable(skb);
1963         struct iphdr *iph = ip_hdr(skb);
1964         struct flowi4 fl4 = {
1965                 .daddr = iph->daddr,
1966                 .saddr = iph->saddr,
1967                 .flowi4_tos = RT_TOS(iph->tos),
1968                 .flowi4_oif = (rt_is_output_route(rt) ?
1969                                skb->dev->ifindex : 0),
1970                 .flowi4_iif = (rt_is_output_route(rt) ?
1971                                LOOPBACK_IFINDEX :
1972                                skb->dev->ifindex),
1973                 .flowi4_mark = skb->mark,
1974         };
1975         struct mr_table *mrt;
1976         int err;
1977
1978         err = ipmr_fib_lookup(net, &fl4, &mrt);
1979         if (err)
1980                 return ERR_PTR(err);
1981         return mrt;
1982 }
1983
1984 /* Multicast packets for forwarding arrive here
1985  * Called with rcu_read_lock();
1986  */
1987 int ip_mr_input(struct sk_buff *skb)
1988 {
1989         struct mfc_cache *cache;
1990         struct net *net = dev_net(skb->dev);
1991         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1992         struct mr_table *mrt;
1993         struct net_device *dev;
1994
1995         /* skb->dev passed in is the loX master dev for vrfs.
1996          * As there are no vifs associated with loopback devices,
1997          * get the proper interface that does have a vif associated with it.
1998          */
1999         dev = skb->dev;
2000         if (netif_is_l3_master(skb->dev)) {
2001                 dev = dev_get_by_index_rcu(net, IPCB(skb)->iif);
2002                 if (!dev) {
2003                         kfree_skb(skb);
2004                         return -ENODEV;
2005                 }
2006         }
2007
2008         /* Packet is looped back after forward, it should not be
2009          * forwarded second time, but still can be delivered locally.
2010          */
2011         if (IPCB(skb)->flags & IPSKB_FORWARDED)
2012                 goto dont_forward;
2013
2014         mrt = ipmr_rt_fib_lookup(net, skb);
2015         if (IS_ERR(mrt)) {
2016                 kfree_skb(skb);
2017                 return PTR_ERR(mrt);
2018         }
2019         if (!local) {
2020                 if (IPCB(skb)->opt.router_alert) {
2021                         if (ip_call_ra_chain(skb))
2022                                 return 0;
2023                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
2024                         /* IGMPv1 (and broken IGMPv2 implementations sort of
2025                          * Cisco IOS <= 11.2(8)) do not put router alert
2026                          * option to IGMP packets destined to routable
2027                          * groups. It is very bad, because it means
2028                          * that we can forward NO IGMP messages.
2029                          */
2030                         struct sock *mroute_sk;
2031
2032                         mroute_sk = rcu_dereference(mrt->mroute_sk);
2033                         if (mroute_sk) {
2034                                 nf_reset(skb);
2035                                 raw_rcv(mroute_sk, skb);
2036                                 return 0;
2037                         }
2038                     }
2039         }
2040
2041         /* already under rcu_read_lock() */
2042         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
2043         if (!cache) {
2044                 int vif = ipmr_find_vif(mrt, dev);
2045
2046                 if (vif >= 0)
2047                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
2048                                                     vif);
2049         }
2050
2051         /* No usable cache entry */
2052         if (!cache) {
2053                 int vif;
2054
2055                 if (local) {
2056                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
2057                         ip_local_deliver(skb);
2058                         if (!skb2)
2059                                 return -ENOBUFS;
2060                         skb = skb2;
2061                 }
2062
2063                 read_lock(&mrt_lock);
2064                 vif = ipmr_find_vif(mrt, dev);
2065                 if (vif >= 0) {
2066                         int err2 = ipmr_cache_unresolved(mrt, vif, skb, dev);
2067                         read_unlock(&mrt_lock);
2068
2069                         return err2;
2070                 }
2071                 read_unlock(&mrt_lock);
2072                 kfree_skb(skb);
2073                 return -ENODEV;
2074         }
2075
2076         read_lock(&mrt_lock);
2077         ip_mr_forward(net, mrt, dev, skb, cache, local);
2078         read_unlock(&mrt_lock);
2079
2080         if (local)
2081                 return ip_local_deliver(skb);
2082
2083         return 0;
2084
2085 dont_forward:
2086         if (local)
2087                 return ip_local_deliver(skb);
2088         kfree_skb(skb);
2089         return 0;
2090 }
2091
2092 #ifdef CONFIG_IP_PIMSM_V1
2093 /* Handle IGMP messages of PIMv1 */
2094 int pim_rcv_v1(struct sk_buff *skb)
2095 {
2096         struct igmphdr *pim;
2097         struct net *net = dev_net(skb->dev);
2098         struct mr_table *mrt;
2099
2100         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2101                 goto drop;
2102
2103         pim = igmp_hdr(skb);
2104
2105         mrt = ipmr_rt_fib_lookup(net, skb);
2106         if (IS_ERR(mrt))
2107                 goto drop;
