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 <asm/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 struct ipmr_rule {
  72         struct fib_rule         common;
  73 };
  74
  75 struct ipmr_result {
  76         struct mr_table         *mrt;
  77 };
  78
  79 /* Big lock, protecting vif table, mrt cache and mroute socket state.
  80  * Note that the changes are semaphored via rtnl_lock.
  81  */
  82
  83 static DEFINE_RWLOCK(mrt_lock);
  84
  85 /* Multicast router control variables */
  86
  87 /* Special spinlock for queue of unresolved entries */
  88 static DEFINE_SPINLOCK(mfc_unres_lock);
  89
  90 /* We return to original Alan's scheme. Hash table of resolved
  91  * entries is changed only in process context and protected
  92  * with weak lock mrt_lock. Queue of unresolved entries is protected
  93  * with strong spinlock mfc_unres_lock.
  94  *
  95  * In this case data path is free of exclusive locks at all.
  96  */
  97
  98 static struct kmem_cache *mrt_cachep __read_mostly;
  99
 100 static struct mr_table *ipmr_new_table(struct net *net, u32 id);
 101 static void ipmr_free_table(struct mr_table *mrt);
 102
 103 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
 104                           struct sk_buff *skb, struct mfc_cache *cache,
 105                           int local);
 106 static int ipmr_cache_report(struct mr_table *mrt,
 107                              struct sk_buff *pkt, vifi_t vifi, int assert);
 108 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
 109                               struct mfc_cache *c, struct rtmsg *rtm);
 110 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
 111                                  int cmd);
 112 static void mroute_clean_tables(struct mr_table *mrt, bool all);
 113 static void ipmr_expire_process(unsigned long arg);
 114
 115 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 116 #define ipmr_for_each_table(mrt, net) \
 117         list_for_each_entry_rcu(mrt, &net->ipv4.mr_tables, list)
 118
 119 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 120 {
 121         struct mr_table *mrt;
 122
 123         ipmr_for_each_table(mrt, net) {
 124                 if (mrt->id == id)
 125                         return mrt;
 126         }
 127         return NULL;
 128 }
 129
 130 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 131                            struct mr_table **mrt)
 132 {
 133         int err;
 134         struct ipmr_result res;
 135         struct fib_lookup_arg arg = {
 136                 .result = &res,
 137                 .flags = FIB_LOOKUP_NOREF,
 138         };
 139
 140         err = fib_rules_lookup(net->ipv4.mr_rules_ops,
 141                                flowi4_to_flowi(flp4), 0, &arg);
 142         if (err < 0)
 143                 return err;
 144         *mrt = res.mrt;
 145         return 0;
 146 }
 147
 148 static int ipmr_rule_action(struct fib_rule *rule, struct flowi *flp,
 149                             int flags, struct fib_lookup_arg *arg)
 150 {
 151         struct ipmr_result *res = arg->result;
 152         struct mr_table *mrt;
 153
 154         switch (rule->action) {
 155         case FR_ACT_TO_TBL:
 156                 break;
 157         case FR_ACT_UNREACHABLE:
 158                 return -ENETUNREACH;
 159         case FR_ACT_PROHIBIT:
 160                 return -EACCES;
 161         case FR_ACT_BLACKHOLE:
 162         default:
 163                 return -EINVAL;
 164         }
 165
 166         mrt = ipmr_get_table(rule->fr_net, rule->table);
 167         if (!mrt)
 168                 return -EAGAIN;
 169         res->mrt = mrt;
 170         return 0;
 171 }
 172
 173 static int ipmr_rule_match(struct fib_rule *rule, struct flowi *fl, int flags)
 174 {
 175         return 1;
 176 }
 177
 178 static const struct nla_policy ipmr_rule_policy[FRA_MAX + 1] = {
 179         FRA_GENERIC_POLICY,
 180 };
 181
 182 static int ipmr_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 183                                struct fib_rule_hdr *frh, struct nlattr **tb)
 184 {
 185         return 0;
 186 }
 187
 188 static int ipmr_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
 189                              struct nlattr **tb)
 190 {
 191         return 1;
 192 }
 193
 194 static int ipmr_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
 195                           struct fib_rule_hdr *frh)
 196 {
 197         frh->dst_len = 0;
 198         frh->src_len = 0;
 199         frh->tos     = 0;
 200         return 0;
 201 }
 202
 203 static const struct fib_rules_ops __net_initconst ipmr_rules_ops_template = {
 204         .family         = RTNL_FAMILY_IPMR,
 205         .rule_size      = sizeof(struct ipmr_rule),
 206         .addr_size      = sizeof(u32),
 207         .action         = ipmr_rule_action,
 208         .match          = ipmr_rule_match,
 209         .configure      = ipmr_rule_configure,
 210         .compare        = ipmr_rule_compare,
 211         .fill           = ipmr_rule_fill,
 212         .nlgroup        = RTNLGRP_IPV4_RULE,
 213         .policy         = ipmr_rule_policy,
 214         .owner          = THIS_MODULE,
 215 };
 216
 217 static int __net_init ipmr_rules_init(struct net *net)
 218 {
 219         struct fib_rules_ops *ops;
 220         struct mr_table *mrt;
 221         int err;
 222
 223         ops = fib_rules_register(&ipmr_rules_ops_template, net);
 224         if (IS_ERR(ops))
 225                 return PTR_ERR(ops);
 226
 227         INIT_LIST_HEAD(&net->ipv4.mr_tables);
 228
 229         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 230         if (IS_ERR(mrt)) {
 231                 err = PTR_ERR(mrt);
 232                 goto err1;
 233         }
 234
 235         err = fib_default_rule_add(ops, 0x7fff, RT_TABLE_DEFAULT, 0);
 236         if (err < 0)
 237                 goto err2;
 238
 239         net->ipv4.mr_rules_ops = ops;
 240         return 0;
 241
 242 err2:
 243         ipmr_free_table(mrt);
 244 err1:
 245         fib_rules_unregister(ops);
 246         return err;
 247 }
 248
 249 static void __net_exit ipmr_rules_exit(struct net *net)
 250 {
 251         struct mr_table *mrt, *next;
 252
 253         rtnl_lock();
 254         list_for_each_entry_safe(mrt, next, &net->ipv4.mr_tables, list) {
 255                 list_del(&mrt->list);
 256                 ipmr_free_table(mrt);
 257         }
 258         fib_rules_unregister(net->ipv4.mr_rules_ops);
 259         rtnl_unlock();
 260 }
 261 #else
 262 #define ipmr_for_each_table(mrt, net) \
 263         for (mrt = net->ipv4.mrt; mrt; mrt = NULL)
 264
 265 static struct mr_table *ipmr_get_table(struct net *net, u32 id)
 266 {
 267         return net->ipv4.mrt;
 268 }
 269
 270 static int ipmr_fib_lookup(struct net *net, struct flowi4 *flp4,
 271                            struct mr_table **mrt)
 272 {
 273         *mrt = net->ipv4.mrt;
 274         return 0;
 275 }
 276
 277 static int __net_init ipmr_rules_init(struct net *net)
 278 {
 279         struct mr_table *mrt;
 280
 281         mrt = ipmr_new_table(net, RT_TABLE_DEFAULT);
 282         if (IS_ERR(mrt))
 283                 return PTR_ERR(mrt);
 284         net->ipv4.mrt = mrt;
 285         return 0;
 286 }
 287
 288 static void __net_exit ipmr_rules_exit(struct net *net)
 289 {
 290         rtnl_lock();
 291         ipmr_free_table(net->ipv4.mrt);
 292         net->ipv4.mrt = NULL;
 293         rtnl_unlock();
 294 }
 295 #endif
 296
 297 static struct mr_table *ipmr_new_table(struct net *net, u32 id)
 298 {
 299         struct mr_table *mrt;
 300         unsigned int i;
 301
 302         /* "pimreg%u" should not exceed 16 bytes (IFNAMSIZ) */
 303         if (id != RT_TABLE_DEFAULT && id >= 1000000000)
 304                 return ERR_PTR(-EINVAL);
 305
 306         mrt = ipmr_get_table(net, id);
 307         if (mrt)
 308                 return mrt;
 309
 310         mrt = kzalloc(sizeof(*mrt), GFP_KERNEL);
 311         if (!mrt)
 312                 return ERR_PTR(-ENOMEM);
 313         write_pnet(&mrt->net, net);
 314         mrt->id = id;
 315
 316         /* Forwarding cache */
 317         for (i = 0; i < MFC_LINES; i++)
 318                 INIT_LIST_HEAD(&mrt->mfc_cache_array[i]);
 319
 320         INIT_LIST_HEAD(&mrt->mfc_unres_queue);
 321
 322         setup_timer(&mrt->ipmr_expire_timer, ipmr_expire_process,
 323                     (unsigned long)mrt);
 324
 325         mrt->mroute_reg_vif_num = -1;
 326 #ifdef CONFIG_IP_MROUTE_MULTIPLE_TABLES
 327         list_add_tail_rcu(&mrt->list, &net->ipv4.mr_tables);
 328 #endif
 329         return mrt;
 330 }
 331
 332 static void ipmr_free_table(struct mr_table *mrt)
 333 {
 334         del_timer_sync(&mrt->ipmr_expire_timer);
 335         mroute_clean_tables(mrt, true);
 336         kfree(mrt);
 337 }
 338
 339 /* Service routines creating virtual interfaces: DVMRP tunnels and PIMREG */
 340
 341 static void ipmr_del_tunnel(struct net_device *dev, struct vifctl *v)
 342 {
 343         struct net *net = dev_net(dev);
 344
 345         dev_close(dev);
 346
 347         dev = __dev_get_by_name(net, "tunl0");
 348         if (dev) {
 349                 const struct net_device_ops *ops = dev->netdev_ops;
 350                 struct ifreq ifr;
 351                 struct ip_tunnel_parm p;
 352
 353                 memset(&p, 0, sizeof(p));
 354                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 355                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 356                 p.iph.version = 4;
 357                 p.iph.ihl = 5;
 358                 p.iph.protocol = IPPROTO_IPIP;
 359                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 360                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 361
 362                 if (ops->ndo_do_ioctl) {
 363                         mm_segment_t oldfs = get_fs();
 364
 365                         set_fs(KERNEL_DS);
 366                         ops->ndo_do_ioctl(dev, &ifr, SIOCDELTUNNEL);
 367                         set_fs(oldfs);
 368                 }
 369         }
 370 }
 371
 372 /* Initialize ipmr pimreg/tunnel in_device */
 373 static bool ipmr_init_vif_indev(const struct net_device *dev)
 374 {
 375         struct in_device *in_dev;
 376
 377         ASSERT_RTNL();
 378
 379         in_dev = __in_dev_get_rtnl(dev);
 380         if (!in_dev)
 381                 return false;
 382         ipv4_devconf_setall(in_dev);
 383         neigh_parms_data_state_setall(in_dev->arp_parms);
 384         IPV4_DEVCONF(in_dev->cnf, RP_FILTER) = 0;
 385
 386         return true;
 387 }
 388
 389 static struct net_device *ipmr_new_tunnel(struct net *net, struct vifctl *v)
 390 {
 391         struct net_device  *dev;
 392
 393         dev = __dev_get_by_name(net, "tunl0");
 394
 395         if (dev) {
 396                 const struct net_device_ops *ops = dev->netdev_ops;
 397                 int err;
 398                 struct ifreq ifr;
 399                 struct ip_tunnel_parm p;
 400
 401                 memset(&p, 0, sizeof(p));
 402                 p.iph.daddr = v->vifc_rmt_addr.s_addr;
 403                 p.iph.saddr = v->vifc_lcl_addr.s_addr;
 404                 p.iph.version = 4;
 405                 p.iph.ihl = 5;
 406                 p.iph.protocol = IPPROTO_IPIP;
 407                 sprintf(p.name, "dvmrp%d", v->vifc_vifi);
 408                 ifr.ifr_ifru.ifru_data = (__force void __user *)&p;
 409
 410                 if (ops->ndo_do_ioctl) {
 411                         mm_segment_t oldfs = get_fs();
 412
 413                         set_fs(KERNEL_DS);
 414                         err = ops->ndo_do_ioctl(dev, &ifr, SIOCADDTUNNEL);
 415                         set_fs(oldfs);
 416                 } else {
 417                         err = -EOPNOTSUPP;
 418                 }
 419                 dev = NULL;
 420
 421                 if (err == 0 &&
 422                     (dev = __dev_get_by_name(net, p.name)) != NULL) {
 423                         dev->flags |= IFF_MULTICAST;
 424                         if (!ipmr_init_vif_indev(dev))
 425                                 goto failure;
 426                         if (dev_open(dev))
 427                                 goto failure;
 428                         dev_hold(dev);
 429                 }
 430         }
 431         return dev;
 432
 433 failure:
 434         unregister_netdevice(dev);
 435         return NULL;
 436 }
 437
 438 #if defined(CONFIG_IP_PIMSM_V1) || defined(CONFIG_IP_PIMSM_V2)
 439 static netdev_tx_t reg_vif_xmit(struct sk_buff *skb, struct net_device *dev)
 440 {
 441         struct net *net = dev_net(dev);
 442         struct mr_table *mrt;
 443         struct flowi4 fl4 = {
 444                 .flowi4_oif     = dev->ifindex,
 445                 .flowi4_iif     = skb->skb_iif ? : LOOPBACK_IFINDEX,
 446                 .flowi4_mark    = skb->mark,
 447         };
 448         int err;
 449
 450         err = ipmr_fib_lookup(net, &fl4, &mrt);
 451         if (err < 0) {
 452                 kfree_skb(skb);
 453                 return err;
 454         }
 455
 456         read_lock(&mrt_lock);
 457         dev->stats.tx_bytes += skb->len;
 458         dev->stats.tx_packets++;
 459         ipmr_cache_report(mrt, skb, mrt->mroute_reg_vif_num, IGMPMSG_WHOLEPKT);
 460         read_unlock(&mrt_lock);
