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