2108         if (!mrt->mroute_do_pim ||
2109             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2110                 goto drop;
2111
2112         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2113 drop:
2114                 kfree_skb(skb);
2115         }
2116         return 0;
2117 }
2118 #endif
2119
2120 #ifdef CONFIG_IP_PIMSM_V2
2121 static int pim_rcv(struct sk_buff *skb)
2122 {
2123         struct pimreghdr *pim;
2124         struct net *net = dev_net(skb->dev);
2125         struct mr_table *mrt;
2126
2127         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2128                 goto drop;
2129
2130         pim = (struct pimreghdr *)skb_transport_header(skb);
2131         if (pim->type != ((PIM_VERSION << 4) | (PIM_TYPE_REGISTER)) ||
2132             (pim->flags & PIM_NULL_REGISTER) ||
2133             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2134              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2135                 goto drop;
2136
2137         mrt = ipmr_rt_fib_lookup(net, skb);
2138         if (IS_ERR(mrt))
2139                 goto drop;
2140         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2141 drop:
2142                 kfree_skb(skb);
2143         }
2144         return 0;
2145 }
2146 #endif
2147
2148 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2149                               struct mfc_cache *c, struct rtmsg *rtm)
2150 {
2151         struct rta_mfc_stats mfcs;
2152         struct nlattr *mp_attr;
2153         struct rtnexthop *nhp;
2154         unsigned long lastuse;
2155         int ct;
2156
2157         /* If cache is unresolved, don't try to parse IIF and OIF */
2158         if (c->mfc_parent >= MAXVIFS) {
2159                 rtm->rtm_flags |= RTNH_F_UNRESOLVED;
2160                 return -ENOENT;
2161         }
2162
2163         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2164             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2165                 return -EMSGSIZE;
2166
2167         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2168                 return -EMSGSIZE;
2169
2170         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2171                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2172                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2173                                 nla_nest_cancel(skb, mp_attr);
2174                                 return -EMSGSIZE;
2175                         }
2176
2177                         nhp->rtnh_flags = 0;
2178                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2179                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2180                         nhp->rtnh_len = sizeof(*nhp);
2181                 }
2182         }
2183
2184         nla_nest_end(skb, mp_attr);
2185
2186         lastuse = READ_ONCE(c->mfc_un.res.lastuse);
2187         lastuse = time_after_eq(jiffies, lastuse) ? jiffies - lastuse : 0;
2188
2189         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2190         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2191         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2192         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) ||
2193             nla_put_u64_64bit(skb, RTA_EXPIRES, jiffies_to_clock_t(lastuse),
2194                               RTA_PAD))
2195                 return -EMSGSIZE;
2196
2197         rtm->rtm_type = RTN_MULTICAST;
2198         return 1;
2199 }
2200
2201 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2202                    __be32 saddr, __be32 daddr,
2203                    struct rtmsg *rtm, u32 portid)
2204 {
2205         struct mfc_cache *cache;
2206         struct mr_table *mrt;
2207         int err;
2208
2209         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2210         if (!mrt)
2211                 return -ENOENT;
2212
2213         rcu_read_lock();
2214         cache = ipmr_cache_find(mrt, saddr, daddr);
2215         if (!cache && skb->dev) {
2216                 int vif = ipmr_find_vif(mrt, skb->dev);
2217
2218                 if (vif >= 0)
2219                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2220         }
2221         if (!cache) {
2222                 struct sk_buff *skb2;
2223                 struct iphdr *iph;
2224                 struct net_device *dev;
2225                 int vif = -1;
2226
2227                 dev = skb->dev;
2228                 read_lock(&mrt_lock);
2229                 if (dev)
2230                         vif = ipmr_find_vif(mrt, dev);
2231                 if (vif < 0) {
2232                         read_unlock(&mrt_lock);
2233                         rcu_read_unlock();
2234                         return -ENODEV;
2235                 }
2236                 skb2 = skb_clone(skb, GFP_ATOMIC);
2237                 if (!skb2) {
2238                         read_unlock(&mrt_lock);
2239                         rcu_read_unlock();
2240                         return -ENOMEM;
2241                 }
2242
2243                 NETLINK_CB(skb2).portid = portid;
2244                 skb_push(skb2, sizeof(struct iphdr));
2245                 skb_reset_network_header(skb2);
2246                 iph = ip_hdr(skb2);
2247                 iph->ihl = sizeof(struct iphdr) >> 2;
2248                 iph->saddr = saddr;
2249                 iph->daddr = daddr;
2250                 iph->version = 0;