 461         kfree_skb(skb);
 462         return NETDEV_TX_OK;
 463 }
 464
 465 static int reg_vif_get_iflink(const struct net_device *dev)
 466 {
 467         return 0;
 468 }
 469
 470 static const struct net_device_ops reg_vif_netdev_ops = {
 471         .ndo_start_xmit = reg_vif_xmit,
 472         .ndo_get_iflink = reg_vif_get_iflink,
 473 };
 474
 475 static void reg_vif_setup(struct net_device *dev)
 476 {
 477         dev->type               = ARPHRD_PIMREG;
 478         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 8;
 479         dev->flags              = IFF_NOARP;
 480         dev->netdev_ops         = &reg_vif_netdev_ops;
 481         dev->destructor         = free_netdev;
 482         dev->features           |= NETIF_F_NETNS_LOCAL;
 483 }
 484
 485 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 486 {
 487         struct net_device *dev;
 488         char name[IFNAMSIZ];
 489
 490         if (mrt->id == RT_TABLE_DEFAULT)
 491                 sprintf(name, "pimreg");
 492         else
 493                 sprintf(name, "pimreg%u", mrt->id);
 494
 495         dev = alloc_netdev(0, name, NET_NAME_UNKNOWN, reg_vif_setup);
 496
 497         if (!dev)
 498                 return NULL;
 499
 500         dev_net_set(dev, net);
 501
 502         if (register_netdevice(dev)) {
 503                 free_netdev(dev);
 504                 return NULL;
 505         }
 506
 507         if (!ipmr_init_vif_indev(dev))
 508                 goto failure;
 509         if (dev_open(dev))
 510                 goto failure;
 511
 512         dev_hold(dev);
 513
 514         return dev;
 515
 516 failure:
 517         unregister_netdevice(dev);
 518         return NULL;
 519 }
 520
 521 /* called with rcu_read_lock() */
 522 static int __pim_rcv(struct mr_table *mrt, struct sk_buff *skb,
 523                      unsigned int pimlen)
 524 {
 525         struct net_device *reg_dev = NULL;
 526         struct iphdr *encap;
 527
 528         encap = (struct iphdr *)(skb_transport_header(skb) + pimlen);
 529         /* Check that:
 530          * a. packet is really sent to a multicast group
 531          * b. packet is not a NULL-REGISTER
 532          * c. packet is not truncated
 533          */
 534         if (!ipv4_is_multicast(encap->daddr) ||
 535             encap->tot_len == 0 ||
 536             ntohs(encap->tot_len) + pimlen > skb->len)
 537                 return 1;
 538
 539         read_lock(&mrt_lock);
 540         if (mrt->mroute_reg_vif_num >= 0)
 541                 reg_dev = mrt->vif_table[mrt->mroute_reg_vif_num].dev;
 542         read_unlock(&mrt_lock);
 543
 544         if (!reg_dev)
 545                 return 1;
 546
 547         skb->mac_header = skb->network_header;
 548         skb_pull(skb, (u8 *)encap - skb->data);
 549         skb_reset_network_header(skb);
 550         skb->protocol = htons(ETH_P_IP);
 551         skb->ip_summed = CHECKSUM_NONE;
 552
 553         skb_tunnel_rx(skb, reg_dev, dev_net(reg_dev));
 554
 555         netif_rx(skb);
 556
 557         return NET_RX_SUCCESS;
 558 }
 559 #else
 560 static struct net_device *ipmr_reg_vif(struct net *net, struct mr_table *mrt)
 561 {
 562         return NULL;
 563 }
 564 #endif
 565
 566 /**
 567  *      vif_delete - Delete a VIF entry
 568  *      @notify: Set to 1, if the caller is a notifier_call
 569  */
 570 static int vif_delete(struct mr_table *mrt, int vifi, int notify,
 571                       struct list_head *head)
 572 {
 573         struct vif_device *v;
 574         struct net_device *dev;
 575         struct in_device *in_dev;
 576
 577         if (vifi < 0 || vifi >= mrt->maxvif)
 578                 return -EADDRNOTAVAIL;
 579
 580         v = &mrt->vif_table[vifi];
 581
 582         write_lock_bh(&mrt_lock);
 583         dev = v->dev;
 584         v->dev = NULL;
 585
 586         if (!dev) {
 587                 write_unlock_bh(&mrt_lock);
 588                 return -EADDRNOTAVAIL;
 589         }
 590
 591         if (vifi == mrt->mroute_reg_vif_num)
 592                 mrt->mroute_reg_vif_num = -1;
 593
 594         if (vifi + 1 == mrt->maxvif) {
 595                 int tmp;
 596
 597                 for (tmp = vifi - 1; tmp >= 0; tmp--) {
 598                         if (VIF_EXISTS(mrt, tmp))
 599                                 break;
 600                 }
 601                 mrt->maxvif = tmp+1;
 602         }
 603
 604         write_unlock_bh(&mrt_lock);
 605
 606         dev_set_allmulti(dev, -1);
 607
 608         in_dev = __in_dev_get_rtnl(dev);
 609         if (in_dev) {
 610                 IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)--;
 611                 inet_netconf_notify_devconf(dev_net(dev),
 612                                             NETCONFA_MC_FORWARDING,
 613                                             dev->ifindex, &in_dev->cnf);
 614                 ip_rt_multicast_event(in_dev);
 615         }
 616
 617         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER) && !notify)
 618                 unregister_netdevice_queue(dev, head);
 619
 620         dev_put(dev);
 621         return 0;
 622 }
 623
 624 static void ipmr_cache_free_rcu(struct rcu_head *head)
 625 {
 626         struct mfc_cache *c = container_of(head, struct mfc_cache, rcu);
 627
 628         kmem_cache_free(mrt_cachep, c);
 629 }
 630
 631 static inline void ipmr_cache_free(struct mfc_cache *c)
 632 {
 633         call_rcu(&c->rcu, ipmr_cache_free_rcu);
 634 }
 635
 636 /* Destroy an unresolved cache entry, killing queued skbs
 637  * and reporting error to netlink readers.
 638  */
 639 static void ipmr_destroy_unres(struct mr_table *mrt, struct mfc_cache *c)
 640 {
 641         struct net *net = read_pnet(&mrt->net);
 642         struct sk_buff *skb;
 643         struct nlmsgerr *e;
 644
 645         atomic_dec(&mrt->cache_resolve_queue_len);
 646
 647         while ((skb = skb_dequeue(&c->mfc_un.unres.unresolved))) {
 648                 if (ip_hdr(skb)->version == 0) {
 649                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 650                         nlh->nlmsg_type = NLMSG_ERROR;
 651                         nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 652                         skb_trim(skb, nlh->nlmsg_len);
 653                         e = nlmsg_data(nlh);
 654                         e->error = -ETIMEDOUT;
 655                         memset(&e->msg, 0, sizeof(e->msg));
 656
 657                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 658                 } else {
 659                         kfree_skb(skb);
 660                 }
 661         }
 662
 663         ipmr_cache_free(c);
 664 }
 665
 666 /* Timer process for the unresolved queue. */
 667 static void ipmr_expire_process(unsigned long arg)
 668 {
 669         struct mr_table *mrt = (struct mr_table *)arg;
 670         unsigned long now;
 671         unsigned long expires;
 672         struct mfc_cache *c, *next;
 673
 674         if (!spin_trylock(&mfc_unres_lock)) {
 675                 mod_timer(&mrt->ipmr_expire_timer, jiffies+HZ/10);
 676                 return;
 677         }
 678
 679         if (list_empty(&mrt->mfc_unres_queue))
 680                 goto out;
 681
 682         now = jiffies;
 683         expires = 10*HZ;
 684
 685         list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
 686                 if (time_after(c->mfc_un.unres.expires, now)) {
 687                         unsigned long interval = c->mfc_un.unres.expires - now;
 688                         if (interval < expires)
 689                                 expires = interval;
 690                         continue;
 691                 }
 692
 693                 list_del(&c->list);
 694                 mroute_netlink_event(mrt, c, RTM_DELROUTE);
 695                 ipmr_destroy_unres(mrt, c);
 696         }
 697
 698         if (!list_empty(&mrt->mfc_unres_queue))
 699                 mod_timer(&mrt->ipmr_expire_timer, jiffies + expires);
 700
 701 out:
 702         spin_unlock(&mfc_unres_lock);
 703 }
 704
 705 /* Fill oifs list. It is called under write locked mrt_lock. */
 706 static void ipmr_update_thresholds(struct mr_table *mrt, struct mfc_cache *cache,
 707                                    unsigned char *ttls)
 708 {
 709         int vifi;
 710
 711         cache->mfc_un.res.minvif = MAXVIFS;
 712         cache->mfc_un.res.maxvif = 0;
 713         memset(cache->mfc_un.res.ttls, 255, MAXVIFS);
 714
 715         for (vifi = 0; vifi < mrt->maxvif; vifi++) {
 716                 if (VIF_EXISTS(mrt, vifi) &&
 717                     ttls[vifi] && ttls[vifi] < 255) {
 718                         cache->mfc_un.res.ttls[vifi] = ttls[vifi];
 719                         if (cache->mfc_un.res.minvif > vifi)
 720                                 cache->mfc_un.res.minvif = vifi;
 721                         if (cache->mfc_un.res.maxvif <= vifi)
 722                                 cache->mfc_un.res.maxvif = vifi + 1;
 723                 }
 724         }
 725 }
 726
 727 static int vif_add(struct net *net, struct mr_table *mrt,
 728                    struct vifctl *vifc, int mrtsock)
 729 {
 730         int vifi = vifc->vifc_vifi;
 731         struct vif_device *v = &mrt->vif_table[vifi];
 732         struct net_device *dev;
 733         struct in_device *in_dev;
 734         int err;
 735
 736         /* Is vif busy ? */
 737         if (VIF_EXISTS(mrt, vifi))
 738                 return -EADDRINUSE;
 739
 740         switch (vifc->vifc_flags) {
 741         case VIFF_REGISTER:
 742                 if (!ipmr_pimsm_enabled())
 743                         return -EINVAL;
 744                 /* Special Purpose VIF in PIM
 745                  * All the packets will be sent to the daemon
 746                  */
 747                 if (mrt->mroute_reg_vif_num >= 0)
 748                         return -EADDRINUSE;
 749                 dev = ipmr_reg_vif(net, mrt);
 750                 if (!dev)
 751                         return -ENOBUFS;
 752                 err = dev_set_allmulti(dev, 1);
 753                 if (err) {
 754                         unregister_netdevice(dev);
 755                         dev_put(dev);
 756                         return err;
 757                 }
 758                 break;
 759         case VIFF_TUNNEL:
 760                 dev = ipmr_new_tunnel(net, vifc);
 761                 if (!dev)
 762                         return -ENOBUFS;
 763                 err = dev_set_allmulti(dev, 1);
 764                 if (err) {
 765                         ipmr_del_tunnel(dev, vifc);
 766                         dev_put(dev);
 767                         return err;
 768                 }
 769                 break;
 770         case VIFF_USE_IFINDEX:
 771         case 0:
 772                 if (vifc->vifc_flags == VIFF_USE_IFINDEX) {
 773                         dev = dev_get_by_index(net, vifc->vifc_lcl_ifindex);
 774                         if (dev && !__in_dev_get_rtnl(dev)) {
 775                                 dev_put(dev);
 776                                 return -EADDRNOTAVAIL;
 777                         }
 778                 } else {
 779                         dev = ip_dev_find(net, vifc->vifc_lcl_addr.s_addr);
 780                 }
 781                 if (!dev)
 782                         return -EADDRNOTAVAIL;
 783                 err = dev_set_allmulti(dev, 1);
 784                 if (err) {
 785                         dev_put(dev);
 786                         return err;
 787                 }
 788                 break;
 789         default:
 790                 return -EINVAL;
 791         }
 792
 793         in_dev = __in_dev_get_rtnl(dev);
 794         if (!in_dev) {
 795                 dev_put(dev);
 796                 return -EADDRNOTAVAIL;
 797         }
 798         IPV4_DEVCONF(in_dev->cnf, MC_FORWARDING)++;
 799         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING, dev->ifindex,
 800                                     &in_dev->cnf);
 801         ip_rt_multicast_event(in_dev);
 802
 803         /* Fill in the VIF structures */
 804
 805         v->rate_limit = vifc->vifc_rate_limit;
 806         v->local = vifc->vifc_lcl_addr.s_addr;
 807         v->remote = vifc->vifc_rmt_addr.s_addr;
 808         v->flags = vifc->vifc_flags;
 809         if (!mrtsock)
 810                 v->flags |= VIFF_STATIC;
 811         v->threshold = vifc->vifc_threshold;
 812         v->bytes_in = 0;
 813         v->bytes_out = 0;
 814         v->pkt_in = 0;
 815         v->pkt_out = 0;
 816         v->link = dev->ifindex;
 817         if (v->flags & (VIFF_TUNNEL | VIFF_REGISTER))
 818                 v->link = dev_get_iflink(dev);
 819
 820         /* And finish update writing critical data */
 821         write_lock_bh(&mrt_lock);
 822         v->dev = dev;
 823         if (v->flags & VIFF_REGISTER)
 824                 mrt->mroute_reg_vif_num = vifi;
 825         if (vifi+1 > mrt->maxvif)
 826                 mrt->maxvif = vifi+1;
 827         write_unlock_bh(&mrt_lock);
 828         return 0;
 829 }
 830
 831 /* called with rcu_read_lock() */
 832 static struct mfc_cache *ipmr_cache_find(struct mr_table *mrt,
 833                                          __be32 origin,
 834                                          __be32 mcastgrp)
 835 {
 836         int line = MFC_HASH(mcastgrp, origin);
