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