2251                 err = ipmr_cache_unresolved(mrt, vif, skb2, dev);
2252                 read_unlock(&mrt_lock);
2253                 rcu_read_unlock();
2254                 return err;
2255         }
2256
2257         read_lock(&mrt_lock);
2258         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2259         read_unlock(&mrt_lock);
2260         rcu_read_unlock();
2261         return err;
2262 }
2263
2264 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2265                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2266                             int flags)
2267 {
2268         struct nlmsghdr *nlh;
2269         struct rtmsg *rtm;
2270         int err;
2271
2272         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2273         if (!nlh)
2274                 return -EMSGSIZE;
2275
2276         rtm = nlmsg_data(nlh);
2277         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2278         rtm->rtm_dst_len  = 32;
2279         rtm->rtm_src_len  = 32;
2280         rtm->rtm_tos      = 0;
2281         rtm->rtm_table    = mrt->id;
2282         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2283                 goto nla_put_failure;
2284         rtm->rtm_type     = RTN_MULTICAST;
2285         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2286         if (c->mfc_flags & MFC_STATIC)
2287                 rtm->rtm_protocol = RTPROT_STATIC;
2288         else
2289                 rtm->rtm_protocol = RTPROT_MROUTED;
2290         rtm->rtm_flags    = 0;
2291
2292         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2293             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2294                 goto nla_put_failure;
2295         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2296         /* do not break the dump if cache is unresolved */
2297         if (err < 0 && err != -ENOENT)
2298                 goto nla_put_failure;
2299
2300         nlmsg_end(skb, nlh);
2301         return 0;
2302
2303 nla_put_failure:
2304         nlmsg_cancel(skb, nlh);
2305         return -EMSGSIZE;
2306 }
2307
2308 static size_t mroute_msgsize(bool unresolved, int maxvif)
2309 {
2310         size_t len =
2311                 NLMSG_ALIGN(sizeof(struct rtmsg))
2312                 + nla_total_size(4)     /* RTA_TABLE */
2313                 + nla_total_size(4)     /* RTA_SRC */
2314                 + nla_total_size(4)     /* RTA_DST */
2315                 ;
2316
2317         if (!unresolved)
2318                 len = len
2319                       + nla_total_size(4)       /* RTA_IIF */
2320                       + nla_total_size(0)       /* RTA_MULTIPATH */
2321                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2322                                                 /* RTA_MFC_STATS */
2323                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2324                 ;
2325
2326         return len;
2327 }
2328
2329 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2330                                  int cmd)
2331 {
2332         struct net *net = read_pnet(&mrt->net);
2333         struct sk_buff *skb;
2334         int err = -ENOBUFS;
2335
2336         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2337                         GFP_ATOMIC);
2338         if (!skb)
2339                 goto errout;
2340
2341         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2342         if (err < 0)
2343                 goto errout;
2344
2345         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2346         return;
2347
2348 errout:
2349         kfree_skb(skb);
2350         if (err < 0)
2351                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2352 }
2353
2354 static size_t igmpmsg_netlink_msgsize(size_t payloadlen)
2355 {
2356         size_t len =
2357                 NLMSG_ALIGN(sizeof(struct rtgenmsg))
2358                 + nla_total_size(1)     /* IPMRA_CREPORT_MSGTYPE */
2359                 + nla_total_size(4)     /* IPMRA_CREPORT_VIF_ID */
2360                 + nla_total_size(4)     /* IPMRA_CREPORT_SRC_ADDR */
2361                 + nla_total_size(4)     /* IPMRA_CREPORT_DST_ADDR */
2362                                         /* IPMRA_CREPORT_PKT */
2363                 + nla_total_size(payloadlen)
2364                 ;
2365
2366         return len;
2367 }
2368
2369 static void igmpmsg_netlink_event(struct mr_table *mrt, struct sk_buff *pkt)
2370 {
2371         struct net *net = read_pnet(&mrt->net);
2372         struct nlmsghdr *nlh;
2373         struct rtgenmsg *rtgenm;
2374         struct igmpmsg *msg;
2375         struct sk_buff *skb;
2376         struct nlattr *nla;
2377         int payloadlen;
2378
2379         payloadlen = pkt->len - sizeof(struct igmpmsg);
2380         msg = (struct igmpmsg *)skb_network_header(pkt);
2381
2382         skb = nlmsg_new(igmpmsg_netlink_msgsize(payloadlen), GFP_ATOMIC);
2383         if (!skb)
2384                 goto errout;
2385
2386         nlh = nlmsg_put(skb, 0, 0, RTM_NEWCACHEREPORT,
2387                         sizeof(struct rtgenmsg), 0);
2388         if (!nlh)
2389                 goto errout;
2390         rtgenm = nlmsg_data(nlh);
2391         rtgenm->rtgen_family = RTNL_FAMILY_IPMR;
2392         if (nla_put_u8(skb, IPMRA_CREPORT_MSGTYPE, msg->im_msgtype) ||
2393             nla_put_u32(skb, IPMRA_CREPORT_VIF_ID, msg->im_vif) ||
2394             nla_put_in_addr(skb, IPMRA_CREPORT_SRC_ADDR,
2395                             msg->im_src.s_addr) ||