 837         struct mfc_cache *c;
 838
 839         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list) {
 840                 if (c->mfc_origin == origin && c->mfc_mcastgrp == mcastgrp)
 841                         return c;
 842         }
 843         return NULL;
 844 }
 845
 846 /* Look for a (*,*,oif) entry */
 847 static struct mfc_cache *ipmr_cache_find_any_parent(struct mr_table *mrt,
 848                                                     int vifi)
 849 {
 850         int line = MFC_HASH(htonl(INADDR_ANY), htonl(INADDR_ANY));
 851         struct mfc_cache *c;
 852
 853         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 854                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 855                     c->mfc_mcastgrp == htonl(INADDR_ANY) &&
 856                     c->mfc_un.res.ttls[vifi] < 255)
 857                         return c;
 858
 859         return NULL;
 860 }
 861
 862 /* Look for a (*,G) entry */
 863 static struct mfc_cache *ipmr_cache_find_any(struct mr_table *mrt,
 864                                              __be32 mcastgrp, int vifi)
 865 {
 866         int line = MFC_HASH(mcastgrp, htonl(INADDR_ANY));
 867         struct mfc_cache *c, *proxy;
 868
 869         if (mcastgrp == htonl(INADDR_ANY))
 870                 goto skip;
 871
 872         list_for_each_entry_rcu(c, &mrt->mfc_cache_array[line], list)
 873                 if (c->mfc_origin == htonl(INADDR_ANY) &&
 874                     c->mfc_mcastgrp == mcastgrp) {
 875                         if (c->mfc_un.res.ttls[vifi] < 255)
 876                                 return c;
 877
 878                         /* It's ok if the vifi is part of the static tree */
 879                         proxy = ipmr_cache_find_any_parent(mrt,
 880                                                            c->mfc_parent);
 881                         if (proxy && proxy->mfc_un.res.ttls[vifi] < 255)
 882                                 return c;
 883                 }
 884
 885 skip:
 886         return ipmr_cache_find_any_parent(mrt, vifi);
 887 }
 888
 889 /* Allocate a multicast cache entry */
 890 static struct mfc_cache *ipmr_cache_alloc(void)
 891 {
 892         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_KERNEL);
 893
 894         if (c) {
 895                 c->mfc_un.res.last_assert = jiffies - MFC_ASSERT_THRESH - 1;
 896                 c->mfc_un.res.minvif = MAXVIFS;
 897         }
 898         return c;
 899 }
 900
 901 static struct mfc_cache *ipmr_cache_alloc_unres(void)
 902 {
 903         struct mfc_cache *c = kmem_cache_zalloc(mrt_cachep, GFP_ATOMIC);
 904
 905         if (c) {
 906                 skb_queue_head_init(&c->mfc_un.unres.unresolved);
 907                 c->mfc_un.unres.expires = jiffies + 10*HZ;
 908         }
 909         return c;
 910 }
 911
 912 /* A cache entry has gone into a resolved state from queued */
 913 static void ipmr_cache_resolve(struct net *net, struct mr_table *mrt,
 914                                struct mfc_cache *uc, struct mfc_cache *c)
 915 {
 916         struct sk_buff *skb;
 917         struct nlmsgerr *e;
 918
 919         /* Play the pending entries through our router */
 920         while ((skb = __skb_dequeue(&uc->mfc_un.unres.unresolved))) {
 921                 if (ip_hdr(skb)->version == 0) {
 922                         struct nlmsghdr *nlh = (struct nlmsghdr *)skb_pull(skb, sizeof(struct iphdr));
 923
 924                         if (__ipmr_fill_mroute(mrt, skb, c, nlmsg_data(nlh)) > 0) {
 925                                 nlh->nlmsg_len = skb_tail_pointer(skb) -
 926                                                  (u8 *)nlh;
 927                         } else {
 928                                 nlh->nlmsg_type = NLMSG_ERROR;
 929                                 nlh->nlmsg_len = nlmsg_msg_size(sizeof(struct nlmsgerr));
 930                                 skb_trim(skb, nlh->nlmsg_len);
 931                                 e = nlmsg_data(nlh);
 932                                 e->error = -EMSGSIZE;
 933                                 memset(&e->msg, 0, sizeof(e->msg));
 934                         }
 935
 936                         rtnl_unicast(skb, net, NETLINK_CB(skb).portid);
 937                 } else {
 938                         ip_mr_forward(net, mrt, skb, c, 0);
 939                 }
 940         }
 941 }
 942
 943 /* Bounce a cache query up to mrouted. We could use netlink for this but mrouted
 944  * expects the following bizarre scheme.
 945  *
 946  * Called under mrt_lock.
 947  */
 948 static int ipmr_cache_report(struct mr_table *mrt,
 949                              struct sk_buff *pkt, vifi_t vifi, int assert)
 950 {
 951         const int ihl = ip_hdrlen(pkt);
 952         struct sock *mroute_sk;
 953         struct igmphdr *igmp;
 954         struct igmpmsg *msg;
 955         struct sk_buff *skb;
 956         int ret;
 957
 958         if (assert == IGMPMSG_WHOLEPKT)
 959                 skb = skb_realloc_headroom(pkt, sizeof(struct iphdr));
 960         else
 961                 skb = alloc_skb(128, GFP_ATOMIC);
 962
 963         if (!skb)
 964                 return -ENOBUFS;
 965
 966         if (assert == IGMPMSG_WHOLEPKT) {
 967                 /* Ugly, but we have no choice with this interface.
 968                  * Duplicate old header, fix ihl, length etc.
 969                  * And all this only to mangle msg->im_msgtype and
 970                  * to set msg->im_mbz to "mbz" :-)
 971                  */
 972                 skb_push(skb, sizeof(struct iphdr));
 973                 skb_reset_network_header(skb);
 974                 skb_reset_transport_header(skb);
 975                 msg = (struct igmpmsg *)skb_network_header(skb);
 976                 memcpy(msg, skb_network_header(pkt), sizeof(struct iphdr));
 977                 msg->im_msgtype = IGMPMSG_WHOLEPKT;
 978                 msg->im_mbz = 0;
 979                 msg->im_vif = mrt->mroute_reg_vif_num;
 980                 ip_hdr(skb)->ihl = sizeof(struct iphdr) >> 2;
 981                 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(pkt)->tot_len) +
 982                                              sizeof(struct iphdr));
 983         } else {
 984                 /* Copy the IP header */
 985                 skb_set_network_header(skb, skb->len);
 986                 skb_put(skb, ihl);
 987                 skb_copy_to_linear_data(skb, pkt->data, ihl);
 988                 /* Flag to the kernel this is a route add */
 989                 ip_hdr(skb)->protocol = 0;
 990                 msg = (struct igmpmsg *)skb_network_header(skb);
 991                 msg->im_vif = vifi;
 992                 skb_dst_set(skb, dst_clone(skb_dst(pkt)));
 993                 /* Add our header */
 994                 igmp = (struct igmphdr *)skb_put(skb, sizeof(struct igmphdr));
 995                 igmp->type = assert;
 996                 msg->im_msgtype = assert;
 997                 igmp->code = 0;
 998                 ip_hdr(skb)->tot_len = htons(skb->len); /* Fix the length */
 999                 skb->transport_header = skb->network_header;
1000         }
1001
1002         rcu_read_lock();
1003         mroute_sk = rcu_dereference(mrt->mroute_sk);
1004         if (!mroute_sk) {
1005                 rcu_read_unlock();
1006                 kfree_skb(skb);
1007                 return -EINVAL;
1008         }
1009
1010         /* Deliver to mrouted */
1011         ret = sock_queue_rcv_skb(mroute_sk, skb);
1012         rcu_read_unlock();
1013         if (ret < 0) {
1014                 net_warn_ratelimited("mroute: pending queue full, dropping entries\n");
1015                 kfree_skb(skb);
1016         }
1017
1018         return ret;
1019 }
1020
1021 /* Queue a packet for resolution. It gets locked cache entry! */
1022 static int ipmr_cache_unresolved(struct mr_table *mrt, vifi_t vifi,
1023                                  struct sk_buff *skb)
1024 {
1025         bool found = false;
1026         int err;
1027         struct mfc_cache *c;
1028         const struct iphdr *iph = ip_hdr(skb);
1029
1030         spin_lock_bh(&mfc_unres_lock);
1031         list_for_each_entry(c, &mrt->mfc_unres_queue, list) {
1032                 if (c->mfc_mcastgrp == iph->daddr &&
1033                     c->mfc_origin == iph->saddr) {
1034                         found = true;
1035                         break;
1036                 }
1037         }
1038
1039         if (!found) {
1040                 /* Create a new entry if allowable */
1041                 if (atomic_read(&mrt->cache_resolve_queue_len) >= 10 ||
1042                     (c = ipmr_cache_alloc_unres()) == NULL) {
1043                         spin_unlock_bh(&mfc_unres_lock);
1044
1045                         kfree_skb(skb);
1046                         return -ENOBUFS;
1047                 }
1048
1049                 /* Fill in the new cache entry */
1050                 c->mfc_parent   = -1;
1051                 c->mfc_origin   = iph->saddr;
1052                 c->mfc_mcastgrp = iph->daddr;
1053
1054                 /* Reflect first query at mrouted. */
1055                 err = ipmr_cache_report(mrt, skb, vifi, IGMPMSG_NOCACHE);
1056                 if (err < 0) {
1057                         /* If the report failed throw the cache entry
1058                            out - Brad Parker
1059                          */
1060                         spin_unlock_bh(&mfc_unres_lock);
1061
1062                         ipmr_cache_free(c);
1063                         kfree_skb(skb);
1064                         return err;
1065                 }
1066
1067                 atomic_inc(&mrt->cache_resolve_queue_len);
1068                 list_add(&c->list, &mrt->mfc_unres_queue);
1069                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1070
1071                 if (atomic_read(&mrt->cache_resolve_queue_len) == 1)
1072                         mod_timer(&mrt->ipmr_expire_timer, c->mfc_un.unres.expires);
1073         }
1074
1075         /* See if we can append the packet */
1076         if (c->mfc_un.unres.unresolved.qlen > 3) {
1077                 kfree_skb(skb);
1078                 err = -ENOBUFS;
1079         } else {
1080                 skb_queue_tail(&c->mfc_un.unres.unresolved, skb);
1081                 err = 0;
1082         }
1083
1084         spin_unlock_bh(&mfc_unres_lock);
1085         return err;
1086 }
1087
1088 /* MFC cache manipulation by user space mroute daemon */
1089
1090 static int ipmr_mfc_delete(struct mr_table *mrt, struct mfcctl *mfc, int parent)
1091 {
1092         int line;
1093         struct mfc_cache *c, *next;
1094
1095         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1096
1097         list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[line], list) {
1098                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1099                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1100                     (parent == -1 || parent == c->mfc_parent)) {
1101                         list_del_rcu(&c->list);
1102                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1103                         ipmr_cache_free(c);
1104                         return 0;
1105                 }
1106         }
1107         return -ENOENT;
1108 }
1109
1110 static int ipmr_mfc_add(struct net *net, struct mr_table *mrt,
1111                         struct mfcctl *mfc, int mrtsock, int parent)
1112 {
1113         bool found = false;
1114         int line;
1115         struct mfc_cache *uc, *c;
1116
1117         if (mfc->mfcc_parent >= MAXVIFS)
1118                 return -ENFILE;
1119
1120         line = MFC_HASH(mfc->mfcc_mcastgrp.s_addr, mfc->mfcc_origin.s_addr);
1121
1122         list_for_each_entry(c, &mrt->mfc_cache_array[line], list) {
1123                 if (c->mfc_origin == mfc->mfcc_origin.s_addr &&
1124                     c->mfc_mcastgrp == mfc->mfcc_mcastgrp.s_addr &&
1125                     (parent == -1 || parent == c->mfc_parent)) {
1126                         found = true;
1127                         break;
1128                 }
1129         }
1130
1131         if (found) {
1132                 write_lock_bh(&mrt_lock);
1133                 c->mfc_parent = mfc->mfcc_parent;
1134                 ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1135                 if (!mrtsock)
1136                         c->mfc_flags |= MFC_STATIC;
1137                 write_unlock_bh(&mrt_lock);
1138                 mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1139                 return 0;
1140         }
1141
1142         if (mfc->mfcc_mcastgrp.s_addr != htonl(INADDR_ANY) &&
1143             !ipv4_is_multicast(mfc->mfcc_mcastgrp.s_addr))
1144                 return -EINVAL;
1145
1146         c = ipmr_cache_alloc();
1147         if (!c)
1148                 return -ENOMEM;
1149
1150         c->mfc_origin = mfc->mfcc_origin.s_addr;
1151         c->mfc_mcastgrp = mfc->mfcc_mcastgrp.s_addr;
1152         c->mfc_parent = mfc->mfcc_parent;
1153         ipmr_update_thresholds(mrt, c, mfc->mfcc_ttls);
1154         if (!mrtsock)
1155                 c->mfc_flags |= MFC_STATIC;
1156
1157         list_add_rcu(&c->list, &mrt->mfc_cache_array[line]);
1158
1159         /* Check to see if we resolved a queued list. If so we
1160          * need to send on the frames and tidy up.
