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