2396             nla_put_in_addr(skb, IPMRA_CREPORT_DST_ADDR,
2397                             msg->im_dst.s_addr))
2398                 goto nla_put_failure;
2399
2400         nla = nla_reserve(skb, IPMRA_CREPORT_PKT, payloadlen);
2401         if (!nla || skb_copy_bits(pkt, sizeof(struct igmpmsg),
2402                                   nla_data(nla), payloadlen))
2403                 goto nla_put_failure;
2404
2405         nlmsg_end(skb, nlh);
2406
2407         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE_R, NULL, GFP_ATOMIC);
2408         return;
2409
2410 nla_put_failure:
2411         nlmsg_cancel(skb, nlh);
2412 errout:
2413         kfree_skb(skb);
2414         rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE_R, -ENOBUFS);
2415 }
2416
2417 static int ipmr_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2418                              struct netlink_ext_ack *extack)
2419 {
2420         struct net *net = sock_net(in_skb->sk);
2421         struct nlattr *tb[RTA_MAX + 1];
2422         struct sk_buff *skb = NULL;
2423         struct mfc_cache *cache;
2424         struct mr_table *mrt;
2425         struct rtmsg *rtm;
2426         __be32 src, grp;
2427         u32 tableid;
2428         int err;
2429
2430         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX,
2431                           rtm_ipv4_policy, extack);
2432         if (err < 0)
2433                 goto errout;
2434
2435         rtm = nlmsg_data(nlh);
2436
2437         src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2438         grp = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2439         tableid = tb[RTA_TABLE] ? nla_get_u32(tb[RTA_TABLE]) : 0;
2440
2441         mrt = ipmr_get_table(net, tableid ? tableid : RT_TABLE_DEFAULT);
2442         if (!mrt) {
2443                 err = -ENOENT;
2444                 goto errout_free;
2445         }
2446
2447         /* entries are added/deleted only under RTNL */
2448         rcu_read_lock();
2449         cache = ipmr_cache_find(mrt, src, grp);
2450         rcu_read_unlock();
2451         if (!cache) {
2452                 err = -ENOENT;
2453                 goto errout_free;
2454         }
2455
2456         skb = nlmsg_new(mroute_msgsize(false, mrt->maxvif), GFP_KERNEL);
2457         if (!skb) {
2458                 err = -ENOBUFS;
2459                 goto errout_free;
2460         }
2461
2462         err = ipmr_fill_mroute(mrt, skb, NETLINK_CB(in_skb).portid,
2463                                nlh->nlmsg_seq, cache,
2464                                RTM_NEWROUTE, 0);
2465         if (err < 0)
2466                 goto errout_free;
2467
2468         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2469
2470 errout:
2471         return err;
2472
2473 errout_free:
2474         kfree_skb(skb);
2475         goto errout;
2476 }
2477
2478 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2479 {
2480         struct net *net = sock_net(skb->sk);
2481         struct mr_table *mrt;
2482         struct mfc_cache *mfc;
2483         unsigned int t = 0, s_t;
2484         unsigned int e = 0, s_e;
2485
2486         s_t = cb->args[0];
2487         s_e = cb->args[1];
2488
2489         rcu_read_lock();
2490         ipmr_for_each_table(mrt, net) {
2491                 if (t < s_t)
2492                         goto next_table;
2493                 list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list) {
2494                         if (e < s_e)
2495                                 goto next_entry;
2496                         if (ipmr_fill_mroute(mrt, skb,
2497                                              NETLINK_CB(cb->skb).portid,
2498                                              cb->nlh->nlmsg_seq,
2499                                              mfc, RTM_NEWROUTE,
2500                                              NLM_F_MULTI) < 0)
2501                                 goto done;
2502 next_entry:
2503                         e++;
2504                 }
2505
2506                 spin_lock_bh(&mfc_unres_lock);
2507                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2508                         if (e < s_e)
2509                                 goto next_entry2;
2510                         if (ipmr_fill_mroute(mrt, skb,
2511                                              NETLINK_CB(cb->skb).portid,
2512                                              cb->nlh->nlmsg_seq,
2513                                              mfc, RTM_NEWROUTE,
2514                                              NLM_F_MULTI) < 0) {
2515                                 spin_unlock_bh(&mfc_unres_lock);
2516                                 goto done;
2517                         }
2518 next_entry2:
2519                         e++;
2520                 }
2521                 spin_unlock_bh(&mfc_unres_lock);
2522                 e = 0;
2523                 s_e = 0;
2524 next_table:
2525                 t++;
2526         }
2527 done:
2528         rcu_read_unlock();
2529
2530         cb->args[1] = e;
2531         cb->args[0] = t;
2532
2533         return skb->len;
2534 }
2535
2536 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2537         [RTA_SRC]       = { .type = NLA_U32 },
2538         [RTA_DST]       = { .type = NLA_U32 },
2539         [RTA_IIF]       = { .type = NLA_U32 },
2540         [RTA_TABLE]     = { .type = NLA_U32 },
2541         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2542 };