1161          */
1162         found = false;
1163         spin_lock_bh(&mfc_unres_lock);
1164         list_for_each_entry(uc, &mrt->mfc_unres_queue, list) {
1165                 if (uc->mfc_origin == c->mfc_origin &&
1166                     uc->mfc_mcastgrp == c->mfc_mcastgrp) {
1167                         list_del(&uc->list);
1168                         atomic_dec(&mrt->cache_resolve_queue_len);
1169                         found = true;
1170                         break;
1171                 }
1172         }
1173         if (list_empty(&mrt->mfc_unres_queue))
1174                 del_timer(&mrt->ipmr_expire_timer);
1175         spin_unlock_bh(&mfc_unres_lock);
1176
1177         if (found) {
1178                 ipmr_cache_resolve(net, mrt, uc, c);
1179                 ipmr_cache_free(uc);
1180         }
1181         mroute_netlink_event(mrt, c, RTM_NEWROUTE);
1182         return 0;
1183 }
1184
1185 /* Close the multicast socket, and clear the vif tables etc */
1186 static void mroute_clean_tables(struct mr_table *mrt, bool all)
1187 {
1188         int i;
1189         LIST_HEAD(list);
1190         struct mfc_cache *c, *next;
1191
1192         /* Shut down all active vif entries */
1193         for (i = 0; i < mrt->maxvif; i++) {
1194                 if (!all && (mrt->vif_table[i].flags & VIFF_STATIC))
1195                         continue;
1196                 vif_delete(mrt, i, 0, &list);
1197         }
1198         unregister_netdevice_many(&list);
1199
1200         /* Wipe the cache */
1201         for (i = 0; i < MFC_LINES; i++) {
1202                 list_for_each_entry_safe(c, next, &mrt->mfc_cache_array[i], list) {
1203                         if (!all && (c->mfc_flags & MFC_STATIC))
1204                                 continue;
1205                         list_del_rcu(&c->list);
1206                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1207                         ipmr_cache_free(c);
1208                 }
1209         }
1210
1211         if (atomic_read(&mrt->cache_resolve_queue_len) != 0) {
1212                 spin_lock_bh(&mfc_unres_lock);
1213                 list_for_each_entry_safe(c, next, &mrt->mfc_unres_queue, list) {
1214                         list_del(&c->list);
1215                         mroute_netlink_event(mrt, c, RTM_DELROUTE);
1216                         ipmr_destroy_unres(mrt, c);
1217                 }
1218                 spin_unlock_bh(&mfc_unres_lock);
1219         }
1220 }
1221
1222 /* called from ip_ra_control(), before an RCU grace period,
1223  * we dont need to call synchronize_rcu() here
1224  */
1225 static void mrtsock_destruct(struct sock *sk)
1226 {
1227         struct net *net = sock_net(sk);
1228         struct mr_table *mrt;
1229
1230         rtnl_lock();
1231         ipmr_for_each_table(mrt, net) {
1232                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1233                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)--;
1234                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1235                                                     NETCONFA_IFINDEX_ALL,
1236                                                     net->ipv4.devconf_all);
1237                         RCU_INIT_POINTER(mrt->mroute_sk, NULL);
1238                         mroute_clean_tables(mrt, false);
1239                 }
1240         }
1241         rtnl_unlock();
1242 }
1243
1244 /* Socket options and virtual interface manipulation. The whole
1245  * virtual interface system is a complete heap, but unfortunately
1246  * that's how BSD mrouted happens to think. Maybe one day with a proper
1247  * MOSPF/PIM router set up we can clean this up.
1248  */
1249
1250 int ip_mroute_setsockopt(struct sock *sk, int optname, char __user *optval,
1251                          unsigned int optlen)
1252 {
1253         struct net *net = sock_net(sk);
1254         int val, ret = 0, parent = 0;
1255         struct mr_table *mrt;
1256         struct vifctl vif;
1257         struct mfcctl mfc;
1258         u32 uval;
1259
1260         /* There's one exception to the lock - MRT_DONE which needs to unlock */
1261         rtnl_lock();
1262         if (sk->sk_type != SOCK_RAW ||
1263             inet_sk(sk)->inet_num != IPPROTO_IGMP) {
1264                 ret = -EOPNOTSUPP;
1265                 goto out_unlock;
1266         }
1267
1268         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1269         if (!mrt) {
1270                 ret = -ENOENT;
1271                 goto out_unlock;
1272         }
1273         if (optname != MRT_INIT) {
1274                 if (sk != rcu_access_pointer(mrt->mroute_sk) &&
1275                     !ns_capable(net->user_ns, CAP_NET_ADMIN)) {
1276                         ret = -EACCES;
1277                         goto out_unlock;
1278                 }
1279         }
1280
1281         switch (optname) {
1282         case MRT_INIT:
1283                 if (optlen != sizeof(int)) {
1284                         ret = -EINVAL;
1285                         break;
1286                 }
1287                 if (rtnl_dereference(mrt->mroute_sk)) {
1288                         ret = -EADDRINUSE;
1289                         break;
1290                 }
1291
1292                 ret = ip_ra_control(sk, 1, mrtsock_destruct);
1293                 if (ret == 0) {
1294                         rcu_assign_pointer(mrt->mroute_sk, sk);
1295                         IPV4_DEVCONF_ALL(net, MC_FORWARDING)++;
1296                         inet_netconf_notify_devconf(net, NETCONFA_MC_FORWARDING,
1297                                                     NETCONFA_IFINDEX_ALL,
1298                                                     net->ipv4.devconf_all);
1299                 }
1300                 break;
1301         case MRT_DONE:
1302                 if (sk != rcu_access_pointer(mrt->mroute_sk)) {
1303                         ret = -EACCES;
1304                 } else {
1305                         /* We need to unlock here because mrtsock_destruct takes
1306                          * care of rtnl itself and we can't change that due to
1307                          * the IP_ROUTER_ALERT setsockopt which runs without it.
1308                          */
1309                         rtnl_unlock();
1310                         ret = ip_ra_control(sk, 0, NULL);
1311                         goto out;
1312                 }
1313                 break;
1314         case MRT_ADD_VIF:
1315         case MRT_DEL_VIF:
1316                 if (optlen != sizeof(vif)) {
1317                         ret = -EINVAL;
1318                         break;
1319                 }
1320                 if (copy_from_user(&vif, optval, sizeof(vif))) {
1321                         ret = -EFAULT;
1322                         break;
1323                 }
1324                 if (vif.vifc_vifi >= MAXVIFS) {
1325                         ret = -ENFILE;
1326                         break;
1327                 }
1328                 if (optname == MRT_ADD_VIF) {
1329                         ret = vif_add(net, mrt, &vif,
1330                                       sk == rtnl_dereference(mrt->mroute_sk));
1331                 } else {
1332                         ret = vif_delete(mrt, vif.vifc_vifi, 0, NULL);
1333                 }
1334                 break;
1335         /* Manipulate the forwarding caches. These live
1336          * in a sort of kernel/user symbiosis.
1337          */
1338         case MRT_ADD_MFC:
1339         case MRT_DEL_MFC:
1340                 parent = -1;
1341         case MRT_ADD_MFC_PROXY:
1342         case MRT_DEL_MFC_PROXY:
1343                 if (optlen != sizeof(mfc)) {
1344                         ret = -EINVAL;
1345                         break;
1346                 }
1347                 if (copy_from_user(&mfc, optval, sizeof(mfc))) {
1348                         ret = -EFAULT;
1349                         break;
1350                 }
1351                 if (parent == 0)
1352                         parent = mfc.mfcc_parent;
1353                 if (optname == MRT_DEL_MFC || optname == MRT_DEL_MFC_PROXY)
1354                         ret = ipmr_mfc_delete(mrt, &mfc, parent);
1355                 else
1356                         ret = ipmr_mfc_add(net, mrt, &mfc,
1357                                            sk == rtnl_dereference(mrt->mroute_sk),
1358                                            parent);
1359                 break;
1360         /* Control PIM assert. */
1361         case MRT_ASSERT:
1362                 if (optlen != sizeof(val)) {
1363                         ret = -EINVAL;
1364                         break;
1365                 }
1366                 if (get_user(val, (int __user *)optval)) {
1367                         ret = -EFAULT;
1368                         break;
1369                 }
1370                 mrt->mroute_do_assert = val;
1371                 break;
1372         case MRT_PIM:
1373                 if (!ipmr_pimsm_enabled()) {
1374                         ret = -ENOPROTOOPT;
1375                         break;
1376                 }
1377                 if (optlen != sizeof(val)) {
1378                         ret = -EINVAL;
1379                         break;
1380                 }
1381                 if (get_user(val, (int __user *)optval)) {
1382                         ret = -EFAULT;
1383                         break;
1384                 }
1385
1386                 val = !!val;
1387                 if (val != mrt->mroute_do_pim) {
1388                         mrt->mroute_do_pim = val;
1389                         mrt->mroute_do_assert = val;
1390                 }
1391                 break;
1392         case MRT_TABLE:
1393                 if (!IS_BUILTIN(CONFIG_IP_MROUTE_MULTIPLE_TABLES)) {
1394                         ret = -ENOPROTOOPT;
1395                         break;
1396                 }
1397                 if (optlen != sizeof(uval)) {
1398                         ret = -EINVAL;
1399                         break;
1400                 }
1401                 if (get_user(uval, (u32 __user *)optval)) {
1402                         ret = -EFAULT;
1403                         break;
1404                 }
1405
1406                 if (sk == rtnl_dereference(mrt->mroute_sk)) {
1407                         ret = -EBUSY;
1408                 } else {
1409                         mrt = ipmr_new_table(net, uval);
1410                         if (IS_ERR(mrt))
1411                                 ret = PTR_ERR(mrt);
1412                         else
1413                                 raw_sk(sk)->ipmr_table = uval;
1414                 }
1415                 break;
1416         /* Spurious command, or MRT_VERSION which you cannot set. */
1417         default:
1418                 ret = -ENOPROTOOPT;
1419         }
1420 out_unlock:
1421         rtnl_unlock();
1422 out:
1423         return ret;
1424 }
1425
1426 /* Getsock opt support for the multicast routing system. */
1427 int ip_mroute_getsockopt(struct sock *sk, int optname, char __user *optval, int __user *optlen)
1428 {
1429         int olr;
1430         int val;
1431         struct net *net = sock_net(sk);
1432         struct mr_table *mrt;
1433
1434         if (sk->sk_type != SOCK_RAW ||
1435             inet_sk(sk)->inet_num != IPPROTO_IGMP)
1436                 return -EOPNOTSUPP;
1437
1438         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1439         if (!mrt)
1440                 return -ENOENT;
1441
1442         switch (optname) {
1443         case MRT_VERSION:
1444                 val = 0x0305;
1445                 break;
1446         case MRT_PIM:
1447                 if (!ipmr_pimsm_enabled())
1448                         return -ENOPROTOOPT;
1449                 val = mrt->mroute_do_pim;
1450                 break;
1451         case MRT_ASSERT:
1452                 val = mrt->mroute_do_assert;
1453                 break;
1454         default:
1455                 return -ENOPROTOOPT;
1456         }
1457
1458         if (get_user(olr, optlen))
1459                 return -EFAULT;
1460         olr = min_t(unsigned int, olr, sizeof(int));
1461         if (olr < 0)
1462                 return -EINVAL;
1463         if (put_user(olr, optlen))
1464                 return -EFAULT;
1465         if (copy_to_user(optval, &val, olr))
1466                 return -EFAULT;
1467         return 0;
1468 }
1469
1470 /* The IP multicast ioctl support routines. */
1471 int ipmr_ioctl(struct sock *sk, int cmd, void __user *arg)
1472 {
1473         struct sioc_sg_req sr;
1474         struct sioc_vif_req vr;
1475         struct vif_device *vif;
1476         struct mfc_cache *c;
1477         struct net *net = sock_net(sk);
1478         struct mr_table *mrt;
1479
1480         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1481         if (!mrt)
1482                 return -ENOENT;
1483
1484         switch (cmd) {
1485         case SIOCGETVIFCNT:
1486                 if (copy_from_user(&vr, arg, sizeof(vr)))
1487                         return -EFAULT;
1488                 if (vr.vifi >= mrt->maxvif)
1489                         return -EINVAL;
1490                 read_lock(&mrt_lock);
1491                 vif = &mrt->vif_table[vr.vifi];
1492                 if (VIF_EXISTS(mrt, vr.vifi)) {
1493                         vr.icount = vif->pkt_in;
1494                         vr.ocount = vif->pkt_out;
1495                         vr.ibytes = vif->bytes_in;
1496                         vr.obytes = vif->bytes_out;
1497                         read_unlock(&mrt_lock);
1498
1499                         if (copy_to_user(arg, &vr, sizeof(vr)))
1500                                 return -EFAULT;
1501                         return 0;
1502                 }
1503                 read_unlock(&mrt_lock);
1504                 return -EADDRNOTAVAIL;
1505         case SIOCGETSGCNT:
1506                 if (copy_from_user(&sr, arg, sizeof(sr)))
1507                         return -EFAULT;
1508
1509                 rcu_read_lock();
1510                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1511                 if (c) {
1512                         sr.pktcnt = c->mfc_un.res.pkt;
1513                         sr.bytecnt = c->mfc_un.res.bytes;
1514                         sr.wrong_if = c->mfc_un.res.wrong_if;
1515                         rcu_read_unlock();
1516
1517                         if (copy_to_user(arg, &sr, sizeof(sr)))
1518                                 return -EFAULT;
1519                         return 0;
1520                 }
1521                 rcu_read_unlock();
1522                 return -EADDRNOTAVAIL;
1523         default:
1524                 return -ENOIOCTLCMD;
1525         }
1526 }
1527
1528 #ifdef CONFIG_COMPAT
1529 struct compat_sioc_sg_req {
1530         struct in_addr src;
1531         struct in_addr grp;
1532         compat_ulong_t pktcnt;
1533         compat_ulong_t bytecnt;
1534         compat_ulong_t wrong_if;
1535 };
1536
1537 struct compat_sioc_vif_req {
1538         vifi_t  vifi;           /* Which iface */
1539         compat_ulong_t icount;
1540         compat_ulong_t ocount;
1541         compat_ulong_t ibytes;
1542         compat_ulong_t obytes;
1543 };
1544
1545 int ipmr_compat_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
1546 {
1547         struct compat_sioc_sg_req sr;
1548         struct compat_sioc_vif_req vr;
1549         struct vif_device *vif;
1550         struct mfc_cache *c;
1551         struct net *net = sock_net(sk);
1552         struct mr_table *mrt;
1553
1554         mrt = ipmr_get_table(net, raw_sk(sk)->ipmr_table ? : RT_TABLE_DEFAULT);
1555         if (!mrt)
1556                 return -ENOENT;
1557
1558         switch (cmd) {
1559         case SIOCGETVIFCNT:
1560                 if (copy_from_user(&vr, arg, sizeof(vr)))
1561                         return -EFAULT;
1562                 if (vr.vifi >= mrt->maxvif)
1563                         return -EINVAL;
1564                 read_lock(&mrt_lock);
1565                 vif = &mrt->vif_table[vr.vifi];
1566                 if (VIF_EXISTS(mrt, vr.vifi)) {
1567                         vr.icount = vif->pkt_in;
1568                         vr.ocount = vif->pkt_out;
1569                         vr.ibytes = vif->bytes_in;
1570                         vr.obytes = vif->bytes_out;
1571                         read_unlock(&mrt_lock);
1572
1573                         if (copy_to_user(arg, &vr, sizeof(vr)))
1574                                 return -EFAULT;
1575                         return 0;
1576                 }
1577                 read_unlock(&mrt_lock);
1578                 return -EADDRNOTAVAIL;
1579         case SIOCGETSGCNT:
1580                 if (copy_from_user(&sr, arg, sizeof(sr)))
1581                         return -EFAULT;
1582
1583                 rcu_read_lock();
1584                 c = ipmr_cache_find(mrt, sr.src.s_addr, sr.grp.s_addr);
1585                 if (c) {
1586                         sr.pktcnt = c->mfc_un.res.pkt;
1587                         sr.bytecnt = c->mfc_un.res.bytes;
1588                         sr.wrong_if = c->mfc_un.res.wrong_if;
1589                         rcu_read_unlock();
1590
1591                         if (copy_to_user(arg, &sr, sizeof(sr)))
1592                                 return -EFAULT;
1593                         return 0;
1594                 }
1595                 rcu_read_unlock();
1596                 return -EADDRNOTAVAIL;
1597         default:
1598                 return -ENOIOCTLCMD;
1599         }
1600 }
1601 #endif
1602
1603 static int ipmr_device_event(struct notifier_block *this, unsigned long event, void *ptr)
1604 {
1605         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1606         struct net *net = dev_net(dev);
1607         struct mr_table *mrt;
1608         struct vif_device *v;
1609         int ct;
1610
1611         if (event != NETDEV_UNREGISTER)
1612                 return NOTIFY_DONE;
1613
1614         ipmr_for_each_table(mrt, net) {
1615                 v = &mrt->vif_table[0];
1616                 for (ct = 0; ct < mrt->maxvif; ct++, v++) {
1617                         if (v->dev == dev)
1618                                 vif_delete(mrt, ct, 1, NULL);
1619                 }
1620         }
1621         return NOTIFY_DONE;
1622 }
1623
1624 static struct notifier_block ip_mr_notifier = {
1625         .notifier_call = ipmr_device_event,
1626 };
1627
1628 /* Encapsulate a packet by attaching a valid IPIP header to it.
1629  * This avoids tunnel drivers and other mess and gives us the speed so
1630  * important for multicast video.
1631  */
1632 static void ip_encap(struct net *net, struct sk_buff *skb,
1633                      __be32 saddr, __be32 daddr)
1634 {
1635         struct iphdr *iph;
1636         const struct iphdr *old_iph = ip_hdr(skb);
1637
1638         skb_push(skb, sizeof(struct iphdr));
1639         skb->transport_header = skb->network_header;
1640         skb_reset_network_header(skb);
1641         iph = ip_hdr(skb);
1642
1643         iph->version    =       4;
1644         iph->tos        =       old_iph->tos;
1645         iph->ttl        =       old_iph->ttl;
1646         iph->frag_off   =       0;
1647         iph->daddr      =       daddr;
1648         iph->saddr      =       saddr;
1649         iph->protocol   =       IPPROTO_IPIP;
1650         iph->ihl        =       5;
1651         iph->tot_len    =       htons(skb->len);
1652         ip_select_ident(net, skb, NULL);
1653         ip_send_check(iph);
1654
1655         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1656         nf_reset(skb);
1657 }
1658
1659 static inline int ipmr_forward_finish(struct net *net, struct sock *sk,
1660                                       struct sk_buff *skb)
1661 {
1662         struct ip_options *opt = &(IPCB(skb)->opt);
1663
1664         IP_INC_STATS(net, IPSTATS_MIB_OUTFORWDATAGRAMS);
1665         IP_ADD_STATS(net, IPSTATS_MIB_OUTOCTETS, skb->len);
1666
1667         if (unlikely(opt->optlen))
1668                 ip_forward_options(skb);
1669
1670         return dst_output(net, sk, skb);
1671 }
1672
1673 /* Processing handlers for ipmr_forward */
1674
1675 static void ipmr_queue_xmit(struct net *net, struct mr_table *mrt,
1676                             struct sk_buff *skb, struct mfc_cache *c, int vifi)
1677 {
1678         const struct iphdr *iph = ip_hdr(skb);
1679         struct vif_device *vif = &mrt->vif_table[vifi];
1680         struct net_device *dev;
1681         struct rtable *rt;
1682         struct flowi4 fl4;
1683         int    encap = 0;
1684
1685         if (!vif->dev)
1686                 goto out_free;
1687
1688         if (vif->flags & VIFF_REGISTER) {
1689                 vif->pkt_out++;
1690                 vif->bytes_out += skb->len;
1691                 vif->dev->stats.tx_bytes += skb->len;
1692                 vif->dev->stats.tx_packets++;
1693                 ipmr_cache_report(mrt, skb, vifi, IGMPMSG_WHOLEPKT);
1694                 goto out_free;
1695         }
1696
1697         if (vif->flags & VIFF_TUNNEL) {
1698                 rt = ip_route_output_ports(net, &fl4, NULL,
1699                                            vif->remote, vif->local,
1700                                            0, 0,
1701                                            IPPROTO_IPIP,
1702                                            RT_TOS(iph->tos), vif->link);
1703                 if (IS_ERR(rt))
1704                         goto out_free;
1705                 encap = sizeof(struct iphdr);
1706         } else {
1707                 rt = ip_route_output_ports(net, &fl4, NULL, iph->daddr, 0,
1708                                            0, 0,
1709                                            IPPROTO_IPIP,
1710                                            RT_TOS(iph->tos), vif->link);
1711                 if (IS_ERR(rt))
1712                         goto out_free;
1713         }
1714
1715         dev = rt->dst.dev;
1716
1717         if (skb->len+encap > dst_mtu(&rt->dst) && (ntohs(iph->frag_off) & IP_DF)) {
1718                 /* Do not fragment multicasts. Alas, IPv4 does not
1719                  * allow to send ICMP, so that packets will disappear
1720                  * to blackhole.
1721                  */
1722                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
1723                 ip_rt_put(rt);
1724                 goto out_free;
1725         }
1726
1727         encap += LL_RESERVED_SPACE(dev) + rt->dst.header_len;
1728
1729         if (skb_cow(skb, encap)) {
1730                 ip_rt_put(rt);
1731                 goto out_free;
1732         }
1733
1734         vif->pkt_out++;
1735         vif->bytes_out += skb->len;
1736
1737         skb_dst_drop(skb);
1738         skb_dst_set(skb, &rt->dst);
1739         ip_decrease_ttl(ip_hdr(skb));
1740
1741         /* FIXME: forward and output firewalls used to be called here.
1742          * What do we do with netfilter? -- RR
1743          */
1744         if (vif->flags & VIFF_TUNNEL) {
1745                 ip_encap(net, skb, vif->local, vif->remote);
1746                 /* FIXME: extra output firewall step used to be here. --RR */
1747                 vif->dev->stats.tx_packets++;
1748                 vif->dev->stats.tx_bytes += skb->len;
1749         }
1750
1751         IPCB(skb)->flags |= IPSKB_FORWARDED;
1752
1753         /* RFC1584 teaches, that DVMRP/PIM router must deliver packets locally
1754          * not only before forwarding, but after forwarding on all output
1755          * interfaces. It is clear, if mrouter runs a multicasting
1756          * program, it should receive packets not depending to what interface
1757          * program is joined.
1758          * If we will not make it, the program will have to join on all
1759          * interfaces. On the other hand, multihoming host (or router, but
1760          * not mrouter) cannot join to more than one interface - it will
1761          * result in receiving multiple packets.
1762          */
1763         NF_HOOK(NFPROTO_IPV4, NF_INET_FORWARD,
1764                 net, NULL, skb, skb->dev, dev,
1765                 ipmr_forward_finish);
1766         return;
1767
1768 out_free:
1769         kfree_skb(skb);
1770 }
1771
1772 static int ipmr_find_vif(struct mr_table *mrt, struct net_device *dev)
1773 {
1774         int ct;
1775
1776         for (ct = mrt->maxvif-1; ct >= 0; ct--) {
1777                 if (mrt->vif_table[ct].dev == dev)
1778                         break;
1779         }
1780         return ct;
1781 }
1782
1783 /* "local" means that we should preserve one skb (for local delivery) */
1784 static void ip_mr_forward(struct net *net, struct mr_table *mrt,
1785                           struct sk_buff *skb, struct mfc_cache *cache,
1786                           int local)
1787 {
1788         int psend = -1;
1789         int vif, ct;
1790         int true_vifi = ipmr_find_vif(mrt, skb->dev);
1791
1792         vif = cache->mfc_parent;
1793         cache->mfc_un.res.pkt++;
1794         cache->mfc_un.res.bytes += skb->len;
1795
1796         if (cache->mfc_origin == htonl(INADDR_ANY) && true_vifi >= 0) {
1797                 struct mfc_cache *cache_proxy;
1798
1799                 /* For an (*,G) entry, we only check that the incomming
1800                  * interface is part of the static tree.
1801                  */
1802                 cache_proxy = ipmr_cache_find_any_parent(mrt, vif);
1803                 if (cache_proxy &&
1804                     cache_proxy->mfc_un.res.ttls[true_vifi] < 255)
1805                         goto forward;
1806         }
1807
1808         /* Wrong interface: drop packet and (maybe) send PIM assert. */
1809         if (mrt->vif_table[vif].dev != skb->dev) {
1810                 if (rt_is_output_route(skb_rtable(skb))) {
1811                         /* It is our own packet, looped back.
1812                          * Very complicated situation...
1813                          *
1814                          * The best workaround until routing daemons will be
1815                          * fixed is not to redistribute packet, if it was
1816                          * send through wrong interface. It means, that
1817                          * multicast applications WILL NOT work for
1818                          * (S,G), which have default multicast route pointing
1819                          * to wrong oif. In any case, it is not a good
1820                          * idea to use multicasting applications on router.
1821                          */
1822                         goto dont_forward;
1823                 }
1824
1825                 cache->mfc_un.res.wrong_if++;
1826
1827                 if (true_vifi >= 0 && mrt->mroute_do_assert &&
1828                     /* pimsm uses asserts, when switching from RPT to SPT,
1829                      * so that we cannot check that packet arrived on an oif.
1830                      * It is bad, but otherwise we would need to move pretty
1831                      * large chunk of pimd to kernel. Ough... --ANK
1832                      */
1833                     (mrt->mroute_do_pim ||
1834                      cache->mfc_un.res.ttls[true_vifi] < 255) &&
1835                     time_after(jiffies,
1836                                cache->mfc_un.res.last_assert + MFC_ASSERT_THRESH)) {
1837                         cache->mfc_un.res.last_assert = jiffies;
1838                         ipmr_cache_report(mrt, skb, true_vifi, IGMPMSG_WRONGVIF);
1839                 }
1840                 goto dont_forward;
1841         }
1842
1843 forward:
1844         mrt->vif_table[vif].pkt_in++;
1845         mrt->vif_table[vif].bytes_in += skb->len;
1846
1847         /* Forward the frame */
1848         if (cache->mfc_origin == htonl(INADDR_ANY) &&
1849             cache->mfc_mcastgrp == htonl(INADDR_ANY)) {
1850                 if (true_vifi >= 0 &&
1851                     true_vifi != cache->mfc_parent &&
1852                     ip_hdr(skb)->ttl >
1853                                 cache->mfc_un.res.ttls[cache->mfc_parent]) {
1854                         /* It's an (*,*) entry and the packet is not coming from
1855                          * the upstream: forward the packet to the upstream
1856                          * only.
1857                          */
1858                         psend = cache->mfc_parent;
1859                         goto last_forward;
1860                 }
1861                 goto dont_forward;
1862         }
1863         for (ct = cache->mfc_un.res.maxvif - 1;
1864              ct >= cache->mfc_un.res.minvif; ct--) {
1865                 /* For (*,G) entry, don't forward to the incoming interface */
1866                 if ((cache->mfc_origin != htonl(INADDR_ANY) ||
1867                      ct != true_vifi) &&
1868                     ip_hdr(skb)->ttl > cache->mfc_un.res.ttls[ct]) {
1869                         if (psend != -1) {
1870                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1871
1872                                 if (skb2)
1873                                         ipmr_queue_xmit(net, mrt, skb2, cache,
1874                                                         psend);
1875                         }
1876                         psend = ct;
1877                 }
1878         }
1879 last_forward:
1880         if (psend != -1) {
1881                 if (local) {
1882                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1883
1884                         if (skb2)
1885                                 ipmr_queue_xmit(net, mrt, skb2, cache, psend);
1886                 } else {
1887                         ipmr_queue_xmit(net, mrt, skb, cache, psend);
1888                         return;
1889                 }
1890         }
1891
1892 dont_forward:
1893         if (!local)
1894                 kfree_skb(skb);
1895 }
1896
1897 static struct mr_table *ipmr_rt_fib_lookup(struct net *net, struct sk_buff *skb)
1898 {
1899         struct rtable *rt = skb_rtable(skb);
1900         struct iphdr *iph = ip_hdr(skb);
1901         struct flowi4 fl4 = {
1902                 .daddr = iph->daddr,
1903                 .saddr = iph->saddr,
1904                 .flowi4_tos = RT_TOS(iph->tos),
1905                 .flowi4_oif = (rt_is_output_route(rt) ?
1906                                skb->dev->ifindex : 0),
1907                 .flowi4_iif = (rt_is_output_route(rt) ?
1908                                LOOPBACK_IFINDEX :
1909                                skb->dev->ifindex),
1910                 .flowi4_mark = skb->mark,
1911         };
1912         struct mr_table *mrt;
1913         int err;
1914
1915         err = ipmr_fib_lookup(net, &fl4, &mrt);
1916         if (err)
1917                 return ERR_PTR(err);
1918         return mrt;
1919 }
1920
1921 /* Multicast packets for forwarding arrive here
1922  * Called with rcu_read_lock();
1923  */
1924 int ip_mr_input(struct sk_buff *skb)
1925 {
1926         struct mfc_cache *cache;
1927         struct net *net = dev_net(skb->dev);
1928         int local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
1929         struct mr_table *mrt;
1930
1931         /* Packet is looped back after forward, it should not be
1932          * forwarded second time, but still can be delivered locally.
1933          */
1934         if (IPCB(skb)->flags & IPSKB_FORWARDED)
1935                 goto dont_forward;
1936
1937         mrt = ipmr_rt_fib_lookup(net, skb);
1938         if (IS_ERR(mrt)) {
1939                 kfree_skb(skb);
1940                 return PTR_ERR(mrt);
1941         }
1942         if (!local) {
1943                 if (IPCB(skb)->opt.router_alert) {
1944                         if (ip_call_ra_chain(skb))
1945                                 return 0;
1946                 } else if (ip_hdr(skb)->protocol == IPPROTO_IGMP) {
1947                         /* IGMPv1 (and broken IGMPv2 implementations sort of
1948                          * Cisco IOS <= 11.2(8)) do not put router alert
1949                          * option to IGMP packets destined to routable
1950                          * groups. It is very bad, because it means
1951                          * that we can forward NO IGMP messages.
1952                          */
1953                         struct sock *mroute_sk;
1954
1955                         mroute_sk = rcu_dereference(mrt->mroute_sk);
1956                         if (mroute_sk) {
1957                                 nf_reset(skb);
1958                                 raw_rcv(mroute_sk, skb);
1959                                 return 0;
1960                         }
1961                     }
1962         }
1963
1964         /* already under rcu_read_lock() */
1965         cache = ipmr_cache_find(mrt, ip_hdr(skb)->saddr, ip_hdr(skb)->daddr);
1966         if (!cache) {
1967                 int vif = ipmr_find_vif(mrt, skb->dev);
1968
1969                 if (vif >= 0)
1970                         cache = ipmr_cache_find_any(mrt, ip_hdr(skb)->daddr,
1971                                                     vif);
1972         }
1973
1974         /* No usable cache entry */
1975         if (!cache) {
1976                 int vif;
1977
1978                 if (local) {
1979                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1980                         ip_local_deliver(skb);
1981                         if (!skb2)
1982                                 return -ENOBUFS;
1983                         skb = skb2;
1984                 }
1985
1986                 read_lock(&mrt_lock);
1987                 vif = ipmr_find_vif(mrt, skb->dev);
1988                 if (vif >= 0) {
1989                         int err2 = ipmr_cache_unresolved(mrt, vif, skb);
1990                         read_unlock(&mrt_lock);
1991
1992                         return err2;
1993                 }
1994                 read_unlock(&mrt_lock);
1995                 kfree_skb(skb);
1996                 return -ENODEV;
1997         }
1998
1999         read_lock(&mrt_lock);
2000         ip_mr_forward(net, mrt, skb, cache, local);
2001         read_unlock(&mrt_lock);
2002
2003         if (local)
2004                 return ip_local_deliver(skb);
2005
2006         return 0;
2007
2008 dont_forward:
2009         if (local)
2010                 return ip_local_deliver(skb);
2011         kfree_skb(skb);
2012         return 0;
2013 }
2014
2015 #ifdef CONFIG_IP_PIMSM_V1
2016 /* Handle IGMP messages of PIMv1 */
2017 int pim_rcv_v1(struct sk_buff *skb)
2018 {
2019         struct igmphdr *pim;
2020         struct net *net = dev_net(skb->dev);
2021         struct mr_table *mrt;
2022
2023         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2024                 goto drop;
2025
2026         pim = igmp_hdr(skb);
2027
2028         mrt = ipmr_rt_fib_lookup(net, skb);
2029         if (IS_ERR(mrt))
2030                 goto drop;
2031         if (!mrt->mroute_do_pim ||
2032             pim->group != PIM_V1_VERSION || pim->code != PIM_V1_REGISTER)
2033                 goto drop;
2034
2035         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2036 drop:
2037                 kfree_skb(skb);
2038         }
2039         return 0;
2040 }
2041 #endif
2042
2043 #ifdef CONFIG_IP_PIMSM_V2
2044 static int pim_rcv(struct sk_buff *skb)
2045 {
2046         struct pimreghdr *pim;
2047         struct net *net = dev_net(skb->dev);
2048         struct mr_table *mrt;
2049
2050         if (!pskb_may_pull(skb, sizeof(*pim) + sizeof(struct iphdr)))
2051                 goto drop;
2052
2053         pim = (struct pimreghdr *)skb_transport_header(skb);
2054         if (pim->type != ((PIM_VERSION << 4) | (PIM_REGISTER)) ||
2055             (pim->flags & PIM_NULL_REGISTER) ||
2056             (ip_compute_csum((void *)pim, sizeof(*pim)) != 0 &&
2057              csum_fold(skb_checksum(skb, 0, skb->len, 0))))
2058                 goto drop;
2059
2060         mrt = ipmr_rt_fib_lookup(net, skb);
2061         if (IS_ERR(mrt))
2062                 goto drop;
2063         if (__pim_rcv(mrt, skb, sizeof(*pim))) {
2064 drop:
2065                 kfree_skb(skb);
2066         }
2067         return 0;
2068 }
2069 #endif
2070
2071 static int __ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2072                               struct mfc_cache *c, struct rtmsg *rtm)
2073 {
2074         int ct;
2075         struct rtnexthop *nhp;
2076         struct nlattr *mp_attr;
2077         struct rta_mfc_stats mfcs;
2078
2079         /* If cache is unresolved, don't try to parse IIF and OIF */
2080         if (c->mfc_parent >= MAXVIFS)
2081                 return -ENOENT;
2082
2083         if (VIF_EXISTS(mrt, c->mfc_parent) &&
2084             nla_put_u32(skb, RTA_IIF, mrt->vif_table[c->mfc_parent].dev->ifindex) < 0)
2085                 return -EMSGSIZE;
2086
2087         if (!(mp_attr = nla_nest_start(skb, RTA_MULTIPATH)))
2088                 return -EMSGSIZE;
2089
2090         for (ct = c->mfc_un.res.minvif; ct < c->mfc_un.res.maxvif; ct++) {
2091                 if (VIF_EXISTS(mrt, ct) && c->mfc_un.res.ttls[ct] < 255) {
2092                         if (!(nhp = nla_reserve_nohdr(skb, sizeof(*nhp)))) {
2093                                 nla_nest_cancel(skb, mp_attr);
2094                                 return -EMSGSIZE;
2095                         }
2096
2097                         nhp->rtnh_flags = 0;
2098                         nhp->rtnh_hops = c->mfc_un.res.ttls[ct];
2099                         nhp->rtnh_ifindex = mrt->vif_table[ct].dev->ifindex;
2100                         nhp->rtnh_len = sizeof(*nhp);
2101                 }
2102         }
2103
2104         nla_nest_end(skb, mp_attr);
2105
2106         mfcs.mfcs_packets = c->mfc_un.res.pkt;
2107         mfcs.mfcs_bytes = c->mfc_un.res.bytes;
2108         mfcs.mfcs_wrong_if = c->mfc_un.res.wrong_if;
2109         if (nla_put_64bit(skb, RTA_MFC_STATS, sizeof(mfcs), &mfcs, RTA_PAD) < 0)
2110                 return -EMSGSIZE;
2111
2112         rtm->rtm_type = RTN_MULTICAST;
2113         return 1;
2114 }
2115
2116 int ipmr_get_route(struct net *net, struct sk_buff *skb,
2117                    __be32 saddr, __be32 daddr,
2118                    struct rtmsg *rtm, int nowait)
2119 {
2120         struct mfc_cache *cache;
2121         struct mr_table *mrt;
2122         int err;
2123
2124         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2125         if (!mrt)
2126                 return -ENOENT;
2127
2128         rcu_read_lock();
2129         cache = ipmr_cache_find(mrt, saddr, daddr);
2130         if (!cache && skb->dev) {
2131                 int vif = ipmr_find_vif(mrt, skb->dev);
2132
2133                 if (vif >= 0)