2543
2544 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2545 {
2546         switch (rtm_protocol) {
2547         case RTPROT_STATIC:
2548         case RTPROT_MROUTED:
2549                 return true;
2550         }
2551         return false;
2552 }
2553
2554 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2555 {
2556         struct rtnexthop *rtnh = nla_data(nla);
2557         int remaining = nla_len(nla), vifi = 0;
2558
2559         while (rtnh_ok(rtnh, remaining)) {
2560                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2561                 if (++vifi == MAXVIFS)
2562                         break;
2563                 rtnh = rtnh_next(rtnh, &remaining);
2564         }
2565
2566         return remaining > 0 ? -EINVAL : vifi;
2567 }
2568
2569 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2570 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2571                             struct mfcctl *mfcc, int *mrtsock,
2572                             struct mr_table **mrtret,
2573                             struct netlink_ext_ack *extack)
2574 {
2575         struct net_device *dev = NULL;
2576         u32 tblid = RT_TABLE_DEFAULT;
2577         struct mr_table *mrt;
2578         struct nlattr *attr;
2579         struct rtmsg *rtm;
2580         int ret, rem;
2581
2582         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy,
2583                              extack);
2584         if (ret < 0)
2585                 goto out;
2586         rtm = nlmsg_data(nlh);
2587
2588         ret = -EINVAL;
2589         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2590             rtm->rtm_type != RTN_MULTICAST ||
2591             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2592             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2593                 goto out;
2594
2595         memset(mfcc, 0, sizeof(*mfcc));
2596         mfcc->mfcc_parent = -1;
2597         ret = 0;
2598         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2599                 switch (nla_type(attr)) {
2600                 case RTA_SRC:
2601                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2602                         break;
2603                 case RTA_DST:
2604                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2605                         break;
2606                 case RTA_IIF:
2607                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2608                         if (!dev) {
2609                                 ret = -ENODEV;
2610                                 goto out;
2611                         }
2612                         break;
2613                 case RTA_MULTIPATH:
2614                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2615                                 ret = -EINVAL;
2616                                 goto out;
2617                         }
2618                         break;
2619                 case RTA_PREFSRC:
2620                         ret = 1;
2621                         break;
2622                 case RTA_TABLE:
2623                         tblid = nla_get_u32(attr);
2624                         break;
2625                 }
2626         }
2627         mrt = ipmr_get_table(net, tblid);
2628         if (!mrt) {
2629                 ret = -ENOENT;
2630                 goto out;
2631         }
2632         *mrtret = mrt;
2633         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2634         if (dev)
2635                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2636
2637 out:
2638         return ret;
2639 }
2640
2641 /* takes care of both newroute and delroute */
2642 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh,
2643                           struct netlink_ext_ack *extack)
2644 {
2645         struct net *net = sock_net(skb->sk);
2646         int ret, mrtsock, parent;
2647         struct mr_table *tbl;
2648         struct mfcctl mfcc;
2649
2650         mrtsock = 0;
2651         tbl = NULL;
2652         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl, extack);
2653         if (ret < 0)
2654                 return ret;
2655
2656         parent = ret ? mfcc.mfcc_parent : -1;
2657         if (nlh->nlmsg_type == RTM_NEWROUTE)
2658                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2659         else
2660                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2661 }
2662
2663 static bool ipmr_fill_table(struct mr_table *mrt, struct sk_buff *skb)
2664 {
2665         u32 queue_len = atomic_read(&mrt->cache_resolve_queue_len);
2666
2667         if (nla_put_u32(skb, IPMRA_TABLE_ID, mrt->id) ||
2668             nla_put_u32(skb, IPMRA_TABLE_CACHE_RES_QUEUE_LEN, queue_len) ||
2669             nla_put_s32(skb, IPMRA_TABLE_MROUTE_REG_VIF_NUM,
2670                         mrt->mroute_reg_vif_num) ||
2671             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_ASSERT,
2672                        mrt->mroute_do_assert) ||
2673             nla_put_u8(skb, IPMRA_TABLE_MROUTE_DO_PIM, mrt->mroute_do_pim))
2674                 return false;
2675
2676         return true;
2677 }
2678
2679 static bool ipmr_fill_vif(struct mr_table *mrt, u32 vifid, struct sk_buff *skb)
2680 {
2681         struct nlattr *vif_nest;
2682         struct vif_device *vif;
2683
2684         /* if the VIF doesn't exist just continue */
2685         if (!VIF_EXISTS(mrt, vifid))
2686                 return true;