2134                         cache = ipmr_cache_find_any(mrt, daddr, vif);
2135         }
2136         if (!cache) {
2137                 struct sk_buff *skb2;
2138                 struct iphdr *iph;
2139                 struct net_device *dev;
2140                 int vif = -1;
2141
2142                 if (nowait) {
2143                         rcu_read_unlock();
2144                         return -EAGAIN;
2145                 }
2146
2147                 dev = skb->dev;
2148                 read_lock(&mrt_lock);
2149                 if (dev)
2150                         vif = ipmr_find_vif(mrt, dev);
2151                 if (vif < 0) {
2152                         read_unlock(&mrt_lock);
2153                         rcu_read_unlock();
2154                         return -ENODEV;
2155                 }
2156                 skb2 = skb_clone(skb, GFP_ATOMIC);
2157                 if (!skb2) {
2158                         read_unlock(&mrt_lock);
2159                         rcu_read_unlock();
2160                         return -ENOMEM;
2161                 }
2162
2163                 skb_push(skb2, sizeof(struct iphdr));
2164                 skb_reset_network_header(skb2);
2165                 iph = ip_hdr(skb2);
2166                 iph->ihl = sizeof(struct iphdr) >> 2;
2167                 iph->saddr = saddr;
2168                 iph->daddr = daddr;
2169                 iph->version = 0;
2170                 err = ipmr_cache_unresolved(mrt, vif, skb2);
2171                 read_unlock(&mrt_lock);
2172                 rcu_read_unlock();
2173                 return err;
2174         }
2175
2176         read_lock(&mrt_lock);
2177         err = __ipmr_fill_mroute(mrt, skb, cache, rtm);
2178         read_unlock(&mrt_lock);
2179         rcu_read_unlock();
2180         return err;
2181 }
2182
2183 static int ipmr_fill_mroute(struct mr_table *mrt, struct sk_buff *skb,
2184                             u32 portid, u32 seq, struct mfc_cache *c, int cmd,
2185                             int flags)
2186 {
2187         struct nlmsghdr *nlh;
2188         struct rtmsg *rtm;
2189         int err;
2190
2191         nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rtm), flags);
2192         if (!nlh)
2193                 return -EMSGSIZE;
2194
2195         rtm = nlmsg_data(nlh);
2196         rtm->rtm_family   = RTNL_FAMILY_IPMR;
2197         rtm->rtm_dst_len  = 32;
2198         rtm->rtm_src_len  = 32;
2199         rtm->rtm_tos      = 0;
2200         rtm->rtm_table    = mrt->id;
2201         if (nla_put_u32(skb, RTA_TABLE, mrt->id))
2202                 goto nla_put_failure;
2203         rtm->rtm_type     = RTN_MULTICAST;
2204         rtm->rtm_scope    = RT_SCOPE_UNIVERSE;
2205         if (c->mfc_flags & MFC_STATIC)
2206                 rtm->rtm_protocol = RTPROT_STATIC;
2207         else
2208                 rtm->rtm_protocol = RTPROT_MROUTED;
2209         rtm->rtm_flags    = 0;
2210
2211         if (nla_put_in_addr(skb, RTA_SRC, c->mfc_origin) ||
2212             nla_put_in_addr(skb, RTA_DST, c->mfc_mcastgrp))
2213                 goto nla_put_failure;
2214         err = __ipmr_fill_mroute(mrt, skb, c, rtm);
2215         /* do not break the dump if cache is unresolved */
2216         if (err < 0 && err != -ENOENT)
2217                 goto nla_put_failure;
2218
2219         nlmsg_end(skb, nlh);
2220         return 0;
2221
2222 nla_put_failure:
2223         nlmsg_cancel(skb, nlh);
2224         return -EMSGSIZE;
2225 }
2226
2227 static size_t mroute_msgsize(bool unresolved, int maxvif)
2228 {
2229         size_t len =
2230                 NLMSG_ALIGN(sizeof(struct rtmsg))
2231                 + nla_total_size(4)     /* RTA_TABLE */
2232                 + nla_total_size(4)     /* RTA_SRC */
2233                 + nla_total_size(4)     /* RTA_DST */
2234                 ;
2235
2236         if (!unresolved)
2237                 len = len
2238                       + nla_total_size(4)       /* RTA_IIF */
2239                       + nla_total_size(0)       /* RTA_MULTIPATH */
2240                       + maxvif * NLA_ALIGN(sizeof(struct rtnexthop))
2241                                                 /* RTA_MFC_STATS */
2242                       + nla_total_size_64bit(sizeof(struct rta_mfc_stats))
2243                 ;
2244
2245         return len;
2246 }
2247
2248 static void mroute_netlink_event(struct mr_table *mrt, struct mfc_cache *mfc,
2249                                  int cmd)
2250 {
2251         struct net *net = read_pnet(&mrt->net);
2252         struct sk_buff *skb;
2253         int err = -ENOBUFS;
2254
2255         skb = nlmsg_new(mroute_msgsize(mfc->mfc_parent >= MAXVIFS, mrt->maxvif),
2256                         GFP_ATOMIC);
2257         if (!skb)
2258                 goto errout;
2259
2260         err = ipmr_fill_mroute(mrt, skb, 0, 0, mfc, cmd, 0);
2261         if (err < 0)
2262                 goto errout;
2263
2264         rtnl_notify(skb, net, 0, RTNLGRP_IPV4_MROUTE, NULL, GFP_ATOMIC);
2265         return;
2266
2267 errout:
2268         kfree_skb(skb);
2269         if (err < 0)
2270                 rtnl_set_sk_err(net, RTNLGRP_IPV4_MROUTE, err);
2271 }
2272
2273 static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
2274 {
2275         struct net *net = sock_net(skb->sk);
2276         struct mr_table *mrt;
2277         struct mfc_cache *mfc;
2278         unsigned int t = 0, s_t;
2279         unsigned int h = 0, s_h;
2280         unsigned int e = 0, s_e;
2281
2282         s_t = cb->args[0];
2283         s_h = cb->args[1];
2284         s_e = cb->args[2];
2285
2286         rcu_read_lock();
2287         ipmr_for_each_table(mrt, net) {
2288                 if (t < s_t)
2289                         goto next_table;
2290                 if (t > s_t)
2291                         s_h = 0;
2292                 for (h = s_h; h < MFC_LINES; h++) {
2293                         list_for_each_entry_rcu(mfc, &mrt->mfc_cache_array[h], list) {
2294                                 if (e < s_e)
2295                                         goto next_entry;
2296                                 if (ipmr_fill_mroute(mrt, skb,
2297                                                      NETLINK_CB(cb->skb).portid,
2298                                                      cb->nlh->nlmsg_seq,
2299                                                      mfc, RTM_NEWROUTE,
2300                                                      NLM_F_MULTI) < 0)
2301                                         goto done;
2302 next_entry:
2303                                 e++;
2304                         }
2305                         e = s_e = 0;
2306                 }
2307                 spin_lock_bh(&mfc_unres_lock);
2308                 list_for_each_entry(mfc, &mrt->mfc_unres_queue, list) {
2309                         if (e < s_e)
2310                                 goto next_entry2;
2311                         if (ipmr_fill_mroute(mrt, skb,
2312                                              NETLINK_CB(cb->skb).portid,
2313                                              cb->nlh->nlmsg_seq,
2314                                              mfc, RTM_NEWROUTE,
2315                                              NLM_F_MULTI) < 0) {
2316                                 spin_unlock_bh(&mfc_unres_lock);
2317                                 goto done;
2318                         }
2319 next_entry2:
2320                         e++;
2321                 }
2322                 spin_unlock_bh(&mfc_unres_lock);
2323                 e = s_e = 0;
2324                 s_h = 0;
2325 next_table:
2326                 t++;
2327         }
2328 done:
2329         rcu_read_unlock();
2330
2331         cb->args[2] = e;
2332         cb->args[1] = h;
2333         cb->args[0] = t;
2334
2335         return skb->len;
2336 }
2337
2338 static const struct nla_policy rtm_ipmr_policy[RTA_MAX + 1] = {
2339         [RTA_SRC]       = { .type = NLA_U32 },
2340         [RTA_DST]       = { .type = NLA_U32 },
2341         [RTA_IIF]       = { .type = NLA_U32 },
2342         [RTA_TABLE]     = { .type = NLA_U32 },
2343         [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
2344 };
2345
2346 static bool ipmr_rtm_validate_proto(unsigned char rtm_protocol)
2347 {
2348         switch (rtm_protocol) {
2349         case RTPROT_STATIC:
2350         case RTPROT_MROUTED:
2351                 return true;
2352         }
2353         return false;
2354 }
2355
2356 static int ipmr_nla_get_ttls(const struct nlattr *nla, struct mfcctl *mfcc)
2357 {
2358         struct rtnexthop *rtnh = nla_data(nla);
2359         int remaining = nla_len(nla), vifi = 0;
2360
2361         while (rtnh_ok(rtnh, remaining)) {
2362                 mfcc->mfcc_ttls[vifi] = rtnh->rtnh_hops;
2363                 if (++vifi == MAXVIFS)
2364                         break;
2365                 rtnh = rtnh_next(rtnh, &remaining);
2366         }
2367
2368         return remaining > 0 ? -EINVAL : vifi;
2369 }
2370
2371 /* returns < 0 on error, 0 for ADD_MFC and 1 for ADD_MFC_PROXY */
2372 static int rtm_to_ipmr_mfcc(struct net *net, struct nlmsghdr *nlh,
2373                             struct mfcctl *mfcc, int *mrtsock,
2374                             struct mr_table **mrtret)
2375 {
2376         struct net_device *dev = NULL;
2377         u32 tblid = RT_TABLE_DEFAULT;
2378         struct mr_table *mrt;
2379         struct nlattr *attr;
2380         struct rtmsg *rtm;
2381         int ret, rem;
2382
2383         ret = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipmr_policy);
2384         if (ret < 0)
2385                 goto out;
2386         rtm = nlmsg_data(nlh);
2387
2388         ret = -EINVAL;
2389         if (rtm->rtm_family != RTNL_FAMILY_IPMR || rtm->rtm_dst_len != 32 ||
2390             rtm->rtm_type != RTN_MULTICAST ||
2391             rtm->rtm_scope != RT_SCOPE_UNIVERSE ||
2392             !ipmr_rtm_validate_proto(rtm->rtm_protocol))
2393                 goto out;
2394
2395         memset(mfcc, 0, sizeof(*mfcc));
2396         mfcc->mfcc_parent = -1;
2397         ret = 0;
2398         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), rem) {
2399                 switch (nla_type(attr)) {
2400                 case RTA_SRC:
2401                         mfcc->mfcc_origin.s_addr = nla_get_be32(attr);
2402                         break;
2403                 case RTA_DST:
2404                         mfcc->mfcc_mcastgrp.s_addr = nla_get_be32(attr);
2405                         break;
2406                 case RTA_IIF:
2407                         dev = __dev_get_by_index(net, nla_get_u32(attr));
2408                         if (!dev) {
2409                                 ret = -ENODEV;
2410                                 goto out;
2411                         }
2412                         break;
2413                 case RTA_MULTIPATH:
2414                         if (ipmr_nla_get_ttls(attr, mfcc) < 0) {
2415                                 ret = -EINVAL;
2416                                 goto out;
2417                         }
2418                         break;
2419                 case RTA_PREFSRC:
2420                         ret = 1;
2421                         break;
2422                 case RTA_TABLE:
2423                         tblid = nla_get_u32(attr);
2424                         break;
2425                 }
2426         }
2427         mrt = ipmr_get_table(net, tblid);
2428         if (!mrt) {
2429                 ret = -ENOENT;
2430                 goto out;
2431         }
2432         *mrtret = mrt;
2433         *mrtsock = rtm->rtm_protocol == RTPROT_MROUTED ? 1 : 0;
2434         if (dev)
2435                 mfcc->mfcc_parent = ipmr_find_vif(mrt, dev);
2436
2437 out:
2438         return ret;
2439 }
2440
2441 /* takes care of both newroute and delroute */
2442 static int ipmr_rtm_route(struct sk_buff *skb, struct nlmsghdr *nlh)
2443 {
2444         struct net *net = sock_net(skb->sk);
2445         int ret, mrtsock, parent;
2446         struct mr_table *tbl;
2447         struct mfcctl mfcc;
2448
2449         mrtsock = 0;
2450         tbl = NULL;
2451         ret = rtm_to_ipmr_mfcc(net, nlh, &mfcc, &mrtsock, &tbl);
2452         if (ret < 0)
2453                 return ret;
2454
2455         parent = ret ? mfcc.mfcc_parent : -1;
2456         if (nlh->nlmsg_type == RTM_NEWROUTE)
2457                 return ipmr_mfc_add(net, tbl, &mfcc, mrtsock, parent);
2458         else
2459                 return ipmr_mfc_delete(tbl, &mfcc, parent);
2460 }
2461
2462 #ifdef CONFIG_PROC_FS
2463 /* The /proc interfaces to multicast routing :
2464  * /proc/net/ip_mr_cache & /proc/net/ip_mr_vif
2465  */
2466 struct ipmr_vif_iter {
2467         struct seq_net_private p;
2468         struct mr_table *mrt;
2469         int ct;
2470 };
2471
2472 static struct vif_device *ipmr_vif_seq_idx(struct net *net,
2473                                            struct ipmr_vif_iter *iter,
2474                                            loff_t pos)
2475 {
2476         struct mr_table *mrt = iter->mrt;
2477
2478         for (iter->ct = 0; iter->ct < mrt->maxvif; ++iter->ct) {
2479                 if (!VIF_EXISTS(mrt, iter->ct))
2480                         continue;
2481                 if (pos-- == 0)
2482                         return &mrt->vif_table[iter->ct];
2483         }
2484         return NULL;
2485 }
2486
2487 static void *ipmr_vif_seq_start(struct seq_file *seq, loff_t *pos)
2488         __acquires(mrt_lock)
2489 {
2490         struct ipmr_vif_iter *iter = seq->private;
2491         struct net *net = seq_file_net(seq);
2492         struct mr_table *mrt;
2493
2494         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2495         if (!mrt)
2496                 return ERR_PTR(-ENOENT);
2497
2498         iter->mrt = mrt;
2499
2500         read_lock(&mrt_lock);
2501         return *pos ? ipmr_vif_seq_idx(net, seq->private, *pos - 1)
2502                 : SEQ_START_TOKEN;
2503 }
2504
2505 static void *ipmr_vif_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2506 {
2507         struct ipmr_vif_iter *iter = seq->private;
2508         struct net *net = seq_file_net(seq);
2509         struct mr_table *mrt = iter->mrt;
2510
2511         ++*pos;
2512         if (v == SEQ_START_TOKEN)
2513                 return ipmr_vif_seq_idx(net, iter, 0);
2514
2515         while (++iter->ct < mrt->maxvif) {
2516                 if (!VIF_EXISTS(mrt, iter->ct))
2517                         continue;
2518                 return &mrt->vif_table[iter->ct];
2519         }
2520         return NULL;
2521 }
2522
2523 static void ipmr_vif_seq_stop(struct seq_file *seq, void *v)
2524         __releases(mrt_lock)
2525 {
2526         read_unlock(&mrt_lock);
2527 }
2528
2529 static int ipmr_vif_seq_show(struct seq_file *seq, void *v)
2530 {
2531         struct ipmr_vif_iter *iter = seq->private;
2532         struct mr_table *mrt = iter->mrt;
2533
2534         if (v == SEQ_START_TOKEN) {
2535                 seq_puts(seq,
2536                          "Interface      BytesIn  PktsIn  BytesOut PktsOut Flags Local    Remote\n");
2537         } else {
2538                 const struct vif_device *vif = v;
2539                 const char *name =  vif->dev ? vif->dev->name : "none";
2540
2541                 seq_printf(seq,
2542                            "%2Zd %-10s %8ld %7ld  %8ld %7ld %05X %08X %08X\n",
2543                            vif - mrt->vif_table,
2544                            name, vif->bytes_in, vif->pkt_in,
2545                            vif->bytes_out, vif->pkt_out,
2546                            vif->flags, vif->local, vif->remote);
2547         }
2548         return 0;
2549 }
2550
2551 static const struct seq_operations ipmr_vif_seq_ops = {
2552         .start = ipmr_vif_seq_start,
2553         .next  = ipmr_vif_seq_next,
2554         .stop  = ipmr_vif_seq_stop,
2555         .show  = ipmr_vif_seq_show,
2556 };
2557
2558 static int ipmr_vif_open(struct inode *inode, struct file *file)
2559 {
2560         return seq_open_net(inode, file, &ipmr_vif_seq_ops,
2561                             sizeof(struct ipmr_vif_iter));
2562 }
2563
2564 static const struct file_operations ipmr_vif_fops = {
2565         .owner   = THIS_MODULE,
2566         .open    = ipmr_vif_open,
2567         .read    = seq_read,
2568         .llseek  = seq_lseek,
2569         .release = seq_release_net,
2570 };
2571
2572 struct ipmr_mfc_iter {
2573         struct seq_net_private p;
2574         struct mr_table *mrt;
2575         struct list_head *cache;
2576         int ct;
2577 };
2578
2579
2580 static struct mfc_cache *ipmr_mfc_seq_idx(struct net *net,
2581                                           struct ipmr_mfc_iter *it, loff_t pos)
2582 {
2583         struct mr_table *mrt = it->mrt;
2584         struct mfc_cache *mfc;
2585
2586         rcu_read_lock();
2587         for (it->ct = 0; it->ct < MFC_LINES; it->ct++) {
2588                 it->cache = &mrt->mfc_cache_array[it->ct];
2589                 list_for_each_entry_rcu(mfc, it->cache, list)
2590                         if (pos-- == 0)
2591                                 return mfc;
2592         }
2593         rcu_read_unlock();
2594
2595         spin_lock_bh(&mfc_unres_lock);
2596         it->cache = &mrt->mfc_unres_queue;
2597         list_for_each_entry(mfc, it->cache, list)
2598                 if (pos-- == 0)
2599                         return mfc;
2600         spin_unlock_bh(&mfc_unres_lock);
2601
2602         it->cache = NULL;
2603         return NULL;
2604 }
2605
2606
2607 static void *ipmr_mfc_seq_start(struct seq_file *seq, loff_t *pos)
2608 {
2609         struct ipmr_mfc_iter *it = seq->private;
2610         struct net *net = seq_file_net(seq);
2611         struct mr_table *mrt;
2612
2613         mrt = ipmr_get_table(net, RT_TABLE_DEFAULT);
2614         if (!mrt)
2615                 return ERR_PTR(-ENOENT);
2616
2617         it->mrt = mrt;
2618         it->cache = NULL;
2619         it->ct = 0;
2620         return *pos ? ipmr_mfc_seq_idx(net, seq->private, *pos - 1)
2621                 : SEQ_START_TOKEN;
2622 }
2623
2624 static void *ipmr_mfc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2625 {
2626         struct mfc_cache *mfc = v;
2627         struct ipmr_mfc_iter *it = seq->private;
2628         struct net *net = seq_file_net(seq);
2629         struct mr_table *mrt = it->mrt;
2630
2631         ++*pos;
2632
2633         if (v == SEQ_START_TOKEN)
2634                 return ipmr_mfc_seq_idx(net, seq->private, 0);
2635
2636         if (mfc->list.next != it->cache)
2637                 return list_entry(mfc->list.next, struct mfc_cache, list);
2638
2639         if (it->cache == &mrt->mfc_unres_queue)
2640                 goto end_of_list;
2641
2642         BUG_ON(it->cache != &mrt->mfc_cache_array[it->ct]);
2643
2644         while (++it->ct < MFC_LINES) {
2645                 it->cache = &mrt->mfc_cache_array[it->ct];
2646                 if (list_empty(it->cache))
2647                         continue;
2648                 return list_first_entry(it->cache, struct mfc_cache, list);
2649         }
2650
2651         /* exhausted cache_array, show unresolved */
2652         rcu_read_unlock();
2653         it->cache = &mrt->mfc_unres_queue;
2654         it->ct = 0;
2655
2656         spin_lock_bh(&mfc_unres_lock);
2657         if (!list_empty(it->cache))
2658                 return list_first_entry(it->cache, struct mfc_cache, list);
2659
2660 end_of_list:
2661         spin_unlock_bh(&mfc_unres_lock);
2662         it->cache = NULL;
2663
2664         return NULL;
2665 }
2666
2667 static void ipmr_mfc_seq_stop(struct seq_file *seq, void *v)
2668 {
2669         struct ipmr_mfc_iter *it = seq->private;
2670         struct mr_table *mrt = it->mrt;
2671
2672         if (it->cache == &mrt->mfc_unres_queue)
2673                 spin_unlock_bh(&mfc_unres_lock);
2674         else if (it->cache == &mrt->mfc_cache_array[it->ct])
2675                 rcu_read_unlock();
2676 }
2677
2678 static int ipmr_mfc_seq_show(struct seq_file *seq, void *v)
2679 {
2680         int n;
2681
2682         if (v == SEQ_START_TOKEN) {
2683                 seq_puts(seq,
2684                  "Group    Origin   Iif     Pkts    Bytes    Wrong Oifs\n");
2685         } else {
2686                 const struct mfc_cache *mfc = v;
2687                 const struct ipmr_mfc_iter *it = seq->private;
2688                 const struct mr_table *mrt = it->mrt;
2689
2690                 seq_printf(seq, "%08X %08X %-3hd",
2691                            (__force u32) mfc->mfc_mcastgrp,
2692                            (__force u32) mfc->mfc_origin,
2693                            mfc->mfc_parent);
2694
2695                 if (it->cache != &mrt->mfc_unres_queue) {
2696                         seq_printf(seq, " %8lu %8lu %8lu",
2697                                    mfc->mfc_un.res.pkt,
2698                                    mfc->mfc_un.res.bytes,
2699                                    mfc->mfc_un.res.wrong_if);
2700                         for (n = mfc->mfc_un.res.minvif;
2701                              n < mfc->mfc_un.res.maxvif; n++) {
2702                                 if (VIF_EXISTS(mrt, n) &&
2703                                     mfc->mfc_un.res.ttls[n] < 255)
2704                                         seq_printf(seq,
2705                                            " %2d:%-3d",
2706                                            n, mfc->mfc_un.res.ttls[n]);
2707                         }
2708                 } else {
2709                         /* unresolved mfc_caches don't contain
2710                          * pkt, bytes and wrong_if values
2711                          */
2712                         seq_printf(seq, " %8lu %8lu %8lu", 0ul, 0ul, 0ul);
2713                 }
2714                 seq_putc(seq, '\n');
2715         }
2716         return 0;
2717 }
2718
2719 static const struct seq_operations ipmr_mfc_seq_ops = {
2720         .start = ipmr_mfc_seq_start,
2721         .next  = ipmr_mfc_seq_next,
2722         .stop  = ipmr_mfc_seq_stop,
2723         .show  = ipmr_mfc_seq_show,
2724 };
2725
2726 static int ipmr_mfc_open(struct inode *inode, struct file *file)
2727 {
2728         return seq_open_net(inode, file, &ipmr_mfc_seq_ops,
2729                             sizeof(struct ipmr_mfc_iter));
2730 }
2731
2732 static const struct file_operations ipmr_mfc_fops = {
2733         .owner   = THIS_MODULE,
2734         .open    = ipmr_mfc_open,
2735         .read    = seq_read,
2736         .llseek  = seq_lseek,
2737         .release = seq_release_net,
2738 };
2739 #endif
2740
2741 #ifdef CONFIG_IP_PIMSM_V2
2742 static const struct net_protocol pim_protocol = {
2743         .handler        =       pim_rcv,
2744         .netns_ok       =       1,
2745 };
2746 #endif
2747
2748 /* Setup for IP multicast routing */
2749 static int __net_init ipmr_net_init(struct net *net)
2750 {
2751         int err;
2752
2753         err = ipmr_rules_init(net);
2754         if (err < 0)
2755                 goto fail;
2756
2757 #ifdef CONFIG_PROC_FS
2758         err = -ENOMEM;
2759         if (!proc_create("ip_mr_vif", 0, net->proc_net, &ipmr_vif_fops))
2760                 goto proc_vif_fail;
2761         if (!proc_create("ip_mr_cache", 0, net->proc_net, &ipmr_mfc_fops))
2762                 goto proc_cache_fail;
2763 #endif
2764         return 0;
2765
2766 #ifdef CONFIG_PROC_FS
2767 proc_cache_fail:
2768         remove_proc_entry("ip_mr_vif", net->proc_net);
2769 proc_vif_fail:
2770         ipmr_rules_exit(net);
2771 #endif
2772 fail:
2773         return err;
2774 }
2775
2776 static void __net_exit ipmr_net_exit(struct net *net)
2777 {
2778 #ifdef CONFIG_PROC_FS
2779         remove_proc_entry("ip_mr_cache", net->proc_net);
2780         remove_proc_entry("ip_mr_vif", net->proc_net);
2781 #endif
2782         ipmr_rules_exit(net);
2783 }
2784
2785 static struct pernet_operations ipmr_net_ops = {
2786         .init = ipmr_net_init,
2787         .exit = ipmr_net_exit,
2788 };
2789
2790 int __init ip_mr_init(void)
2791 {
2792         int err;
2793
2794         mrt_cachep = kmem_cache_create("ip_mrt_cache",
2795                                        sizeof(struct mfc_cache),
2796                                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
2797                                        NULL);
2798
2799         err = register_pernet_subsys(&ipmr_net_ops);
2800         if (err)
2801                 goto reg_pernet_fail;
2802
2803         err = register_netdevice_notifier(&ip_mr_notifier);
2804         if (err)
2805                 goto reg_notif_fail;
2806 #ifdef CONFIG_IP_PIMSM_V2
2807         if (inet_add_protocol(&pim_protocol, IPPROTO_PIM) < 0) {
2808                 pr_err("%s: can't add PIM protocol\n", __func__);
2809                 err = -EAGAIN;
2810                 goto add_proto_fail;
2811         }
2812 #endif
2813         rtnl_register(RTNL_FAMILY_IPMR, RTM_GETROUTE,
2814                       NULL, ipmr_rtm_dumproute, NULL);
2815         rtnl_register(RTNL_FAMILY_IPMR, RTM_NEWROUTE,
2816                       ipmr_rtm_route, NULL, NULL);
2817         rtnl_register(RTNL_FAMILY_IPMR, RTM_DELROUTE,
2818                       ipmr_rtm_route, NULL, NULL);
2819         return 0;
2820
2821 #ifdef CONFIG_IP_PIMSM_V2
2822 add_proto_fail:
2823         unregister_netdevice_notifier(&ip_mr_notifier);
2824 #endif
2825 reg_notif_fail:
2826         unregister_pernet_subsys(&ipmr_net_ops);
2827 reg_pernet_fail:
2828         kmem_cache_destroy(mrt_cachep);
2829         return err;
2830 }