2687
2688         vif = &mrt->vif_table[vifid];
2689         vif_nest = nla_nest_start(skb, IPMRA_VIF);
2690         if (!vif_nest)
2691                 return false;
2692         if (nla_put_u32(skb, IPMRA_VIFA_IFINDEX, vif->dev->ifindex) ||
2693             nla_put_u32(skb, IPMRA_VIFA_VIF_ID, vifid) ||
2694             nla_put_u16(skb, IPMRA_VIFA_FLAGS, vif->flags) ||
2695             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_IN, vif->bytes_in,
2696                               IPMRA_VIFA_PAD) ||
2697             nla_put_u64_64bit(skb, IPMRA_VIFA_BYTES_OUT, vif->bytes_out,
2698                               IPMRA_VIFA_PAD) ||
2699             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_IN, vif->pkt_in,
2700                               IPMRA_VIFA_PAD) ||
2701             nla_put_u64_64bit(skb, IPMRA_VIFA_PACKETS_OUT, vif->pkt_out,
2702                               IPMRA_VIFA_PAD) ||
2703             nla_put_be32(skb, IPMRA_VIFA_LOCAL_ADDR, vif->local) ||
2704             nla_put_be32(skb, IPMRA_VIFA_REMOTE_ADDR, vif->remote)) {
2705                 nla_nest_cancel(skb, vif_nest);
2706                 return false;
2707         }
2708         nla_nest_end(skb, vif_nest);
2709
2710         return true;
2711 }
2712
2713 static int ipmr_rtm_dumplink(struct sk_buff *skb, struct netlink_callback *cb)
2714 {
2715         struct net *net = sock_net(skb->sk);
2716         struct nlmsghdr *nlh = NULL;
2717         unsigned int t = 0, s_t;
2718         unsigned int e = 0, s_e;
2719         struct mr_table *mrt;
2720
2721         s_t = cb->args[0];
2722         s_e = cb->args[1];
2723
2724         ipmr_for_each_table(mrt, net) {
2725                 struct nlattr *vifs, *af;
2726                 struct ifinfomsg *hdr;
2727                 u32 i;
2728
2729                 if (t < s_t)
2730                         goto skip_table;
2731                 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
2732                                 cb->nlh->nlmsg_seq, RTM_NEWLINK,
2733                                 sizeof(*hdr), NLM_F_MULTI);
2734                 if (!nlh)
2735                         break;
2736
2737                 hdr = nlmsg_data(nlh);
2738                 memset(hdr, 0, sizeof(*hdr));
2739                 hdr->ifi_family = RTNL_FAMILY_IPMR;
2740
2741                 af = nla_nest_start(skb, IFLA_AF_SPEC);
2742                 if (!af) {
2743                         nlmsg_cancel(skb, nlh);
2744                         goto out;
2745                 }
2746
2747                 if (!ipmr_fill_table(mrt, skb)) {
2748                         nlmsg_cancel(skb, nlh);
2749                         goto out;
2750                 }
2751
2752                 vifs = nla_nest_start(skb, IPMRA_TABLE_VIFS);
2753                 if (!vifs) {
2754                         nla_nest_end(skb, af);
2755                         nlmsg_end(skb, nlh);
2756                         goto out;
2757                 }
2758                 for (i = 0; i < mrt->maxvif; i++) {
2759                         if (e < s_e)
2760                                 goto skip_entry;
2761                         if (!ipmr_fill_vif(mrt, i, skb)) {
2762                                 nla_nest_end(skb, vifs);
2763                                 nla_nest_end(skb, af);
2764                                 nlmsg_end(skb, nlh);
2765                                 goto out;
2766                         }
2767 skip_entry:
2768                         e++;
2769                 }
2770                 s_e = 0;
2771                 e = 0;
2772                 nla_nest_end(skb, vifs);
2773                 nla_nest_end(skb, af);
2774                 nlmsg_end(skb, nlh);
2775 skip_table:
2776                 t++;
2777         }
2778
2779 out:
2780         cb->args[1] = e;
2781         cb->args[0] = t;
2782
2783         return skb->len;
2784 }
2785
2786 #ifdef CONFIG_PROC_FS
2787 /* The /proc interfaces to multicast routing :
2788  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2789  */
2790 struct ipmr_vif_iter {
2791         struct seq_net_private p;
2792         struct mr_table *mrt;
2793         int ct;
2794 };
2795
2796 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2797                                            struct ipmr_vif_iter *iter,
2798                                            loff_t pos)
2799 {
2800         struct mr_table *mrt = iter->mrt;
2801
2802         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2803                 if (!VIF_EXISTS(mrt, iter->ct))
2804                         continue;
2805                 if (pos-- == 0)
2806                         return &mrt->vif_table[iter->ct];
2807         }
2808         return NULL;
2809 }
2810
2811 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2812         __acquires(mrt_lock)
2813 {
2814         struct ipmr_vif_iter *iter = seq->private;
2815         struct net *net = seq_file_net(seq);
2816         struct mr_table *mrt;
2817
2818         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2819         if (!mrt)
2820                 return ERR_PTR(-ENOENT);
2821
2822         iter->mrt = mrt;
2823
2824         read_lock(&mrt_lock);
2825         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2826                 : SEQ_START_TOKEN;
2827 }
2828
2829 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2830 {
2831         struct ipmr_vif_iter *iter = seq->private;
2832         struct net *net = seq_file_net(seq);
2833         struct mr_table *mrt = iter->mrt;
2834
2835         ++*pos;
2836         if (v == SEQ_START_TOKEN)
2837                 return ipmr_vif_seq_idx(net, iter, 0);
2838
2839         while (++iter->ct < mrt->maxvif) {
2840                 if (!VIF_EXISTS(mrt, iter->ct))
2841                         continue;
2842                 return &mrt->vif_table[iter->ct];
2843         }
2844         return NULL;
2845 }
2846
2847 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2848         __releases(mrt_lock)
2849 {
2850         read_unlock(&mrt_lock);
2851 }
2852
2853 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2854 {
2855         struct ipmr_vif_iter *iter = seq->private;
2856         struct mr_table *mrt = iter->mrt;
2857
2858         if (v == SEQ_START_TOKEN) {
2859                 seq_puts(seq,
2860                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2861         } else {
2862                 const struct vif_device *vif = v;
2863                 const char *name =  vif->dev ? vif->dev->name : "none";
2864
2865                 seq_printf(seq,
2866                            "%2zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2867                            vif - mrt->vif_table,
2868                            name, vif->bytes_in, vif->pkt_in,
2869                            vif->bytes_out, vif->pkt_out,
2870                            vif->flags, vif->local, vif->remote);
2871         }
2872         return 0;
2873 }
2874
2875 static const struct seq_operations ipmr_vif_seq_ops = {
2876         .start = ipmr_vif_seq_start,
2877         .next  = ipmr_vif_seq_next,
2878         .stop  = ipmr_vif_seq_stop,
2879         .show  = ipmr_vif_seq_show,
2880 };
2881
2882 static int ipmr_vif_open(struct inode *inode, struct file *file)
2883 {
2884         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2885                             sizeof(struct ipmr_vif_iter));
2886 }
2887
2888 static const struct file_operations ipmr_vif_fops = {
2889         .owner   = THIS_MODULE,
2890         .open    = ipmr_vif_open,
2891         .read    = seq_read,
2892         .llseek  = seq_lseek,
2893         .release = seq_release_net,
2894 };
2895
2896 struct ipmr_mfc_iter {
2897         struct seq_net_private p;
2898         struct mr_table *mrt;
2899         struct list_head *cache;
2900 };
2901
2902 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2903                                           struct ipmr_mfc_iter *it, loff_t pos)
2904 {
2905         struct mr_table *mrt = it->mrt;
2906         struct mfc_cache *mfc;
2907
2908         rcu_read_lock();
2909         it->cache = &mrt->mfc_cache_list;
2910         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list)
2911                 if (pos-- == 0)
2912                         return mfc;
2913         rcu_read_unlock();
2914
2915         spin_lock_bh(&mfc_unres_lock);
2916         it->cache = &mrt->mfc_unres_queue;
2917         list_for_each_entry(mfc, it->cache, list)
2918                 if (pos-- == 0)
2919                         return mfc;
2920         spin_unlock_bh(&mfc_unres_lock);
2921
2922         it->cache = NULL;
2923         return NULL;
2924 }
2925
2926
2927 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2928 {
2929         struct ipmr_mfc_iter *it = seq->private;
2930         struct net *net = seq_file_net(seq);
2931         struct mr_table *mrt;
2932
2933         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2934         if (!mrt)
2935                 return ERR_PTR(-ENOENT);
2936
2937         it->mrt = mrt;
2938         it->cache = NULL;
2939         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2940                 : SEQ_START_TOKEN;
2941 }
2942
2943 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2944 {
2945         struct ipmr_mfc_iter *it = seq->private;
2946         struct net *net = seq_file_net(seq);
2947         struct mr_table *mrt = it->mrt;
2948         struct mfc_cache *mfc = v;
2949
2950         ++*pos;
2951
2952         if (v == SEQ_START_TOKEN)
2953                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2954
2955         if (mfc->list.next != it->cache)
2956                 return list_entry(mfc->list.next, struct mfc_cache, list);
2957
2958         if (it->cache == &mrt->mfc_unres_queue)
2959                 goto end_of_list;
2960
2961         /* exhausted cache_array, show unresolved */
2962         rcu_read_unlock();
2963         it->cache = &mrt->mfc_unres_queue;
2964
2965         spin_lock_bh(&mfc_unres_lock);
2966         if (!list_empty(it->cache))
2967                 return list_first_entry(it->cache, struct mfc_cache, list);
2968
2969 end_of_list:
2970         spin_unlock_bh(&mfc_unres_lock);
2971         it->cache = NULL;
2972
2973         return NULL;
2974 }
2975
2976 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2977 {
2978         struct ipmr_mfc_iter *it = seq->private;
2979         struct mr_table *mrt = it->mrt;
2980
2981         if (it->cache == &mrt->mfc_unres_queue)
2982                 spin_unlock_bh(&mfc_unres_lock);
2983         else if (it->cache == &mrt->mfc_cache_list)
2984                 rcu_read_unlock();
2985 }
2986
2987 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2988 {
2989         int n;
2990
2991         if (v == SEQ_START_TOKEN) {
2992                 seq_puts(seq,
2993                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2994         } else {
2995                 const struct mfc_cache *mfc = v;
2996                 const struct ipmr_mfc_iter *it = seq->private;
2997                 const struct mr_table *mrt = it->mrt;
2998
2999                 seq_printf(seq, "%08X %08X %-3hd",
3000                            (__force u32) mfc->mfc_mcastgrp,
3001                            (__force u32) mfc->mfc_origin,
3002                            mfc->mfc_parent);
3003
3004                 if (it->cache != &mrt->mfc_unres_queue) {
3005                         seq_printf(seq, " %8lu %8lu %8lu",
3006                                    mfc->mfc_un.res.pkt,
3007                                    mfc->mfc_un.res.bytes,
3008                                    mfc->mfc_un.res.wrong_if);
3009                         for (n = mfc->mfc_un.res.minvif;
3010                              n < mfc->mfc_un.res.maxvif; n++) {
3011                                 if (VIF_EXISTS(mrt, n) &&
3012                                     mfc->mfc_un.res.ttls[n] < 255)
3013                                         seq_printf(seq,
3014                                            " %2d:%-3d",
3015                                            n, mfc->mfc_un.res.ttls[n]);
3016                         }
3017                 } else {
3018                         /* unresolved mfc_caches don't contain
3019                          * pkt, bytes and wrong_if values
3020                          */
3021                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
3022                 }
3023                 seq_putc(seq, '\n');
3024         }
3025         return 0;
3026 }
3027
3028 static const struct seq_operations ipmr_mfc_seq_ops = {
3029         .start = ipmr_mfc_seq_start,
3030         .next  = ipmr_mfc_seq_next,
3031         .stop  = ipmr_mfc_seq_stop,
3032         .show  = ipmr_mfc_seq_show,
3033 };
3034
3035 static int ipmr_mfc_open(struct inode *inode, struct file *file)
3036 {
3037         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
3038                             sizeof(struct ipmr_mfc_iter));
3039 }
3040
3041 static const struct file_operations ipmr_mfc_fops = {
3042         .owner   = THIS_MODULE,
3043         .open    = ipmr_mfc_open,
3044         .read    = seq_read,
3045         .llseek  = seq_lseek,
3046         .release = seq_release_net,
3047 };
3048 #endif
3049
3050 #ifdef CONFIG_IP_PIMSM_V2
3051 static const struct net_protocol pim_protocol = {
3052         .handler        =       pim_rcv,
3053         .netns_ok       =       1,
3054 };
3055 #endif
3056
3057 /* Setup for IP multicast routing */
3058 static int __net_init ipmr_net_init(struct net *net)
3059 {
3060         int err;
3061
3062         err = ipmr_rules_init(net);
3063         if (err < 0)
3064                 goto fail;
3065
3066 #ifdef CONFIG_PROC_FS
3067         err = -ENOMEM;
3068         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
3069                 goto proc_vif_fail;
3070         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
3071                 goto proc_cache_fail;
3072 #endif
3073         return 0;
3074
3075 #ifdef CONFIG_PROC_FS
3076 proc_cache_fail:
3077         remove_proc_entry("ip_mr_vif", net->proc_net);
3078 proc_vif_fail:
3079         ipmr_rules_exit(net);
3080 #endif
3081 fail:
3082         return err;
3083 }
3084
3085 static void __net_exit ipmr_net_exit(struct net *net)
3086 {
3087 #ifdef CONFIG_PROC_FS
3088         remove_proc_entry("ip_mr_cache", net->proc_net);
3089         remove_proc_entry("ip_mr_vif", net->proc_net);
3090 #endif
3091         ipmr_rules_exit(net);
3092 }
3093
3094 static struct pernet_operations ipmr_net_ops = {
3095         .init = ipmr_net_init,
3096         .exit = ipmr_net_exit,
3097 };
3098
3099 int __init ip_mr_init(void)
3100 {
3101         int err;
3102
3103         mrt_cachep = kmem_cache_create("ip_mrt_cache",
3104                                        sizeof(struct mfc_cache),
3105                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
3106                                        NULL);
3107
3108         err = register_pernet_subsys(&ipmr_net_ops);
3109         if (err)
3110                 goto reg_pernet_fail;
3111
3112         err = register_netdevice_notifier(&ip_mr_notifier);
3113         if (err)
3114                 goto reg_notif_fail;
3115 #ifdef CONFIG_IP_PIMSM_V2
3116         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
3117                 pr_err("%s: can't add PIM protocol\n", __func__);
3118                 err = -EAGAIN;
3119                 goto add_proto_fail;
3120         }
3121 #endif
3122         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
3123                       ipmr_rtm_getroute, ipmr_rtm_dumproute, 0);
3124         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
3125                       ipmr_rtm_route, NULL, 0);
3126         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
3127                       ipmr_rtm_route, NULL, 0);
3128
3129         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETLINK,
3130                       NULL, ipmr_rtm_dumplink, 0);
3131         return 0;
3132
3133 #ifdef CONFIG_IP_PIMSM_V2
3134 add_proto_fail:
3135         unregister_netdevice_notifier(&ip_mr_notifier);
3136 #endif
3137 reg_notif_fail:
3138         unregister_pernet_subsys(&ipmr_net_ops);
3139 reg_pernet_fail:
3140         kmem_cache_destroy(mrt_cachep);
3141         return err;
3142 }